blob: 7d3fe571d15fc704b35955014cabfeda6b107c43 [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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280atomic_t tcp_orphan_count = ATOMIC_INIT(0);
281
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -0700282EXPORT_SYMBOL_GPL(tcp_orphan_count);
283
David S. Millerb8059ea2006-03-25 01:36:56 -0800284int sysctl_tcp_mem[3] __read_mostly;
285int sysctl_tcp_wmem[3] __read_mostly;
286int sysctl_tcp_rmem[3] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287
288EXPORT_SYMBOL(sysctl_tcp_mem);
289EXPORT_SYMBOL(sysctl_tcp_rmem);
290EXPORT_SYMBOL(sysctl_tcp_wmem);
291
292atomic_t tcp_memory_allocated; /* Current allocated memory. */
293atomic_t tcp_sockets_allocated; /* Current number of TCP sockets. */
294
295EXPORT_SYMBOL(tcp_memory_allocated);
296EXPORT_SYMBOL(tcp_sockets_allocated);
297
298/*
Jens Axboe9c55e012007-11-06 23:30:13 -0800299 * TCP splice context
300 */
301struct tcp_splice_state {
302 struct pipe_inode_info *pipe;
303 size_t len;
304 unsigned int flags;
305};
306
307/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 * Pressure flag: try to collapse.
309 * Technical note: it is used by multiple contexts non atomically.
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800310 * All the __sk_mem_schedule() is of this nature: accounting
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 * is strict, actions are advisory and have some latency.
312 */
Eric Dumazet4103f8c2007-03-27 13:58:31 -0700313int tcp_memory_pressure __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314
315EXPORT_SYMBOL(tcp_memory_pressure);
316
Pavel Emelyanov5c52ba12008-07-16 20:28:10 -0700317void tcp_enter_memory_pressure(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318{
319 if (!tcp_memory_pressure) {
Pavel Emelyanov4e673442008-07-16 20:30:14 -0700320 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 tcp_memory_pressure = 1;
322 }
323}
324
325EXPORT_SYMBOL(tcp_enter_memory_pressure);
326
327/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 * Wait for a TCP event.
329 *
330 * Note that we don't need to lock the socket, as the upper poll layers
331 * take care of normal races (between the test and the event) and we don't
332 * go look at any of the socket buffers directly.
333 */
334unsigned int tcp_poll(struct file *file, struct socket *sock, poll_table *wait)
335{
336 unsigned int mask;
337 struct sock *sk = sock->sk;
338 struct tcp_sock *tp = tcp_sk(sk);
339
340 poll_wait(file, sk->sk_sleep, wait);
341 if (sk->sk_state == TCP_LISTEN)
Arnaldo Carvalho de Melodc40c7b2005-08-23 21:52:58 -0700342 return inet_csk_listen_poll(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343
344 /* Socket is not locked. We are protected from async events
Will Newton70efce22008-07-16 20:13:43 -0700345 * by poll logic and correct handling of state changes
346 * made by other threads is impossible in any case.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 */
348
349 mask = 0;
350 if (sk->sk_err)
351 mask = POLLERR;
352
353 /*
354 * POLLHUP is certainly not done right. But poll() doesn't
355 * have a notion of HUP in just one direction, and for a
356 * socket the read side is more interesting.
357 *
358 * Some poll() documentation says that POLLHUP is incompatible
359 * with the POLLOUT/POLLWR flags, so somebody should check this
360 * all. But careful, it tends to be safer to return too many
361 * bits than too few, and you can easily break real applications
362 * if you don't tell them that something has hung up!
363 *
364 * Check-me.
365 *
366 * Check number 1. POLLHUP is _UNMASKABLE_ event (see UNIX98 and
367 * our fs/select.c). It means that after we received EOF,
368 * poll always returns immediately, making impossible poll() on write()
369 * in state CLOSE_WAIT. One solution is evident --- to set POLLHUP
370 * if and only if shutdown has been made in both directions.
371 * Actually, it is interesting to look how Solaris and DUX
Will Newton70efce22008-07-16 20:13:43 -0700372 * solve this dilemma. I would prefer, if POLLHUP were maskable,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 * then we could set it on SND_SHUTDOWN. BTW examples given
374 * in Stevens' books assume exactly this behaviour, it explains
Will Newton70efce22008-07-16 20:13:43 -0700375 * why POLLHUP is incompatible with POLLOUT. --ANK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 *
377 * NOTE. Check for TCP_CLOSE is added. The goal is to prevent
378 * blocking on fresh not-connected or disconnected socket. --ANK
379 */
380 if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == TCP_CLOSE)
381 mask |= POLLHUP;
382 if (sk->sk_shutdown & RCV_SHUTDOWN)
Davide Libenzif348d702006-03-25 03:07:39 -0800383 mask |= POLLIN | POLLRDNORM | POLLRDHUP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384
385 /* Connected? */
386 if ((1 << sk->sk_state) & ~(TCPF_SYN_SENT | TCPF_SYN_RECV)) {
David S. Millerc7004482008-10-06 10:43:54 -0700387 int target = sock_rcvlowat(sk, 0, INT_MAX);
388
389 if (tp->urg_seq == tp->copied_seq &&
390 !sock_flag(sk, SOCK_URGINLINE) &&
391 tp->urg_data)
392 target--;
393
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 /* Potential race condition. If read of tp below will
395 * escape above sk->sk_state, we can be illegally awaken
396 * in SYN_* states. */
David S. Millerc7004482008-10-06 10:43:54 -0700397 if (tp->rcv_nxt - tp->copied_seq >= target)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 mask |= POLLIN | POLLRDNORM;
399
400 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
401 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
402 mask |= POLLOUT | POLLWRNORM;
403 } else { /* send SIGIO later */
404 set_bit(SOCK_ASYNC_NOSPACE,
405 &sk->sk_socket->flags);
406 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
407
408 /* Race breaker. If space is freed after
409 * wspace test but before the flags are set,
410 * IO signal will be lost.
411 */
412 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
413 mask |= POLLOUT | POLLWRNORM;
414 }
415 }
416
417 if (tp->urg_data & TCP_URG_VALID)
418 mask |= POLLPRI;
419 }
420 return mask;
421}
422
423int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
424{
425 struct tcp_sock *tp = tcp_sk(sk);
426 int answ;
427
428 switch (cmd) {
429 case SIOCINQ:
430 if (sk->sk_state == TCP_LISTEN)
431 return -EINVAL;
432
433 lock_sock(sk);
434 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
435 answ = 0;
436 else if (sock_flag(sk, SOCK_URGINLINE) ||
437 !tp->urg_data ||
438 before(tp->urg_seq, tp->copied_seq) ||
439 !before(tp->urg_seq, tp->rcv_nxt)) {
440 answ = tp->rcv_nxt - tp->copied_seq;
441
442 /* Subtract 1, if FIN is in queue. */
443 if (answ && !skb_queue_empty(&sk->sk_receive_queue))
444 answ -=
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700445 tcp_hdr((struct sk_buff *)sk->sk_receive_queue.prev)->fin;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 } else
447 answ = tp->urg_seq - tp->copied_seq;
448 release_sock(sk);
449 break;
450 case SIOCATMARK:
451 answ = tp->urg_data && tp->urg_seq == tp->copied_seq;
452 break;
453 case SIOCOUTQ:
454 if (sk->sk_state == TCP_LISTEN)
455 return -EINVAL;
456
457 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
458 answ = 0;
459 else
460 answ = tp->write_seq - tp->snd_una;
461 break;
462 default:
463 return -ENOIOCTLCMD;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700464 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
466 return put_user(answ, (int __user *)arg);
467}
468
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb)
470{
471 TCP_SKB_CB(skb)->flags |= TCPCB_FLAG_PSH;
472 tp->pushed_seq = tp->write_seq;
473}
474
475static inline int forced_push(struct tcp_sock *tp)
476{
477 return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1));
478}
479
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700480static inline void skb_entail(struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700482 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo352d4802006-11-17 19:59:12 -0200483 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
484
485 skb->csum = 0;
486 tcb->seq = tcb->end_seq = tp->write_seq;
487 tcb->flags = TCPCB_FLAG_ACK;
488 tcb->sacked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 skb_header_release(skb);
David S. Millerfe067e82007-03-07 12:12:44 -0800490 tcp_add_write_queue_tail(sk, skb);
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800491 sk->sk_wmem_queued += skb->truesize;
492 sk_mem_charge(sk, skb->truesize);
David S. Miller89ebd192005-08-23 10:13:06 -0700493 if (tp->nonagle & TCP_NAGLE_PUSH)
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900494 tp->nonagle &= ~TCP_NAGLE_PUSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495}
496
497static inline void tcp_mark_urg(struct tcp_sock *tp, int flags,
498 struct sk_buff *skb)
499{
500 if (flags & MSG_OOB) {
501 tp->urg_mode = 1;
502 tp->snd_up = tp->write_seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 }
504}
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;
525
526 return skb_splice_bits(skb, offset, tss->pipe, tss->len, tss->flags);
527}
528
529static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss)
530{
531 /* Store TCP splice context information in read_descriptor_t. */
532 read_descriptor_t rd_desc = {
533 .arg.data = tss,
534 };
535
536 return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv);
537}
538
539/**
540 * tcp_splice_read - splice data from TCP socket to a pipe
541 * @sock: socket to splice from
542 * @ppos: position (not valid)
543 * @pipe: pipe to splice to
544 * @len: number of bytes to splice
545 * @flags: splice modifier flags
546 *
547 * Description:
548 * Will read pages from given socket and fill them into a pipe.
549 *
550 **/
551ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos,
552 struct pipe_inode_info *pipe, size_t len,
553 unsigned int flags)
554{
555 struct sock *sk = sock->sk;
556 struct tcp_splice_state tss = {
557 .pipe = pipe,
558 .len = len,
559 .flags = flags,
560 };
561 long timeo;
562 ssize_t spliced;
563 int ret;
564
565 /*
566 * We can't seek on a socket input
567 */
568 if (unlikely(*ppos))
569 return -ESPIPE;
570
571 ret = spliced = 0;
572
573 lock_sock(sk);
574
575 timeo = sock_rcvtimeo(sk, flags & SPLICE_F_NONBLOCK);
576 while (tss.len) {
577 ret = __tcp_splice_read(sk, &tss);
578 if (ret < 0)
579 break;
580 else if (!ret) {
581 if (spliced)
582 break;
583 if (flags & SPLICE_F_NONBLOCK) {
584 ret = -EAGAIN;
585 break;
586 }
587 if (sock_flag(sk, SOCK_DONE))
588 break;
589 if (sk->sk_err) {
590 ret = sock_error(sk);
591 break;
592 }
593 if (sk->sk_shutdown & RCV_SHUTDOWN)
594 break;
595 if (sk->sk_state == TCP_CLOSE) {
596 /*
597 * This occurs when user tries to read
598 * from never connected socket.
599 */
600 if (!sock_flag(sk, SOCK_DONE))
601 ret = -ENOTCONN;
602 break;
603 }
604 if (!timeo) {
605 ret = -EAGAIN;
606 break;
607 }
608 sk_wait_data(sk, &timeo);
609 if (signal_pending(current)) {
610 ret = sock_intr_errno(timeo);
611 break;
612 }
613 continue;
614 }
615 tss.len -= ret;
616 spliced += ret;
617
618 release_sock(sk);
619 lock_sock(sk);
620
621 if (sk->sk_err || sk->sk_state == TCP_CLOSE ||
622 (sk->sk_shutdown & RCV_SHUTDOWN) || !timeo ||
623 signal_pending(current))
624 break;
625 }
626
627 release_sock(sk);
628
629 if (spliced)
630 return spliced;
631
632 return ret;
633}
634
Pavel Emelyanovdf97c702007-11-29 21:22:33 +1100635struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp)
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100636{
637 struct sk_buff *skb;
638
639 /* The TCP header must be at least 32-bit aligned. */
640 size = ALIGN(size, 4);
641
642 skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp);
643 if (skb) {
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800644 if (sk_wmem_schedule(sk, skb->truesize)) {
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100645 /*
646 * Make sure that we have exactly size bytes
647 * available to the caller, no more, no less.
