?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[CPUFREQ] Fix stale cpufreq_cpu_governor pointer
[linux-2.6/mini2440.git] / net / ipv4 / tcp.c
blob98440ad82558064ce4eb7ee9ad525f9725413739
1 /*
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 *
8 * Authors: Ross Biro
9 * 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
248 #include <linux/kernel.h>
249 #include <linux/module.h>
250 #include <linux/types.h>
251 #include <linux/fcntl.h>
252 #include <linux/poll.h>
253 #include <linux/init.h>
254 #include <linux/fs.h>
255 #include <linux/skbuff.h>
256 #include <linux/scatterlist.h>
257 #include <linux/splice.h>
258 #include <linux/net.h>
259 #include <linux/socket.h>
260 #include <linux/random.h>
261 #include <linux/bootmem.h>
262 #include <linux/highmem.h>
263 #include <linux/swap.h>
264 #include <linux/cache.h>
265 #include <linux/err.h>
266 #include <linux/crypto.h>
267
268 #include <net/icmp.h>
269 #include <net/tcp.h>
270 #include <net/xfrm.h>
271 #include <net/ip.h>
272 #include <net/netdma.h>
273 #include <net/sock.h>
274
275 #include <asm/uaccess.h>
276 #include <asm/ioctls.h>
277
278 int sysctl_tcp_fin_timeout __read_mostly = TCP_FIN_TIMEOUT;
279
280 struct percpu_counter tcp_orphan_count;
281 EXPORT_SYMBOL_GPL(tcp_orphan_count);
282
283 int sysctl_tcp_mem[3] __read_mostly;
284 int sysctl_tcp_wmem[3] __read_mostly;
285 int sysctl_tcp_rmem[3] __read_mostly;
286
287 EXPORT_SYMBOL(sysctl_tcp_mem);
288 EXPORT_SYMBOL(sysctl_tcp_rmem);
289 EXPORT_SYMBOL(sysctl_tcp_wmem);
290
291 atomic_t tcp_memory_allocated; /* Current allocated memory. */
292 EXPORT_SYMBOL(tcp_memory_allocated);
293
294 /*
295 * Current number of TCP sockets.
296 */
297 struct percpu_counter tcp_sockets_allocated;
298 EXPORT_SYMBOL(tcp_sockets_allocated);
299
300 /*
301 * TCP splice context
302 */
303 struct tcp_splice_state {
304 struct pipe_inode_info *pipe;
305 size_t len;
306 unsigned int flags;
307 };
308
309 /*
310 * Pressure flag: try to collapse.
311 * Technical note: it is used by multiple contexts non atomically.
312 * All the __sk_mem_schedule() is of this nature: accounting
313 * is strict, actions are advisory and have some latency.
314 */
315 int tcp_memory_pressure __read_mostly;
316
317 EXPORT_SYMBOL(tcp_memory_pressure);
318
319 void tcp_enter_memory_pressure(struct sock *sk)
320 {
321 if (!tcp_memory_pressure) {
322 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES);
323 tcp_memory_pressure = 1;
324 }
325 }
326
327 EXPORT_SYMBOL(tcp_enter_memory_pressure);
328
329 /* Convert seconds to retransmits based on initial and max timeout */
330 static u8 secs_to_retrans(int seconds, int timeout, int rto_max)
331 {
332 u8 res = 0;
333
334 if (seconds > 0) {
335 int period = timeout;
336
337 res = 1;
338 while (seconds > period && res < 255) {
339 res++;
340 timeout <<= 1;
341 if (timeout > rto_max)
342 timeout = rto_max;
343 period += timeout;
344 }
345 }
346 return res;
347 }
348
349 /* Convert retransmits to seconds based on initial and max timeout */
350 static int retrans_to_secs(u8 retrans, int timeout, int rto_max)
351 {
352 int period = 0;
353
354 if (retrans > 0) {
355 period = timeout;
356 while (--retrans) {
357 timeout <<= 1;
358 if (timeout > rto_max)
359 timeout = rto_max;
360 period += timeout;
361 }
362 }
363 return period;
364 }
365
366 /*
367 * Wait for a TCP event.
368 *
369 * Note that we don't need to lock the socket, as the upper poll layers
370 * take care of normal races (between the test and the event) and we don't
371 * go look at any of the socket buffers directly.
372 */
373 unsigned int tcp_poll(struct file *file, struct socket *sock, poll_table *wait)
374 {
375 unsigned int mask;
376 struct sock *sk = sock->sk;
377 struct tcp_sock *tp = tcp_sk(sk);
378
379 sock_poll_wait(file, sk->sk_sleep, wait);
380 if (sk->sk_state == TCP_LISTEN)
381 return inet_csk_listen_poll(sk);
382
383 /* Socket is not locked. We are protected from async events
384 * by poll logic and correct handling of state changes
385 * made by other threads is impossible in any case.
386 */
387
388 mask = 0;
389 if (sk->sk_err)
390 mask = POLLERR;
391
392 /*
393 * POLLHUP is certainly not done right. But poll() doesn't
394 * have a notion of HUP in just one direction, and for a
395 * socket the read side is more interesting.
396 *
397 * Some poll() documentation says that POLLHUP is incompatible
398 * with the POLLOUT/POLLWR flags, so somebody should check this
399 * all. But careful, it tends to be safer to return too many
400 * bits than too few, and you can easily break real applications
401 * if you don't tell them that something has hung up!
402 *
403 * Check-me.
404 *
405 * Check number 1. POLLHUP is _UNMASKABLE_ event (see UNIX98 and
406 * our fs/select.c). It means that after we received EOF,
407 * poll always returns immediately, making impossible poll() on write()
408 * in state CLOSE_WAIT. One solution is evident --- to set POLLHUP
409 * if and only if shutdown has been made in both directions.
410 * Actually, it is interesting to look how Solaris and DUX
411 * solve this dilemma. I would prefer, if POLLHUP were maskable,
412 * then we could set it on SND_SHUTDOWN. BTW examples given
413 * in Stevens' books assume exactly this behaviour, it explains
414 * why POLLHUP is incompatible with POLLOUT. --ANK
415 *
416 * NOTE. Check for TCP_CLOSE is added. The goal is to prevent
417 * blocking on fresh not-connected or disconnected socket. --ANK
418 */
419 if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == TCP_CLOSE)
420 mask |= POLLHUP;
421 if (sk->sk_shutdown & RCV_SHUTDOWN)
422 mask |= POLLIN | POLLRDNORM | POLLRDHUP;
423
424 /* Connected? */
425 if ((1 << sk->sk_state) & ~(TCPF_SYN_SENT | TCPF_SYN_RECV)) {
426 int target = sock_rcvlowat(sk, 0, INT_MAX);
427
428 if (tp->urg_seq == tp->copied_seq &&
429 !sock_flag(sk, SOCK_URGINLINE) &&
430 tp->urg_data)
431 target--;
432
433 /* Potential race condition. If read of tp below will
434 * escape above sk->sk_state, we can be illegally awaken
435 * in SYN_* states. */
436 if (tp->rcv_nxt - tp->copied_seq >= target)
437 mask |= POLLIN | POLLRDNORM;
438
439 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
440 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
441 mask |= POLLOUT | POLLWRNORM;
442 } else { /* send SIGIO later */
443 set_bit(SOCK_ASYNC_NOSPACE,
444 &sk->sk_socket->flags);
445 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
446
447 /* Race breaker. If space is freed after
448 * wspace test but before the flags are set,
449 * IO signal will be lost.
450 */
451 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
452 mask |= POLLOUT | POLLWRNORM;
453 }
454 }
455
456 if (tp->urg_data & TCP_URG_VALID)
457 mask |= POLLPRI;
458 }
459 return mask;
460 }
461
462 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
463 {
464 struct tcp_sock *tp = tcp_sk(sk);
465 int answ;
466
467 switch (cmd) {
468 case SIOCINQ:
469 if (sk->sk_state == TCP_LISTEN)
470 return -EINVAL;
471
472 lock_sock(sk);
473 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
474 answ = 0;
475 else if (sock_flag(sk, SOCK_URGINLINE) ||
476 !tp->urg_data ||
477 before(tp->urg_seq, tp->copied_seq) ||
478 !before(tp->urg_seq, tp->rcv_nxt)) {
479 struct sk_buff *skb;
480
481 answ = tp->rcv_nxt - tp->copied_seq;
482
483 /* Subtract 1, if FIN is in queue. */
484 skb = skb_peek_tail(&sk->sk_receive_queue);
485 if (answ && skb)
486 answ -= tcp_hdr(skb)->fin;
487 } else
488 answ = tp->urg_seq - tp->copied_seq;
489 release_sock(sk);
490 break;
491 case SIOCATMARK:
492 answ = tp->urg_data && tp->urg_seq == tp->copied_seq;
493 break;
494 case SIOCOUTQ:
495 if (sk->sk_state == TCP_LISTEN)
496 return -EINVAL;
497
498 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
499 answ = 0;
500 else
501 answ = tp->write_seq - tp->snd_una;
502 break;
503 default:
504 return -ENOIOCTLCMD;
505 }
506
507 return put_user(answ, (int __user *)arg);
508 }
509
510 static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb)
511 {
512 TCP_SKB_CB(skb)->flags |= TCPCB_FLAG_PSH;
513 tp->pushed_seq = tp->write_seq;
514 }
515
516 static inline int forced_push(struct tcp_sock *tp)
517 {
518 return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1));
519 }
520
521 static inline void skb_entail(struct sock *sk, struct sk_buff *skb)
522 {
523 struct tcp_sock *tp = tcp_sk(sk);
524 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
525
526 skb->csum = 0;
527 tcb->seq = tcb->end_seq = tp->write_seq;
528 tcb->flags = TCPCB_FLAG_ACK;
529 tcb->sacked = 0;
530 skb_header_release(skb);
531 tcp_add_write_queue_tail(sk, skb);
532 sk->sk_wmem_queued += skb->truesize;
533 sk_mem_charge(sk, skb->truesize);
534 if (tp->nonagle & TCP_NAGLE_PUSH)
535 tp->nonagle &= ~TCP_NAGLE_PUSH;
536 }
537
538 static inline void tcp_mark_urg(struct tcp_sock *tp, int flags,
539 struct sk_buff *skb)
540 {
541 if (flags & MSG_OOB)
542 tp->snd_up = tp->write_seq;
543 }
544
545 static inline void tcp_push(struct sock *sk, int flags, int mss_now,
546 int nonagle)
547 {
548 struct tcp_sock *tp = tcp_sk(sk);
549
550 if (tcp_send_head(sk)) {
551 struct sk_buff *skb = tcp_write_queue_tail(sk);
552 if (!(flags & MSG_MORE) || forced_push(tp))
553 tcp_mark_push(tp, skb);
554 tcp_mark_urg(tp, flags, skb);
555 __tcp_push_pending_frames(sk, mss_now,
556 (flags & MSG_MORE) ? TCP_NAGLE_CORK : nonagle);
557 }
558 }
559
560 static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
561 unsigned int offset, size_t len)
562 {
563 struct tcp_splice_state *tss = rd_desc->arg.data;
564 int ret;
565
566 ret = skb_splice_bits(skb, offset, tss->pipe, min(rd_desc->count, len),
567 tss->flags);
568 if (ret > 0)
569 rd_desc->count -= ret;
570 return ret;
571 }
572
573 static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss)
574 {
575 /* Store TCP splice context information in read_descriptor_t. */
576 read_descriptor_t rd_desc = {
577 .arg.data = tss,
578 .count = tss->len,
579 };
580
581 return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv);
582 }
583
584 /**
585 * tcp_splice_read - splice data from TCP socket to a pipe
586 * @sock: socket to splice from
587 * @ppos: position (not valid)
588 * @pipe: pipe to splice to
589 * @len: number of bytes to splice
590 * @flags: splice modifier flags
591 *
592 * Description:
593 * Will read pages from given socket and fill them into a pipe.
