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.
6 * Implementation of the Transmission Control Protocol(TCP).
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>
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
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
37 * Alan Cox : tcp option processing.
38 * Alan Cox : Reset tweaked (still not 100%) [Had
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
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
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
55 * Alan Cox : Tidied tcp_data to avoid a potential
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
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
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
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
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
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
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
110 * Michael Pall : recv(...,MSG_OOB) never blocks in the
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
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
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
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
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
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
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
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
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
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.
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.
213 * Description of States:
215 * TCP_SYN_SENT sent a connection request, waiting for ack
217 * TCP_SYN_RECV received a connection request, sent ack,
218 * waiting for final ack in three-way handshake.
220 * TCP_ESTABLISHED connection established
222 * TCP_FIN_WAIT1 our side has shutdown, waiting to complete
223 * transmission of remaining buffered data
225 * TCP_FIN_WAIT2 all buffered data sent, waiting for remote
228 * TCP_CLOSING both sides have shutdown but we still have
229 * data we have to finish sending
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)
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)
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
245 * TCP_CLOSE socket is finished
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>
268 #include <net/icmp.h>
270 #include <net/xfrm.h>
272 #include <net/netdma.h>
273 #include <net/sock.h>
275 #include <asm/uaccess.h>
276 #include <asm/ioctls.h>
278 int sysctl_tcp_fin_timeout __read_mostly
= TCP_FIN_TIMEOUT
;
280 struct percpu_counter tcp_orphan_count
;
281 EXPORT_SYMBOL_GPL(tcp_orphan_count
);
283 int sysctl_tcp_mem
[3] __read_mostly
;
284 int sysctl_tcp_wmem
[3] __read_mostly
;
285 int sysctl_tcp_rmem
[3] __read_mostly
;
287 EXPORT_SYMBOL(sysctl_tcp_mem
);
288 EXPORT_SYMBOL(sysctl_tcp_rmem
);
289 EXPORT_SYMBOL(sysctl_tcp_wmem
);
291 atomic_t tcp_memory_allocated
; /* Current allocated memory. */
292 EXPORT_SYMBOL(tcp_memory_allocated
);
295 * Current number of TCP sockets.
297 struct percpu_counter tcp_sockets_allocated
;
298 EXPORT_SYMBOL(tcp_sockets_allocated
);
303 struct tcp_splice_state
{
304 struct pipe_inode_info
*pipe
;
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.
315 int tcp_memory_pressure __read_mostly
;
317 EXPORT_SYMBOL(tcp_memory_pressure
);
319 void tcp_enter_memory_pressure(struct sock
*sk
)
321 if (!tcp_memory_pressure
) {
322 NET_INC_STATS(sock_net(sk
), LINUX_MIB_TCPMEMORYPRESSURES
);
323 tcp_memory_pressure
= 1;
327 EXPORT_SYMBOL(tcp_enter_memory_pressure
);
330 * Wait for a TCP event.
332 * Note that we don't need to lock the socket, as the upper poll layers
333 * take care of normal races (between the test and the event) and we don't
334 * go look at any of the socket buffers directly.
336 unsigned int tcp_poll(struct file
*file
, struct socket
*sock
, poll_table
*wait
)
339 struct sock
*sk
= sock
->sk
;
340 struct tcp_sock
*tp
= tcp_sk(sk
);
342 poll_wait(file
, sk
->sk_sleep
, wait
);
343 if (sk
->sk_state
== TCP_LISTEN
)
344 return inet_csk_listen_poll(sk
);
346 /* Socket is not locked. We are protected from async events
347 * by poll logic and correct handling of state changes
348 * made by other threads is impossible in any case.
356 * POLLHUP is certainly not done right. But poll() doesn't
357 * have a notion of HUP in just one direction, and for a
358 * socket the read side is more interesting.
360 * Some poll() documentation says that POLLHUP is incompatible
361 * with the POLLOUT/POLLWR flags, so somebody should check this
362 * all. But careful, it tends to be safer to return too many
363 * bits than too few, and you can easily break real applications
364 * if you don't tell them that something has hung up!
368 * Check number 1. POLLHUP is _UNMASKABLE_ event (see UNIX98 and
369 * our fs/select.c). It means that after we received EOF,
370 * poll always returns immediately, making impossible poll() on write()
371 * in state CLOSE_WAIT. One solution is evident --- to set POLLHUP
372 * if and only if shutdown has been made in both directions.
373 * Actually, it is interesting to look how Solaris and DUX
374 * solve this dilemma. I would prefer, if POLLHUP were maskable,
375 * then we could set it on SND_SHUTDOWN. BTW examples given
376 * in Stevens' books assume exactly this behaviour, it explains
377 * why POLLHUP is incompatible with POLLOUT. --ANK
379 * NOTE. Check for TCP_CLOSE is added. The goal is to prevent
380 * blocking on fresh not-connected or disconnected socket. --ANK
382 if (sk
->sk_shutdown
== SHUTDOWN_MASK
|| sk
->sk_state
== TCP_CLOSE
)
384 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
385 mask
|= POLLIN
| POLLRDNORM
| POLLRDHUP
;
388 if ((1 << sk
->sk_state
) & ~(TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
389 int target
= sock_rcvlowat(sk
, 0, INT_MAX
);
391 if (tp
->urg_seq
== tp
->copied_seq
&&
392 !sock_flag(sk
, SOCK_URGINLINE
) &&
396 /* Potential race condition. If read of tp below will
397 * escape above sk->sk_state, we can be illegally awaken
398 * in SYN_* states. */
399 if (tp
->rcv_nxt
- tp
->copied_seq
>= target
)
400 mask
|= POLLIN
| POLLRDNORM
;
402 if (!(sk
->sk_shutdown
& SEND_SHUTDOWN
)) {
403 if (sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
)) {
404 mask
|= POLLOUT
| POLLWRNORM
;
405 } else { /* send SIGIO later */
406 set_bit(SOCK_ASYNC_NOSPACE
,
407 &sk
->sk_socket
->flags
);
408 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
410 /* Race breaker. If space is freed after
411 * wspace test but before the flags are set,
412 * IO signal will be lost.
414 if (sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
))
415 mask
|= POLLOUT
| POLLWRNORM
;
419 if (tp
->urg_data
& TCP_URG_VALID
)
425 int tcp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
)
427 struct tcp_sock
*tp
= tcp_sk(sk
);
432 if (sk
->sk_state
== TCP_LISTEN
)
436 if ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
))
438 else if (sock_flag(sk
, SOCK_URGINLINE
) ||
440 before(tp
->urg_seq
, tp
->copied_seq
) ||
441 !before(tp
->urg_seq
, tp
->rcv_nxt
)) {
442 answ
= tp
->rcv_nxt
- tp
->copied_seq
;
444 /* Subtract 1, if FIN is in queue. */
445 if (answ
&& !skb_queue_empty(&sk
->sk_receive_queue
))
447 tcp_hdr((struct sk_buff
*)sk
->sk_receive_queue
.prev
)->fin
;
449 answ
= tp
->urg_seq
- tp
->copied_seq
;
453 answ
= tp
->urg_data
&& tp
->urg_seq
== tp
->copied_seq
;
456 if (sk
->sk_state
== TCP_LISTEN
)
459 if ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
))
462 answ
= tp
->write_seq
- tp
->snd_una
;
468 return put_user(answ
, (int __user
*)arg
);
471 static inline void tcp_mark_push(struct tcp_sock
*tp
, struct sk_buff
*skb
)
473 TCP_SKB_CB(skb
)->flags
|= TCPCB_FLAG_PSH
;
474 tp
->pushed_seq
= tp
->write_seq
;
477 static inline int forced_push(struct tcp_sock
*tp
)
479 return after(tp
->write_seq
, tp
->pushed_seq
+ (tp
->max_window
>> 1));
482 static inline void skb_entail(struct sock
*sk
, struct sk_buff
*skb
)
484 struct tcp_sock
*tp
= tcp_sk(sk
);
485 struct tcp_skb_cb
*tcb
= TCP_SKB_CB(skb
);
488 tcb
->seq
= tcb
->end_seq
= tp
->write_seq
;
489 tcb
->flags
= TCPCB_FLAG_ACK
;
491 skb_header_release(skb
);
492 tcp_add_write_queue_tail(sk
, skb
);
493 sk
->sk_wmem_queued
+= skb
->truesize
;
494 sk_mem_charge(sk
, skb
->truesize
);
495 if (tp
->nonagle
& TCP_NAGLE_PUSH
)
496 tp
->nonagle
&= ~TCP_NAGLE_PUSH
;
499 static inline void tcp_mark_urg(struct tcp_sock
*tp
, int flags
,
503 tp
->snd_up
= tp
->write_seq
;
506 static inline void tcp_push(struct sock
*sk
, int flags
, int mss_now
,
509 struct tcp_sock
*tp
= tcp_sk(sk
);
511 if (tcp_send_head(sk
)) {
512 struct sk_buff
*skb
= tcp_write_queue_tail(sk
);
513 if (!(flags
& MSG_MORE
) || forced_push(tp
))
514 tcp_mark_push(tp
, skb
);
515 tcp_mark_urg(tp
, flags
, skb
);
516 __tcp_push_pending_frames(sk
, mss_now
,
517 (flags
& MSG_MORE
) ? TCP_NAGLE_CORK
: nonagle
);
521 static int tcp_splice_data_recv(read_descriptor_t
*rd_desc
, struct sk_buff
*skb
,
522 unsigned int offset
, size_t len
)
524 struct tcp_splice_state
*tss
= rd_desc
->arg
.data
;
527 ret
= skb_splice_bits(skb
, offset
, tss
->pipe
, min(rd_desc
->count
, len
),
530 rd_desc
->count
-= ret
;
534 static int __tcp_splice_read(struct sock
*sk
, struct tcp_splice_state
*tss
)
536 /* Store TCP splice context information in read_descriptor_t. */
537 read_descriptor_t rd_desc
= {
542 return tcp_read_sock(sk
, &rd_desc
, tcp_splice_data_recv
);
546 * tcp_splice_read - splice data from TCP socket to a pipe
547 * @sock: socket to splice from
548 * @ppos: position (not valid)
549 * @pipe: pipe to splice to
550 * @len: number of bytes to splice
551 * @flags: splice modifier flags
554 * Will read pages from given socket and fill them into a pipe.