648 */
649 skb_reserve(skb, skb_tailroom(skb) - size);
650 return skb;
651 }
652 __kfree_skb(skb);
653 } else {
Pavel Emelyanov5c52ba12008-07-16 20:28:10 -0700654 sk->sk_prot->enter_memory_pressure(sk);
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100655 sk_stream_moderate_sndbuf(sk);
656 }
657 return NULL;
658}
659
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660static ssize_t do_tcp_sendpages(struct sock *sk, struct page **pages, int poffset,
661 size_t psize, int flags)
662{
663 struct tcp_sock *tp = tcp_sk(sk);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700664 int mss_now, size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 int err;
666 ssize_t copied;
667 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
668
669 /* Wait for a connection to finish. */
670 if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))
671 if ((err = sk_stream_wait_connect(sk, &timeo)) != 0)
672 goto out_err;
673
674 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
675
676 mss_now = tcp_current_mss(sk, !(flags&MSG_OOB));
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700677 size_goal = tp->xmit_size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 copied = 0;
679
680 err = -EPIPE;
681 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
682 goto do_error;
683
684 while (psize > 0) {
David S. Millerfe067e82007-03-07 12:12:44 -0800685 struct sk_buff *skb = tcp_write_queue_tail(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 struct page *page = pages[poffset / PAGE_SIZE];
687 int copy, i, can_coalesce;
688 int offset = poffset % PAGE_SIZE;
689 int size = min_t(size_t, psize, PAGE_SIZE - offset);
690
David S. Millerfe067e82007-03-07 12:12:44 -0800691 if (!tcp_send_head(sk) || (copy = size_goal - skb->len) <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692new_segment:
693 if (!sk_stream_memory_free(sk))
694 goto wait_for_sndbuf;
695
Pavel Emelyanovdf97c702007-11-29 21:22:33 +1100696 skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 if (!skb)
698 goto wait_for_memory;
699
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700700 skb_entail(sk, skb);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700701 copy = size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 }
703
704 if (copy > size)
705 copy = size;
706
707 i = skb_shinfo(skb)->nr_frags;
708 can_coalesce = skb_can_coalesce(skb, i, page, offset);
709 if (!can_coalesce && i >= MAX_SKB_FRAGS) {
710 tcp_mark_push(tp, skb);
711 goto new_segment;
712 }
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800713 if (!sk_wmem_schedule(sk, copy))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 goto wait_for_memory;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900715
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 if (can_coalesce) {
717 skb_shinfo(skb)->frags[i - 1].size += copy;
718 } else {
719 get_page(page);
720 skb_fill_page_desc(skb, i, page, offset, copy);
721 }
722
723 skb->len += copy;
724 skb->data_len += copy;
725 skb->truesize += copy;
726 sk->sk_wmem_queued += copy;
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800727 sk_mem_charge(sk, copy);
Patrick McHardy84fa7932006-08-29 16:44:56 -0700728 skb->ip_summed = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 tp->write_seq += copy;
730 TCP_SKB_CB(skb)->end_seq += copy;
Herbert Xu79671682006-06-22 02:40:14 -0700731 skb_shinfo(skb)->gso_segs = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
733 if (!copied)
734 TCP_SKB_CB(skb)->flags &= ~TCPCB_FLAG_PSH;
735
736 copied += copy;
737 poffset += copy;
738 if (!(psize -= copy))
739 goto out;
740
Herbert Xu69d15062008-03-22 15:47:05 -0700741 if (skb->len < size_goal || (flags & MSG_OOB))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 continue;
743
744 if (forced_push(tp)) {
745 tcp_mark_push(tp, skb);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700746 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
David S. Millerfe067e82007-03-07 12:12:44 -0800747 } else if (skb == tcp_send_head(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 tcp_push_one(sk, mss_now);
749 continue;
750
751wait_for_sndbuf:
752 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
753wait_for_memory:
754 if (copied)
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700755 tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756
757 if ((err = sk_stream_wait_memory(sk, &timeo)) != 0)
758 goto do_error;
759
760 mss_now = tcp_current_mss(sk, !(flags&MSG_OOB));
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700761 size_goal = tp->xmit_size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 }
763
764out:
765 if (copied)
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700766 tcp_push(sk, flags, mss_now, tp->nonagle);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 return copied;
768
769do_error:
770 if (copied)
771 goto out;
772out_err:
773 return sk_stream_error(sk, flags, err);
774}
775
776ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset,
777 size_t size, int flags)
778{
779 ssize_t res;
780 struct sock *sk = sock->sk;
781
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 if (!(sk->sk_route_caps & NETIF_F_SG) ||
Herbert Xu8648b302006-06-17 22:06:05 -0700783 !(sk->sk_route_caps & NETIF_F_ALL_CSUM))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 return sock_no_sendpage(sock, page, offset, size, flags);
785
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 lock_sock(sk);
787 TCP_CHECK_TIMER(sk);
788 res = do_tcp_sendpages(sk, &page, offset, size, flags);
789 TCP_CHECK_TIMER(sk);
790 release_sock(sk);
791 return res;
792}
793
794#define TCP_PAGE(sk) (sk->sk_sndmsg_page)
795#define TCP_OFF(sk) (sk->sk_sndmsg_off)
796
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700797static inline int select_size(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700799 struct tcp_sock *tp = tcp_sk(sk);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700800 int tmp = tp->mss_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
David S. Millerb4e26f52005-07-05 15:20:27 -0700802 if (sk->sk_route_caps & NETIF_F_SG) {
Herbert Xubcd76112006-06-30 13:36:35 -0700803 if (sk_can_gso(sk))
David S. Millerb4e26f52005-07-05 15:20:27 -0700804 tmp = 0;
805 else {
806 int pgbreak = SKB_MAX_HEAD(MAX_TCP_HEADER);
807
808 if (tmp >= pgbreak &&
809 tmp <= pgbreak + (MAX_SKB_FRAGS - 1) * PAGE_SIZE)
810 tmp = pgbreak;
811 }
812 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 return tmp;
815}
816
David S. Miller3516ffb2007-08-02 19:23:56 -0700817int tcp_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 size_t size)
819{
David S. Miller3516ffb2007-08-02 19:23:56 -0700820 struct sock *sk = sock->sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 struct iovec *iov;
822 struct tcp_sock *tp = tcp_sk(sk);
823 struct sk_buff *skb;
824 int iovlen, flags;
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700825 int mss_now, size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 int err, copied;
827 long timeo;
828
829 lock_sock(sk);
830 TCP_CHECK_TIMER(sk);
831
832 flags = msg->msg_flags;
833 timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
834
835 /* Wait for a connection to finish. */
836 if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))
837 if ((err = sk_stream_wait_connect(sk, &timeo)) != 0)
838 goto out_err;
839
840 /* This should be in poll */
841 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
842
843 mss_now = tcp_current_mss(sk, !(flags&MSG_OOB));
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700844 size_goal = tp->xmit_size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845
846 /* Ok commence sending. */
847 iovlen = msg->msg_iovlen;
848 iov = msg->msg_iov;
849 copied = 0;
850
851 err = -EPIPE;
852 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
853 goto do_error;
854
855 while (--iovlen >= 0) {
856 int seglen = iov->iov_len;
857 unsigned char __user *from = iov->iov_base;
858
859 iov++;
860
861 while (seglen > 0) {
862 int copy;
863
David S. Millerfe067e82007-03-07 12:12:44 -0800864 skb = tcp_write_queue_tail(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865
David S. Millerfe067e82007-03-07 12:12:44 -0800866 if (!tcp_send_head(sk) ||
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700867 (copy = size_goal - skb->len) <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
869new_segment:
870 /* Allocate new segment. If the interface is SG,
871 * allocate skb fitting to single page.
872 */
873 if (!sk_stream_memory_free(sk))
874 goto wait_for_sndbuf;
875
Pavel Emelyanovdf97c702007-11-29 21:22:33 +1100876 skb = sk_stream_alloc_skb(sk, select_size(sk),
877 sk->sk_allocation);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 if (!skb)
879 goto wait_for_memory;
880
881 /*
882 * Check whether we can use HW checksum.
883 */
Herbert Xu8648b302006-06-17 22:06:05 -0700884 if (sk->sk_route_caps & NETIF_F_ALL_CSUM)
Patrick McHardy84fa7932006-08-29 16:44:56 -0700885 skb->ip_summed = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700887 skb_entail(sk, skb);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700888 copy = size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 }
890
891 /* Try to append data to the end of skb. */
892 if (copy > seglen)
893 copy = seglen;
894
895 /* Where to copy to? */
896 if (skb_tailroom(skb) > 0) {
897 /* We have some space in skb head. Superb! */
898 if (copy > skb_tailroom(skb))
899 copy = skb_tailroom(skb);
900 if ((err = skb_add_data(skb, from, copy)) != 0)
901 goto do_fault;
902 } else {
903 int merge = 0;
904 int i = skb_shinfo(skb)->nr_frags;
905 struct page *page = TCP_PAGE(sk);
906 int off = TCP_OFF(sk);
907
908 if (skb_can_coalesce(skb, i, page, off) &&
909 off != PAGE_SIZE) {
910 /* We can extend the last page
911 * fragment. */
912 merge = 1;
913 } else if (i == MAX_SKB_FRAGS ||
914 (!i &&
915 !(sk->sk_route_caps & NETIF_F_SG))) {
916 /* Need to add new fragment and cannot
917 * do this because interface is non-SG,
918 * or because all the page slots are
919 * busy. */
920 tcp_mark_push(tp, skb);
921 goto new_segment;
922 } else if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 if (off == PAGE_SIZE) {
924 put_page(page);
925 TCP_PAGE(sk) = page = NULL;
Herbert Xufb5f5e62005-09-05 18:55:48 -0700926 off = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 }
Herbert Xuef015782005-09-01 17:48:59 -0700928 } else
Herbert Xufb5f5e62005-09-05 18:55:48 -0700929 off = 0;
Herbert Xuef015782005-09-01 17:48:59 -0700930
931 if (copy > PAGE_SIZE - off)
932 copy = PAGE_SIZE - off;
933
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800934 if (!sk_wmem_schedule(sk, copy))
Herbert Xuef015782005-09-01 17:48:59 -0700935 goto wait_for_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936
937 if (!page) {
938 /* Allocate new cache page. */
939 if (!(page = sk_stream_alloc_page(sk)))
940 goto wait_for_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 }
942
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 /* Time to copy data. We are close to
944 * the end! */
945 err = skb_copy_to_page(sk, from, skb, page,
946 off, copy);
947 if (err) {
948 /* If this page was new, give it to the
949 * socket so it does not get leaked.