594 *
595 **/
596 ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos,
597 struct pipe_inode_info *pipe, size_t len,
598 unsigned int flags)
599 {
600 struct sock *sk = sock->sk;
601 struct tcp_splice_state tss = {
602 .pipe = pipe,
603 .len = len,
604 .flags = flags,
605 };
606 long timeo;
607 ssize_t spliced;
608 int ret;
609
610 /*
611 * We can't seek on a socket input
612 */
613 if (unlikely(*ppos))
614 return -ESPIPE;
615
616 ret = spliced = 0;
617
618 lock_sock(sk);
619
620 timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK);
621 while (tss.len) {
622 ret = __tcp_splice_read(sk, &tss);
623 if (ret < 0)
624 break;
625 else if (!ret) {
626 if (spliced)
627 break;
628 if (sock_flag(sk, SOCK_DONE))
629 break;
630 if (sk->sk_err) {
631 ret = sock_error(sk);
632 break;
633 }
634 if (sk->sk_shutdown & RCV_SHUTDOWN)
635 break;
636 if (sk->sk_state == TCP_CLOSE) {
637 /*
638 * This occurs when user tries to read
639 * from never connected socket.
640 */
641 if (!sock_flag(sk, SOCK_DONE))
642 ret = -ENOTCONN;
643 break;
644 }
645 if (!timeo) {
646 ret = -EAGAIN;
647 break;
648 }
649 sk_wait_data(sk, &timeo);
650 if (signal_pending(current)) {
651 ret = sock_intr_errno(timeo);
652 break;
653 }
654 continue;
655 }
656 tss.len -= ret;
657 spliced += ret;
658
659 if (!timeo)
660 break;
661 release_sock(sk);
662 lock_sock(sk);
663
664 if (sk->sk_err || sk->sk_state == TCP_CLOSE ||
665 (sk->sk_shutdown & RCV_SHUTDOWN) ||
666 signal_pending(current))
667 break;
668 }
669
670 release_sock(sk);
671
672 if (spliced)
673 return spliced;
674
675 return ret;
676 }
677
678 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp)
679 {
680 struct sk_buff *skb;
681
682 /* The TCP header must be at least 32-bit aligned. */
683 size = ALIGN(size, 4);
684
685 skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp);
686 if (skb) {
687 if (sk_wmem_schedule(sk, skb->truesize)) {
688 /*
689 * Make sure that we have exactly size bytes
690 * available to the caller, no more, no less.
691 */
692 skb_reserve(skb, skb_tailroom(skb) - size);
693 return skb;
694 }
695 __kfree_skb(skb);
696 } else {
697 sk->sk_prot->enter_memory_pressure(sk);
698 sk_stream_moderate_sndbuf(sk);
699 }
700 return NULL;
701 }
702
703 static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now,
704 int large_allowed)
705 {
706 struct tcp_sock *tp = tcp_sk(sk);
707 u32 xmit_size_goal, old_size_goal;
708
709 xmit_size_goal = mss_now;
710
711 if (large_allowed && sk_can_gso(sk)) {
712 xmit_size_goal = ((sk->sk_gso_max_size - 1) -
713 inet_csk(sk)->icsk_af_ops->net_header_len -
714 inet_csk(sk)->icsk_ext_hdr_len -
715 tp->tcp_header_len);
716
717 xmit_size_goal = tcp_bound_to_half_wnd(tp, xmit_size_goal);
718
719 /* We try hard to avoid divides here */
720 old_size_goal = tp->xmit_size_goal_segs * mss_now;
721
722 if (likely(old_size_goal <= xmit_size_goal &&
723 old_size_goal + mss_now > xmit_size_goal)) {
724 xmit_size_goal = old_size_goal;
725 } else {
726 tp->xmit_size_goal_segs = xmit_size_goal / mss_now;
727 xmit_size_goal = tp->xmit_size_goal_segs * mss_now;
728 }
729 }
730
731 return max(xmit_size_goal, mss_now);
732 }
733
734 static int tcp_send_mss(struct sock *sk, int *size_goal, int flags)
735 {
736 int mss_now;
737
738 mss_now = tcp_current_mss(sk);
739 *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB));
740
741 return mss_now;
742 }
743
744 static ssize_t do_tcp_sendpages(struct sock *sk, struct page **pages, int poffset,
745 size_t psize, int flags)
746 {
747 struct tcp_sock *tp = tcp_sk(sk);
748 int mss_now, size_goal;
749 int err;
750 ssize_t copied;
751 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
752
753 /* Wait for a connection to finish. */
754 if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))
755 if ((err = sk_stream_wait_connect(sk, &timeo)) != 0)
756 goto out_err;
757
758 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
759
760 mss_now = tcp_send_mss(sk, &size_goal, flags);
761 copied = 0;
762
763 err = -EPIPE;
764 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
765 goto out_err;
766
767 while (psize > 0) {
768 struct sk_buff *skb = tcp_write_queue_tail(sk);
769 struct page *page = pages[poffset / PAGE_SIZE];
770 int copy, i, can_coalesce;
771 int offset = poffset % PAGE_SIZE;
772 int size = min_t(size_t, psize, PAGE_SIZE - offset);
773
774 if (!tcp_send_head(sk) || (copy = size_goal - skb->len) <= 0) {
775 new_segment:
776 if (!sk_stream_memory_free(sk))
777 goto wait_for_sndbuf;
778
779 skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation);
780 if (!skb)
781 goto wait_for_memory;
782
783 skb_entail(sk, skb);
784 copy = size_goal;
785 }
786
787 if (copy > size)
788 copy = size;
789
790 i = skb_shinfo(skb)->nr_frags;
791 can_coalesce = skb_can_coalesce(skb, i, page, offset);
792 if (!can_coalesce && i >= MAX_SKB_FRAGS) {
793 tcp_mark_push(tp, skb);
794 goto new_segment;
795 }
796 if (!sk_wmem_schedule(sk, copy))
797 goto wait_for_memory;
798
799 if (can_coalesce) {
800 skb_shinfo(skb)->frags[i - 1].size += copy;
801 } else {
802 get_page(page);
803 skb_fill_page_desc(skb, i, page, offset, copy);
804 }
805
806 skb->len += copy;
807 skb->data_len += copy;
808 skb->truesize += copy;
809 sk->sk_wmem_queued += copy;
810 sk_mem_charge(sk, copy);
811 skb->ip_summed = CHECKSUM_PARTIAL;
812 tp->write_seq += copy;
813 TCP_SKB_CB(skb)->end_seq += copy;
814 skb_shinfo(skb)->gso_segs = 0;
815
816 if (!copied)
817 TCP_SKB_CB(skb)->flags &= ~TCPCB_FLAG_PSH;
818
819 copied += copy;
820 poffset += copy;
821 if (!(psize -= copy))
822 goto out;
823
824 if (skb->len < size_goal || (flags & MSG_OOB))
825 continue;
826
827 if (forced_push(tp)) {
828 tcp_mark_push(tp, skb);
829 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
830 } else if (skb == tcp_send_head(sk))
831 tcp_push_one(sk, mss_now);
832 continue;
833
834 wait_for_sndbuf:
835 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
836 wait_for_memory:
837 if (copied)
838 tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH);
839
840 if ((err = sk_stream_wait_memory(sk, &timeo)) != 0)
841 goto do_error;
842
843 mss_now = tcp_send_mss(sk, &size_goal, flags);
844 }
845
846 out:
847 if (copied)
848 tcp_push(sk, flags, mss_now, tp->nonagle);
849 return copied;
850
851 do_error:
852 if (copied)
853 goto out;
854 out_err:
855 return sk_stream_error(sk, flags, err);
856 }
857
858 ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset,
859 size_t size, int flags)
860 {
861 ssize_t res;
862 struct sock *sk = sock->sk;
863
864 if (!(sk->sk_route_caps & NETIF_F_SG) ||
865 !(sk->sk_route_caps & NETIF_F_ALL_CSUM))
866 return sock_no_sendpage(sock, page, offset, size, flags);
867
868 lock_sock(sk);
869 TCP_CHECK_TIMER(sk);
870 res = do_tcp_sendpages(sk, &page, offset, size, flags);
871 TCP_CHECK_TIMER(sk);
872 release_sock(sk);
873 return res;
874 }
875
876 #define TCP_PAGE(sk) (sk->sk_sndmsg_page)
877 #define TCP_OFF(sk) (sk->sk_sndmsg_off)
878
879 static inline int select_size(struct sock *sk)
880 {
881 struct tcp_sock *tp = tcp_sk(sk);
882 int tmp = tp->mss_cache;
883
884 if (sk->sk_route_caps & NETIF_F_SG) {
885 if (sk_can_gso(sk))
886 tmp = 0;
887 else {
888 int pgbreak = SKB_MAX_HEAD(MAX_TCP_HEADER);
889
890 if (tmp >= pgbreak &&
891 tmp <= pgbreak + (MAX_SKB_FRAGS - 1) * PAGE_SIZE)
892 tmp = pgbreak;
893 }
894 }
895
896 return tmp;
897 }
898
899 int tcp_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
900 size_t size)
901 {
902 struct sock *sk = sock->sk;
903 struct iovec *iov;
904 struct tcp_sock *tp = tcp_sk(sk);
905 struct sk_buff *skb;
906 int iovlen, flags;
907 int mss_now, size_goal;
908 int err, copied;
909 long timeo;
910
911 lock_sock(sk);
912 TCP_CHECK_TIMER(sk);
913
914 flags = msg->msg_flags;
915 timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
916
917 /* Wait for a connection to finish. */
918 if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))
919 if ((err = sk_stream_wait_connect(sk, &timeo)) != 0)
920 goto out_err;
921
922 /* This should be in poll */
923 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
924
925 mss_now = tcp_send_mss(sk, &size_goal, flags);
926
927 /* Ok commence sending. */
928 iovlen = msg->msg_iovlen;
929 iov = msg->msg_iov;
930 copied = 0;
931
932 err = -EPIPE;
933 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
934 goto out_err;
935
936 while (--iovlen >= 0) {
937 int seglen = iov->iov_len;
938 unsigned char __user *from = iov->iov_base;
939
940 iov++;
941
942 while (seglen > 0) {
943 int copy = 0;
944 int max = size_goal;
945
946 skb = tcp_write_queue_tail(sk);
947 if (tcp_send_head(sk)) {
948 if (skb->ip_summed == CHECKSUM_NONE)
949 max = mss_now;
950 copy = max - skb->len;
951 }
952
953 if (copy <= 0) {
954 new_segment:
955 /* Allocate new segment. If the interface is SG,
956 * allocate skb fitting to single page.
957 */
958 if (!sk_stream_memory_free(sk))
959 goto wait_for_sndbuf;
960
961 skb = sk_stream_alloc_skb(sk, select_size(sk),
962 sk->sk_allocation);
963 if (!skb)
964 goto wait_for_memory;
965
966 /*
967 * Check whether we can use HW checksum.