557 ssize_t
tcp_splice_read(struct socket
*sock
, loff_t
*ppos
,
558 struct pipe_inode_info
*pipe
, size_t len
,
561 struct sock
*sk
= sock
->sk
;
562 struct tcp_splice_state tss
= {
572 * We can't seek on a socket input
581 timeo
= sock_rcvtimeo(sk
, flags
& SPLICE_F_NONBLOCK
);
583 ret
= __tcp_splice_read(sk
, &tss
);
589 if (sock_flag(sk
, SOCK_DONE
))
592 ret
= sock_error(sk
);
595 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
597 if (sk
->sk_state
== TCP_CLOSE
) {
599 * This occurs when user tries to read
600 * from never connected socket.
602 if (!sock_flag(sk
, SOCK_DONE
))
610 sk_wait_data(sk
, &timeo
);
611 if (signal_pending(current
)) {
612 ret
= sock_intr_errno(timeo
);
625 if (sk
->sk_err
|| sk
->sk_state
== TCP_CLOSE
||
626 (sk
->sk_shutdown
& RCV_SHUTDOWN
) ||
627 signal_pending(current
))
639 struct sk_buff
*sk_stream_alloc_skb(struct sock
*sk
, int size
, gfp_t gfp
)
643 /* The TCP header must be at least 32-bit aligned. */
644 size
= ALIGN(size
, 4);
646 skb
= alloc_skb_fclone(size
+ sk
->sk_prot
->max_header
, gfp
);
648 if (sk_wmem_schedule(sk
, skb
->truesize
)) {
650 * Make sure that we have exactly size bytes
651 * available to the caller, no more, no less.
653 skb_reserve(skb
, skb_tailroom(skb
) - size
);
658 sk
->sk_prot
->enter_memory_pressure(sk
);
659 sk_stream_moderate_sndbuf(sk
);
664 static unsigned int tcp_xmit_size_goal(struct sock
*sk
, u32 mss_now
,
667 struct tcp_sock
*tp
= tcp_sk(sk
);
668 u32 xmit_size_goal
, old_size_goal
;
670 xmit_size_goal
= mss_now
;
672 if (large_allowed
&& sk_can_gso(sk
)) {
673 xmit_size_goal
= ((sk
->sk_gso_max_size
- 1) -
674 inet_csk(sk
)->icsk_af_ops
->net_header_len
-
675 inet_csk(sk
)->icsk_ext_hdr_len
-
678 xmit_size_goal
= tcp_bound_to_half_wnd(tp
, xmit_size_goal
);
680 /* We try hard to avoid divides here */
681 old_size_goal
= tp
->xmit_size_goal_segs
* mss_now
;
683 if (likely(old_size_goal
<= xmit_size_goal
&&
684 old_size_goal
+ mss_now
> xmit_size_goal
)) {
685 xmit_size_goal
= old_size_goal
;
687 tp
->xmit_size_goal_segs
= xmit_size_goal
/ mss_now
;
688 xmit_size_goal
= tp
->xmit_size_goal_segs
* mss_now
;
692 return max(xmit_size_goal
, mss_now
);
695 static int tcp_send_mss(struct sock
*sk
, int *size_goal
, int flags
)
699 mss_now
= tcp_current_mss(sk
);
700 *size_goal
= tcp_xmit_size_goal(sk
, mss_now
, !(flags
& MSG_OOB
));
705 static ssize_t
do_tcp_sendpages(struct sock
*sk
, struct page
**pages
, int poffset
,
706 size_t psize
, int flags
)
708 struct tcp_sock
*tp
= tcp_sk(sk
);
709 int mss_now
, size_goal
;
712 long timeo
= sock_sndtimeo(sk
, flags
& MSG_DONTWAIT
);
714 /* Wait for a connection to finish. */
715 if ((1 << sk
->sk_state
) & ~(TCPF_ESTABLISHED
| TCPF_CLOSE_WAIT
))
716 if ((err
= sk_stream_wait_connect(sk
, &timeo
)) != 0)
719 clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
721 mss_now
= tcp_send_mss(sk
, &size_goal
, flags
);
725 if (sk
->sk_err
|| (sk
->sk_shutdown
& SEND_SHUTDOWN
))
729 struct sk_buff
*skb
= tcp_write_queue_tail(sk
);
730 struct page
*page
= pages
[poffset
/ PAGE_SIZE
];
731 int copy
, i
, can_coalesce
;
732 int offset
= poffset
% PAGE_SIZE
;
733 int size
= min_t(size_t, psize
, PAGE_SIZE
- offset
);
735 if (!tcp_send_head(sk
) || (copy
= size_goal
- skb
->len
) <= 0) {
737 if (!sk_stream_memory_free(sk
))
738 goto wait_for_sndbuf
;
740 skb
= sk_stream_alloc_skb(sk
, 0, sk
->sk_allocation
);
742 goto wait_for_memory
;
751 i
= skb_shinfo(skb
)->nr_frags
;
752 can_coalesce
= skb_can_coalesce(skb
, i
, page
, offset
);
753 if (!can_coalesce
&& i
>= MAX_SKB_FRAGS
) {
754 tcp_mark_push(tp
, skb
);
757 if (!sk_wmem_schedule(sk
, copy
))
758 goto wait_for_memory
;
761 skb_shinfo(skb
)->frags
[i
- 1].size
+= copy
;
764 skb_fill_page_desc(skb
, i
, page
, offset
, copy
);
768 skb
->data_len
+= copy
;
769 skb
->truesize
+= copy
;
770 sk
->sk_wmem_queued
+= copy
;
771 sk_mem_charge(sk
, copy
);
772 skb
->ip_summed
= CHECKSUM_PARTIAL
;
773 tp
->write_seq
+= copy
;
774 TCP_SKB_CB(skb
)->end_seq
+= copy
;
775 skb_shinfo(skb
)->gso_segs
= 0;
778 TCP_SKB_CB(skb
)->flags
&= ~TCPCB_FLAG_PSH
;
782 if (!(psize
-= copy
))
785 if (skb
->len
< size_goal
|| (flags
& MSG_OOB
))
788 if (forced_push(tp
)) {
789 tcp_mark_push(tp
, skb
);
790 __tcp_push_pending_frames(sk
, mss_now
, TCP_NAGLE_PUSH
);
791 } else if (skb
== tcp_send_head(sk
))
792 tcp_push_one(sk
, mss_now
);
796 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
799 tcp_push(sk
, flags
& ~MSG_MORE
, mss_now
, TCP_NAGLE_PUSH
);
801 if ((err
= sk_stream_wait_memory(sk
, &timeo
)) != 0)
804 mss_now
= tcp_send_mss(sk
, &size_goal
, flags
);
809 tcp_push(sk
, flags
, mss_now
, tp
->nonagle
);
816 return sk_stream_error(sk
, flags
, err
);
819 ssize_t
tcp_sendpage(struct socket
*sock
, struct page
*page
, int offset
,
820 size_t size
, int flags
)
823 struct sock
*sk
= sock
->sk
;
825 if (!(sk
->sk_route_caps
& NETIF_F_SG
) ||
826 !(sk
->sk_route_caps
& NETIF_F_ALL_CSUM
))
827 return sock_no_sendpage(sock
, page
, offset
, size
, flags
);
831 res
= do_tcp_sendpages(sk
, &page
, offset
, size
, flags
);
837 #define TCP_PAGE(sk) (sk->sk_sndmsg_page)
838 #define TCP_OFF(sk) (sk->sk_sndmsg_off)
840 static inline int select_size(struct sock
*sk
)
842 struct tcp_sock
*tp
= tcp_sk(sk
);
843 int tmp
= tp
->mss_cache
;
845 if (sk
->sk_route_caps
& NETIF_F_SG
) {
849 int pgbreak
= SKB_MAX_HEAD(MAX_TCP_HEADER
);
851 if (tmp
>= pgbreak
&&
852 tmp
<= pgbreak
+ (MAX_SKB_FRAGS
- 1) * PAGE_SIZE
)
860 int tcp_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*msg
,
863 struct sock
*sk
= sock
->sk
;
865 struct tcp_sock
*tp
= tcp_sk(sk
);
868 int mss_now
, size_goal
;
875 flags
= msg
->msg_flags
;
876 timeo
= sock_sndtimeo(sk
, flags
& MSG_DONTWAIT
);
878 /* Wait for a connection to finish. */
879 if ((1 << sk
->sk_state
) & ~(TCPF_ESTABLISHED
| TCPF_CLOSE_WAIT
))
880 if ((err
= sk_stream_wait_connect(sk
, &timeo
)) != 0)
883 /* This should be in poll */
884 clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
886 mss_now
= tcp_send_mss(sk
, &size_goal
, flags
);
888 /* Ok commence sending. */
889 iovlen
= msg
->msg_iovlen
;
894 if (sk
->sk_err
|| (sk
->sk_shutdown
& SEND_SHUTDOWN
))
897 while (--iovlen
>= 0) {
898 int seglen
= iov
->iov_len
;
899 unsigned char __user
*from
= iov
->iov_base
;
906 skb
= tcp_write_queue_tail(sk
);
908 if (!tcp_send_head(sk
) ||
909 (copy
= size_goal
- skb
->len
) <= 0) {
912 /* Allocate new segment. If the interface is SG,
913 * allocate skb fitting to single page.