950 */
951 if (!TCP_PAGE(sk)) {
952 TCP_PAGE(sk) = page;
953 TCP_OFF(sk) = 0;
954 }
955 goto do_error;
956 }
957
958 /* Update the skb. */
959 if (merge) {
960 skb_shinfo(skb)->frags[i - 1].size +=
961 copy;
962 } else {
963 skb_fill_page_desc(skb, i, page, off, copy);
964 if (TCP_PAGE(sk)) {
965 get_page(page);
966 } else if (off + copy < PAGE_SIZE) {
967 get_page(page);
968 TCP_PAGE(sk) = page;
969 }
970 }
971
972 TCP_OFF(sk) = off + copy;
973 }
974
975 if (!copied)
976 TCP_SKB_CB(skb)->flags &= ~TCPCB_FLAG_PSH;
977
978 tp->write_seq += copy;
979 TCP_SKB_CB(skb)->end_seq += copy;
Herbert Xu79671682006-06-22 02:40:14 -0700980 skb_shinfo(skb)->gso_segs = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981
982 from += copy;
983 copied += copy;
984 if ((seglen -= copy) == 0 && iovlen == 0)
985 goto out;
986
Herbert Xu69d15062008-03-22 15:47:05 -0700987 if (skb->len < size_goal || (flags & MSG_OOB))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 continue;
989
990 if (forced_push(tp)) {
991 tcp_mark_push(tp, skb);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700992 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
David S. Millerfe067e82007-03-07 12:12:44 -0800993 } else if (skb == tcp_send_head(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 tcp_push_one(sk, mss_now);
995 continue;
996
997wait_for_sndbuf:
998 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
999wait_for_memory:
1000 if (copied)
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001001 tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002
1003 if ((err = sk_stream_wait_memory(sk, &timeo)) != 0)
1004 goto do_error;
1005
1006 mss_now = tcp_current_mss(sk, !(flags&MSG_OOB));
David S. Millerc1b4a7e2005-07-05 15:24:38 -07001007 size_goal = tp->xmit_size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 }
1009 }
1010
1011out:
1012 if (copied)
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001013 tcp_push(sk, flags, mss_now, tp->nonagle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 TCP_CHECK_TIMER(sk);
1015 release_sock(sk);
1016 return copied;
1017
1018do_fault:
1019 if (!skb->len) {
David S. Millerfe067e82007-03-07 12:12:44 -08001020 tcp_unlink_write_queue(skb, sk);
1021 /* It is the one place in all of TCP, except connection
1022 * reset, where we can be unlinking the send_head.
1023 */
1024 tcp_check_send_head(sk, skb);
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001025 sk_wmem_free_skb(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 }
1027
1028do_error:
1029 if (copied)
1030 goto out;
1031out_err:
1032 err = sk_stream_error(sk, flags, err);
1033 TCP_CHECK_TIMER(sk);
1034 release_sock(sk);
1035 return err;
1036}
1037
1038/*
1039 * Handle reading urgent data. BSD has very simple semantics for
1040 * this, no blocking and very strange errors 8)
1041 */
1042
1043static int tcp_recv_urg(struct sock *sk, long timeo,
1044 struct msghdr *msg, int len, int flags,
1045 int *addr_len)
1046{
1047 struct tcp_sock *tp = tcp_sk(sk);
1048
1049 /* No URG data to read. */
1050 if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data ||
1051 tp->urg_data == TCP_URG_READ)
1052 return -EINVAL; /* Yes this is right ! */
1053
1054 if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE))
1055 return -ENOTCONN;
1056
1057 if (tp->urg_data & TCP_URG_VALID) {
1058 int err = 0;
1059 char c = tp->urg_data;
1060
1061 if (!(flags & MSG_PEEK))
1062 tp->urg_data = TCP_URG_READ;
1063
1064 /* Read urgent data. */
1065 msg->msg_flags |= MSG_OOB;
1066
1067 if (len > 0) {
1068 if (!(flags & MSG_TRUNC))
1069 err = memcpy_toiovec(msg->msg_iov, &c, 1);
1070 len = 1;
1071 } else
1072 msg->msg_flags |= MSG_TRUNC;
1073
1074 return err ? -EFAULT : len;
1075 }
1076
1077 if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
1078 return 0;
1079
1080 /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and
1081 * the available implementations agree in this case:
1082 * this call should never block, independent of the
1083 * blocking state of the socket.
1084 * Mike <pall@rz.uni-karlsruhe.de>
1085 */
1086 return -EAGAIN;
1087}
1088
1089/* Clean up the receive buffer for full frames taken by the user,
1090 * then send an ACK if necessary. COPIED is the number of bytes
1091 * tcp_recvmsg has given to the user so far, it speeds up the
1092 * calculation of whether or not we must ACK for the sake of
1093 * a window update.
1094 */
Chris Leech0e4b4992006-05-23 18:00:16 -07001095void tcp_cleanup_rbuf(struct sock *sk, int copied)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096{
1097 struct tcp_sock *tp = tcp_sk(sk);
1098 int time_to_ack = 0;
1099
1100#if TCP_DEBUG
1101 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
1102
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001103 WARN_ON(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104#endif
1105
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001106 if (inet_csk_ack_scheduled(sk)) {
1107 const struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 /* Delayed ACKs frequently hit locked sockets during bulk
1109 * receive. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001110 if (icsk->icsk_ack.blocked ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 /* Once-per-two-segments ACK was not sent by tcp_input.c */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001112 tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 /*
1114 * If this read emptied read buffer, we send ACK, if
1115 * connection is not bidirectional, user drained
1116 * receive buffer and there was a small segment
1117 * in queue.
1118 */
Alexey Kuznetsov1ef96962006-09-19 12:52:50 -07001119 (copied > 0 &&
1120 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) ||
1121 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) &&
1122 !icsk->icsk_ack.pingpong)) &&
1123 !atomic_read(&sk->sk_rmem_alloc)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 time_to_ack = 1;
1125 }
1126
1127 /* We send an ACK if we can now advertise a non-zero window
1128 * which has been raised "significantly".
1129 *
1130 * Even if window raised up to infinity, do not send window open ACK
1131 * in states, where we will not receive more. It is useless.
1132 */
1133 if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1134 __u32 rcv_window_now = tcp_receive_window(tp);
1135
1136 /* Optimize, __tcp_select_window() is not cheap. */
1137 if (2*rcv_window_now <= tp->window_clamp) {
1138 __u32 new_window = __tcp_select_window(sk);
1139
1140 /* Send ACK now, if this read freed lots of space
1141 * in our buffer. Certainly, new_window is new window.
1142 * We can advertise it now, if it is not less than current one.
1143 * "Lots" means "at least twice" here.
1144 */
1145 if (new_window && new_window >= 2 * rcv_window_now)
1146 time_to_ack = 1;
1147 }
1148 }
1149 if (time_to_ack)
1150 tcp_send_ack(sk);
1151}
1152
1153static void tcp_prequeue_process(struct sock *sk)
1154{
1155 struct sk_buff *skb;
1156 struct tcp_sock *tp = tcp_sk(sk);
1157
Pavel Emelyanov6f67c812008-07-16 20:31:39 -07001158 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPPREQUEUED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159
1160 /* RX process wants to run with disabled BHs, though it is not
1161 * necessary */
1162 local_bh_disable();
1163 while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
Peter Zijlstrac57943a2008-10-07 14:18:42 -07001164 sk_backlog_rcv(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 local_bh_enable();
1166
1167 /* Clear memory counter. */
1168 tp->ucopy.memory = 0;
1169}
1170
1171static inline struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off)
1172{
1173 struct sk_buff *skb;
1174 u32 offset;
1175
1176 skb_queue_walk(&sk->sk_receive_queue, skb) {
1177 offset = seq - TCP_SKB_CB(skb)->seq;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001178 if (tcp_hdr(skb)->syn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 offset--;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001180 if (offset < skb->len || tcp_hdr(skb)->fin) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 *off = offset;
1182 return skb;
1183 }
1184 }
1185 return NULL;
1186}
1187
1188/*
1189 * This routine provides an alternative to tcp_recvmsg() for routines
1190 * that would like to handle copying from skbuffs directly in 'sendfile'
1191 * fashion.
1192 * Note:
1193 * - It is assumed that the socket was locked by the caller.
1194 * - The routine does not block.
1195 * - At present, there is no support for reading OOB data
1196 * or for 'peeking' the socket using this routine
1197 * (although both would be easy to implement).
1198 */
1199int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
1200 sk_read_actor_t recv_actor)
1201{
1202 struct sk_buff *skb;
1203 struct tcp_sock *tp = tcp_sk(sk);
1204 u32 seq = tp->copied_seq;
1205 u32 offset;
1206 int copied = 0;
1207
1208 if (sk->sk_state == TCP_LISTEN)
1209 return -ENOTCONN;
1210 while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) {
1211 if (offset < skb->len) {
Octavian Purdila374e7b52008-07-03 03:31:21 -07001212 int used;
1213 size_t len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214
1215 len = skb->len - offset;
1216 /* Stop reading if we hit a patch of urgent data */
1217 if (tp->urg_data) {
1218 u32 urg_offset = tp->urg_seq - seq;
1219 if (urg_offset < len)
1220 len = urg_offset;
1221 if (!len)
1222 break;
1223 }
1224 used = recv_actor(desc, skb, offset, len);
Jens Axboeddb61a52007-06-23 23:07:50 -07001225 if (used < 0) {
1226 if (!copied)
1227 copied = used;
1228 break;
1229 } else if (used <= len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 seq += used;
1231 copied += used;
1232 offset += used;
1233 }
Octavian Purdila293ad602008-06-04 15:45:58 -07001234 /*
1235 * If recv_actor drops the lock (e.g. TCP splice
1236 * receive) the skb pointer might be invalid when
1237 * getting here: tcp_collapse might have deleted it
1238 * while aggregating skbs from the socket queue.
1239 */
1240 skb = tcp_recv_skb(sk, seq-1, &offset);
1241 if (!skb || (offset+1 != skb->len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 break;
1243 }
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001244 if (tcp_hdr(skb)->fin) {
Chris Leech624d1162006-05-23 18:01:28 -07001245 sk_eat_skb(sk, skb, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 ++seq;
1247 break;
1248 }
Chris Leech624d1162006-05-23 18:01:28 -07001249 sk_eat_skb(sk, skb, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 if (!desc->count)
1251 break;
1252 }
1253 tp->copied_seq = seq;
1254
1255 tcp_rcv_space_adjust(sk);
1256
1257 /* Clean up data we have read: This will do ACK frames. */
Jens Axboeddb61a52007-06-23 23:07:50 -07001258 if (copied > 0)
Chris Leech0e4b4992006-05-23 18:00:16 -07001259 tcp_cleanup_rbuf(sk, copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 return copied;
1261}
1262
1263/*
1264 * This routine copies from a sock struct into the user buffer.
1265 *
1266 * Technical note: in 2.3 we work on _locked_ socket, so that
1267 * tricks with *seq access order and skb->users are not required.
1268 * Probably, code can be easily improved even more.
1269 */
1270
1271int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1272 size_t len, int nonblock, int flags, int *addr_len)
1273{
1274 struct tcp_sock *tp = tcp_sk(sk);
1275 int copied = 0;
1276 u32 peek_seq;
1277 u32 *seq;
1278 unsigned long used;
1279 int err;
1280 int target; /* Read at least this many bytes */
1281 long timeo;
1282 struct task_struct *user_recv = NULL;
Chris Leech1a2449a2006-05-23 18:05:53 -07001283 int copied_early = 0;
Chris Leech2b1244a2007-03-08 09:57:36 -08001284 struct sk_buff *skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285
1286 lock_sock(sk);
1287
1288 TCP_CHECK_TIMER(sk);
1289
1290 err = -ENOTCONN;
1291 if (sk->sk_state == TCP_LISTEN)
1292 goto out;
1293
1294 timeo = sock_rcvtimeo(sk, nonblock);
1295
1296 /* Urgent data needs to be handled specially. */
1297 if (flags & MSG_OOB)
1298 goto recv_urg;
1299
1300 seq = &tp->copied_seq;
1301 if (flags & MSG_PEEK) {
1302 peek_seq = tp->copied_seq;
1303 seq = &peek_seq;
1304 }
1305
1306 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1307
Chris Leech1a2449a2006-05-23 18:05:53 -07001308#ifdef CONFIG_NET_DMA
1309 tp->ucopy.dma_chan = NULL;
1310 preempt_disable();
Chris Leech2b1244a2007-03-08 09:57:36 -08001311 skb = skb_peek_tail(&sk->sk_receive_queue);
Andrew Mortone00c5d82007-03-08 09:57:36 -08001312 {
1313 int available = 0;
1314
1315 if (skb)
1316 available = TCP_SKB_CB(skb)->seq + skb->len - (*seq);
1317 if ((available < target) &&
1318 (len > sysctl_tcp_dma_copybreak) && !(flags & MSG_PEEK) &&
1319 !sysctl_tcp_low_latency &&
1320 __get_cpu_var(softnet_data).net_dma) {
1321 preempt_enable_no_resched();
1322 tp->ucopy.pinned_list =
1323 dma_pin_iovec_pages(msg->msg_iov, len);
1324 } else {
1325 preempt_enable_no_resched();
1326 }
1327 }
Chris Leech1a2449a2006-05-23 18:05:53 -07001328#endif
1329
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 u32 offset;
1332
1333 /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */
1334 if (tp->urg_data && tp->urg_seq == *seq) {
1335 if (copied)
1336 break;
1337 if (signal_pending(current)) {
1338 copied = timeo ? sock_intr_errno(timeo) : -EAGAIN;
1339 break;
1340 }
1341 }
1342
1343 /* Next get a buffer. */
1344
1345 skb = skb_peek(&sk->sk_receive_queue);
1346 do {
1347 if (!skb)
1348 break;
1349
1350 /* Now that we have two receive queues this
1351 * shouldn't happen.