968 */
969 if (sk->sk_route_caps & NETIF_F_ALL_CSUM)
970 skb->ip_summed = CHECKSUM_PARTIAL;
971
972 skb_entail(sk, skb);
973 copy = size_goal;
974 max = size_goal;
975 }
976
977 /* Try to append data to the end of skb. */
978 if (copy > seglen)
979 copy = seglen;
980
981 /* Where to copy to? */
982 if (skb_tailroom(skb) > 0) {
983 /* We have some space in skb head. Superb! */
984 if (copy > skb_tailroom(skb))
985 copy = skb_tailroom(skb);
986 if ((err = skb_add_data(skb, from, copy)) != 0)
987 goto do_fault;
988 } else {
989 int merge = 0;
990 int i = skb_shinfo(skb)->nr_frags;
991 struct page *page = TCP_PAGE(sk);
992 int off = TCP_OFF(sk);
993
994 if (skb_can_coalesce(skb, i, page, off) &&
995 off != PAGE_SIZE) {
996 /* We can extend the last page
997 * fragment. */
998 merge = 1;
999 } else if (i == MAX_SKB_FRAGS ||
1000 (!i &&
1001 !(sk->sk_route_caps & NETIF_F_SG))) {
1002 /* Need to add new fragment and cannot
1003 * do this because interface is non-SG,
1004 * or because all the page slots are
1005 * busy. */
1006 tcp_mark_push(tp, skb);
1007 goto new_segment;
1008 } else if (page) {
1009 if (off == PAGE_SIZE) {
1010 put_page(page);
1011 TCP_PAGE(sk) = page = NULL;
1012 off = 0;
1013 }
1014 } else
1015 off = 0;
1016
1017 if (copy > PAGE_SIZE - off)
1018 copy = PAGE_SIZE - off;
1019
1020 if (!sk_wmem_schedule(sk, copy))
1021 goto wait_for_memory;
1022
1023 if (!page) {
1024 /* Allocate new cache page. */
1025 if (!(page = sk_stream_alloc_page(sk)))
1026 goto wait_for_memory;
1027 }
1028
1029 /* Time to copy data. We are close to
1030 * the end! */
1031 err = skb_copy_to_page(sk, from, skb, page,
1032 off, copy);
1033 if (err) {
1034 /* If this page was new, give it to the
1035 * socket so it does not get leaked.
1036 */
1037 if (!TCP_PAGE(sk)) {
1038 TCP_PAGE(sk) = page;
1039 TCP_OFF(sk) = 0;
1040 }
1041 goto do_error;
1042 }
1043
1044 /* Update the skb. */
1045 if (merge) {
1046 skb_shinfo(skb)->frags[i - 1].size +=
1047 copy;
1048 } else {
1049 skb_fill_page_desc(skb, i, page, off, copy);
1050 if (TCP_PAGE(sk)) {
1051 get_page(page);
1052 } else if (off + copy < PAGE_SIZE) {
1053 get_page(page);
1054 TCP_PAGE(sk) = page;
1055 }
1056 }
1057
1058 TCP_OFF(sk) = off + copy;
1059 }
1060
1061 if (!copied)
1062 TCP_SKB_CB(skb)->flags &= ~TCPCB_FLAG_PSH;
1063
1064 tp->write_seq += copy;
1065 TCP_SKB_CB(skb)->end_seq += copy;
1066 skb_shinfo(skb)->gso_segs = 0;
1067
1068 from += copy;
1069 copied += copy;
1070 if ((seglen -= copy) == 0 && iovlen == 0)
1071 goto out;
1072
1073 if (skb->len < max || (flags & MSG_OOB))
1074 continue;
1075
1076 if (forced_push(tp)) {
1077 tcp_mark_push(tp, skb);
1078 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
1079 } else if (skb == tcp_send_head(sk))
1080 tcp_push_one(sk, mss_now);
1081 continue;
1082
1083 wait_for_sndbuf:
1084 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1085 wait_for_memory:
1086 if (copied)
1087 tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH);
1088
1089 if ((err = sk_stream_wait_memory(sk, &timeo)) != 0)
1090 goto do_error;
1091
1092 mss_now = tcp_send_mss(sk, &size_goal, flags);
1093 }
1094 }
1095
1096 out:
1097 if (copied)
1098 tcp_push(sk, flags, mss_now, tp->nonagle);
1099 TCP_CHECK_TIMER(sk);
1100 release_sock(sk);
1101 return copied;
1102
1103 do_fault:
1104 if (!skb->len) {
1105 tcp_unlink_write_queue(skb, sk);
1106 /* It is the one place in all of TCP, except connection
1107 * reset, where we can be unlinking the send_head.
1108 */
1109 tcp_check_send_head(sk, skb);
1110 sk_wmem_free_skb(sk, skb);
1111 }
1112
1113 do_error:
1114 if (copied)
1115 goto out;
1116 out_err:
1117 err = sk_stream_error(sk, flags, err);
1118 TCP_CHECK_TIMER(sk);
1119 release_sock(sk);
1120 return err;
1121 }
1122
1123 /*
1124 * Handle reading urgent data. BSD has very simple semantics for
1125 * this, no blocking and very strange errors 8)
1126 */
1127
1128 static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags)
1129 {
1130 struct tcp_sock *tp = tcp_sk(sk);
1131
1132 /* No URG data to read. */
1133 if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data ||
1134 tp->urg_data == TCP_URG_READ)
1135 return -EINVAL; /* Yes this is right ! */
1136
1137 if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE))
1138 return -ENOTCONN;
1139
1140 if (tp->urg_data & TCP_URG_VALID) {
1141 int err = 0;
1142 char c = tp->urg_data;
1143
1144 if (!(flags & MSG_PEEK))
1145 tp->urg_data = TCP_URG_READ;
1146
1147 /* Read urgent data. */
1148 msg->msg_flags |= MSG_OOB;
1149
1150 if (len > 0) {
1151 if (!(flags & MSG_TRUNC))
1152 err = memcpy_toiovec(msg->msg_iov, &c, 1);
1153 len = 1;
1154 } else
1155 msg->msg_flags |= MSG_TRUNC;
1156
1157 return err ? -EFAULT : len;
1158 }
1159
1160 if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
1161 return 0;
1162
1163 /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and
1164 * the available implementations agree in this case:
1165 * this call should never block, independent of the
1166 * blocking state of the socket.
1167 * Mike <pall@rz.uni-karlsruhe.de>
1168 */
1169 return -EAGAIN;
1170 }
1171
1172 /* Clean up the receive buffer for full frames taken by the user,
1173 * then send an ACK if necessary. COPIED is the number of bytes
1174 * tcp_recvmsg has given to the user so far, it speeds up the
1175 * calculation of whether or not we must ACK for the sake of
1176 * a window update.
1177 */
1178 void tcp_cleanup_rbuf(struct sock *sk, int copied)
1179 {
1180 struct tcp_sock *tp = tcp_sk(sk);
1181 int time_to_ack = 0;
1182
1183 #if TCP_DEBUG
1184 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
1185
1186 WARN_ON(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq));
1187 #endif
1188
1189 if (inet_csk_ack_scheduled(sk)) {
1190 const struct inet_connection_sock *icsk = inet_csk(sk);
1191 /* Delayed ACKs frequently hit locked sockets during bulk
1192 * receive. */
1193 if (icsk->icsk_ack.blocked ||
1194 /* Once-per-two-segments ACK was not sent by tcp_input.c */
1195 tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss ||
1196 /*
1197 * If this read emptied read buffer, we send ACK, if
1198 * connection is not bidirectional, user drained
1199 * receive buffer and there was a small segment
1200 * in queue.
1201 */
1202 (copied > 0 &&
1203 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) ||
1204 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) &&
1205 !icsk->icsk_ack.pingpong)) &&
1206 !atomic_read(&sk->sk_rmem_alloc)))
1207 time_to_ack = 1;
1208 }
1209
1210 /* We send an ACK if we can now advertise a non-zero window
1211 * which has been raised "significantly".
1212 *
1213 * Even if window raised up to infinity, do not send window open ACK
1214 * in states, where we will not receive more. It is useless.
1215 */
1216 if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1217 __u32 rcv_window_now = tcp_receive_window(tp);
1218
1219 /* Optimize, __tcp_select_window() is not cheap. */
1220 if (2*rcv_window_now <= tp->window_clamp) {
1221 __u32 new_window = __tcp_select_window(sk);
1222
1223 /* Send ACK now, if this read freed lots of space
1224 * in our buffer. Certainly, new_window is new window.
1225 * We can advertise it now, if it is not less than current one.
1226 * "Lots" means "at least twice" here.
1227 */
1228 if (new_window && new_window >= 2 * rcv_window_now)
1229 time_to_ack = 1;
1230 }
1231 }
1232 if (time_to_ack)
1233 tcp_send_ack(sk);
1234 }
1235
1236 static void tcp_prequeue_process(struct sock *sk)
1237 {
1238 struct sk_buff *skb;
1239 struct tcp_sock *tp = tcp_sk(sk);
1240
1241 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPPREQUEUED);
1242
1243 /* RX process wants to run with disabled BHs, though it is not
1244 * necessary */
1245 local_bh_disable();
1246 while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
1247 sk_backlog_rcv(sk, skb);
1248 local_bh_enable();
1249
1250 /* Clear memory counter. */
1251 tp->ucopy.memory = 0;
1252 }
1253
1254 static inline struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off)
1255 {
1256 struct sk_buff *skb;
1257 u32 offset;
1258
1259 skb_queue_walk(&sk->sk_receive_queue, skb) {
1260 offset = seq - TCP_SKB_CB(skb)->seq;
1261 if (tcp_hdr(skb)->syn)
1262 offset--;
1263 if (offset < skb->len || tcp_hdr(skb)->fin) {
1264 *off = offset;
1265 return skb;
1266 }
1267 }
1268 return NULL;
1269 }
1270
1271 /*
1272 * This routine provides an alternative to tcp_recvmsg() for routines
1273 * that would like to handle copying from skbuffs directly in 'sendfile'
1274 * fashion.
1275 * Note:
1276 * - It is assumed that the socket was locked by the caller.
1277 * - The routine does not block.
1278 * - At present, there is no support for reading OOB data
1279 * or for 'peeking' the socket using this routine
1280 * (although both would be easy to implement).
1281 */
1282 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
1283 sk_read_actor_t recv_actor)
1284 {
1285 struct sk_buff *skb;
1286 struct tcp_sock *tp = tcp_sk(sk);
1287 u32 seq = tp->copied_seq;
1288 u32 offset;
1289 int copied = 0;
1290
1291 if (sk->sk_state == TCP_LISTEN)
1292 return -ENOTCONN;
1293 while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) {
1294 if (offset < skb->len) {
1295 int used;
1296 size_t len;
1297
1298 len = skb->len - offset;
1299 /* Stop reading if we hit a patch of urgent data */
1300 if (tp->urg_data) {
1301 u32 urg_offset = tp->urg_seq - seq;
1302 if (urg_offset < len)
1303 len = urg_offset;
1304 if (!len)
1305 break;
1306 }
1307 used = recv_actor(desc, skb, offset, len);
1308 if (used < 0) {
1309 if (!copied)
1310 copied = used;
1311 break;
1312 } else if (used <= len) {
1313 seq += used;
1314 copied += used;
1315 offset += used;
1316 }
1317 /*
1318 * If recv_actor drops the lock (e.g. TCP splice
1319 * receive) the skb pointer might be invalid when
1320 * getting here: tcp_collapse might have deleted it
1321 * while aggregating skbs from the socket queue.
1322 */
1323 skb = tcp_recv_skb(sk, seq-1, &offset);
1324 if (!skb || (offset+1 != skb->len))
1325 break;
1326 }
1327 if (tcp_hdr(skb)->fin) {
1328 sk_eat_skb(sk, skb, 0);
1329 ++seq;
1330 break;
1331 }
1332 sk_eat_skb(sk, skb, 0);
1333 if (!desc->count)
1334 break;
1335 }
1336 tp->copied_seq = seq;
1337
1338 tcp_rcv_space_adjust(sk);
1339
1340 /* Clean up data we have read: This will do ACK frames. */
1341 if (copied > 0)
1342 tcp_cleanup_rbuf(sk, copied);
1343 return copied;
1344 }
1345
1346 /*
1347 * This routine copies from a sock struct into the user buffer.
1348 *
1349 * Technical note: in 2.3 we work on _locked_ socket, so that
1350 * tricks with *seq access order and skb->users are not required.
1351 * Probably, code can be easily improved even more.