915 if (!sk_stream_memory_free(sk
))
916 goto wait_for_sndbuf
;
918 skb
= sk_stream_alloc_skb(sk
, select_size(sk
),
921 goto wait_for_memory
;
924 * Check whether we can use HW checksum.
926 if (sk
->sk_route_caps
& NETIF_F_ALL_CSUM
)
927 skb
->ip_summed
= CHECKSUM_PARTIAL
;
933 /* Try to append data to the end of skb. */
937 /* Where to copy to? */
938 if (skb_tailroom(skb
) > 0) {
939 /* We have some space in skb head. Superb! */
940 if (copy
> skb_tailroom(skb
))
941 copy
= skb_tailroom(skb
);
942 if ((err
= skb_add_data(skb
, from
, copy
)) != 0)
946 int i
= skb_shinfo(skb
)->nr_frags
;
947 struct page
*page
= TCP_PAGE(sk
);
948 int off
= TCP_OFF(sk
);
950 if (skb_can_coalesce(skb
, i
, page
, off
) &&
952 /* We can extend the last page
955 } else if (i
== MAX_SKB_FRAGS
||
957 !(sk
->sk_route_caps
& NETIF_F_SG
))) {
958 /* Need to add new fragment and cannot
959 * do this because interface is non-SG,
960 * or because all the page slots are
962 tcp_mark_push(tp
, skb
);
965 if (off
== PAGE_SIZE
) {
967 TCP_PAGE(sk
) = page
= NULL
;
973 if (copy
> PAGE_SIZE
- off
)
974 copy
= PAGE_SIZE
- off
;
976 if (!sk_wmem_schedule(sk
, copy
))
977 goto wait_for_memory
;
980 /* Allocate new cache page. */
981 if (!(page
= sk_stream_alloc_page(sk
)))
982 goto wait_for_memory
;
985 /* Time to copy data. We are close to
987 err
= skb_copy_to_page(sk
, from
, skb
, page
,
990 /* If this page was new, give it to the
991 * socket so it does not get leaked.
1000 /* Update the skb. */
1002 skb_shinfo(skb
)->frags
[i
- 1].size
+=
1005 skb_fill_page_desc(skb
, i
, page
, off
, copy
);
1008 } else if (off
+ copy
< PAGE_SIZE
) {
1010 TCP_PAGE(sk
) = page
;
1014 TCP_OFF(sk
) = off
+ copy
;
1018 TCP_SKB_CB(skb
)->flags
&= ~TCPCB_FLAG_PSH
;
1020 tp
->write_seq
+= copy
;
1021 TCP_SKB_CB(skb
)->end_seq
+= copy
;
1022 skb_shinfo(skb
)->gso_segs
= 0;
1026 if ((seglen
-= copy
) == 0 && iovlen
== 0)
1029 if (skb
->len
< size_goal
|| (flags
& MSG_OOB
))
1032 if (forced_push(tp
)) {
1033 tcp_mark_push(tp
, skb
);
1034 __tcp_push_pending_frames(sk
, mss_now
, TCP_NAGLE_PUSH
);
1035 } else if (skb
== tcp_send_head(sk
))
1036 tcp_push_one(sk
, mss_now
);
1040 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1043 tcp_push(sk
, flags
& ~MSG_MORE
, mss_now
, TCP_NAGLE_PUSH
);
1045 if ((err
= sk_stream_wait_memory(sk
, &timeo
)) != 0)
1048 mss_now
= tcp_send_mss(sk
, &size_goal
, flags
);
1054 tcp_push(sk
, flags
, mss_now
, tp
->nonagle
);
1055 TCP_CHECK_TIMER(sk
);
1061 tcp_unlink_write_queue(skb
, sk
);
1062 /* It is the one place in all of TCP, except connection
1063 * reset, where we can be unlinking the send_head.
1065 tcp_check_send_head(sk
, skb
);
1066 sk_wmem_free_skb(sk
, skb
);
1073 err
= sk_stream_error(sk
, flags
, err
);
1074 TCP_CHECK_TIMER(sk
);
1080 * Handle reading urgent data. BSD has very simple semantics for
1081 * this, no blocking and very strange errors 8)
1084 static int tcp_recv_urg(struct sock
*sk
, struct msghdr
*msg
, int len
, int flags
)
1086 struct tcp_sock
*tp
= tcp_sk(sk
);
1088 /* No URG data to read. */
1089 if (sock_flag(sk
, SOCK_URGINLINE
) || !tp
->urg_data
||
1090 tp
->urg_data
== TCP_URG_READ
)
1091 return -EINVAL
; /* Yes this is right ! */
1093 if (sk
->sk_state
== TCP_CLOSE
&& !sock_flag(sk
, SOCK_DONE
))
1096 if (tp
->urg_data
& TCP_URG_VALID
) {
1098 char c
= tp
->urg_data
;
1100 if (!(flags
& MSG_PEEK
))
1101 tp
->urg_data
= TCP_URG_READ
;
1103 /* Read urgent data. */
1104 msg
->msg_flags
|= MSG_OOB
;
1107 if (!(flags
& MSG_TRUNC
))
1108 err
= memcpy_toiovec(msg
->msg_iov
, &c
, 1);
1111 msg
->msg_flags
|= MSG_TRUNC
;
1113 return err
? -EFAULT
: len
;
1116 if (sk
->sk_state
== TCP_CLOSE
|| (sk
->sk_shutdown
& RCV_SHUTDOWN
))
1119 /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and
1120 * the available implementations agree in this case:
1121 * this call should never block, independent of the
1122 * blocking state of the socket.
1123 * Mike <pall@rz.uni-karlsruhe.de>
1128 /* Clean up the receive buffer for full frames taken by the user,
1129 * then send an ACK if necessary. COPIED is the number of bytes
1130 * tcp_recvmsg has given to the user so far, it speeds up the
1131 * calculation of whether or not we must ACK for the sake of
1134 void tcp_cleanup_rbuf(struct sock
*sk
, int copied
)
1136 struct tcp_sock
*tp
= tcp_sk(sk
);
1137 int time_to_ack
= 0;
1140 struct sk_buff
*skb
= skb_peek(&sk
->sk_receive_queue
);
1142 WARN_ON(skb
&& !before(tp
->copied_seq
, TCP_SKB_CB(skb
)->end_seq
));
1145 if (inet_csk_ack_scheduled(sk
)) {
1146 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
1147 /* Delayed ACKs frequently hit locked sockets during bulk
1149 if (icsk
->icsk_ack
.blocked
||
1150 /* Once-per-two-segments ACK was not sent by tcp_input.c */
1151 tp
->rcv_nxt
- tp
->rcv_wup
> icsk
->icsk_ack
.rcv_mss
||
1153 * If this read emptied read buffer, we send ACK, if
1154 * connection is not bidirectional, user drained
1155 * receive buffer and there was a small segment
1159 ((icsk
->icsk_ack
.pending
& ICSK_ACK_PUSHED2
) ||
1160 ((icsk
->icsk_ack
.pending
& ICSK_ACK_PUSHED
) &&
1161 !icsk
->icsk_ack
.pingpong
)) &&
1162 !atomic_read(&sk
->sk_rmem_alloc
)))
1166 /* We send an ACK if we can now advertise a non-zero window
1167 * which has been raised "significantly".
1169 * Even if window raised up to infinity, do not send window open ACK
1170 * in states, where we will not receive more. It is useless.
1172 if (copied
> 0 && !time_to_ack
&& !(sk
->sk_shutdown
& RCV_SHUTDOWN
)) {
1173 __u32 rcv_window_now
= tcp_receive_window(tp
);
1175 /* Optimize, __tcp_select_window() is not cheap. */
1176 if (2*rcv_window_now
<= tp
->window_clamp
) {
1177 __u32 new_window
= __tcp_select_window(sk
);
1179 /* Send ACK now, if this read freed lots of space
1180 * in our buffer. Certainly, new_window is new window.
1181 * We can advertise it now, if it is not less than current one.
1182 * "Lots" means "at least twice" here.