1352 */
1353 if (before(*seq, TCP_SKB_CB(skb)->seq)) {
1354 printk(KERN_INFO "recvmsg bug: copied %X "
1355 "seq %X\n", *seq, TCP_SKB_CB(skb)->seq);
1356 break;
1357 }
1358 offset = *seq - TCP_SKB_CB(skb)->seq;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001359 if (tcp_hdr(skb)->syn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 offset--;
1361 if (offset < skb->len)
1362 goto found_ok_skb;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001363 if (tcp_hdr(skb)->fin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 goto found_fin_ok;
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001365 WARN_ON(!(flags & MSG_PEEK));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 skb = skb->next;
1367 } while (skb != (struct sk_buff *)&sk->sk_receive_queue);
1368
1369 /* Well, if we have backlog, try to process it now yet. */
1370
1371 if (copied >= target && !sk->sk_backlog.tail)
1372 break;
1373
1374 if (copied) {
1375 if (sk->sk_err ||
1376 sk->sk_state == TCP_CLOSE ||
1377 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1378 !timeo ||
1379 signal_pending(current) ||
1380 (flags & MSG_PEEK))
1381 break;
1382 } else {
1383 if (sock_flag(sk, SOCK_DONE))
1384 break;
1385
1386 if (sk->sk_err) {
1387 copied = sock_error(sk);
1388 break;
1389 }
1390
1391 if (sk->sk_shutdown & RCV_SHUTDOWN)
1392 break;
1393
1394 if (sk->sk_state == TCP_CLOSE) {
1395 if (!sock_flag(sk, SOCK_DONE)) {
1396 /* This occurs when user tries to read
1397 * from never connected socket.
1398 */
1399 copied = -ENOTCONN;
1400 break;
1401 }
1402 break;
1403 }
1404
1405 if (!timeo) {
1406 copied = -EAGAIN;
1407 break;
1408 }
1409
1410 if (signal_pending(current)) {
1411 copied = sock_intr_errno(timeo);
1412 break;
1413 }
1414 }
1415
Chris Leech0e4b4992006-05-23 18:00:16 -07001416 tcp_cleanup_rbuf(sk, copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417
David S. Miller7df55122005-06-18 23:01:10 -07001418 if (!sysctl_tcp_low_latency && tp->ucopy.task == user_recv) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 /* Install new reader */
1420 if (!user_recv && !(flags & (MSG_TRUNC | MSG_PEEK))) {
1421 user_recv = current;
1422 tp->ucopy.task = user_recv;
1423 tp->ucopy.iov = msg->msg_iov;
1424 }
1425
1426 tp->ucopy.len = len;
1427
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001428 WARN_ON(tp->copied_seq != tp->rcv_nxt &&
1429 !(flags & (MSG_PEEK | MSG_TRUNC)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430
1431 /* Ugly... If prequeue is not empty, we have to
1432 * process it before releasing socket, otherwise
1433 * order will be broken at second iteration.
1434 * More elegant solution is required!!!
1435 *
1436 * Look: we have the following (pseudo)queues:
1437 *
1438 * 1. packets in flight
1439 * 2. backlog
1440 * 3. prequeue
1441 * 4. receive_queue
1442 *
1443 * Each queue can be processed only if the next ones
1444 * are empty. At this point we have empty receive_queue.
1445 * But prequeue _can_ be not empty after 2nd iteration,
1446 * when we jumped to start of loop because backlog
1447 * processing added something to receive_queue.
1448 * We cannot release_sock(), because backlog contains
1449 * packets arrived _after_ prequeued ones.
1450 *
1451 * Shortly, algorithm is clear --- to process all
1452 * the queues in order. We could make it more directly,
1453 * requeueing packets from backlog to prequeue, if
1454 * is not empty. It is more elegant, but eats cycles,
1455 * unfortunately.
1456 */
David S. Millerb03efcf2005-07-08 14:57:23 -07001457 if (!skb_queue_empty(&tp->ucopy.prequeue))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 goto do_prequeue;
1459
1460 /* __ Set realtime policy in scheduler __ */
1461 }
1462
1463 if (copied >= target) {
1464 /* Do not sleep, just process backlog. */
1465 release_sock(sk);
1466 lock_sock(sk);
1467 } else
1468 sk_wait_data(sk, &timeo);
1469
Chris Leech1a2449a2006-05-23 18:05:53 -07001470#ifdef CONFIG_NET_DMA
1471 tp->ucopy.wakeup = 0;
1472#endif
1473
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 if (user_recv) {
1475 int chunk;
1476
1477 /* __ Restore normal policy in scheduler __ */
1478
1479 if ((chunk = len - tp->ucopy.len) != 0) {
Pavel Emelyanoved880982008-07-16 20:32:45 -07001480 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMBACKLOG, chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 len -= chunk;
1482 copied += chunk;
1483 }
1484
1485 if (tp->rcv_nxt == tp->copied_seq &&
David S. Millerb03efcf2005-07-08 14:57:23 -07001486 !skb_queue_empty(&tp->ucopy.prequeue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487do_prequeue:
1488 tcp_prequeue_process(sk);
1489
1490 if ((chunk = len - tp->ucopy.len) != 0) {
Pavel Emelyanoved880982008-07-16 20:32:45 -07001491 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 len -= chunk;
1493 copied += chunk;
1494 }
1495 }
1496 }
1497 if ((flags & MSG_PEEK) && peek_seq != tp->copied_seq) {
1498 if (net_ratelimit())
1499 printk(KERN_DEBUG "TCP(%s:%d): Application bug, race in MSG_PEEK.\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001500 current->comm, task_pid_nr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 peek_seq = tp->copied_seq;
1502 }
1503 continue;
1504
1505 found_ok_skb:
1506 /* Ok so how much can we use? */
1507 used = skb->len - offset;
1508 if (len < used)
1509 used = len;
1510
1511 /* Do we have urgent data here? */
1512 if (tp->urg_data) {
1513 u32 urg_offset = tp->urg_seq - *seq;
1514 if (urg_offset < used) {
1515 if (!urg_offset) {
1516 if (!sock_flag(sk, SOCK_URGINLINE)) {
1517 ++*seq;
1518 offset++;
1519 used--;
1520 if (!used)
1521 goto skip_copy;
1522 }
1523 } else
1524 used = urg_offset;
1525 }
1526 }
1527
1528 if (!(flags & MSG_TRUNC)) {
Chris Leech1a2449a2006-05-23 18:05:53 -07001529#ifdef CONFIG_NET_DMA
1530 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1531 tp->ucopy.dma_chan = get_softnet_dma();
1532
1533 if (tp->ucopy.dma_chan) {
1534 tp->ucopy.dma_cookie = dma_skb_copy_datagram_iovec(
1535 tp->ucopy.dma_chan, skb, offset,
1536 msg->msg_iov, used,
1537 tp->ucopy.pinned_list);
1538
1539 if (tp->ucopy.dma_cookie < 0) {
1540
1541 printk(KERN_ALERT "dma_cookie < 0\n");
1542
1543 /* Exception. Bailout! */
1544 if (!copied)
1545 copied = -EFAULT;
1546 break;
1547 }
1548 if ((offset + used) == skb->len)
1549 copied_early = 1;
1550
1551 } else
1552#endif
1553 {
1554 err = skb_copy_datagram_iovec(skb, offset,
1555 msg->msg_iov, used);
1556 if (err) {
1557 /* Exception. Bailout! */
1558 if (!copied)
1559 copied = -EFAULT;
1560 break;
1561 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 }
1563 }
1564
1565 *seq += used;
1566 copied += used;
1567 len -= used;
1568
1569 tcp_rcv_space_adjust(sk);
1570
1571skip_copy:
1572 if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) {
1573 tp->urg_data = 0;
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001574 tcp_fast_path_check(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 }
1576 if (used + offset < skb->len)
1577 continue;
1578
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001579 if (tcp_hdr(skb)->fin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 goto found_fin_ok;
Chris Leech1a2449a2006-05-23 18:05:53 -07001581 if (!(flags & MSG_PEEK)) {
1582 sk_eat_skb(sk, skb, copied_early);
1583 copied_early = 0;
1584 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 continue;
1586
1587 found_fin_ok:
1588 /* Process the FIN. */
1589 ++*seq;
Chris Leech1a2449a2006-05-23 18:05:53 -07001590 if (!(flags & MSG_PEEK)) {
1591 sk_eat_skb(sk, skb, copied_early);
1592 copied_early = 0;
1593 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 break;
1595 } while (len > 0);
1596
1597 if (user_recv) {
David S. Millerb03efcf2005-07-08 14:57:23 -07001598 if (!skb_queue_empty(&tp->ucopy.prequeue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 int chunk;
1600
1601 tp->ucopy.len = copied > 0 ? len : 0;
1602
1603 tcp_prequeue_process(sk);
1604
1605 if (copied > 0 && (chunk = len - tp->ucopy.len) != 0) {
Pavel Emelyanoved880982008-07-16 20:32:45 -07001606 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 len -= chunk;
1608 copied += chunk;
1609 }
1610 }
1611
1612 tp->ucopy.task = NULL;
1613 tp->ucopy.len = 0;
1614 }
1615
Chris Leech1a2449a2006-05-23 18:05:53 -07001616#ifdef CONFIG_NET_DMA
1617 if (tp->ucopy.dma_chan) {
Chris Leech1a2449a2006-05-23 18:05:53 -07001618 dma_cookie_t done, used;
1619
1620 dma_async_memcpy_issue_pending(tp->ucopy.dma_chan);
1621
1622 while (dma_async_memcpy_complete(tp->ucopy.dma_chan,
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001623 tp->ucopy.dma_cookie, &done,
1624 &used) == DMA_IN_PROGRESS) {
Chris Leech1a2449a2006-05-23 18:05:53 -07001625 /* do partial cleanup of sk_async_wait_queue */
1626 while ((skb = skb_peek(&sk->sk_async_wait_queue)) &&
1627 (dma_async_is_complete(skb->dma_cookie, done,
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001628 used) == DMA_SUCCESS)) {
Chris Leech1a2449a2006-05-23 18:05:53 -07001629 __skb_dequeue(&sk->sk_async_wait_queue);
1630 kfree_skb(skb);
1631 }
1632 }
1633
1634 /* Safe to free early-copied skbs now */
1635 __skb_queue_purge(&sk->sk_async_wait_queue);
1636 dma_chan_put(tp->ucopy.dma_chan);
1637 tp->ucopy.dma_chan = NULL;
1638 }
1639 if (tp->ucopy.pinned_list) {
1640 dma_unpin_iovec_pages(tp->ucopy.pinned_list);
1641 tp->ucopy.pinned_list = NULL;
1642 }
1643#endif
1644
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 /* According to UNIX98, msg_name/msg_namelen are ignored
1646 * on connected socket. I was just happy when found this 8) --ANK
1647 */
1648
1649 /* Clean up data we have read: This will do ACK frames. */
Chris Leech0e4b4992006-05-23 18:00:16 -07001650 tcp_cleanup_rbuf(sk, copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651
1652 TCP_CHECK_TIMER(sk);
1653 release_sock(sk);
1654 return copied;
1655
1656out:
1657 TCP_CHECK_TIMER(sk);
1658 release_sock(sk);
1659 return err;
1660
1661recv_urg:
1662 err = tcp_recv_urg(sk, timeo, msg, len, flags, addr_len);
1663 goto out;
1664}
1665
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001666void tcp_set_state(struct sock *sk, int state)
1667{
1668 int oldstate = sk->sk_state;
1669
1670 switch (state) {
1671 case TCP_ESTABLISHED:
1672 if (oldstate != TCP_ESTABLISHED)
Pavel Emelyanov81cc8a72008-07-16 20:22:04 -07001673 TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001674 break;
1675
1676 case TCP_CLOSE:
1677 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
Pavel Emelyanov81cc8a72008-07-16 20:22:04 -07001678 TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001679
1680 sk->sk_prot->unhash(sk);
1681 if (inet_csk(sk)->icsk_bind_hash &&
1682 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -08001683 inet_put_port(sk);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001684 /* fall through */
1685 default:
1686 if (oldstate==TCP_ESTABLISHED)
Pavel Emelyanov74688e42008-07-16 20:22:46 -07001687 TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001688 }
1689
1690 /* Change state AFTER socket is unhashed to avoid closed
1691 * socket sitting in hash tables.