1352 */
1353
1354 int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1355 size_t len, int nonblock, int flags, int *addr_len)
1356 {
1357 struct tcp_sock *tp = tcp_sk(sk);
1358 int copied = 0;
1359 u32 peek_seq;
1360 u32 *seq;
1361 unsigned long used;
1362 int err;
1363 int target; /* Read at least this many bytes */
1364 long timeo;
1365 struct task_struct *user_recv = NULL;
1366 int copied_early = 0;
1367 struct sk_buff *skb;
1368 u32 urg_hole = 0;
1369
1370 lock_sock(sk);
1371
1372 TCP_CHECK_TIMER(sk);
1373
1374 err = -ENOTCONN;
1375 if (sk->sk_state == TCP_LISTEN)
1376 goto out;
1377
1378 timeo = sock_rcvtimeo(sk, nonblock);
1379
1380 /* Urgent data needs to be handled specially. */
1381 if (flags & MSG_OOB)
1382 goto recv_urg;
1383
1384 seq = &tp->copied_seq;
1385 if (flags & MSG_PEEK) {
1386 peek_seq = tp->copied_seq;
1387 seq = &peek_seq;
1388 }
1389
1390 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1391
1392 #ifdef CONFIG_NET_DMA
1393 tp->ucopy.dma_chan = NULL;
1394 preempt_disable();
1395 skb = skb_peek_tail(&sk->sk_receive_queue);
1396 {
1397 int available = 0;
1398
1399 if (skb)
1400 available = TCP_SKB_CB(skb)->seq + skb->len - (*seq);
1401 if ((available < target) &&
1402 (len > sysctl_tcp_dma_copybreak) && !(flags & MSG_PEEK) &&
1403 !sysctl_tcp_low_latency &&
1404 dma_find_channel(DMA_MEMCPY)) {
1405 preempt_enable_no_resched();
1406 tp->ucopy.pinned_list =
1407 dma_pin_iovec_pages(msg->msg_iov, len);
1408 } else {
1409 preempt_enable_no_resched();
1410 }
1411 }
1412 #endif
1413
1414 do {
1415 u32 offset;
1416
1417 /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */
1418 if (tp->urg_data && tp->urg_seq == *seq) {
1419 if (copied)
1420 break;
1421 if (signal_pending(current)) {
1422 copied = timeo ? sock_intr_errno(timeo) : -EAGAIN;
1423 break;
1424 }
1425 }
1426
1427 /* Next get a buffer. */
1428
1429 skb_queue_walk(&sk->sk_receive_queue, skb) {
1430 /* Now that we have two receive queues this
1431 * shouldn't happen.
1432 */
1433 if (before(*seq, TCP_SKB_CB(skb)->seq)) {
1434 printk(KERN_INFO "recvmsg bug: copied %X "
1435 "seq %X\n", *seq, TCP_SKB_CB(skb)->seq);
1436 break;
1437 }
1438 offset = *seq - TCP_SKB_CB(skb)->seq;
1439 if (tcp_hdr(skb)->syn)
1440 offset--;
1441 if (offset < skb->len)
1442 goto found_ok_skb;
1443 if (tcp_hdr(skb)->fin)
1444 goto found_fin_ok;
1445 WARN(!(flags & MSG_PEEK), KERN_INFO "recvmsg bug 2: "
1446 "copied %X seq %X\n", *seq,
1447 TCP_SKB_CB(skb)->seq);
1448 }
1449
1450 /* Well, if we have backlog, try to process it now yet. */
1451
1452 if (copied >= target && !sk->sk_backlog.tail)
1453 break;
1454
1455 if (copied) {
1456 if (sk->sk_err ||
1457 sk->sk_state == TCP_CLOSE ||
1458 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1459 !timeo ||
1460 signal_pending(current))
1461 break;
1462 } else {
1463 if (sock_flag(sk, SOCK_DONE))
1464 break;
1465
1466 if (sk->sk_err) {
1467 copied = sock_error(sk);
1468 break;
1469 }
1470
1471 if (sk->sk_shutdown & RCV_SHUTDOWN)
1472 break;
1473
1474 if (sk->sk_state == TCP_CLOSE) {
1475 if (!sock_flag(sk, SOCK_DONE)) {
1476 /* This occurs when user tries to read
1477 * from never connected socket.
1478 */
1479 copied = -ENOTCONN;
1480 break;
1481 }
1482 break;
1483 }
1484
1485 if (!timeo) {
1486 copied = -EAGAIN;
1487 break;
1488 }
1489
1490 if (signal_pending(current)) {
1491 copied = sock_intr_errno(timeo);
1492 break;
1493 }
1494 }
1495
1496 tcp_cleanup_rbuf(sk, copied);
1497
1498 if (!sysctl_tcp_low_latency && tp->ucopy.task == user_recv) {
1499 /* Install new reader */
1500 if (!user_recv && !(flags & (MSG_TRUNC | MSG_PEEK))) {
1501 user_recv = current;
1502 tp->ucopy.task = user_recv;
1503 tp->ucopy.iov = msg->msg_iov;
1504 }
1505
1506 tp->ucopy.len = len;
1507
1508 WARN_ON(tp->copied_seq != tp->rcv_nxt &&
1509 !(flags & (MSG_PEEK | MSG_TRUNC)));
1510
1511 /* Ugly... If prequeue is not empty, we have to
1512 * process it before releasing socket, otherwise
1513 * order will be broken at second iteration.
1514 * More elegant solution is required!!!
1515 *
1516 * Look: we have the following (pseudo)queues:
1517 *
1518 * 1. packets in flight
1519 * 2. backlog
1520 * 3. prequeue
1521 * 4. receive_queue
1522 *
1523 * Each queue can be processed only if the next ones
1524 * are empty. At this point we have empty receive_queue.
1525 * But prequeue _can_ be not empty after 2nd iteration,
1526 * when we jumped to start of loop because backlog
1527 * processing added something to receive_queue.
1528 * We cannot release_sock(), because backlog contains
1529 * packets arrived _after_ prequeued ones.
1530 *
1531 * Shortly, algorithm is clear --- to process all
1532 * the queues in order. We could make it more directly,
1533 * requeueing packets from backlog to prequeue, if
1534 * is not empty. It is more elegant, but eats cycles,
1535 * unfortunately.
1536 */
1537 if (!skb_queue_empty(&tp->ucopy.prequeue))
1538 goto do_prequeue;
1539
1540 /* __ Set realtime policy in scheduler __ */
1541 }
1542
1543 if (copied >= target) {
1544 /* Do not sleep, just process backlog. */
1545 release_sock(sk);
1546 lock_sock(sk);
1547 } else
1548 sk_wait_data(sk, &timeo);
1549
1550 #ifdef CONFIG_NET_DMA
1551 tp->ucopy.wakeup = 0;
1552 #endif
1553
1554 if (user_recv) {
1555 int chunk;
1556
1557 /* __ Restore normal policy in scheduler __ */
1558
1559 if ((chunk = len - tp->ucopy.len) != 0) {
1560 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMBACKLOG, chunk);
1561 len -= chunk;
1562 copied += chunk;
1563 }
1564
1565 if (tp->rcv_nxt == tp->copied_seq &&
1566 !skb_queue_empty(&tp->ucopy.prequeue)) {
1567 do_prequeue:
1568 tcp_prequeue_process(sk);
1569
1570 if ((chunk = len - tp->ucopy.len) != 0) {
1571 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
1572 len -= chunk;
1573 copied += chunk;
1574 }
1575 }
1576 }
1577 if ((flags & MSG_PEEK) &&
1578 (peek_seq - copied - urg_hole != tp->copied_seq)) {
1579 if (net_ratelimit())
1580 printk(KERN_DEBUG "TCP(%s:%d): Application bug, race in MSG_PEEK.\n",
1581 current->comm, task_pid_nr(current));
1582 peek_seq = tp->copied_seq;
1583 }
1584 continue;
1585
1586 found_ok_skb:
1587 /* Ok so how much can we use? */
1588 used = skb->len - offset;
1589 if (len < used)
1590 used = len;
1591
1592 /* Do we have urgent data here? */
1593 if (tp->urg_data) {
1594 u32 urg_offset = tp->urg_seq - *seq;
1595 if (urg_offset < used) {
1596 if (!urg_offset) {
1597 if (!sock_flag(sk, SOCK_URGINLINE)) {
1598 ++*seq;
1599 urg_hole++;
1600 offset++;
1601 used--;
1602 if (!used)
1603 goto skip_copy;
1604 }
1605 } else
1606 used = urg_offset;
1607 }
1608 }
1609
1610 if (!(flags & MSG_TRUNC)) {
1611 #ifdef CONFIG_NET_DMA
1612 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1613 tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
1614
1615 if (tp->ucopy.dma_chan) {
1616 tp->ucopy.dma_cookie = dma_skb_copy_datagram_iovec(
1617 tp->ucopy.dma_chan, skb, offset,
1618 msg->msg_iov, used,
1619 tp->ucopy.pinned_list);
1620
1621 if (tp->ucopy.dma_cookie < 0) {
1622
1623 printk(KERN_ALERT "dma_cookie < 0\n");
1624
1625 /* Exception. Bailout! */
1626 if (!copied)
1627 copied = -EFAULT;
1628 break;
1629 }
1630 if ((offset + used) == skb->len)
1631 copied_early = 1;
1632
1633 } else
1634 #endif
1635 {
1636 err = skb_copy_datagram_iovec(skb, offset,
1637 msg->msg_iov, used);
1638 if (err) {
1639 /* Exception. Bailout! */
1640 if (!copied)
1641 copied = -EFAULT;
1642 break;
1643 }
1644 }
1645 }
1646
1647 *seq += used;
1648 copied += used;
1649 len -= used;
1650
1651 tcp_rcv_space_adjust(sk);
1652
1653 skip_copy:
1654 if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) {
1655 tp->urg_data = 0;
1656 tcp_fast_path_check(sk);
1657 }
1658 if (used + offset < skb->len)
1659 continue;
1660
1661 if (tcp_hdr(skb)->fin)
1662 goto found_fin_ok;
1663 if (!(flags & MSG_PEEK)) {
1664 sk_eat_skb(sk, skb, copied_early);
1665 copied_early = 0;
1666 }
1667 continue;
1668
1669 found_fin_ok:
1670 /* Process the FIN. */
1671 ++*seq;
1672 if (!(flags & MSG_PEEK)) {
1673 sk_eat_skb(sk, skb, copied_early);
1674 copied_early = 0;
1675 }
1676 break;
1677 } while (len > 0);
1678
1679 if (user_recv) {
1680 if (!skb_queue_empty(&tp->ucopy.prequeue)) {
1681 int chunk;
1682
1683 tp->ucopy.len = copied > 0 ? len : 0;
1684
1685 tcp_prequeue_process(sk);
1686
1687 if (copied > 0 && (chunk = len - tp->ucopy.len) != 0) {
1688 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
1689 len -= chunk;
1690 copied += chunk;
1691 }
1692 }
1693
1694 tp->ucopy.task = NULL;
1695 tp->ucopy.len = 0;
1696 }
1697
1698 #ifdef CONFIG_NET_DMA
1699 if (tp->ucopy.dma_chan) {
1700 dma_cookie_t done, used;
1701
1702 dma_async_memcpy_issue_pending(tp->ucopy.dma_chan);
1703
1704 while (dma_async_memcpy_complete(tp->ucopy.dma_chan,
1705 tp->ucopy.dma_cookie, &done,
1706 &used) == DMA_IN_PROGRESS) {
1707 /* do partial cleanup of sk_async_wait_queue */
1708 while ((skb = skb_peek(&sk->sk_async_wait_queue)) &&
1709 (dma_async_is_complete(skb->dma_cookie, done,
1710 used) == DMA_SUCCESS)) {
1711 __skb_dequeue(&sk->sk_async_wait_queue);
1712 kfree_skb(skb);
1713 }
1714 }
1715
1716 /* Safe to free early-copied skbs now */
1717 __skb_queue_purge(&sk->sk_async_wait_queue);
1718 tp->ucopy.dma_chan = NULL;
1719 }
1720 if (tp->ucopy.pinned_list) {
1721 dma_unpin_iovec_pages(tp->ucopy.pinned_list);
1722 tp->ucopy.pinned_list = NULL;
1723 }
1724 #endif
1725
1726 /* According to UNIX98, msg_name/msg_namelen are ignored
1727 * on connected socket. I was just happy when found this 8) --ANK
1728 */
1729
1730 /* Clean up data we have read: This will do ACK frames. */
1731 tcp_cleanup_rbuf(sk, copied);
1732
1733 TCP_CHECK_TIMER(sk);
1734 release_sock(sk);
1735 return copied;
1736
1737 out:
1738 TCP_CHECK_TIMER(sk);
1739 release_sock(sk);
1740 return err;
1741
1742 recv_urg:
1743 err = tcp_recv_urg(sk, msg, len, flags);
1744 goto out;
1745 }
1746
1747 void tcp_set_state(struct sock *sk, int state)
1748 {
1749 int oldstate = sk->sk_state;
1750
1751 switch (state) {
1752 case TCP_ESTABLISHED:
1753 if (oldstate != TCP_ESTABLISHED)
1754 TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
1755 break;
1756
1757 case TCP_CLOSE:
1758 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
1759 TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS);
1760
1761 sk->sk_prot->unhash(sk);
1762 if (inet_csk(sk)->icsk_bind_hash &&
1763 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1764 inet_put_port(sk);
1765 /* fall through */
1766 default:
1767 if (oldstate == TCP_ESTABLISHED)
1768 TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
1769 }
1770
1771 /* Change state AFTER socket is unhashed to avoid closed
1772 * socket sitting in hash tables.