1184 if (new_window
&& new_window
>= 2 * rcv_window_now
)
1192 static void tcp_prequeue_process(struct sock
*sk
)
1194 struct sk_buff
*skb
;
1195 struct tcp_sock
*tp
= tcp_sk(sk
);
1197 NET_INC_STATS_USER(sock_net(sk
), LINUX_MIB_TCPPREQUEUED
);
1199 /* RX process wants to run with disabled BHs, though it is not
1202 while ((skb
= __skb_dequeue(&tp
->ucopy
.prequeue
)) != NULL
)
1203 sk_backlog_rcv(sk
, skb
);
1206 /* Clear memory counter. */
1207 tp
->ucopy
.memory
= 0;
1210 static inline struct sk_buff
*tcp_recv_skb(struct sock
*sk
, u32 seq
, u32
*off
)
1212 struct sk_buff
*skb
;
1215 skb_queue_walk(&sk
->sk_receive_queue
, skb
) {
1216 offset
= seq
- TCP_SKB_CB(skb
)->seq
;
1217 if (tcp_hdr(skb
)->syn
)
1219 if (offset
< skb
->len
|| tcp_hdr(skb
)->fin
) {
1228 * This routine provides an alternative to tcp_recvmsg() for routines
1229 * that would like to handle copying from skbuffs directly in 'sendfile'
1232 * - It is assumed that the socket was locked by the caller.
1233 * - The routine does not block.
1234 * - At present, there is no support for reading OOB data
1235 * or for 'peeking' the socket using this routine
1236 * (although both would be easy to implement).
1238 int tcp_read_sock(struct sock
*sk
, read_descriptor_t
*desc
,
1239 sk_read_actor_t recv_actor
)
1241 struct sk_buff
*skb
;
1242 struct tcp_sock
*tp
= tcp_sk(sk
);
1243 u32 seq
= tp
->copied_seq
;
1247 if (sk
->sk_state
== TCP_LISTEN
)
1249 while ((skb
= tcp_recv_skb(sk
, seq
, &offset
)) != NULL
) {
1250 if (offset
< skb
->len
) {
1254 len
= skb
->len
- offset
;
1255 /* Stop reading if we hit a patch of urgent data */
1257 u32 urg_offset
= tp
->urg_seq
- seq
;
1258 if (urg_offset
< len
)
1263 used
= recv_actor(desc
, skb
, offset
, len
);
1268 } else if (used
<= len
) {
1274 * If recv_actor drops the lock (e.g. TCP splice
1275 * receive) the skb pointer might be invalid when
1276 * getting here: tcp_collapse might have deleted it
1277 * while aggregating skbs from the socket queue.
1279 skb
= tcp_recv_skb(sk
, seq
-1, &offset
);
1280 if (!skb
|| (offset
+1 != skb
->len
))
1283 if (tcp_hdr(skb
)->fin
) {
1284 sk_eat_skb(sk
, skb
, 0);
1288 sk_eat_skb(sk
, skb
, 0);
1292 tp
->copied_seq
= seq
;
1294 tcp_rcv_space_adjust(sk
);
1296 /* Clean up data we have read: This will do ACK frames. */
1298 tcp_cleanup_rbuf(sk
, copied
);
1303 * This routine copies from a sock struct into the user buffer.
1305 * Technical note: in 2.3 we work on _locked_ socket, so that
1306 * tricks with *seq access order and skb->users are not required.
1307 * Probably, code can be easily improved even more.
1310 int tcp_recvmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
1311 size_t len
, int nonblock
, int flags
, int *addr_len
)
1313 struct tcp_sock
*tp
= tcp_sk(sk
);
1319 int target
; /* Read at least this many bytes */
1321 struct task_struct
*user_recv
= NULL
;
1322 int copied_early
= 0;
1323 struct sk_buff
*skb
;
1327 TCP_CHECK_TIMER(sk
);
1330 if (sk
->sk_state
== TCP_LISTEN
)
1333 timeo
= sock_rcvtimeo(sk
, nonblock
);
1335 /* Urgent data needs to be handled specially. */
1336 if (flags
& MSG_OOB
)
1339 seq
= &tp
->copied_seq
;
1340 if (flags
& MSG_PEEK
) {
1341 peek_seq
= tp
->copied_seq
;
1345 target
= sock_rcvlowat(sk
, flags
& MSG_WAITALL
, len
);
1347 #ifdef CONFIG_NET_DMA
1348 tp
->ucopy
.dma_chan
= NULL
;
1350 skb
= skb_peek_tail(&sk
->sk_receive_queue
);
1355 available
= TCP_SKB_CB(skb
)->seq
+ skb
->len
- (*seq
);
1356 if ((available
< target
) &&
1357 (len
> sysctl_tcp_dma_copybreak
) && !(flags
& MSG_PEEK
) &&
1358 !sysctl_tcp_low_latency
&&
1359 dma_find_channel(DMA_MEMCPY
)) {
1360 preempt_enable_no_resched();
1361 tp
->ucopy
.pinned_list
=
1362 dma_pin_iovec_pages(msg
->msg_iov
, len
);
1364 preempt_enable_no_resched();
1372 /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */
1373 if (tp
->urg_data
&& tp
->urg_seq
== *seq
) {
1376 if (signal_pending(current
)) {
1377 copied
= timeo
? sock_intr_errno(timeo
) : -EAGAIN
;
1382 /* Next get a buffer. */
1384 skb
= skb_peek(&sk
->sk_receive_queue
);
1389 /* Now that we have two receive queues this
1392 if (before(*seq
, TCP_SKB_CB(skb
)->seq
)) {
1393 printk(KERN_INFO
"recvmsg bug: copied %X "
1394 "seq %X\n", *seq
, TCP_SKB_CB(skb
)->seq
);
1397 offset
= *seq
- TCP_SKB_CB(skb
)->seq
;
1398 if (tcp_hdr(skb
)->syn
)
1400 if (offset
< skb
->len
)
1402 if (tcp_hdr(skb
)->fin
)
1404 WARN_ON(!(flags
& MSG_PEEK
));
1406 } while (skb
!= (struct sk_buff
*)&sk
->sk_receive_queue
);
1408 /* Well, if we have backlog, try to process it now yet. */
1410 if (copied
>= target
&& !sk
->sk_backlog
.tail
)
1415 sk
->sk_state
== TCP_CLOSE
||
1416 (sk
->sk_shutdown
& RCV_SHUTDOWN
) ||
1418 signal_pending(current
))
1421 if (sock_flag(sk
, SOCK_DONE
))
1425 copied
= sock_error(sk
);
1429 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
1432 if (sk
->sk_state
== TCP_CLOSE
) {
1433 if (!sock_flag(sk
, SOCK_DONE
)) {
1434 /* This occurs when user tries to read
1435 * from never connected socket.
1448 if (signal_pending(current
)) {
1449 copied
= sock_intr_errno(timeo
);
1454 tcp_cleanup_rbuf(sk
, copied
);
1456 if (!sysctl_tcp_low_latency
&& tp
->ucopy
.task
== user_recv
) {
1457 /* Install new reader */
1458 if (!user_recv
&& !(flags
& (MSG_TRUNC
| MSG_PEEK
))) {
1459 user_recv
= current
;
1460 tp
->ucopy
.task
= user_recv
;
1461 tp
->ucopy
.iov
= msg
->msg_iov
;
1464 tp
->ucopy
.len
= len
;
1466 WARN_ON(tp
->copied_seq
!= tp
->rcv_nxt
&&
1467 !(flags
& (MSG_PEEK
| MSG_TRUNC
)));
1469 /* Ugly... If prequeue is not empty, we have to
1470 * process it before releasing socket, otherwise
1471 * order will be broken at second iteration.
1472 * More elegant solution is required!!!
1474 * Look: we have the following (pseudo)queues:
1476 * 1. packets in flight
1481 * Each queue can be processed only if the next ones
1482 * are empty. At this point we have empty receive_queue.
1483 * But prequeue _can_ be not empty after 2nd iteration,
1484 * when we jumped to start of loop because backlog
1485 * processing added something to receive_queue.
1486 * We cannot release_sock(), because backlog contains
1487 * packets arrived _after_ prequeued ones.