1692 */
1693 sk->sk_state = state;
1694
1695#ifdef STATE_TRACE
1696 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
1697#endif
1698}
1699EXPORT_SYMBOL_GPL(tcp_set_state);
1700
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701/*
1702 * State processing on a close. This implements the state shift for
1703 * sending our FIN frame. Note that we only send a FIN for some
1704 * states. A shutdown() may have already sent the FIN, or we may be
1705 * closed.
1706 */
1707
Arjan van de Ven9b5b5cf2005-11-29 16:21:38 -08001708static const unsigned char new_state[16] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 /* current state: new state: action: */
1710 /* (Invalid) */ TCP_CLOSE,
1711 /* TCP_ESTABLISHED */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
1712 /* TCP_SYN_SENT */ TCP_CLOSE,
1713 /* TCP_SYN_RECV */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
1714 /* TCP_FIN_WAIT1 */ TCP_FIN_WAIT1,
1715 /* TCP_FIN_WAIT2 */ TCP_FIN_WAIT2,
1716 /* TCP_TIME_WAIT */ TCP_CLOSE,
1717 /* TCP_CLOSE */ TCP_CLOSE,
1718 /* TCP_CLOSE_WAIT */ TCP_LAST_ACK | TCP_ACTION_FIN,
1719 /* TCP_LAST_ACK */ TCP_LAST_ACK,
1720 /* TCP_LISTEN */ TCP_CLOSE,
1721 /* TCP_CLOSING */ TCP_CLOSING,
1722};
1723
1724static int tcp_close_state(struct sock *sk)
1725{
1726 int next = (int)new_state[sk->sk_state];
1727 int ns = next & TCP_STATE_MASK;
1728
1729 tcp_set_state(sk, ns);
1730
1731 return next & TCP_ACTION_FIN;
1732}
1733
1734/*
1735 * Shutdown the sending side of a connection. Much like close except
Satoru SATOH1f29b052008-04-21 02:27:58 -07001736 * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 */
1738
1739void tcp_shutdown(struct sock *sk, int how)
1740{
1741 /* We need to grab some memory, and put together a FIN,
1742 * and then put it into the queue to be sent.
1743 * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92.
1744 */
1745 if (!(how & SEND_SHUTDOWN))
1746 return;
1747
1748 /* If we've already sent a FIN, or it's a closed state, skip this. */
1749 if ((1 << sk->sk_state) &
1750 (TCPF_ESTABLISHED | TCPF_SYN_SENT |
1751 TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) {
1752 /* Clear out any half completed packets. FIN if needed. */
1753 if (tcp_close_state(sk))
1754 tcp_send_fin(sk);
1755 }
1756}
1757
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758void tcp_close(struct sock *sk, long timeout)
1759{
1760 struct sk_buff *skb;
1761 int data_was_unread = 0;
Herbert Xu75c2d9072006-05-03 23:31:35 -07001762 int state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763
1764 lock_sock(sk);
1765 sk->sk_shutdown = SHUTDOWN_MASK;
1766
1767 if (sk->sk_state == TCP_LISTEN) {
1768 tcp_set_state(sk, TCP_CLOSE);
1769
1770 /* Special case. */
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07001771 inet_csk_listen_stop(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772
1773 goto adjudge_to_death;
1774 }
1775
1776 /* We need to flush the recv. buffs. We do this only on the
1777 * descriptor close, not protocol-sourced closes, because the
1778 * reader process may not have drained the data yet!
1779 */
1780 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
1781 u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq -
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001782 tcp_hdr(skb)->fin;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 data_was_unread += len;
1784 __kfree_skb(skb);
1785 }
1786
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001787 sk_mem_reclaim(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788
Gerrit Renker65bb7232007-04-28 21:21:46 -07001789 /* As outlined in RFC 2525, section 2.17, we send a RST here because
1790 * data was lost. To witness the awful effects of the old behavior of
1791 * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk
1792 * GET in an FTP client, suspend the process, wait for the client to
1793 * advertise a zero window, then kill -9 the FTP client, wheee...
1794 * Note: timeout is always zero in such a case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 */
1796 if (data_was_unread) {
1797 /* Unread data was tossed, zap the connection. */
Pavel Emelyanov6f67c812008-07-16 20:31:39 -07001798 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 tcp_set_state(sk, TCP_CLOSE);
1800 tcp_send_active_reset(sk, GFP_KERNEL);
1801 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
1802 /* Check zero linger _after_ checking for unread data. */
1803 sk->sk_prot->disconnect(sk, 0);
Pavel Emelyanov6f67c812008-07-16 20:31:39 -07001804 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONDATA);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 } else if (tcp_close_state(sk)) {
1806 /* We FIN if the application ate all the data before
1807 * zapping the connection.
1808 */
1809
1810 /* RED-PEN. Formally speaking, we have broken TCP state
1811 * machine. State transitions:
1812 *
1813 * TCP_ESTABLISHED -> TCP_FIN_WAIT1
1814 * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible)
1815 * TCP_CLOSE_WAIT -> TCP_LAST_ACK
1816 *
1817 * are legal only when FIN has been sent (i.e. in window),
1818 * rather than queued out of window. Purists blame.
1819 *
1820 * F.e. "RFC state" is ESTABLISHED,
1821 * if Linux state is FIN-WAIT-1, but FIN is still not sent.
1822 *
1823 * The visible declinations are that sometimes
1824 * we enter time-wait state, when it is not required really
1825 * (harmless), do not send active resets, when they are
1826 * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when
1827 * they look as CLOSING or LAST_ACK for Linux)
1828 * Probably, I missed some more holelets.
1829 * --ANK
1830 */
1831 tcp_send_fin(sk);
1832 }
1833
1834 sk_stream_wait_close(sk, timeout);
1835
1836adjudge_to_death:
Herbert Xu75c2d9072006-05-03 23:31:35 -07001837 state = sk->sk_state;
1838 sock_hold(sk);
1839 sock_orphan(sk);
1840 atomic_inc(sk->sk_prot->orphan_count);
1841
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 /* It is the last release_sock in its life. It will remove backlog. */
1843 release_sock(sk);
1844
1845
1846 /* Now socket is owned by kernel and we acquire BH lock
1847 to finish close. No need to check for user refs.
1848 */
1849 local_bh_disable();
1850 bh_lock_sock(sk);
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001851 WARN_ON(sock_owned_by_user(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852
Herbert Xu75c2d9072006-05-03 23:31:35 -07001853 /* Have we already been destroyed by a softirq or backlog? */
1854 if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE)
1855 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856
1857 /* This is a (useful) BSD violating of the RFC. There is a
1858 * problem with TCP as specified in that the other end could
1859 * keep a socket open forever with no application left this end.
1860 * We use a 3 minute timeout (about the same as BSD) then kill
1861 * our end. If they send after that then tough - BUT: long enough
1862 * that we won't make the old 4*rto = almost no time - whoops
1863 * reset mistake.
1864 *
1865 * Nope, it was not mistake. It is really desired behaviour
1866 * f.e. on http servers, when such sockets are useless, but
1867 * consume significant resources. Let's do it with special
1868 * linger2 option. --ANK
1869 */
1870
1871 if (sk->sk_state == TCP_FIN_WAIT2) {
1872 struct tcp_sock *tp = tcp_sk(sk);
1873 if (tp->linger2 < 0) {
1874 tcp_set_state(sk, TCP_CLOSE);
1875 tcp_send_active_reset(sk, GFP_ATOMIC);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07001876 NET_INC_STATS_BH(sock_net(sk),
1877 LINUX_MIB_TCPABORTONLINGER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 } else {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001879 const int tmo = tcp_fin_time(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880
1881 if (tmo > TCP_TIMEWAIT_LEN) {
David S. Miller52499af2006-07-31 22:32:09 -07001882 inet_csk_reset_keepalive_timer(sk,
1883 tmo - TCP_TIMEWAIT_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
1886 goto out;
1887 }
1888 }
1889 }
1890 if (sk->sk_state != TCP_CLOSE) {
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001891 sk_mem_reclaim(sk);
Pavel Emelianove4fd5da2007-05-29 13:19:18 -07001892 if (tcp_too_many_orphans(sk,
1893 atomic_read(sk->sk_prot->orphan_count))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 if (net_ratelimit())
1895 printk(KERN_INFO "TCP: too many of orphaned "
1896 "sockets\n");
1897 tcp_set_state(sk, TCP_CLOSE);
1898 tcp_send_active_reset(sk, GFP_ATOMIC);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07001899 NET_INC_STATS_BH(sock_net(sk),
1900 LINUX_MIB_TCPABORTONMEMORY);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 }
1902 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903
1904 if (sk->sk_state == TCP_CLOSE)
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07001905 inet_csk_destroy_sock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 /* Otherwise, socket is reprieved until protocol close. */
1907
1908out:
1909 bh_unlock_sock(sk);
1910 local_bh_enable();
1911 sock_put(sk);
1912}
1913
1914/* These states need RST on ABORT according to RFC793 */
1915
1916static inline int tcp_need_reset(int state)
1917{
1918 return (1 << state) &
1919 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 |
1920 TCPF_FIN_WAIT2 | TCPF_SYN_RECV);
1921}
1922
1923int tcp_disconnect(struct sock *sk, int flags)
1924{
1925 struct inet_sock *inet = inet_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001926 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927 struct tcp_sock *tp = tcp_sk(sk);
1928 int err = 0;
1929 int old_state = sk->sk_state;
1930
1931 if (old_state != TCP_CLOSE)
1932 tcp_set_state(sk, TCP_CLOSE);
1933
1934 /* ABORT function of RFC793 */
1935 if (old_state == TCP_LISTEN) {
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07001936 inet_csk_listen_stop(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 } else if (tcp_need_reset(old_state) ||
1938 (tp->snd_nxt != tp->write_seq &&
1939 (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) {
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -08001940 /* The last check adjusts for discrepancy of Linux wrt. RFC
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 * states
1942 */
1943 tcp_send_active_reset(sk, gfp_any());
1944 sk->sk_err = ECONNRESET;
1945 } else if (old_state == TCP_SYN_SENT)
1946 sk->sk_err = ECONNRESET;
1947
1948 tcp_clear_xmit_timers(sk);
1949 __skb_queue_purge(&sk->sk_receive_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001950 tcp_write_queue_purge(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 __skb_queue_purge(&tp->out_of_order_queue);
Chris Leech1a2449a2006-05-23 18:05:53 -07001952#ifdef CONFIG_NET_DMA
1953 __skb_queue_purge(&sk->sk_async_wait_queue);
1954#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955
1956 inet->dport = 0;
1957
1958 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1959 inet_reset_saddr(sk);
1960
1961 sk->sk_shutdown = 0;
1962 sock_reset_flag(sk, SOCK_DONE);
1963 tp->srtt = 0;
1964 if ((tp->write_seq += tp->max_window + 2) == 0)
1965 tp->write_seq = 1;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001966 icsk->icsk_backoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 tp->snd_cwnd = 2;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001968 icsk->icsk_probes_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 tp->packets_out = 0;
1970 tp->snd_ssthresh = 0x7fffffff;
1971 tp->snd_cwnd_cnt = 0;
Stephen Hemminger9772efb2005-11-10 17:09:53 -08001972 tp->bytes_acked = 0;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001973 tcp_set_ca_state(sk, TCP_CA_Open);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 tcp_clear_retrans(tp);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001975 inet_csk_delack_init(sk);
David S. Millerfe067e82007-03-07 12:12:44 -08001976 tcp_init_send_head(sk);
Srinivas Ajib40b4f72007-05-03 17:32:28 -07001977 memset(&tp->rx_opt, 0, sizeof(tp->rx_opt));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 __sk_dst_reset(sk);
1979
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001980 WARN_ON(inet->num && !icsk->icsk_bind_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981
1982 sk->sk_error_report(sk);
1983 return err;
1984}
1985
1986/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 * Socket option code for TCP.