1773 */
1774 sk->sk_state = state;
1775
1776 #ifdef STATE_TRACE
1777 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n", sk, statename[oldstate], statename[state]);
1778 #endif
1779 }
1780 EXPORT_SYMBOL_GPL(tcp_set_state);
1781
1782 /*
1783 * State processing on a close. This implements the state shift for
1784 * sending our FIN frame. Note that we only send a FIN for some
1785 * states. A shutdown() may have already sent the FIN, or we may be
1786 * closed.
1787 */
1788
1789 static const unsigned char new_state[16] = {
1790 /* current state: new state: action: */
1791 /* (Invalid) */ TCP_CLOSE,
1792 /* TCP_ESTABLISHED */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
1793 /* TCP_SYN_SENT */ TCP_CLOSE,
1794 /* TCP_SYN_RECV */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
1795 /* TCP_FIN_WAIT1 */ TCP_FIN_WAIT1,
1796 /* TCP_FIN_WAIT2 */ TCP_FIN_WAIT2,
1797 /* TCP_TIME_WAIT */ TCP_CLOSE,
1798 /* TCP_CLOSE */ TCP_CLOSE,
1799 /* TCP_CLOSE_WAIT */ TCP_LAST_ACK | TCP_ACTION_FIN,
1800 /* TCP_LAST_ACK */ TCP_LAST_ACK,
1801 /* TCP_LISTEN */ TCP_CLOSE,
1802 /* TCP_CLOSING */ TCP_CLOSING,
1803 };
1804
1805 static int tcp_close_state(struct sock *sk)
1806 {
1807 int next = (int)new_state[sk->sk_state];
1808 int ns = next & TCP_STATE_MASK;
1809
1810 tcp_set_state(sk, ns);
1811
1812 return next & TCP_ACTION_FIN;
1813 }
1814
1815 /*
1816 * Shutdown the sending side of a connection. Much like close except
1817 * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD).
1818 */
1819
1820 void tcp_shutdown(struct sock *sk, int how)
1821 {
1822 /* We need to grab some memory, and put together a FIN,
1823 * and then put it into the queue to be sent.
1824 * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92.
1825 */
1826 if (!(how & SEND_SHUTDOWN))
1827 return;
1828
1829 /* If we've already sent a FIN, or it's a closed state, skip this. */
1830 if ((1 << sk->sk_state) &
1831 (TCPF_ESTABLISHED | TCPF_SYN_SENT |
1832 TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) {
1833 /* Clear out any half completed packets. FIN if needed. */
1834 if (tcp_close_state(sk))
1835 tcp_send_fin(sk);
1836 }
1837 }
1838
1839 void tcp_close(struct sock *sk, long timeout)
1840 {
1841 struct sk_buff *skb;
1842 int data_was_unread = 0;
1843 int state;
1844
1845 lock_sock(sk);
1846 sk->sk_shutdown = SHUTDOWN_MASK;
1847
1848 if (sk->sk_state == TCP_LISTEN) {
1849 tcp_set_state(sk, TCP_CLOSE);
1850
1851 /* Special case. */
1852 inet_csk_listen_stop(sk);
1853
1854 goto adjudge_to_death;
1855 }
1856
1857 /* We need to flush the recv. buffs. We do this only on the
1858 * descriptor close, not protocol-sourced closes, because the
1859 * reader process may not have drained the data yet!
1860 */
1861 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
1862 u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq -
1863 tcp_hdr(skb)->fin;
1864 data_was_unread += len;
1865 __kfree_skb(skb);
1866 }
1867
1868 sk_mem_reclaim(sk);
1869
1870 /* As outlined in RFC 2525, section 2.17, we send a RST here because
1871 * data was lost. To witness the awful effects of the old behavior of
1872 * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk
1873 * GET in an FTP client, suspend the process, wait for the client to
1874 * advertise a zero window, then kill -9 the FTP client, wheee...
1875 * Note: timeout is always zero in such a case.
1876 */
1877 if (data_was_unread) {
1878 /* Unread data was tossed, zap the connection. */
1879 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE);
1880 tcp_set_state(sk, TCP_CLOSE);
1881 tcp_send_active_reset(sk, sk->sk_allocation);
1882 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
1883 /* Check zero linger _after_ checking for unread data. */
1884 sk->sk_prot->disconnect(sk, 0);
1885 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONDATA);
1886 } else if (tcp_close_state(sk)) {
1887 /* We FIN if the application ate all the data before
1888 * zapping the connection.
1889 */
1890
1891 /* RED-PEN. Formally speaking, we have broken TCP state
1892 * machine. State transitions:
1893 *
1894 * TCP_ESTABLISHED -> TCP_FIN_WAIT1
1895 * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible)
1896 * TCP_CLOSE_WAIT -> TCP_LAST_ACK
1897 *
1898 * are legal only when FIN has been sent (i.e. in window),
1899 * rather than queued out of window. Purists blame.
1900 *
1901 * F.e. "RFC state" is ESTABLISHED,
1902 * if Linux state is FIN-WAIT-1, but FIN is still not sent.
1903 *
1904 * The visible declinations are that sometimes
1905 * we enter time-wait state, when it is not required really
1906 * (harmless), do not send active resets, when they are
1907 * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when
1908 * they look as CLOSING or LAST_ACK for Linux)
1909 * Probably, I missed some more holelets.
1910 * --ANK
1911 */
1912 tcp_send_fin(sk);
1913 }
1914
1915 sk_stream_wait_close(sk, timeout);
1916
1917 adjudge_to_death:
1918 state = sk->sk_state;
1919 sock_hold(sk);
1920 sock_orphan(sk);
1921
1922 /* It is the last release_sock in its life. It will remove backlog. */
1923 release_sock(sk);
1924
1925
1926 /* Now socket is owned by kernel and we acquire BH lock
1927 to finish close. No need to check for user refs.
1928 */
1929 local_bh_disable();
1930 bh_lock_sock(sk);
1931 WARN_ON(sock_owned_by_user(sk));
1932
1933 percpu_counter_inc(sk->sk_prot->orphan_count);
1934
1935 /* Have we already been destroyed by a softirq or backlog? */
1936 if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE)
1937 goto out;
1938
1939 /* This is a (useful) BSD violating of the RFC. There is a
1940 * problem with TCP as specified in that the other end could
1941 * keep a socket open forever with no application left this end.
1942 * We use a 3 minute timeout (about the same as BSD) then kill
1943 * our end. If they send after that then tough - BUT: long enough
1944 * that we won't make the old 4*rto = almost no time - whoops
1945 * reset mistake.
1946 *
1947 * Nope, it was not mistake. It is really desired behaviour
1948 * f.e. on http servers, when such sockets are useless, but
1949 * consume significant resources. Let's do it with special
1950 * linger2 option. --ANK
1951 */
1952
1953 if (sk->sk_state == TCP_FIN_WAIT2) {
1954 struct tcp_sock *tp = tcp_sk(sk);
1955 if (tp->linger2 < 0) {
1956 tcp_set_state(sk, TCP_CLOSE);
1957 tcp_send_active_reset(sk, GFP_ATOMIC);
1958 NET_INC_STATS_BH(sock_net(sk),
1959 LINUX_MIB_TCPABORTONLINGER);
1960 } else {
1961 const int tmo = tcp_fin_time(sk);
1962
1963 if (tmo > TCP_TIMEWAIT_LEN) {
1964 inet_csk_reset_keepalive_timer(sk,
1965 tmo - TCP_TIMEWAIT_LEN);
1966 } else {
1967 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
1968 goto out;
1969 }
1970 }
1971 }
1972 if (sk->sk_state != TCP_CLOSE) {
1973 int orphan_count = percpu_counter_read_positive(
1974 sk->sk_prot->orphan_count);
1975
1976 sk_mem_reclaim(sk);
1977 if (tcp_too_many_orphans(sk, orphan_count)) {
1978 if (net_ratelimit())
1979 printk(KERN_INFO "TCP: too many of orphaned "
1980 "sockets\n");
1981 tcp_set_state(sk, TCP_CLOSE);
1982 tcp_send_active_reset(sk, GFP_ATOMIC);
1983 NET_INC_STATS_BH(sock_net(sk),
1984 LINUX_MIB_TCPABORTONMEMORY);
1985 }
1986 }
1987
1988 if (sk->sk_state == TCP_CLOSE)
1989 inet_csk_destroy_sock(sk);
1990 /* Otherwise, socket is reprieved until protocol close. */
1991
1992 out:
1993 bh_unlock_sock(sk);
1994 local_bh_enable();
1995 sock_put(sk);
1996 }
1997
1998 /* These states need RST on ABORT according to RFC793 */
1999
2000 static inline int tcp_need_reset(int state)
2001 {
2002 return (1 << state) &
2003 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 |
2004 TCPF_FIN_WAIT2 | TCPF_SYN_RECV);
2005 }
2006
2007 int tcp_disconnect(struct sock *sk, int flags)
2008 {
2009 struct inet_sock *inet = inet_sk(sk);
2010 struct inet_connection_sock *icsk = inet_csk(sk);
2011 struct tcp_sock *tp = tcp_sk(sk);
2012 int err = 0;
2013 int old_state = sk->sk_state;
2014
2015 if (old_state != TCP_CLOSE)
2016 tcp_set_state(sk, TCP_CLOSE);
2017
2018 /* ABORT function of RFC793 */
2019 if (old_state == TCP_LISTEN) {
2020 inet_csk_listen_stop(sk);
2021 } else if (tcp_need_reset(old_state) ||
2022 (tp->snd_nxt != tp->write_seq &&
2023 (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) {
2024 /* The last check adjusts for discrepancy of Linux wrt. RFC
2025 * states
2026 */
2027 tcp_send_active_reset(sk, gfp_any());
2028 sk->sk_err = ECONNRESET;
2029 } else if (old_state == TCP_SYN_SENT)
2030 sk->sk_err = ECONNRESET;
2031
2032 tcp_clear_xmit_timers(sk);
2033 __skb_queue_purge(&sk->sk_receive_queue);
2034 tcp_write_queue_purge(sk);
2035 __skb_queue_purge(&tp->out_of_order_queue);
2036 #ifdef CONFIG_NET_DMA
2037 __skb_queue_purge(&sk->sk_async_wait_queue);
2038 #endif
2039
2040 inet->dport = 0;
2041
2042 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
2043 inet_reset_saddr(sk);
2044
2045 sk->sk_shutdown = 0;
2046 sock_reset_flag(sk, SOCK_DONE);
2047 tp->srtt = 0;
2048 if ((tp->write_seq += tp->max_window + 2) == 0)
2049 tp->write_seq = 1;
2050 icsk->icsk_backoff = 0;
2051 tp->snd_cwnd = 2;
2052 icsk->icsk_probes_out = 0;
2053 tp->packets_out = 0;
2054 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
2055 tp->snd_cwnd_cnt = 0;
2056 tp->bytes_acked = 0;
2057 tcp_set_ca_state(sk, TCP_CA_Open);
2058 tcp_clear_retrans(tp);
2059 inet_csk_delack_init(sk);
2060 tcp_init_send_head(sk);
2061 memset(&tp->rx_opt, 0, sizeof(tp->rx_opt));
2062 __sk_dst_reset(sk);
2063
2064 WARN_ON(inet->num && !icsk->icsk_bind_hash);
2065
2066 sk->sk_error_report(sk);
2067 return err;
2068 }
2069
2070 /*
2071 * Socket option code for TCP.