1489 * Shortly, algorithm is clear --- to process all
1490 * the queues in order. We could make it more directly,
1491 * requeueing packets from backlog to prequeue, if
1492 * is not empty. It is more elegant, but eats cycles,
1495 if (!skb_queue_empty(&tp
->ucopy
.prequeue
))
1498 /* __ Set realtime policy in scheduler __ */
1501 if (copied
>= target
) {
1502 /* Do not sleep, just process backlog. */
1506 sk_wait_data(sk
, &timeo
);
1508 #ifdef CONFIG_NET_DMA
1509 tp
->ucopy
.wakeup
= 0;
1515 /* __ Restore normal policy in scheduler __ */
1517 if ((chunk
= len
- tp
->ucopy
.len
) != 0) {
1518 NET_ADD_STATS_USER(sock_net(sk
), LINUX_MIB_TCPDIRECTCOPYFROMBACKLOG
, chunk
);
1523 if (tp
->rcv_nxt
== tp
->copied_seq
&&
1524 !skb_queue_empty(&tp
->ucopy
.prequeue
)) {
1526 tcp_prequeue_process(sk
);
1528 if ((chunk
= len
- tp
->ucopy
.len
) != 0) {
1529 NET_ADD_STATS_USER(sock_net(sk
), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE
, chunk
);
1535 if ((flags
& MSG_PEEK
) && peek_seq
!= tp
->copied_seq
) {
1536 if (net_ratelimit())
1537 printk(KERN_DEBUG
"TCP(%s:%d): Application bug, race in MSG_PEEK.\n",
1538 current
->comm
, task_pid_nr(current
));
1539 peek_seq
= tp
->copied_seq
;
1544 /* Ok so how much can we use? */
1545 used
= skb
->len
- offset
;
1549 /* Do we have urgent data here? */
1551 u32 urg_offset
= tp
->urg_seq
- *seq
;
1552 if (urg_offset
< used
) {
1554 if (!sock_flag(sk
, SOCK_URGINLINE
)) {
1566 if (!(flags
& MSG_TRUNC
)) {
1567 #ifdef CONFIG_NET_DMA
1568 if (!tp
->ucopy
.dma_chan
&& tp
->ucopy
.pinned_list
)
1569 tp
->ucopy
.dma_chan
= dma_find_channel(DMA_MEMCPY
);
1571 if (tp
->ucopy
.dma_chan
) {
1572 tp
->ucopy
.dma_cookie
= dma_skb_copy_datagram_iovec(
1573 tp
->ucopy
.dma_chan
, skb
, offset
,
1575 tp
->ucopy
.pinned_list
);
1577 if (tp
->ucopy
.dma_cookie
< 0) {
1579 printk(KERN_ALERT
"dma_cookie < 0\n");
1581 /* Exception. Bailout! */
1586 if ((offset
+ used
) == skb
->len
)
1592 err
= skb_copy_datagram_iovec(skb
, offset
,
1593 msg
->msg_iov
, used
);
1595 /* Exception. Bailout! */
1607 tcp_rcv_space_adjust(sk
);
1610 if (tp
->urg_data
&& after(tp
->copied_seq
, tp
->urg_seq
)) {
1612 tcp_fast_path_check(sk
);
1614 if (used
+ offset
< skb
->len
)
1617 if (tcp_hdr(skb
)->fin
)
1619 if (!(flags
& MSG_PEEK
)) {
1620 sk_eat_skb(sk
, skb
, copied_early
);
1626 /* Process the FIN. */
1628 if (!(flags
& MSG_PEEK
)) {
1629 sk_eat_skb(sk
, skb
, copied_early
);
1636 if (!skb_queue_empty(&tp
->ucopy
.prequeue
)) {
1639 tp
->ucopy
.len
= copied
> 0 ? len
: 0;
1641 tcp_prequeue_process(sk
);
1643 if (copied
> 0 && (chunk
= len
- tp
->ucopy
.len
) != 0) {
1644 NET_ADD_STATS_USER(sock_net(sk
), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE
, chunk
);
1650 tp
->ucopy
.task
= NULL
;
1654 #ifdef CONFIG_NET_DMA
1655 if (tp
->ucopy
.dma_chan
) {
1656 dma_cookie_t done
, used
;
1658 dma_async_memcpy_issue_pending(tp
->ucopy
.dma_chan
);
1660 while (dma_async_memcpy_complete(tp
->ucopy
.dma_chan
,
1661 tp
->ucopy
.dma_cookie
, &done
,
1662 &used
) == DMA_IN_PROGRESS
) {
1663 /* do partial cleanup of sk_async_wait_queue */
1664 while ((skb
= skb_peek(&sk
->sk_async_wait_queue
)) &&
1665 (dma_async_is_complete(skb
->dma_cookie
, done
,
1666 used
) == DMA_SUCCESS
)) {
1667 __skb_dequeue(&sk
->sk_async_wait_queue
);
1672 /* Safe to free early-copied skbs now */
1673 __skb_queue_purge(&sk
->sk_async_wait_queue
);
1674 tp
->ucopy
.dma_chan
= NULL
;
1676 if (tp
->ucopy
.pinned_list
) {
1677 dma_unpin_iovec_pages(tp
->ucopy
.pinned_list
);
1678 tp
->ucopy
.pinned_list
= NULL
;
1682 /* According to UNIX98, msg_name/msg_namelen are ignored
1683 * on connected socket. I was just happy when found this 8) --ANK
1686 /* Clean up data we have read: This will do ACK frames. */
1687 tcp_cleanup_rbuf(sk
, copied
);
1689 TCP_CHECK_TIMER(sk
);
1694 TCP_CHECK_TIMER(sk
);
1699 err
= tcp_recv_urg(sk
, msg
, len
, flags
);
1703 void tcp_set_state(struct sock
*sk
, int state
)
1705 int oldstate
= sk
->sk_state
;
1708 case TCP_ESTABLISHED
:
1709 if (oldstate
!= TCP_ESTABLISHED
)
1710 TCP_INC_STATS(sock_net(sk
), TCP_MIB_CURRESTAB
);
1714 if (oldstate
== TCP_CLOSE_WAIT
|| oldstate
== TCP_ESTABLISHED
)
1715 TCP_INC_STATS(sock_net(sk
), TCP_MIB_ESTABRESETS
);
1717 sk
->sk_prot
->unhash(sk
);
1718 if (inet_csk(sk
)->icsk_bind_hash
&&
1719 !(sk
->sk_userlocks
& SOCK_BINDPORT_LOCK
))
1723 if (oldstate
== TCP_ESTABLISHED
)
1724 TCP_DEC_STATS(sock_net(sk
), TCP_MIB_CURRESTAB
);
1727 /* Change state AFTER socket is unhashed to avoid closed
1728 * socket sitting in hash tables.
1730 sk
->sk_state
= state
;
1733 SOCK_DEBUG(sk
, "TCP sk=%p, State %s -> %s\n", sk
, statename
[oldstate
], statename
[state
]);
1736 EXPORT_SYMBOL_GPL(tcp_set_state
);
1739 * State processing on a close. This implements the state shift for
1740 * sending our FIN frame. Note that we only send a FIN for some
1741 * states. A shutdown() may have already sent the FIN, or we may be
1745 static const unsigned char new_state
[16] = {
1746 /* current state: new state: action: */
1747 /* (Invalid) */ TCP_CLOSE
,
1748 /* TCP_ESTABLISHED */ TCP_FIN_WAIT1
| TCP_ACTION_FIN
,
1749 /* TCP_SYN_SENT */ TCP_CLOSE
,
1750 /* TCP_SYN_RECV */ TCP_FIN_WAIT1
| TCP_ACTION_FIN
,
1751 /* TCP_FIN_WAIT1 */ TCP_FIN_WAIT1
,
1752 /* TCP_FIN_WAIT2 */ TCP_FIN_WAIT2
,
1753 /* TCP_TIME_WAIT */ TCP_CLOSE
,
1754 /* TCP_CLOSE */ TCP_CLOSE
,
1755 /* TCP_CLOSE_WAIT */ TCP_LAST_ACK
| TCP_ACTION_FIN
,
1756 /* TCP_LAST_ACK */ TCP_LAST_ACK
,
1757 /* TCP_LISTEN */ TCP_CLOSE
,
1758 /* TCP_CLOSING */ TCP_CLOSING
,
1761 static int tcp_close_state(struct sock
*sk
)
1763 int next
= (int)new_state
[sk
->sk_state
];
1764 int ns
= next
& TCP_STATE_MASK
;
1766 tcp_set_state(sk
, ns
);
1768 return next
& TCP_ACTION_FIN
;
1772 * Shutdown the sending side of a connection. Much like close except
1773 * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD).
1776 void tcp_shutdown(struct sock
*sk
, int how
)
1778 /* We need to grab some memory, and put together a FIN,
1779 * and then put it into the queue to be sent.
1780 * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92.
1782 if (!(how
& SEND_SHUTDOWN
))
1785 /* If we've already sent a FIN, or it's a closed state, skip this. */
1786 if ((1 << sk
->sk_state
) &
1787 (TCPF_ESTABLISHED
| TCPF_SYN_SENT
|
1788 TCPF_SYN_RECV
| TCPF_CLOSE_WAIT
)) {
1789 /* Clear out any half completed packets. FIN if needed. */
1790 if (tcp_close_state(sk
))
1795 void tcp_close(struct sock
*sk
, long timeout
)
1797 struct sk_buff
*skb
;
1798 int data_was_unread
= 0;
1802 sk
->sk_shutdown
= SHUTDOWN_MASK
;
1804 if (sk
->sk_state
== TCP_LISTEN
) {
1805 tcp_set_state(sk
, TCP_CLOSE
);
1808 inet_csk_listen_stop(sk
);
1810 goto adjudge_to_death
;
1813 /* We need to flush the recv. buffs. We do this only on the
1814 * descriptor close, not protocol-sourced closes, because the
1815 * reader process may not have drained the data yet!
1817 while ((skb
= __skb_dequeue(&sk
->sk_receive_queue
)) != NULL
) {
1818 u32 len
= TCP_SKB_CB(skb
)->end_seq
- TCP_SKB_CB(skb
)->seq
-
1820 data_was_unread
+= len
;
1826 /* As outlined in RFC 2525, section 2.17, we send a RST here because
1827 * data was lost. To witness the awful effects of the old behavior of
1828 * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk
1829 * GET in an FTP client, suspend the process, wait for the client to
1830 * advertise a zero window, then kill -9 the FTP client, wheee...
1831 * Note: timeout is always zero in such a case.