1988 */
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08001989static int do_tcp_setsockopt(struct sock *sk, int level,
1990 int optname, char __user *optval, int optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991{
1992 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001993 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 int val;
1995 int err = 0;
1996
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07001997 /* This is a string value all the others are int's */
1998 if (optname == TCP_CONGESTION) {
1999 char name[TCP_CA_NAME_MAX];
2000
2001 if (optlen < 1)
2002 return -EINVAL;
2003
2004 val = strncpy_from_user(name, optval,
2005 min(TCP_CA_NAME_MAX-1, optlen));
2006 if (val < 0)
2007 return -EFAULT;
2008 name[val] = 0;
2009
2010 lock_sock(sk);
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002011 err = tcp_set_congestion_control(sk, name);
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002012 release_sock(sk);
2013 return err;
2014 }
2015
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 if (optlen < sizeof(int))
2017 return -EINVAL;
2018
2019 if (get_user(val, (int __user *)optval))
2020 return -EFAULT;
2021
2022 lock_sock(sk);
2023
2024 switch (optname) {
2025 case TCP_MAXSEG:
2026 /* Values greater than interface MTU won't take effect. However
2027 * at the point when this call is done we typically don't yet
2028 * know which interface is going to be used */
2029 if (val < 8 || val > MAX_TCP_WINDOW) {
2030 err = -EINVAL;
2031 break;
2032 }
2033 tp->rx_opt.user_mss = val;
2034 break;
2035
2036 case TCP_NODELAY:
2037 if (val) {
2038 /* TCP_NODELAY is weaker than TCP_CORK, so that
2039 * this option on corked socket is remembered, but
2040 * it is not activated until cork is cleared.
2041 *
2042 * However, when TCP_NODELAY is set we make
2043 * an explicit push, which overrides even TCP_CORK
2044 * for currently queued segments.
2045 */
2046 tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH;
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07002047 tcp_push_pending_frames(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 } else {
2049 tp->nonagle &= ~TCP_NAGLE_OFF;
2050 }
2051 break;
2052
2053 case TCP_CORK:
2054 /* When set indicates to always queue non-full frames.
2055 * Later the user clears this option and we transmit
2056 * any pending partial frames in the queue. This is
2057 * meant to be used alongside sendfile() to get properly
2058 * filled frames when the user (for example) must write
2059 * out headers with a write() call first and then use
2060 * sendfile to send out the data parts.
2061 *
2062 * TCP_CORK can be set together with TCP_NODELAY and it is
2063 * stronger than TCP_NODELAY.
2064 */
2065 if (val) {
2066 tp->nonagle |= TCP_NAGLE_CORK;
2067 } else {
2068 tp->nonagle &= ~TCP_NAGLE_CORK;
2069 if (tp->nonagle&TCP_NAGLE_OFF)
2070 tp->nonagle |= TCP_NAGLE_PUSH;
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07002071 tcp_push_pending_frames(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 }
2073 break;
2074
2075 case TCP_KEEPIDLE:
2076 if (val < 1 || val > MAX_TCP_KEEPIDLE)
2077 err = -EINVAL;
2078 else {
2079 tp->keepalive_time = val * HZ;
2080 if (sock_flag(sk, SOCK_KEEPOPEN) &&
2081 !((1 << sk->sk_state) &
2082 (TCPF_CLOSE | TCPF_LISTEN))) {
2083 __u32 elapsed = tcp_time_stamp - tp->rcv_tstamp;
2084 if (tp->keepalive_time > elapsed)
2085 elapsed = tp->keepalive_time - elapsed;
2086 else
2087 elapsed = 0;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002088 inet_csk_reset_keepalive_timer(sk, elapsed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 }
2090 }
2091 break;
2092 case TCP_KEEPINTVL:
2093 if (val < 1 || val > MAX_TCP_KEEPINTVL)
2094 err = -EINVAL;
2095 else
2096 tp->keepalive_intvl = val * HZ;
2097 break;
2098 case TCP_KEEPCNT:
2099 if (val < 1 || val > MAX_TCP_KEEPCNT)
2100 err = -EINVAL;
2101 else
2102 tp->keepalive_probes = val;
2103 break;
2104 case TCP_SYNCNT:
2105 if (val < 1 || val > MAX_TCP_SYNCNT)
2106 err = -EINVAL;
2107 else
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002108 icsk->icsk_syn_retries = val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 break;
2110
2111 case TCP_LINGER2:
2112 if (val < 0)
2113 tp->linger2 = -1;
2114 else if (val > sysctl_tcp_fin_timeout / HZ)
2115 tp->linger2 = 0;
2116 else
2117 tp->linger2 = val * HZ;
2118 break;
2119
2120 case TCP_DEFER_ACCEPT:
David S. Millerec0a1962008-06-12 16:31:35 -07002121 icsk->icsk_accept_queue.rskq_defer_accept = 0;
2122 if (val > 0) {
2123 /* Translate value in seconds to number of
2124 * retransmits */
2125 while (icsk->icsk_accept_queue.rskq_defer_accept < 32 &&
2126 val > ((TCP_TIMEOUT_INIT / HZ) <<
2127 icsk->icsk_accept_queue.rskq_defer_accept))
2128 icsk->icsk_accept_queue.rskq_defer_accept++;
2129 icsk->icsk_accept_queue.rskq_defer_accept++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 }
2131 break;
2132
2133 case TCP_WINDOW_CLAMP:
2134 if (!val) {
2135 if (sk->sk_state != TCP_CLOSE) {
2136 err = -EINVAL;
2137 break;
2138 }
2139 tp->window_clamp = 0;
2140 } else
2141 tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ?
2142 SOCK_MIN_RCVBUF / 2 : val;
2143 break;
2144
2145 case TCP_QUICKACK:
2146 if (!val) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002147 icsk->icsk_ack.pingpong = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 } else {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002149 icsk->icsk_ack.pingpong = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 if ((1 << sk->sk_state) &
2151 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) &&
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002152 inet_csk_ack_scheduled(sk)) {
2153 icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;
Chris Leech0e4b4992006-05-23 18:00:16 -07002154 tcp_cleanup_rbuf(sk, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 if (!(val & 1))
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002156 icsk->icsk_ack.pingpong = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 }
2158 }
2159 break;
2160
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002161#ifdef CONFIG_TCP_MD5SIG
2162 case TCP_MD5SIG:
2163 /* Read the IP->Key mappings from userspace */
2164 err = tp->af_specific->md5_parse(sk, optval, optlen);
2165 break;
2166#endif
2167
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 default:
2169 err = -ENOPROTOOPT;
2170 break;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -07002171 }
2172
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 release_sock(sk);
2174 return err;
2175}
2176
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002177int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval,
2178 int optlen)
2179{
2180 struct inet_connection_sock *icsk = inet_csk(sk);
2181
2182 if (level != SOL_TCP)
2183 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
2184 optval, optlen);
2185 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2186}
2187
2188#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002189int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
2190 char __user *optval, int optlen)
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002191{
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -08002192 if (level != SOL_TCP)
2193 return inet_csk_compat_setsockopt(sk, level, optname,
2194 optval, optlen);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002195 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2196}
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002197
2198EXPORT_SYMBOL(compat_tcp_setsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002199#endif
2200
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201/* Return information about state of tcp endpoint in API format. */
2202void tcp_get_info(struct sock *sk, struct tcp_info *info)
2203{
2204 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002205 const struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 u32 now = tcp_time_stamp;
2207
2208 memset(info, 0, sizeof(*info));
2209
2210 info->tcpi_state = sk->sk_state;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002211 info->tcpi_ca_state = icsk->icsk_ca_state;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002212 info->tcpi_retransmits = icsk->icsk_retransmits;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002213 info->tcpi_probes = icsk->icsk_probes_out;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002214 info->tcpi_backoff = icsk->icsk_backoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215
2216 if (tp->rx_opt.tstamp_ok)
2217 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
Ilpo Järvinene60402d2007-08-09 15:14:46 +03002218 if (tcp_is_sack(tp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 info->tcpi_options |= TCPI_OPT_SACK;
2220 if (tp->rx_opt.wscale_ok) {
2221 info->tcpi_options |= TCPI_OPT_WSCALE;
2222 info->tcpi_snd_wscale = tp->rx_opt.snd_wscale;
2223 info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09002224 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225
2226 if (tp->ecn_flags&TCP_ECN_OK)
2227 info->tcpi_options |= TCPI_OPT_ECN;
2228
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002229 info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto);
2230 info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato);
David S. Millerc1b4a7e2005-07-05 15:24:38 -07002231 info->tcpi_snd_mss = tp->mss_cache;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002232 info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233
Rick Jones5ee3afb2007-09-18 13:26:31 -07002234 if (sk->sk_state == TCP_LISTEN) {
2235 info->tcpi_unacked = sk->sk_ack_backlog;
2236 info->tcpi_sacked = sk->sk_max_ack_backlog;
2237 } else {
2238 info->tcpi_unacked = tp->packets_out;
2239 info->tcpi_sacked = tp->sacked_out;
2240 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 info->tcpi_lost = tp->lost_out;
2242 info->tcpi_retrans = tp->retrans_out;
2243 info->tcpi_fackets = tp->fackets_out;
2244
2245 info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002246 info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp);
2248
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -08002249 info->tcpi_pmtu = icsk->icsk_pmtu_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 info->tcpi_rcv_ssthresh = tp->rcv_ssthresh;
2251 info->tcpi_rtt = jiffies_to_usecs(tp->srtt)>>3;
2252 info->tcpi_rttvar = jiffies_to_usecs(tp->mdev)>>2;
2253 info->tcpi_snd_ssthresh = tp->snd_ssthresh;
2254 info->tcpi_snd_cwnd = tp->snd_cwnd;
2255 info->tcpi_advmss = tp->advmss;
2256 info->tcpi_reordering = tp->reordering;
2257
2258 info->tcpi_rcv_rtt = jiffies_to_usecs(tp->rcv_rtt_est.rtt)>>3;
2259 info->tcpi_rcv_space = tp->rcvq_space.space;
2260
2261 info->tcpi_total_retrans = tp->total_retrans;
2262}
2263
2264EXPORT_SYMBOL_GPL(tcp_get_info);
2265
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002266static int do_tcp_getsockopt(struct sock *sk, int level,
2267 int optname, char __user *optval, int __user *optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268{
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -07002269 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 struct tcp_sock *tp = tcp_sk(sk);
2271 int val, len;
2272
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 if (get_user(len, optlen))
2274 return -EFAULT;
2275
2276 len = min_t(unsigned int, len, sizeof(int));
2277
2278 if (len < 0)
2279 return -EINVAL;
2280
2281 switch (optname) {
2282 case TCP_MAXSEG:
David S. Millerc1b4a7e2005-07-05 15:24:38 -07002283 val = tp->mss_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
2285 val = tp->rx_opt.user_mss;
2286 break;
2287 case TCP_NODELAY:
2288 val = !!(tp->nonagle&TCP_NAGLE_OFF);
2289 break;
2290 case TCP_CORK:
2291 val = !!(tp->nonagle&TCP_NAGLE_CORK);
2292 break;
2293 case TCP_KEEPIDLE:
2294 val = (tp->keepalive_time ? : sysctl_tcp_keepalive_time) / HZ;
2295 break;
2296 case TCP_KEEPINTVL:
2297 val = (tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl) / HZ;
2298 break;