2072 */
2073 static int do_tcp_setsockopt(struct sock *sk, int level,
2074 int optname, char __user *optval, unsigned int optlen)
2075 {
2076 struct tcp_sock *tp = tcp_sk(sk);
2077 struct inet_connection_sock *icsk = inet_csk(sk);
2078 int val;
2079 int err = 0;
2080
2081 /* This is a string value all the others are int's */
2082 if (optname == TCP_CONGESTION) {
2083 char name[TCP_CA_NAME_MAX];
2084
2085 if (optlen < 1)
2086 return -EINVAL;
2087
2088 val = strncpy_from_user(name, optval,
2089 min_t(long, TCP_CA_NAME_MAX-1, optlen));
2090 if (val < 0)
2091 return -EFAULT;
2092 name[val] = 0;
2093
2094 lock_sock(sk);
2095 err = tcp_set_congestion_control(sk, name);
2096 release_sock(sk);
2097 return err;
2098 }
2099
2100 if (optlen < sizeof(int))
2101 return -EINVAL;
2102
2103 if (get_user(val, (int __user *)optval))
2104 return -EFAULT;
2105
2106 lock_sock(sk);
2107
2108 switch (optname) {
2109 case TCP_MAXSEG:
2110 /* Values greater than interface MTU won't take effect. However
2111 * at the point when this call is done we typically don't yet
2112 * know which interface is going to be used */
2113 if (val < 8 || val > MAX_TCP_WINDOW) {
2114 err = -EINVAL;
2115 break;
2116 }
2117 tp->rx_opt.user_mss = val;
2118 break;
2119
2120 case TCP_NODELAY:
2121 if (val) {
2122 /* TCP_NODELAY is weaker than TCP_CORK, so that
2123 * this option on corked socket is remembered, but
2124 * it is not activated until cork is cleared.
2125 *
2126 * However, when TCP_NODELAY is set we make
2127 * an explicit push, which overrides even TCP_CORK
2128 * for currently queued segments.
2129 */
2130 tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH;
2131 tcp_push_pending_frames(sk);
2132 } else {
2133 tp->nonagle &= ~TCP_NAGLE_OFF;
2134 }
2135 break;
2136
2137 case TCP_CORK:
2138 /* When set indicates to always queue non-full frames.
2139 * Later the user clears this option and we transmit
2140 * any pending partial frames in the queue. This is
2141 * meant to be used alongside sendfile() to get properly
2142 * filled frames when the user (for example) must write
2143 * out headers with a write() call first and then use
2144 * sendfile to send out the data parts.
2145 *
2146 * TCP_CORK can be set together with TCP_NODELAY and it is
2147 * stronger than TCP_NODELAY.
2148 */
2149 if (val) {
2150 tp->nonagle |= TCP_NAGLE_CORK;
2151 } else {
2152 tp->nonagle &= ~TCP_NAGLE_CORK;
2153 if (tp->nonagle&TCP_NAGLE_OFF)
2154 tp->nonagle |= TCP_NAGLE_PUSH;
2155 tcp_push_pending_frames(sk);
2156 }
2157 break;
2158
2159 case TCP_KEEPIDLE:
2160 if (val < 1 || val > MAX_TCP_KEEPIDLE)
2161 err = -EINVAL;
2162 else {
2163 tp->keepalive_time = val * HZ;
2164 if (sock_flag(sk, SOCK_KEEPOPEN) &&
2165 !((1 << sk->sk_state) &
2166 (TCPF_CLOSE | TCPF_LISTEN))) {
2167 __u32 elapsed = tcp_time_stamp - tp->rcv_tstamp;
2168 if (tp->keepalive_time > elapsed)
2169 elapsed = tp->keepalive_time - elapsed;
2170 else
2171 elapsed = 0;
2172 inet_csk_reset_keepalive_timer(sk, elapsed);
2173 }
2174 }
2175 break;
2176 case TCP_KEEPINTVL:
2177 if (val < 1 || val > MAX_TCP_KEEPINTVL)
2178 err = -EINVAL;
2179 else
2180 tp->keepalive_intvl = val * HZ;
2181 break;
2182 case TCP_KEEPCNT:
2183 if (val < 1 || val > MAX_TCP_KEEPCNT)
2184 err = -EINVAL;
2185 else
2186 tp->keepalive_probes = val;
2187 break;
2188 case TCP_SYNCNT:
2189 if (val < 1 || val > MAX_TCP_SYNCNT)
2190 err = -EINVAL;
2191 else
2192 icsk->icsk_syn_retries = val;
2193 break;
2194
2195 case TCP_LINGER2:
2196 if (val < 0)
2197 tp->linger2 = -1;
2198 else if (val > sysctl_tcp_fin_timeout / HZ)
2199 tp->linger2 = 0;
2200 else
2201 tp->linger2 = val * HZ;
2202 break;
2203
2204 case TCP_DEFER_ACCEPT:
2205 /* Translate value in seconds to number of retransmits */
2206 icsk->icsk_accept_queue.rskq_defer_accept =
2207 secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ,
2208 TCP_RTO_MAX / HZ);
2209 break;
2210
2211 case TCP_WINDOW_CLAMP:
2212 if (!val) {
2213 if (sk->sk_state != TCP_CLOSE) {
2214 err = -EINVAL;
2215 break;
2216 }
2217 tp->window_clamp = 0;
2218 } else
2219 tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ?
2220 SOCK_MIN_RCVBUF / 2 : val;
2221 break;
2222
2223 case TCP_QUICKACK:
2224 if (!val) {
2225 icsk->icsk_ack.pingpong = 1;
2226 } else {
2227 icsk->icsk_ack.pingpong = 0;
2228 if ((1 << sk->sk_state) &
2229 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) &&
2230 inet_csk_ack_scheduled(sk)) {
2231 icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;
2232 tcp_cleanup_rbuf(sk, 1);
2233 if (!(val & 1))
2234 icsk->icsk_ack.pingpong = 1;
2235 }
2236 }
2237 break;
2238
2239 #ifdef CONFIG_TCP_MD5SIG
2240 case TCP_MD5SIG:
2241 /* Read the IP->Key mappings from userspace */
2242 err = tp->af_specific->md5_parse(sk, optval, optlen);
2243 break;
2244 #endif
2245
2246 default:
2247 err = -ENOPROTOOPT;
2248 break;
2249 }
2250
2251 release_sock(sk);
2252 return err;
2253 }
2254
2255 int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval,
2256 unsigned int optlen)
2257 {
2258 struct inet_connection_sock *icsk = inet_csk(sk);
2259
2260 if (level != SOL_TCP)
2261 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
2262 optval, optlen);
2263 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2264 }
2265
2266 #ifdef CONFIG_COMPAT
2267 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
2268 char __user *optval, unsigned int optlen)
2269 {
2270 if (level != SOL_TCP)
2271 return inet_csk_compat_setsockopt(sk, level, optname,
2272 optval, optlen);
2273 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2274 }
2275
2276 EXPORT_SYMBOL(compat_tcp_setsockopt);
2277 #endif
2278
2279 /* Return information about state of tcp endpoint in API format. */
2280 void tcp_get_info(struct sock *sk, struct tcp_info *info)
2281 {
2282 struct tcp_sock *tp = tcp_sk(sk);
2283 const struct inet_connection_sock *icsk = inet_csk(sk);
2284 u32 now = tcp_time_stamp;
2285
2286 memset(info, 0, sizeof(*info));
2287
2288 info->tcpi_state = sk->sk_state;
2289 info->tcpi_ca_state = icsk->icsk_ca_state;
2290 info->tcpi_retransmits = icsk->icsk_retransmits;
2291 info->tcpi_probes = icsk->icsk_probes_out;
2292 info->tcpi_backoff = icsk->icsk_backoff;
2293
2294 if (tp->rx_opt.tstamp_ok)
2295 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
2296 if (tcp_is_sack(tp))
2297 info->tcpi_options |= TCPI_OPT_SACK;
2298 if (tp->rx_opt.wscale_ok) {
2299 info->tcpi_options |= TCPI_OPT_WSCALE;
2300 info->tcpi_snd_wscale = tp->rx_opt.snd_wscale;
2301 info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale;
2302 }
2303
2304 if (tp->ecn_flags&TCP_ECN_OK)
2305 info->tcpi_options |= TCPI_OPT_ECN;
2306
2307 info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto);
2308 info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato);
2309 info->tcpi_snd_mss = tp->mss_cache;
2310 info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss;
2311
2312 if (sk->sk_state == TCP_LISTEN) {
2313 info->tcpi_unacked = sk->sk_ack_backlog;
2314 info->tcpi_sacked = sk->sk_max_ack_backlog;
2315 } else {
2316 info->tcpi_unacked = tp->packets_out;
2317 info->tcpi_sacked = tp->sacked_out;
2318 }
2319 info->tcpi_lost = tp->lost_out;
2320 info->tcpi_retrans = tp->retrans_out;
2321 info->tcpi_fackets = tp->fackets_out;
2322
2323 info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime);
2324 info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime);
2325 info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp);
2326
2327 info->tcpi_pmtu = icsk->icsk_pmtu_cookie;
2328 info->tcpi_rcv_ssthresh = tp->rcv_ssthresh;
2329 info->tcpi_rtt = jiffies_to_usecs(tp->srtt)>>3;
2330 info->tcpi_rttvar = jiffies_to_usecs(tp->mdev)>>2;
2331 info->tcpi_snd_ssthresh = tp->snd_ssthresh;
2332 info->tcpi_snd_cwnd = tp->snd_cwnd;
2333 info->tcpi_advmss = tp->advmss;
2334 info->tcpi_reordering = tp->reordering;
2335
2336 info->tcpi_rcv_rtt = jiffies_to_usecs(tp->rcv_rtt_est.rtt)>>3;
2337 info->tcpi_rcv_space = tp->rcvq_space.space;
2338
2339 info->tcpi_total_retrans = tp->total_retrans;
2340 }
2341
2342 EXPORT_SYMBOL_GPL(tcp_get_info);
2343
2344 static int do_tcp_getsockopt(struct sock *sk, int level,
2345 int optname, char __user *optval, int __user *optlen)
2346 {
2347 struct inet_connection_sock *icsk = inet_csk(sk);
2348 struct tcp_sock *tp = tcp_sk(sk);
2349 int val, len;
2350
2351 if (get_user(len, optlen))
2352 return -EFAULT;
2353
2354 len = min_t(unsigned int, len, sizeof(int));
2355
2356 if (len < 0)
2357 return -EINVAL;
2358
2359 switch (optname) {
2360 case TCP_MAXSEG:
2361 val = tp->mss_cache;
2362 if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
2363 val = tp->rx_opt.user_mss;