1833 if (data_was_unread
) {
1834 /* Unread data was tossed, zap the connection. */
1835 NET_INC_STATS_USER(sock_net(sk
), LINUX_MIB_TCPABORTONCLOSE
);
1836 tcp_set_state(sk
, TCP_CLOSE
);
1837 tcp_send_active_reset(sk
, GFP_KERNEL
);
1838 } else if (sock_flag(sk
, SOCK_LINGER
) && !sk
->sk_lingertime
) {
1839 /* Check zero linger _after_ checking for unread data. */
1840 sk
->sk_prot
->disconnect(sk
, 0);
1841 NET_INC_STATS_USER(sock_net(sk
), LINUX_MIB_TCPABORTONDATA
);
1842 } else if (tcp_close_state(sk
)) {
1843 /* We FIN if the application ate all the data before
1844 * zapping the connection.
1847 /* RED-PEN. Formally speaking, we have broken TCP state
1848 * machine. State transitions:
1850 * TCP_ESTABLISHED -> TCP_FIN_WAIT1
1851 * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible)
1852 * TCP_CLOSE_WAIT -> TCP_LAST_ACK
1854 * are legal only when FIN has been sent (i.e. in window),
1855 * rather than queued out of window. Purists blame.
1857 * F.e. "RFC state" is ESTABLISHED,
1858 * if Linux state is FIN-WAIT-1, but FIN is still not sent.
1860 * The visible declinations are that sometimes
1861 * we enter time-wait state, when it is not required really
1862 * (harmless), do not send active resets, when they are
1863 * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when
1864 * they look as CLOSING or LAST_ACK for Linux)
1865 * Probably, I missed some more holelets.
1871 sk_stream_wait_close(sk
, timeout
);
1874 state
= sk
->sk_state
;
1878 /* It is the last release_sock in its life. It will remove backlog. */
1882 /* Now socket is owned by kernel and we acquire BH lock
1883 to finish close. No need to check for user refs.
1887 WARN_ON(sock_owned_by_user(sk
));
1889 percpu_counter_inc(sk
->sk_prot
->orphan_count
);
1891 /* Have we already been destroyed by a softirq or backlog? */
1892 if (state
!= TCP_CLOSE
&& sk
->sk_state
== TCP_CLOSE
)
1895 /* This is a (useful) BSD violating of the RFC. There is a
1896 * problem with TCP as specified in that the other end could
1897 * keep a socket open forever with no application left this end.
1898 * We use a 3 minute timeout (about the same as BSD) then kill
1899 * our end. If they send after that then tough - BUT: long enough
1900 * that we won't make the old 4*rto = almost no time - whoops
1903 * Nope, it was not mistake. It is really desired behaviour
1904 * f.e. on http servers, when such sockets are useless, but
1905 * consume significant resources. Let's do it with special
1906 * linger2 option. --ANK
1909 if (sk
->sk_state
== TCP_FIN_WAIT2
) {
1910 struct tcp_sock
*tp
= tcp_sk(sk
);
1911 if (tp
->linger2
< 0) {
1912 tcp_set_state(sk
, TCP_CLOSE
);
1913 tcp_send_active_reset(sk
, GFP_ATOMIC
);
1914 NET_INC_STATS_BH(sock_net(sk
),
1915 LINUX_MIB_TCPABORTONLINGER
);
1917 const int tmo
= tcp_fin_time(sk
);
1919 if (tmo
> TCP_TIMEWAIT_LEN
) {
1920 inet_csk_reset_keepalive_timer(sk
,
1921 tmo
- TCP_TIMEWAIT_LEN
);
1923 tcp_time_wait(sk
, TCP_FIN_WAIT2
, tmo
);
1928 if (sk
->sk_state
!= TCP_CLOSE
) {
1929 int orphan_count
= percpu_counter_read_positive(
1930 sk
->sk_prot
->orphan_count
);
1933 if (tcp_too_many_orphans(sk
, orphan_count
)) {
1934 if (net_ratelimit())
1935 printk(KERN_INFO
"TCP: too many of orphaned "
1937 tcp_set_state(sk
, TCP_CLOSE
);
1938 tcp_send_active_reset(sk
, GFP_ATOMIC
);
1939 NET_INC_STATS_BH(sock_net(sk
),
1940 LINUX_MIB_TCPABORTONMEMORY
);
1944 if (sk
->sk_state
== TCP_CLOSE
)
1945 inet_csk_destroy_sock(sk
);
1946 /* Otherwise, socket is reprieved until protocol close. */
1954 /* These states need RST on ABORT according to RFC793 */
1956 static inline int tcp_need_reset(int state
)
1958 return (1 << state
) &
1959 (TCPF_ESTABLISHED
| TCPF_CLOSE_WAIT
| TCPF_FIN_WAIT1
|
1960 TCPF_FIN_WAIT2
| TCPF_SYN_RECV
);
1963 int tcp_disconnect(struct sock
*sk
, int flags
)
1965 struct inet_sock
*inet
= inet_sk(sk
);
1966 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1967 struct tcp_sock
*tp
= tcp_sk(sk
);
1969 int old_state
= sk
->sk_state
;
1971 if (old_state
!= TCP_CLOSE
)
1972 tcp_set_state(sk
, TCP_CLOSE
);
1974 /* ABORT function of RFC793 */
1975 if (old_state
== TCP_LISTEN
) {
1976 inet_csk_listen_stop(sk
);
1977 } else if (tcp_need_reset(old_state
) ||
1978 (tp
->snd_nxt
!= tp
->write_seq
&&
1979 (1 << old_state
) & (TCPF_CLOSING
| TCPF_LAST_ACK
))) {
1980 /* The last check adjusts for discrepancy of Linux wrt. RFC
1983 tcp_send_active_reset(sk
, gfp_any());
1984 sk
->sk_err
= ECONNRESET
;
1985 } else if (old_state
== TCP_SYN_SENT
)
1986 sk
->sk_err
= ECONNRESET
;
1988 tcp_clear_xmit_timers(sk
);
1989 __skb_queue_purge(&sk
->sk_receive_queue
);
1990 tcp_write_queue_purge(sk
);
1991 __skb_queue_purge(&tp
->out_of_order_queue
);
1992 #ifdef CONFIG_NET_DMA
1993 __skb_queue_purge(&sk
->sk_async_wait_queue
);
1998 if (!(sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
))
1999 inet_reset_saddr(sk
);
2001 sk
->sk_shutdown
= 0;
2002 sock_reset_flag(sk
, SOCK_DONE
);
2004 if ((tp
->write_seq
+= tp
->max_window
+ 2) == 0)
2006 icsk
->icsk_backoff
= 0;
2008 icsk
->icsk_probes_out
= 0;
2009 tp
->packets_out
= 0;
2010 tp
->snd_ssthresh
= 0x7fffffff;
2011 tp
->snd_cwnd_cnt
= 0;
2012 tp
->bytes_acked
= 0;
2013 tcp_set_ca_state(sk
, TCP_CA_Open
);
2014 tcp_clear_retrans(tp
);
2015 inet_csk_delack_init(sk
);
2016 tcp_init_send_head(sk
);
2017 memset(&tp
->rx_opt
, 0, sizeof(tp
->rx_opt
));
2020 WARN_ON(inet
->num
&& !icsk
->icsk_bind_hash
);
2022 sk
->sk_error_report(sk
);
2027 * Socket option code for TCP.
2029 static int do_tcp_setsockopt(struct sock
*sk
, int level
,
2030 int optname
, char __user
*optval
, int optlen
)
2032 struct tcp_sock
*tp
= tcp_sk(sk
);
2033 struct inet_connection_sock
*icsk
= inet_csk(sk
);
2037 /* This is a string value all the others are int's */
2038 if (optname
== TCP_CONGESTION
) {
2039 char name
[TCP_CA_NAME_MAX
];
2044 val
= strncpy_from_user(name
, optval
,
2045 min(TCP_CA_NAME_MAX
-1, optlen
));
2051 err
= tcp_set_congestion_control(sk
, name
);
2056 if (optlen
< sizeof(int))
2059 if (get_user(val
, (int __user
*)optval
))
2066 /* Values greater than interface MTU won't take effect. However
2067 * at the point when this call is done we typically don't yet
2068 * know which interface is going to be used */
2069 if (val
< 8 || val
> MAX_TCP_WINDOW
) {
2073 tp
->rx_opt
.user_mss
= val
;
2078 /* TCP_NODELAY is weaker than TCP_CORK, so that
2079 * this option on corked socket is remembered, but
2080 * it is not activated until cork is cleared.
2082 * However, when TCP_NODELAY is set we make
2083 * an explicit push, which overrides even TCP_CORK
2084 * for currently queued segments.
2086 tp
->nonagle
|= TCP_NAGLE_OFF
|TCP_NAGLE_PUSH
;
2087 tcp_push_pending_frames(sk
);
2089 tp
->nonagle
&= ~TCP_NAGLE_OFF
;
2094 /* When set indicates to always queue non-full frames.
2095 * Later the user clears this option and we transmit
2096 * any pending partial frames in the queue. This is
2097 * meant to be used alongside sendfile() to get properly
2098 * filled frames when the user (for example) must write
2099 * out headers with a write() call first and then use
2100 * sendfile to send out the data parts.
2102 * TCP_CORK can be set together with TCP_NODELAY and it is
2103 * stronger than TCP_NODELAY.