2299 case TCP_KEEPCNT:
2300 val = tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
2301 break;
2302 case TCP_SYNCNT:
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -07002303 val = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 break;
2305 case TCP_LINGER2:
2306 val = tp->linger2;
2307 if (val >= 0)
2308 val = (val ? : sysctl_tcp_fin_timeout) / HZ;
2309 break;
2310 case TCP_DEFER_ACCEPT:
David S. Millerec0a1962008-06-12 16:31:35 -07002311 val = !icsk->icsk_accept_queue.rskq_defer_accept ? 0 :
2312 ((TCP_TIMEOUT_INIT / HZ) << (icsk->icsk_accept_queue.rskq_defer_accept - 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 break;
2314 case TCP_WINDOW_CLAMP:
2315 val = tp->window_clamp;
2316 break;
2317 case TCP_INFO: {
2318 struct tcp_info info;
2319
2320 if (get_user(len, optlen))
2321 return -EFAULT;
2322
2323 tcp_get_info(sk, &info);
2324
2325 len = min_t(unsigned int, len, sizeof(info));
2326 if (put_user(len, optlen))
2327 return -EFAULT;
2328 if (copy_to_user(optval, &info, len))
2329 return -EFAULT;
2330 return 0;
2331 }
2332 case TCP_QUICKACK:
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -07002333 val = !icsk->icsk_ack.pingpong;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 break;
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002335
2336 case TCP_CONGESTION:
2337 if (get_user(len, optlen))
2338 return -EFAULT;
2339 len = min_t(unsigned int, len, TCP_CA_NAME_MAX);
2340 if (put_user(len, optlen))
2341 return -EFAULT;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002342 if (copy_to_user(optval, icsk->icsk_ca_ops->name, len))
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002343 return -EFAULT;
2344 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 default:
2346 return -ENOPROTOOPT;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -07002347 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348
2349 if (put_user(len, optlen))
2350 return -EFAULT;
2351 if (copy_to_user(optval, &val, len))
2352 return -EFAULT;
2353 return 0;
2354}
2355
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002356int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval,
2357 int __user *optlen)
2358{
2359 struct inet_connection_sock *icsk = inet_csk(sk);
2360
2361 if (level != SOL_TCP)
2362 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
2363 optval, optlen);
2364 return do_tcp_getsockopt(sk, level, optname, optval, optlen);
2365}
2366
2367#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002368int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
2369 char __user *optval, int __user *optlen)
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002370{
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -08002371 if (level != SOL_TCP)
2372 return inet_csk_compat_getsockopt(sk, level, optname,
2373 optval, optlen);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002374 return do_tcp_getsockopt(sk, level, optname, optval, optlen);
2375}
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002376
2377EXPORT_SYMBOL(compat_tcp_getsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002378#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379
Herbert Xu576a30e2006-06-27 13:22:38 -07002380struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07002381{
2382 struct sk_buff *segs = ERR_PTR(-EINVAL);
2383 struct tcphdr *th;
2384 unsigned thlen;
2385 unsigned int seq;
Al Virod3bc23e2006-11-14 21:24:49 -08002386 __be32 delta;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002387 unsigned int oldlen;
2388 unsigned int len;
2389
2390 if (!pskb_may_pull(skb, sizeof(*th)))
2391 goto out;
2392
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07002393 th = tcp_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002394 thlen = th->doff * 4;
2395 if (thlen < sizeof(*th))
2396 goto out;
2397
2398 if (!pskb_may_pull(skb, thlen))
2399 goto out;
2400
Herbert Xu0718bcc2006-06-25 23:55:46 -07002401 oldlen = (u16)~skb->len;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002402 __skb_pull(skb, thlen);
2403
Herbert Xu3820c3f2006-06-29 20:11:25 -07002404 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2405 /* Packet is from an untrusted source, reset gso_segs. */
Herbert Xubbcf4672006-07-03 19:38:35 -07002406 int type = skb_shinfo(skb)->gso_type;
2407 int mss;
Herbert Xu3820c3f2006-06-29 20:11:25 -07002408
Herbert Xubbcf4672006-07-03 19:38:35 -07002409 if (unlikely(type &
2410 ~(SKB_GSO_TCPV4 |
2411 SKB_GSO_DODGY |
2412 SKB_GSO_TCP_ECN |
2413 SKB_GSO_TCPV6 |
2414 0) ||
2415 !(type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))))
2416 goto out;
2417
2418 mss = skb_shinfo(skb)->gso_size;
Ilpo Järvinen172589c2007-08-28 15:50:33 -07002419 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
Herbert Xu3820c3f2006-06-29 20:11:25 -07002420
2421 segs = NULL;
2422 goto out;
2423 }
2424
Herbert Xu576a30e2006-06-27 13:22:38 -07002425 segs = skb_segment(skb, features);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002426 if (IS_ERR(segs))
2427 goto out;
2428
2429 len = skb_shinfo(skb)->gso_size;
Herbert Xu0718bcc2006-06-25 23:55:46 -07002430 delta = htonl(oldlen + (thlen + len));
Herbert Xuf4c50d92006-06-22 03:02:40 -07002431
2432 skb = segs;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07002433 th = tcp_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002434 seq = ntohl(th->seq);
2435
2436 do {
2437 th->fin = th->psh = 0;
2438
Al Virod3bc23e2006-11-14 21:24:49 -08002439 th->check = ~csum_fold((__force __wsum)((__force u32)th->check +
2440 (__force u32)delta));
Patrick McHardy84fa7932006-08-29 16:44:56 -07002441 if (skb->ip_summed != CHECKSUM_PARTIAL)
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07002442 th->check =
2443 csum_fold(csum_partial(skb_transport_header(skb),
2444 thlen, skb->csum));
Herbert Xuf4c50d92006-06-22 03:02:40 -07002445
2446 seq += len;
2447 skb = skb->next;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07002448 th = tcp_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002449
2450 th->seq = htonl(seq);
2451 th->cwr = 0;
2452 } while (skb->next);
2453
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07002454 delta = htonl(oldlen + (skb->tail - skb->transport_header) +
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07002455 skb->data_len);
Al Virod3bc23e2006-11-14 21:24:49 -08002456 th->check = ~csum_fold((__force __wsum)((__force u32)th->check +
2457 (__force u32)delta));
Patrick McHardy84fa7932006-08-29 16:44:56 -07002458 if (skb->ip_summed != CHECKSUM_PARTIAL)
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07002459 th->check = csum_fold(csum_partial(skb_transport_header(skb),
2460 thlen, skb->csum));
Herbert Xuf4c50d92006-06-22 03:02:40 -07002461
2462out:
2463 return segs;
2464}
Herbert Xuadcfc7d2006-06-30 13:36:15 -07002465EXPORT_SYMBOL(tcp_tso_segment);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002466
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002467#ifdef CONFIG_TCP_MD5SIG
2468static unsigned long tcp_md5sig_users;
2469static struct tcp_md5sig_pool **tcp_md5sig_pool;
2470static DEFINE_SPINLOCK(tcp_md5sig_pool_lock);
2471
2472static void __tcp_free_md5sig_pool(struct tcp_md5sig_pool **pool)
2473{
2474 int cpu;
2475 for_each_possible_cpu(cpu) {
2476 struct tcp_md5sig_pool *p = *per_cpu_ptr(pool, cpu);
2477 if (p) {
2478 if (p->md5_desc.tfm)
2479 crypto_free_hash(p->md5_desc.tfm);
2480 kfree(p);
2481 p = NULL;
2482 }
2483 }
2484 free_percpu(pool);
2485}
2486
2487void tcp_free_md5sig_pool(void)
2488{
2489 struct tcp_md5sig_pool **pool = NULL;
2490
David S. Miller2c4f6212007-02-20 23:51:47 -08002491 spin_lock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002492 if (--tcp_md5sig_users == 0) {
2493 pool = tcp_md5sig_pool;
2494 tcp_md5sig_pool = NULL;
2495 }
David S. Miller2c4f6212007-02-20 23:51:47 -08002496 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002497 if (pool)
2498 __tcp_free_md5sig_pool(pool);
2499}
2500
2501EXPORT_SYMBOL(tcp_free_md5sig_pool);
2502
Adrian Bunkf5b99bc2006-11-30 17:22:29 -08002503static struct tcp_md5sig_pool **__tcp_alloc_md5sig_pool(void)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002504{
2505 int cpu;
2506 struct tcp_md5sig_pool **pool;
2507
2508 pool = alloc_percpu(struct tcp_md5sig_pool *);
2509 if (!pool)
2510 return NULL;
2511
2512 for_each_possible_cpu(cpu) {
2513 struct tcp_md5sig_pool *p;
2514 struct crypto_hash *hash;
2515
2516 p = kzalloc(sizeof(*p), GFP_KERNEL);
2517 if (!p)
2518 goto out_free;
2519 *per_cpu_ptr(pool, cpu) = p;
2520
2521 hash = crypto_alloc_hash("md5", 0, CRYPTO_ALG_ASYNC);
2522 if (!hash || IS_ERR(hash))
2523 goto out_free;
2524
2525 p->md5_desc.tfm = hash;
2526 }
2527 return pool;
2528out_free:
2529 __tcp_free_md5sig_pool(pool);
2530 return NULL;
2531}
2532
2533struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void)
2534{
2535 struct tcp_md5sig_pool **pool;
2536 int alloc = 0;
2537
2538retry:
David S. Miller2c4f6212007-02-20 23:51:47 -08002539 spin_lock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002540 pool = tcp_md5sig_pool;
2541 if (tcp_md5sig_users++ == 0) {
2542 alloc = 1;
David S. Miller2c4f6212007-02-20 23:51:47 -08002543 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002544 } else if (!pool) {
2545 tcp_md5sig_users--;
David S. Miller2c4f6212007-02-20 23:51:47 -08002546 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002547 cpu_relax();
2548 goto retry;
2549 } else
David S. Miller2c4f6212007-02-20 23:51:47 -08002550 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002551
2552 if (alloc) {
2553 /* we cannot hold spinlock here because this may sleep. */
2554 struct tcp_md5sig_pool **p = __tcp_alloc_md5sig_pool();
David S. Miller2c4f6212007-02-20 23:51:47 -08002555 spin_lock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002556 if (!p) {
2557 tcp_md5sig_users--;
David S. Miller2c4f6212007-02-20 23:51:47 -08002558 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002559 return NULL;
2560 }
2561 pool = tcp_md5sig_pool;
2562 if (pool) {
2563 /* oops, it has already been assigned. */
David S. Miller2c4f6212007-02-20 23:51:47 -08002564 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002565 __tcp_free_md5sig_pool(p);
2566 } else {
2567 tcp_md5sig_pool = pool = p;
David S. Miller2c4f6212007-02-20 23:51:47 -08002568 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002569 }
2570 }
2571 return pool;
2572}
2573
2574EXPORT_SYMBOL(tcp_alloc_md5sig_pool);
2575
2576struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu)
2577{
2578 struct tcp_md5sig_pool **p;
David S. Miller2c4f6212007-02-20 23:51:47 -08002579 spin_lock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002580 p = tcp_md5sig_pool;
2581 if (p)
2582 tcp_md5sig_users++;
David S. Miller2c4f6212007-02-20 23:51:47 -08002583 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002584 return (p ? *per_cpu_ptr(p, cpu) : NULL);
2585}
2586
2587EXPORT_SYMBOL(__tcp_get_md5sig_pool);
2588
David S. Miller6931ba72006-12-13 16:25:44 -08002589void __tcp_put_md5sig_pool(void)
2590{
2591 tcp_free_md5sig_pool();
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002592}
2593
2594EXPORT_SYMBOL(__tcp_put_md5sig_pool);
Adam Langley49a72df2008-07-19 00:01:42 -07002595
2596int tcp_md5_hash_header(struct tcp_md5sig_pool *hp,
2597 struct tcphdr *th)
2598{
2599 struct scatterlist sg;
2600 int err;
2601
2602 __sum16 old_checksum = th->check;
2603 th->check = 0;
2604 /* options aren't included in the hash */
2605 sg_init_one(&sg, th, sizeof(struct tcphdr));
2606 err = crypto_hash_update(&hp->md5_desc, &sg, sizeof(struct tcphdr));
2607 th->check = old_checksum;
2608 return err;
2609}
2610
2611EXPORT_SYMBOL(tcp_md5_hash_header);
2612
2613int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp,
2614 struct sk_buff *skb, unsigned header_len)
2615{
2616 struct scatterlist sg;
2617 const struct tcphdr *tp = tcp_hdr(skb);
2618 struct hash_desc *desc = &hp->md5_desc;
2619 unsigned i;
2620 const unsigned head_data_len = skb_headlen(skb) > header_len ?