2364 break;
2365 case TCP_NODELAY:
2366 val = !!(tp->nonagle&TCP_NAGLE_OFF);
2367 break;
2368 case TCP_CORK:
2369 val = !!(tp->nonagle&TCP_NAGLE_CORK);
2370 break;
2371 case TCP_KEEPIDLE:
2372 val = keepalive_time_when(tp) / HZ;
2373 break;
2374 case TCP_KEEPINTVL:
2375 val = keepalive_intvl_when(tp) / HZ;
2376 break;
2377 case TCP_KEEPCNT:
2378 val = keepalive_probes(tp);
2379 break;
2380 case TCP_SYNCNT:
2381 val = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries;
2382 break;
2383 case TCP_LINGER2:
2384 val = tp->linger2;
2385 if (val >= 0)
2386 val = (val ? : sysctl_tcp_fin_timeout) / HZ;
2387 break;
2388 case TCP_DEFER_ACCEPT:
2389 val = retrans_to_secs(icsk->icsk_accept_queue.rskq_defer_accept,
2390 TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ);
2391 break;
2392 case TCP_WINDOW_CLAMP:
2393 val = tp->window_clamp;
2394 break;
2395 case TCP_INFO: {
2396 struct tcp_info info;
2397
2398 if (get_user(len, optlen))
2399 return -EFAULT;
2400
2401 tcp_get_info(sk, &info);
2402
2403 len = min_t(unsigned int, len, sizeof(info));
2404 if (put_user(len, optlen))
2405 return -EFAULT;
2406 if (copy_to_user(optval, &info, len))
2407 return -EFAULT;
2408 return 0;
2409 }
2410 case TCP_QUICKACK:
2411 val = !icsk->icsk_ack.pingpong;
2412 break;
2413
2414 case TCP_CONGESTION:
2415 if (get_user(len, optlen))
2416 return -EFAULT;
2417 len = min_t(unsigned int, len, TCP_CA_NAME_MAX);
2418 if (put_user(len, optlen))
2419 return -EFAULT;
2420 if (copy_to_user(optval, icsk->icsk_ca_ops->name, len))
2421 return -EFAULT;
2422 return 0;
2423 default:
2424 return -ENOPROTOOPT;
2425 }
2426
2427 if (put_user(len, optlen))
2428 return -EFAULT;
2429 if (copy_to_user(optval, &val, len))
2430 return -EFAULT;
2431 return 0;
2432 }
2433
2434 int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval,
2435 int __user *optlen)
2436 {
2437 struct inet_connection_sock *icsk = inet_csk(sk);
2438
2439 if (level != SOL_TCP)
2440 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
2441 optval, optlen);
2442 return do_tcp_getsockopt(sk, level, optname, optval, optlen);
2443 }
2444
2445 #ifdef CONFIG_COMPAT
2446 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
2447 char __user *optval, int __user *optlen)
2448 {
2449 if (level != SOL_TCP)
2450 return inet_csk_compat_getsockopt(sk, level, optname,
2451 optval, optlen);
2452 return do_tcp_getsockopt(sk, level, optname, optval, optlen);
2453 }
2454
2455 EXPORT_SYMBOL(compat_tcp_getsockopt);
2456 #endif
2457
2458 struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features)
2459 {
2460 struct sk_buff *segs = ERR_PTR(-EINVAL);
2461 struct tcphdr *th;
2462 unsigned thlen;
2463 unsigned int seq;
2464 __be32 delta;
2465 unsigned int oldlen;
2466 unsigned int mss;
2467
2468 if (!pskb_may_pull(skb, sizeof(*th)))
2469 goto out;
2470
2471 th = tcp_hdr(skb);
2472 thlen = th->doff * 4;
2473 if (thlen < sizeof(*th))
2474 goto out;
2475
2476 if (!pskb_may_pull(skb, thlen))
2477 goto out;
2478
2479 oldlen = (u16)~skb->len;
2480 __skb_pull(skb, thlen);
2481
2482 mss = skb_shinfo(skb)->gso_size;
2483 if (unlikely(skb->len <= mss))
2484 goto out;
2485
2486 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2487 /* Packet is from an untrusted source, reset gso_segs. */
2488 int type = skb_shinfo(skb)->gso_type;
2489
2490 if (unlikely(type &
2491 ~(SKB_GSO_TCPV4 |
2492 SKB_GSO_DODGY |
2493 SKB_GSO_TCP_ECN |
2494 SKB_GSO_TCPV6 |
2495 0) ||
2496 !(type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))))
2497 goto out;
2498
2499 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
2500
2501 segs = NULL;
2502 goto out;
2503 }
2504
2505 segs = skb_segment(skb, features);
2506 if (IS_ERR(segs))
2507 goto out;
2508
2509 delta = htonl(oldlen + (thlen + mss));
2510
2511 skb = segs;
2512 th = tcp_hdr(skb);
2513 seq = ntohl(th->seq);
2514
2515 do {
2516 th->fin = th->psh = 0;
2517
2518 th->check = ~csum_fold((__force __wsum)((__force u32)th->check +
2519 (__force u32)delta));
2520 if (skb->ip_summed != CHECKSUM_PARTIAL)
2521 th->check =
2522 csum_fold(csum_partial(skb_transport_header(skb),
2523 thlen, skb->csum));
2524
2525 seq += mss;
2526 skb = skb->next;
2527 th = tcp_hdr(skb);
2528
2529 th->seq = htonl(seq);
2530 th->cwr = 0;
2531 } while (skb->next);
2532
2533 delta = htonl(oldlen + (skb->tail - skb->transport_header) +
2534 skb->data_len);
2535 th->check = ~csum_fold((__force __wsum)((__force u32)th->check +
2536 (__force u32)delta));
2537 if (skb->ip_summed != CHECKSUM_PARTIAL)
2538 th->check = csum_fold(csum_partial(skb_transport_header(skb),
2539 thlen, skb->csum));
2540
2541 out:
2542 return segs;
2543 }
2544 EXPORT_SYMBOL(tcp_tso_segment);
2545
2546 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2547 {
2548 struct sk_buff **pp = NULL;
2549 struct sk_buff *p;
2550 struct tcphdr *th;
2551 struct tcphdr *th2;
2552 unsigned int len;
2553 unsigned int thlen;
2554 unsigned int flags;
2555 unsigned int mss = 1;
2556 unsigned int hlen;
2557 unsigned int off;
2558 int flush = 1;
2559 int i;
2560
2561 off = skb_gro_offset(skb);
2562 hlen = off + sizeof(*th);
2563 th = skb_gro_header_fast(skb, off);
2564 if (skb_gro_header_hard(skb, hlen)) {
2565 th = skb_gro_header_slow(skb, hlen, off);
2566 if (unlikely(!th))
2567 goto out;
2568 }
2569
2570 thlen = th->doff * 4;
2571 if (thlen < sizeof(*th))
2572 goto out;
2573
2574 hlen = off + thlen;
2575 if (skb_gro_header_hard(skb, hlen)) {
2576 th = skb_gro_header_slow(skb, hlen, off);
2577 if (unlikely(!th))
2578 goto out;
2579 }
2580
2581 skb_gro_pull(skb, thlen);
2582
2583 len = skb_gro_len(skb);
2584 flags = tcp_flag_word(th);
2585
2586 for (; (p = *head); head = &p->next) {
2587 if (!NAPI_GRO_CB(p)->same_flow)
2588 continue;
2589
2590 th2 = tcp_hdr(p);
2591
2592 if (*(u32 *)&th->source ^ *(u32 *)&th2->source) {
2593 NAPI_GRO_CB(p)->same_flow = 0;
2594 continue;
2595 }
2596
2597 goto found;
2598 }
2599
2600 goto out_check_final;
2601
2602 found:
2603 flush = NAPI_GRO_CB(p)->flush;
2604 flush |= flags & TCP_FLAG_CWR;
2605 flush |= (flags ^ tcp_flag_word(th2)) &
2606 ~(TCP_FLAG_CWR | TCP_FLAG_FIN | TCP_FLAG_PSH);
2607 flush |= th->ack_seq ^ th2->ack_seq;
2608 for (i = sizeof(*th); i < thlen; i += 4)
2609 flush |= *(u32 *)((u8 *)th + i) ^
2610 *(u32 *)((u8 *)th2 + i);
2611
2612 mss = skb_shinfo(p)->gso_size;
2613
2614 flush |= (len - 1) >= mss;
2615 flush |= (ntohl(th2->seq) + skb_gro_len(p)) ^ ntohl(th->seq);
2616
2617 if (flush || skb_gro_receive(head, skb)) {
2618 mss = 1;
2619 goto out_check_final;
2620 }
2621
2622 p = *head;
2623 th2 = tcp_hdr(p);
2624 tcp_flag_word(th2) |= flags & (TCP_FLAG_FIN | TCP_FLAG_PSH);
2625
2626 out_check_final:
2627 flush = len < mss;
2628 flush |= flags & (TCP_FLAG_URG | TCP_FLAG_PSH | TCP_FLAG_RST |
2629 TCP_FLAG_SYN | TCP_FLAG_FIN);
2630
2631 if (p && (!NAPI_GRO_CB(skb)->same_flow || flush))
2632 pp = head;
2633
2634 out:
2635 NAPI_GRO_CB(skb)->flush |= flush;
2636
2637 return pp;
2638 }
2639 EXPORT_SYMBOL(tcp_gro_receive);
2640
2641 int tcp_gro_complete(struct sk_buff *skb)
2642 {
2643 struct tcphdr *th = tcp_hdr(skb);
2644
2645 skb->csum_start = skb_transport_header(skb) - skb->head;
2646 skb->csum_offset = offsetof(struct tcphdr, check);
2647 skb->ip_summed = CHECKSUM_PARTIAL;
2648
2649 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
2650
2651 if (th->cwr)
2652 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
2653
2654 return 0;
2655 }
2656 EXPORT_SYMBOL(tcp_gro_complete);
2657
2658 #ifdef CONFIG_TCP_MD5SIG
2659 static unsigned long tcp_md5sig_users;
2660 static struct tcp_md5sig_pool **tcp_md5sig_pool;
2661 static DEFINE_SPINLOCK(tcp_md5sig_pool_lock);
2662
2663 static void __tcp_free_md5sig_pool(struct tcp_md5sig_pool **pool)
2664 {
2665 int cpu;
2666 for_each_possible_cpu(cpu) {
2667 struct tcp_md5sig_pool *p = *per_cpu_ptr(pool, cpu);
2668 if (p) {
2669 if (p->md5_desc.tfm)
2670 crypto_free_hash(p->md5_desc.tfm);
2671 kfree(p);
2672 p = NULL;
2673 }
2674 }
2675 free_percpu(pool);
2676 }
2677
2678 void tcp_free_md5sig_pool(void)
2679 {
2680 struct tcp_md5sig_pool **pool = NULL;
2681
2682 spin_lock_bh(&tcp_md5sig_pool_lock);
2683 if (--tcp_md5sig_users == 0) {
2684 pool = tcp_md5sig_pool;
2685 tcp_md5sig_pool = NULL;
2686 }
2687 spin_unlock_bh(&tcp_md5sig_pool_lock);
2688 if (pool)
2689 __tcp_free_md5sig_pool(pool);
2690 }
2691
2692 EXPORT_SYMBOL(tcp_free_md5sig_pool);
2693
2694 static struct tcp_md5sig_pool **__tcp_alloc_md5sig_pool(struct sock *sk)
2695 {
2696 int cpu;
2697 struct tcp_md5sig_pool **pool;
2698
2699 pool = alloc_percpu(struct tcp_md5sig_pool *);
2700 if (!pool)
2701 return NULL;
2702