2106 tp
->nonagle
|= TCP_NAGLE_CORK
;
2108 tp
->nonagle
&= ~TCP_NAGLE_CORK
;
2109 if (tp
->nonagle
&TCP_NAGLE_OFF
)
2110 tp
->nonagle
|= TCP_NAGLE_PUSH
;
2111 tcp_push_pending_frames(sk
);
2116 if (val
< 1 || val
> MAX_TCP_KEEPIDLE
)
2119 tp
->keepalive_time
= val
* HZ
;
2120 if (sock_flag(sk
, SOCK_KEEPOPEN
) &&
2121 !((1 << sk
->sk_state
) &
2122 (TCPF_CLOSE
| TCPF_LISTEN
))) {
2123 __u32 elapsed
= tcp_time_stamp
- tp
->rcv_tstamp
;
2124 if (tp
->keepalive_time
> elapsed
)
2125 elapsed
= tp
->keepalive_time
- elapsed
;
2128 inet_csk_reset_keepalive_timer(sk
, elapsed
);
2133 if (val
< 1 || val
> MAX_TCP_KEEPINTVL
)
2136 tp
->keepalive_intvl
= val
* HZ
;
2139 if (val
< 1 || val
> MAX_TCP_KEEPCNT
)
2142 tp
->keepalive_probes
= val
;
2145 if (val
< 1 || val
> MAX_TCP_SYNCNT
)
2148 icsk
->icsk_syn_retries
= val
;
2154 else if (val
> sysctl_tcp_fin_timeout
/ HZ
)
2157 tp
->linger2
= val
* HZ
;
2160 case TCP_DEFER_ACCEPT
:
2161 icsk
->icsk_accept_queue
.rskq_defer_accept
= 0;
2163 /* Translate value in seconds to number of
2165 while (icsk
->icsk_accept_queue
.rskq_defer_accept
< 32 &&
2166 val
> ((TCP_TIMEOUT_INIT
/ HZ
) <<
2167 icsk
->icsk_accept_queue
.rskq_defer_accept
))
2168 icsk
->icsk_accept_queue
.rskq_defer_accept
++;
2169 icsk
->icsk_accept_queue
.rskq_defer_accept
++;
2173 case TCP_WINDOW_CLAMP
:
2175 if (sk
->sk_state
!= TCP_CLOSE
) {
2179 tp
->window_clamp
= 0;
2181 tp
->window_clamp
= val
< SOCK_MIN_RCVBUF
/ 2 ?
2182 SOCK_MIN_RCVBUF
/ 2 : val
;
2187 icsk
->icsk_ack
.pingpong
= 1;
2189 icsk
->icsk_ack
.pingpong
= 0;
2190 if ((1 << sk
->sk_state
) &
2191 (TCPF_ESTABLISHED
| TCPF_CLOSE_WAIT
) &&
2192 inet_csk_ack_scheduled(sk
)) {
2193 icsk
->icsk_ack
.pending
|= ICSK_ACK_PUSHED
;
2194 tcp_cleanup_rbuf(sk
, 1);
2196 icsk
->icsk_ack
.pingpong
= 1;
2201 #ifdef CONFIG_TCP_MD5SIG
2203 /* Read the IP->Key mappings from userspace */
2204 err
= tp
->af_specific
->md5_parse(sk
, optval
, optlen
);
2217 int tcp_setsockopt(struct sock
*sk
, int level
, int optname
, char __user
*optval
,
2220 struct inet_connection_sock
*icsk
= inet_csk(sk
);
2222 if (level
!= SOL_TCP
)
2223 return icsk
->icsk_af_ops
->setsockopt(sk
, level
, optname
,
2225 return do_tcp_setsockopt(sk
, level
, optname
, optval
, optlen
);
2228 #ifdef CONFIG_COMPAT
2229 int compat_tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
2230 char __user
*optval
, int optlen
)
2232 if (level
!= SOL_TCP
)
2233 return inet_csk_compat_setsockopt(sk
, level
, optname
,
2235 return do_tcp_setsockopt(sk
, level
, optname
, optval
, optlen
);
2238 EXPORT_SYMBOL(compat_tcp_setsockopt
);
2241 /* Return information about state of tcp endpoint in API format. */
2242 void tcp_get_info(struct sock
*sk
, struct tcp_info
*info
)
2244 struct tcp_sock
*tp
= tcp_sk(sk
);
2245 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
2246 u32 now
= tcp_time_stamp
;
2248 memset(info
, 0, sizeof(*info
));
2250 info
->tcpi_state
= sk
->sk_state
;
2251 info
->tcpi_ca_state
= icsk
->icsk_ca_state
;
2252 info
->tcpi_retransmits
= icsk
->icsk_retransmits
;
2253 info
->tcpi_probes
= icsk
->icsk_probes_out
;
2254 info
->tcpi_backoff
= icsk
->icsk_backoff
;
2256 if (tp
->rx_opt
.tstamp_ok
)
2257 info
->tcpi_options
|= TCPI_OPT_TIMESTAMPS
;
2258 if (tcp_is_sack(tp
))
2259 info
->tcpi_options
|= TCPI_OPT_SACK
;
2260 if (tp
->rx_opt
.wscale_ok
) {
2261 info
->tcpi_options
|= TCPI_OPT_WSCALE
;
2262 info
->tcpi_snd_wscale
= tp
->rx_opt
.snd_wscale
;
2263 info
->tcpi_rcv_wscale
= tp
->rx_opt
.rcv_wscale
;
2266 if (tp
->ecn_flags
&TCP_ECN_OK
)
2267 info
->tcpi_options
|= TCPI_OPT_ECN
;
2269 info
->tcpi_rto
= jiffies_to_usecs(icsk
->icsk_rto
);
2270 info
->tcpi_ato
= jiffies_to_usecs(icsk
->icsk_ack
.ato
);
2271 info
->tcpi_snd_mss
= tp
->mss_cache
;
2272 info
->tcpi_rcv_mss
= icsk
->icsk_ack
.rcv_mss
;
2274 if (sk
->sk_state
== TCP_LISTEN
) {
2275 info
->tcpi_unacked
= sk
->sk_ack_backlog
;
2276 info
->tcpi_sacked
= sk
->sk_max_ack_backlog
;
2278 info
->tcpi_unacked
= tp
->packets_out
;
2279 info
->tcpi_sacked
= tp
->sacked_out
;
2281 info
->tcpi_lost
= tp
->lost_out
;
2282 info
->tcpi_retrans
= tp
->retrans_out
;
2283 info
->tcpi_fackets
= tp
->fackets_out
;
2285 info
->tcpi_last_data_sent
= jiffies_to_msecs(now
- tp
->lsndtime
);
2286 info
->tcpi_last_data_recv
= jiffies_to_msecs(now
- icsk
->icsk_ack
.lrcvtime
);
2287 info
->tcpi_last_ack_recv
= jiffies_to_msecs(now
- tp
->rcv_tstamp
);
2289 info
->tcpi_pmtu
= icsk
->icsk_pmtu_cookie
;
2290 info
->tcpi_rcv_ssthresh
= tp
->rcv_ssthresh
;
2291 info
->tcpi_rtt
= jiffies_to_usecs(tp
->srtt
)>>3;
2292 info
->tcpi_rttvar
= jiffies_to_usecs(tp
->mdev
)>>2;
2293 info
->tcpi_snd_ssthresh
= tp
->snd_ssthresh
;
2294 info
->tcpi_snd_cwnd
= tp
->snd_cwnd
;
2295 info
->tcpi_advmss
= tp
->advmss
;
2296 info
->tcpi_reordering
= tp
->reordering
;
2298 info
->tcpi_rcv_rtt
= jiffies_to_usecs(tp
->rcv_rtt_est
.rtt
)>>3;
2299 info
->tcpi_rcv_space
= tp
->rcvq_space
.space
;
2301 info
->tcpi_total_retrans
= tp
->total_retrans
;
2304 EXPORT_SYMBOL_GPL(tcp_get_info
);
2306 static int do_tcp_getsockopt(struct sock
*sk
, int level
,
2307 int optname
, char __user
*optval
, int __user
*optlen
)
2309 struct inet_connection_sock
*icsk
= inet_csk(sk
);
2310 struct tcp_sock
*tp
= tcp_sk(sk
);
2313 if (get_user(len
, optlen
))
2316 len
= min_t(unsigned int, len
, sizeof(int));
2323 val
= tp
->mss_cache
;
2324 if (!val
&& ((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_LISTEN
)))
2325 val
= tp
->rx_opt
.user_mss
;
2328 val
= !!(tp
->nonagle
&TCP_NAGLE_OFF
);
2331 val
= !!(tp
->nonagle
&TCP_NAGLE_CORK
);
2334 val
= (tp
->keepalive_time
? : sysctl_tcp_keepalive_time
) / HZ
;
2337 val
= (tp
->keepalive_intvl
? : sysctl_tcp_keepalive_intvl
) / HZ
;
2340 val
= tp
->keepalive_probes
? : sysctl_tcp_keepalive_probes
;
2343 val
= icsk
->icsk_syn_retries
? : sysctl_tcp_syn_retries
;
2348 val
= (val
? : sysctl_tcp_fin_timeout
) / HZ
;
2350 case TCP_DEFER_ACCEPT
:
2351 val
= !icsk
->icsk_accept_queue
.rskq_defer_accept
? 0 :
2352 ((TCP_TIMEOUT_INIT
/ HZ
) << (icsk
->icsk_accept_queue
.rskq_defer_accept
- 1));
2354 case TCP_WINDOW_CLAMP
:
2355 val
= tp
->window_clamp
;
2358 struct tcp_info info