2621 skb_headlen(skb) - header_len : 0;
2622 const struct skb_shared_info *shi = skb_shinfo(skb);
2623
2624 sg_init_table(&sg, 1);
2625
2626 sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len);
2627 if (crypto_hash_update(desc, &sg, head_data_len))
2628 return 1;
2629
2630 for (i = 0; i < shi->nr_frags; ++i) {
2631 const struct skb_frag_struct *f = &shi->frags[i];
2632 sg_set_page(&sg, f->page, f->size, f->page_offset);
2633 if (crypto_hash_update(desc, &sg, f->size))
2634 return 1;
2635 }
2636
2637 return 0;
2638}
2639
2640EXPORT_SYMBOL(tcp_md5_hash_skb_data);
2641
2642int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, struct tcp_md5sig_key *key)
2643{
2644 struct scatterlist sg;
2645
2646 sg_init_one(&sg, key->key, key->keylen);
2647 return crypto_hash_update(&hp->md5_desc, &sg, key->keylen);
2648}
2649
2650EXPORT_SYMBOL(tcp_md5_hash_key);
2651
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002652#endif
2653
Andi Kleen4ac02ba2007-04-20 17:11:46 -07002654void tcp_done(struct sock *sk)
2655{
2656 if(sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
Pavel Emelyanov63231bd2008-07-16 20:22:25 -07002657 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
Andi Kleen4ac02ba2007-04-20 17:11:46 -07002658
2659 tcp_set_state(sk, TCP_CLOSE);
2660 tcp_clear_xmit_timers(sk);
2661
2662 sk->sk_shutdown = SHUTDOWN_MASK;
2663
2664 if (!sock_flag(sk, SOCK_DEAD))
2665 sk->sk_state_change(sk);
2666 else
2667 inet_csk_destroy_sock(sk);
2668}
2669EXPORT_SYMBOL_GPL(tcp_done);
2670
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002671extern struct tcp_congestion_ops tcp_reno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672
2673static __initdata unsigned long thash_entries;
2674static int __init set_thash_entries(char *str)
2675{
2676 if (!str)
2677 return 0;
2678 thash_entries = simple_strtoul(str, &str, 0);
2679 return 1;
2680}
2681__setup("thash_entries=", set_thash_entries);
2682
2683void __init tcp_init(void)
2684{
2685 struct sk_buff *skb = NULL;
Miquel van Smoorenburg57413eb2008-06-27 17:23:57 -07002686 unsigned long nr_pages, limit;
John Heffner7b4f4b52006-03-25 01:34:07 -08002687 int order, i, max_share;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688
Pavel Emelyanov1f9e6362007-12-11 02:12:04 -08002689 BUILD_BUG_ON(sizeof(struct tcp_skb_cb) > sizeof(skb->cb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07002691 tcp_hashinfo.bind_bucket_cachep =
2692 kmem_cache_create("tcp_bind_bucket",
2693 sizeof(struct inet_bind_bucket), 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09002694 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696 /* Size and allocate the main established and bind bucket
2697 * hash tables.
2698 *
2699 * The methodology is similar to that of the buffer cache.
2700 */
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07002701 tcp_hashinfo.ehash =
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 alloc_large_system_hash("TCP established",
Arnaldo Carvalho de Melo0f7ff922005-08-09 19:59:44 -07002703 sizeof(struct inet_ehash_bucket),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704 thash_entries,
2705 (num_physpages >= 128 * 1024) ?
Mike Stroyan18955cf2005-11-29 16:12:55 -08002706 13 : 15,
John Heffner9e950ef2006-11-06 23:10:51 -08002707 0,
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07002708 &tcp_hashinfo.ehash_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 NULL,
Jean Delvare0ccfe612007-10-30 00:59:25 -07002710 thash_entries ? 0 : 512 * 1024);
Eric Dumazetdbca9b2752007-02-08 14:16:46 -08002711 tcp_hashinfo.ehash_size = 1 << tcp_hashinfo.ehash_size;
2712 for (i = 0; i < tcp_hashinfo.ehash_size; i++) {
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07002713 INIT_HLIST_HEAD(&tcp_hashinfo.ehash[i].chain);
Eric Dumazetdbca9b2752007-02-08 14:16:46 -08002714 INIT_HLIST_HEAD(&tcp_hashinfo.ehash[i].twchain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 }
Eric Dumazet230140c2007-11-07 02:40:20 -08002716 if (inet_ehash_locks_alloc(&tcp_hashinfo))
2717 panic("TCP: failed to alloc ehash_locks");
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07002718 tcp_hashinfo.bhash =
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719 alloc_large_system_hash("TCP bind",
Arnaldo Carvalho de Melo0f7ff922005-08-09 19:59:44 -07002720 sizeof(struct inet_bind_hashbucket),
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07002721 tcp_hashinfo.ehash_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722 (num_physpages >= 128 * 1024) ?
Mike Stroyan18955cf2005-11-29 16:12:55 -08002723 13 : 15,
John Heffner9e950ef2006-11-06 23:10:51 -08002724 0,
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07002725 &tcp_hashinfo.bhash_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 NULL,
2727 64 * 1024);
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07002728 tcp_hashinfo.bhash_size = 1 << tcp_hashinfo.bhash_size;
2729 for (i = 0; i < tcp_hashinfo.bhash_size; i++) {
2730 spin_lock_init(&tcp_hashinfo.bhash[i].lock);
2731 INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 }
2733
2734 /* Try to be a bit smarter and adjust defaults depending
2735 * on available memory.
2736 */
2737 for (order = 0; ((1 << order) << PAGE_SHIFT) <
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07002738 (tcp_hashinfo.bhash_size * sizeof(struct inet_bind_hashbucket));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739 order++)
2740 ;
Andi Kleene7626482005-06-13 14:24:52 -07002741 if (order >= 4) {
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -07002742 tcp_death_row.sysctl_max_tw_buckets = 180000;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743 sysctl_tcp_max_orphans = 4096 << (order - 4);
2744 sysctl_max_syn_backlog = 1024;
2745 } else if (order < 3) {
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -07002746 tcp_death_row.sysctl_max_tw_buckets >>= (3 - order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747 sysctl_tcp_max_orphans >>= (3 - order);
2748 sysctl_max_syn_backlog = 128;
2749 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750
John Heffner53cdcc02007-03-16 15:04:03 -07002751 /* Set the pressure threshold to be a fraction of global memory that
2752 * is up to 1/2 at 256 MB, decreasing toward zero with the amount of
2753 * memory, with a floor of 128 pages.
2754 */
Miquel van Smoorenburg57413eb2008-06-27 17:23:57 -07002755 nr_pages = totalram_pages - totalhigh_pages;
2756 limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
2757 limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
John Heffner53cdcc02007-03-16 15:04:03 -07002758 limit = max(limit, 128UL);
2759 sysctl_tcp_mem[0] = limit / 4 * 3;
2760 sysctl_tcp_mem[1] = limit;
John Heffner52bf3762006-11-14 20:25:17 -08002761 sysctl_tcp_mem[2] = sysctl_tcp_mem[0] * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762
John Heffner53cdcc02007-03-16 15:04:03 -07002763 /* Set per-socket limits to no more than 1/128 the pressure threshold */
John Heffner7b4f4b52006-03-25 01:34:07 -08002764 limit = ((unsigned long)sysctl_tcp_mem[1]) << (PAGE_SHIFT - 7);
2765 max_share = min(4UL*1024*1024, limit);
2766
Hideo Aoki3ab224b2007-12-31 00:11:19 -08002767 sysctl_tcp_wmem[0] = SK_MEM_QUANTUM;
John Heffner7b4f4b52006-03-25 01:34:07 -08002768 sysctl_tcp_wmem[1] = 16*1024;
2769 sysctl_tcp_wmem[2] = max(64*1024, max_share);
2770
Hideo Aoki3ab224b2007-12-31 00:11:19 -08002771 sysctl_tcp_rmem[0] = SK_MEM_QUANTUM;
John Heffner7b4f4b52006-03-25 01:34:07 -08002772 sysctl_tcp_rmem[1] = 87380;
2773 sysctl_tcp_rmem[2] = max(87380, max_share);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774
2775 printk(KERN_INFO "TCP: Hash tables configured "
2776 "(established %d bind %d)\n",
Eric Dumazetdbca9b2752007-02-08 14:16:46 -08002777 tcp_hashinfo.ehash_size, tcp_hashinfo.bhash_size);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07002778
2779 tcp_register_congestion_control(&tcp_reno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780}
2781
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782EXPORT_SYMBOL(tcp_close);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783EXPORT_SYMBOL(tcp_disconnect);
2784EXPORT_SYMBOL(tcp_getsockopt);
2785EXPORT_SYMBOL(tcp_ioctl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786EXPORT_SYMBOL(tcp_poll);
2787EXPORT_SYMBOL(tcp_read_sock);
2788EXPORT_SYMBOL(tcp_recvmsg);
2789EXPORT_SYMBOL(tcp_sendmsg);
Jens Axboe9c55e012007-11-06 23:30:13 -08002790EXPORT_SYMBOL(tcp_splice_read);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791EXPORT_SYMBOL(tcp_sendpage);
2792EXPORT_SYMBOL(tcp_setsockopt);
2793EXPORT_SYMBOL(tcp_shutdown);