2703 for_each_possible_cpu(cpu) {
2704 struct tcp_md5sig_pool *p;
2705 struct crypto_hash *hash;
2706
2707 p = kzalloc(sizeof(*p), sk->sk_allocation);
2708 if (!p)
2709 goto out_free;
2710 *per_cpu_ptr(pool, cpu) = p;
2711
2712 hash = crypto_alloc_hash("md5", 0, CRYPTO_ALG_ASYNC);
2713 if (!hash || IS_ERR(hash))
2714 goto out_free;
2715
2716 p->md5_desc.tfm = hash;
2717 }
2718 return pool;
2719 out_free:
2720 __tcp_free_md5sig_pool(pool);
2721 return NULL;
2722 }
2723
2724 struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(struct sock *sk)
2725 {
2726 struct tcp_md5sig_pool **pool;
2727 int alloc = 0;
2728
2729 retry:
2730 spin_lock_bh(&tcp_md5sig_pool_lock);
2731 pool = tcp_md5sig_pool;
2732 if (tcp_md5sig_users++ == 0) {
2733 alloc = 1;
2734 spin_unlock_bh(&tcp_md5sig_pool_lock);
2735 } else if (!pool) {
2736 tcp_md5sig_users--;
2737 spin_unlock_bh(&tcp_md5sig_pool_lock);
2738 cpu_relax();
2739 goto retry;
2740 } else
2741 spin_unlock_bh(&tcp_md5sig_pool_lock);
2742
2743 if (alloc) {
2744 /* we cannot hold spinlock here because this may sleep. */
2745 struct tcp_md5sig_pool **p = __tcp_alloc_md5sig_pool(sk);
2746 spin_lock_bh(&tcp_md5sig_pool_lock);
2747 if (!p) {
2748 tcp_md5sig_users--;
2749 spin_unlock_bh(&tcp_md5sig_pool_lock);
2750 return NULL;
2751 }
2752 pool = tcp_md5sig_pool;
2753 if (pool) {
2754 /* oops, it has already been assigned. */
2755 spin_unlock_bh(&tcp_md5sig_pool_lock);
2756 __tcp_free_md5sig_pool(p);
2757 } else {
2758 tcp_md5sig_pool = pool = p;
2759 spin_unlock_bh(&tcp_md5sig_pool_lock);
2760 }
2761 }
2762 return pool;
2763 }
2764
2765 EXPORT_SYMBOL(tcp_alloc_md5sig_pool);
2766
2767 struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu)
2768 {
2769 struct tcp_md5sig_pool **p;
2770 spin_lock_bh(&tcp_md5sig_pool_lock);
2771 p = tcp_md5sig_pool;
2772 if (p)
2773 tcp_md5sig_users++;
2774 spin_unlock_bh(&tcp_md5sig_pool_lock);
2775 return (p ? *per_cpu_ptr(p, cpu) : NULL);
2776 }
2777
2778 EXPORT_SYMBOL(__tcp_get_md5sig_pool);
2779
2780 void __tcp_put_md5sig_pool(void)
2781 {
2782 tcp_free_md5sig_pool();
2783 }
2784
2785 EXPORT_SYMBOL(__tcp_put_md5sig_pool);
2786
2787 int tcp_md5_hash_header(struct tcp_md5sig_pool *hp,
2788 struct tcphdr *th)
2789 {
2790 struct scatterlist sg;
2791 int err;
2792
2793 __sum16 old_checksum = th->check;
2794 th->check = 0;
2795 /* options aren't included in the hash */
2796 sg_init_one(&sg, th, sizeof(struct tcphdr));
2797 err = crypto_hash_update(&hp->md5_desc, &sg, sizeof(struct tcphdr));
2798 th->check = old_checksum;
2799 return err;
2800 }
2801
2802 EXPORT_SYMBOL(tcp_md5_hash_header);
2803
2804 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp,
2805 struct sk_buff *skb, unsigned header_len)
2806 {
2807 struct scatterlist sg;
2808 const struct tcphdr *tp = tcp_hdr(skb);
2809 struct hash_desc *desc = &hp->md5_desc;
2810 unsigned i;
2811 const unsigned head_data_len = skb_headlen(skb) > header_len ?
2812 skb_headlen(skb) - header_len : 0;
2813 const struct skb_shared_info *shi = skb_shinfo(skb);
2814
2815 sg_init_table(&sg, 1);
2816
2817 sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len);
2818 if (crypto_hash_update(desc, &sg, head_data_len))
2819 return 1;
2820
2821 for (i = 0; i < shi->nr_frags; ++i) {
2822 const struct skb_frag_struct *f = &shi->frags[i];
2823 sg_set_page(&sg, f->page, f->size, f->page_offset);
2824 if (crypto_hash_update(desc, &sg, f->size))
2825 return 1;
2826 }
2827
2828 return 0;
2829 }
2830
2831 EXPORT_SYMBOL(tcp_md5_hash_skb_data);
2832
2833 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, struct tcp_md5sig_key *key)
2834 {
2835 struct scatterlist sg;
2836
2837 sg_init_one(&sg, key->key, key->keylen);
2838 return crypto_hash_update(&hp->md5_desc, &sg, key->keylen);
2839 }
2840
2841 EXPORT_SYMBOL(tcp_md5_hash_key);
2842
2843 #endif
2844
2845 void tcp_done(struct sock *sk)
2846 {
2847 if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
2848 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
2849
2850 tcp_set_state(sk, TCP_CLOSE);
2851 tcp_clear_xmit_timers(sk);
2852
2853 sk->sk_shutdown = SHUTDOWN_MASK;
2854
2855 if (!sock_flag(sk, SOCK_DEAD))
2856 sk->sk_state_change(sk);
2857 else
2858 inet_csk_destroy_sock(sk);
2859 }
2860 EXPORT_SYMBOL_GPL(tcp_done);
2861
2862 extern struct tcp_congestion_ops tcp_reno;
2863
2864 static __initdata unsigned long thash_entries;
2865 static int __init set_thash_entries(char *str)
2866 {
2867 if (!str)
2868 return 0;
2869 thash_entries = simple_strtoul(str, &str, 0);
2870 return 1;
2871 }
2872 __setup("thash_entries=", set_thash_entries);
2873
2874 void __init tcp_init(void)
2875 {
2876 struct sk_buff *skb = NULL;
2877 unsigned long nr_pages, limit;
2878 int order, i, max_share;
2879
2880 BUILD_BUG_ON(sizeof(struct tcp_skb_cb) > sizeof(skb->cb));
2881
2882 percpu_counter_init(&tcp_sockets_allocated, 0);
2883 percpu_counter_init(&tcp_orphan_count, 0);
2884 tcp_hashinfo.bind_bucket_cachep =
2885 kmem_cache_create("tcp_bind_bucket",
2886 sizeof(struct inet_bind_bucket), 0,
2887 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2888
2889 /* Size and allocate the main established and bind bucket
2890 * hash tables.
2891 *
2892 * The methodology is similar to that of the buffer cache.
2893 */
2894 tcp_hashinfo.ehash =
2895 alloc_large_system_hash("TCP established",
2896 sizeof(struct inet_ehash_bucket),
2897 thash_entries,
2898 (totalram_pages >= 128 * 1024) ?
2899 13 : 15,
2900 0,
2901 &tcp_hashinfo.ehash_size,
2902 NULL,
2903 thash_entries ? 0 : 512 * 1024);
2904 tcp_hashinfo.ehash_size = 1 << tcp_hashinfo.ehash_size;
2905 for (i = 0; i < tcp_hashinfo.ehash_size; i++) {
2906 INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i);
2907 INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].twchain, i);
2908 }
2909 if (inet_ehash_locks_alloc(&tcp_hashinfo))
2910 panic("TCP: failed to alloc ehash_locks");
2911 tcp_hashinfo.bhash =
2912 alloc_large_system_hash("TCP bind",
2913 sizeof(struct inet_bind_hashbucket),
2914 tcp_hashinfo.ehash_size,
2915 (totalram_pages >= 128 * 1024) ?
2916 13 : 15,
2917 0,
2918 &tcp_hashinfo.bhash_size,
2919 NULL,
2920 64 * 1024);
2921 tcp_hashinfo.bhash_size = 1 << tcp_hashinfo.bhash_size;
2922 for (i = 0; i < tcp_hashinfo.bhash_size; i++) {
2923 spin_lock_init(&tcp_hashinfo.bhash[i].lock);
2924 INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain);
2925 }
2926
2927 /* Try to be a bit smarter and adjust defaults depending
2928 * on available memory.
2929 */
2930 for (order = 0; ((1 << order) << PAGE_SHIFT) <
2931 (tcp_hashinfo.bhash_size * sizeof(struct inet_bind_hashbucket));
2932 order++)
2933 ;
2934 if (order >= 4) {
2935 tcp_death_row.sysctl_max_tw_buckets = 180000;
2936 sysctl_tcp_max_orphans = 4096 << (order - 4);
2937 sysctl_max_syn_backlog = 1024;
2938 } else if (order < 3) {
2939 tcp_death_row.sysctl_max_tw_buckets >>= (3 - order);
2940 sysctl_tcp_max_orphans >>= (3 - order);
2941 sysctl_max_syn_backlog = 128;
2942 }
2943
2944 /* Set the pressure threshold to be a fraction of global memory that
2945 * is up to 1/2 at 256 MB, decreasing toward zero with the amount of
2946 * memory, with a floor of 128 pages.
2947 */
2948 nr_pages = totalram_pages - totalhigh_pages;
2949 limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
2950 limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
2951 limit = max(limit, 128UL);
2952 sysctl_tcp_mem[0] = limit / 4 * 3;
2953 sysctl_tcp_mem[1] = limit;
2954 sysctl_tcp_mem[2] = sysctl_tcp_mem[0] * 2;
2955
2956 /* Set per-socket limits to no more than 1/128 the pressure threshold */
2957 limit = ((unsigned long)sysctl_tcp_mem[1]) << (PAGE_SHIFT - 7);
2958 max_share = min(4UL*1024*1024, limit);
2959
2960 sysctl_tcp_wmem[0] = SK_MEM_QUANTUM;
2961 sysctl_tcp_wmem[1] = 16*1024;
2962 sysctl_tcp_wmem[2] = max(64*1024, max_share);
2963
2964 sysctl_tcp_rmem[0] = SK_MEM_QUANTUM;
2965 sysctl_tcp_rmem[1] = 87380;
2966 sysctl_tcp_rmem[2] = max(87380, max_share);
2967
2968 printk(KERN_INFO "TCP: Hash tables configured "
2969 "(established %d bind %d)\n",
2970 tcp_hashinfo.ehash_size, tcp_hashinfo.bhash_size);
2971
2972 tcp_register_congestion_control(&tcp_reno);
2973 }
2974
2975 EXPORT_SYMBOL(tcp_close);
2976 EXPORT_SYMBOL(tcp_disconnect);
2977 EXPORT_SYMBOL(tcp_getsockopt);
2978 EXPORT_SYMBOL(tcp_ioctl);
2979 EXPORT_SYMBOL(tcp_poll);
2980 EXPORT_SYMBOL(tcp_read_sock);
2981 EXPORT_SYMBOL(tcp_recvmsg);
2982 EXPORT_SYMBOL(tcp_sendmsg);
2983 EXPORT_SYMBOL(tcp_splice_read);
2984 EXPORT_SYMBOL(tcp_sendpage);
2985 EXPORT_SYMBOL(tcp_setsockopt);
2986 EXPORT_SYMBOL(tcp_shutdown);
Support for the Chinese Samsung S3C2440 based development boards