;
2360 if (get_user(len
, optlen
))
2363 tcp_get_info(sk
, &info
);
2365 len
= min_t(unsigned int, len
, sizeof(info
));
2366 if (put_user(len
, optlen
))
2368 if (copy_to_user(optval
, &info
, len
))
2373 val
= !icsk
->icsk_ack
.pingpong
;
2376 case TCP_CONGESTION
:
2377 if (get_user(len
, optlen
))
2379 len
= min_t(unsigned int, len
, TCP_CA_NAME_MAX
);
2380 if (put_user(len
, optlen
))
2382 if (copy_to_user(optval
, icsk
->icsk_ca_ops
->name
, len
))
2386 return -ENOPROTOOPT
;
2389 if (put_user(len
, optlen
))
2391 if (copy_to_user(optval
, &val
, len
))
2396 int tcp_getsockopt(struct sock
*sk
, int level
, int optname
, char __user
*optval
,
2399 struct inet_connection_sock
*icsk
= inet_csk(sk
);
2401 if (level
!= SOL_TCP
)
2402 return icsk
->icsk_af_ops
->getsockopt(sk
, level
, optname
,
2404 return do_tcp_getsockopt(sk
, level
, optname
, optval
, optlen
);
2407 #ifdef CONFIG_COMPAT
2408 int compat_tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
2409 char __user
*optval
, int __user
*optlen
)
2411 if (level
!= SOL_TCP
)
2412 return inet_csk_compat_getsockopt(sk
, level
, optname
,
2414 return do_tcp_getsockopt(sk
, level
, optname
, optval
, optlen
);
2417 EXPORT_SYMBOL(compat_tcp_getsockopt
);
2420 struct sk_buff
*tcp_tso_segment(struct sk_buff
*skb
, int features
)
2422 struct sk_buff
*segs
= ERR_PTR(-EINVAL
);
2427 unsigned int oldlen
;
2430 if (!pskb_may_pull(skb
, sizeof(*th
)))
2434 thlen
= th
->doff
* 4;
2435 if (thlen
< sizeof(*th
))
2438 if (!pskb_may_pull(skb
, thlen
))
2441 oldlen
= (u16
)~skb
->len
;
2442 __skb_pull(skb
, thlen
);
2444 mss
= skb_shinfo(skb
)->gso_size
;
2445 if (unlikely(skb
->len
<= mss
))
2448 if (skb_gso_ok(skb
, features
| NETIF_F_GSO_ROBUST
)) {
2449 /* Packet is from an untrusted source, reset gso_segs. */
2450 int type
= skb_shinfo(skb
)->gso_type
;
2458 !(type
& (SKB_GSO_TCPV4
| SKB_GSO_TCPV6
))))
2461 skb_shinfo(skb
)->gso_segs
= DIV_ROUND_UP(skb
->len
, mss
);
2467 segs
= skb_segment(skb
, features
);
2471 delta
= htonl(oldlen
+ (thlen
+ mss
));
2475 seq
= ntohl(th
->seq
);
2478 th
->fin
= th
->psh
= 0;
2480 th
->check
= ~csum_fold((__force __wsum
)((__force u32
)th
->check
+
2481 (__force u32
)delta
));
2482 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
2484 csum_fold(csum_partial(skb_transport_header(skb
),
2491 th
->seq
= htonl(seq
);
2493 } while (skb
->next
);
2495 delta
= htonl(oldlen
+ (skb
->tail
- skb
->transport_header
) +
2497 th
->check
= ~csum_fold((__force __wsum
)((__force u32
)th
->check
+
2498 (__force u32
)delta
));
2499 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
2500 th
->check
= csum_fold(csum_partial(skb_transport_header(skb
),
2506 EXPORT_SYMBOL(tcp_tso_segment
);
2508 struct sk_buff
**tcp_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2510 struct sk_buff
**pp
= NULL
;
2517 unsigned int mss
= 1;
2521 th
= skb_gro_header(skb
, sizeof(*th
));
2525 thlen
= th
->doff
* 4;
2526 if (thlen
< sizeof(*th
))
2529 th
= skb_gro_header(skb
, thlen
);
2533 skb_gro_pull(skb
, thlen
);
2535 len
= skb_gro_len(skb
);
2536 flags
= tcp_flag_word(th
);
2538 for (; (p
= *head
); head
= &p
->next
) {
2539 if (!NAPI_GRO_CB(p
)->same_flow
)
2544 if ((th
->source
^ th2
->source
) | (th
->dest
^ th2
->dest
)) {
2545 NAPI_GRO_CB(p
)->same_flow
= 0;
2552 goto out_check_final
;
2555 flush
= NAPI_GRO_CB(p
)->flush
;
2556 flush
|= flags
& TCP_FLAG_CWR
;
2557 flush
|= (flags
^ tcp_flag_word(th2
)) &
2558 ~(TCP_FLAG_CWR
| TCP_FLAG_FIN
| TCP_FLAG_PSH
);
2559 flush
|= (th
->ack_seq
^ th2
->ack_seq
) | (th
->window
^ th2
->window
);
2560 for (i
= sizeof(*th
); !flush
&& i
< thlen
; i
+= 4)
2561 flush
|= *(u32
*)((u8
*)th
+ i
) ^
2562 *(u32
*)((u8
*)th2
+ i
);
2564 mss
= skb_shinfo(p
)->gso_size
;
2566 flush
|= (len
> mss
) | !len
;
2567 flush
|= (ntohl(th2
->seq
) + skb_gro_len(p
)) ^ ntohl(th
->seq
);
2569 if (flush
|| skb_gro_receive(head
, skb
)) {
2571 goto out_check_final
;
2576 tcp_flag_word(th2
) |= flags
& (TCP_FLAG_FIN
| TCP_FLAG_PSH
);
2580 flush
|= flags
& (TCP_FLAG_URG
| TCP_FLAG_PSH
| TCP_FLAG_RST
|
2581 TCP_FLAG_SYN
| TCP_FLAG_FIN
);
2583 if (p
&& (!NAPI_GRO_CB(skb
)->same_flow
|| flush
))
2587 NAPI_GRO_CB(skb
)->flush
|= flush
;
2591 EXPORT_SYMBOL(tcp_gro_receive
);
2593 int tcp_gro_complete(struct sk_buff
*skb
)
2595 struct tcphdr
*th
= tcp_hdr(skb
);
2597 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
2598 skb
->csum_offset
= offsetof(struct tcphdr
, check
);
2599 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2601 skb_shinfo(skb
)->gso_segs
= NAPI_GRO_CB(skb
)->count
;
2604 skb_shinfo(skb
)->gso_type
|= SKB_GSO_TCP_ECN
;
2608 EXPORT_SYMBOL(tcp_gro_complete
);
2610 #ifdef CONFIG_TCP_MD5SIG
2611 static unsigned long tcp_md5sig_users
;
2612 static struct tcp_md5sig_pool
**tcp_md5sig_pool
;
2613 static DEFINE_SPINLOCK(tcp_md5sig_pool_lock
);
2615 static void __tcp_free_md5sig_pool(struct tcp_md5sig_pool
**pool
)
2618 for_each_possible_cpu(cpu
) {
2619 struct tcp_md5sig_pool
*p
= *per_cpu_ptr(pool
, cpu
);
2621 if (p
->md5_desc
.tfm
)
2622 crypto_free_hash(p
->md5_desc
.tfm
);
2630 void tcp_free_md5sig_pool(void)
2632 struct tcp_md5sig_pool
**pool
= NULL
;
2634 spin_lock_bh(&tcp_md5sig_pool_lock
);
2635 if (--tcp_md5sig_users
== 0) {
2636 pool
= tcp_md5sig_pool
;
2637 tcp_md5sig_pool
= NULL
;
2639 spin_unlock_bh(&tcp_md5sig_pool_lock
);
2641 __tcp_free_md5sig_pool(pool
);
2644 EXPORT_SYMBOL(tcp_free_md5sig_pool
);
2646 static struct tcp_md5sig_pool
**__tcp_alloc_md5sig_pool(void)
2649 struct tcp_md5sig_pool
**pool
;
2651 pool
= alloc_percpu(struct tcp_md5sig_pool
*);
2655 for_each_possible_cpu(cpu
) {
2656 struct tcp_md5sig_pool
*p
;
2657 struct crypto_hash
*hash
;
2659 p
= kzalloc(sizeof(*p
), GFP_KERNEL
);
2662 *per_cpu_ptr(pool
, cpu
) = p
;
2664 hash
= crypto_alloc_hash("md5", 0, CRYPTO_ALG_ASYNC
);
2665 if (!hash
|| IS_ERR(hash
))
2668 p
->md5_desc
.tfm
= hash
;
2672 __tcp_free_md5sig_pool(pool
);
2676 struct tcp_md5sig_pool
**tcp_alloc_md5sig_pool(void)
2678 struct tcp_md5sig_pool
**pool
;
2682 spin_lock_bh(&tcp_md5sig_pool_lock
);
2683 pool
= tcp_md5sig_pool
;
2684 if (tcp_md5sig_users
++ == 0) {
2686 spin_unlock_bh(&tcp_md5sig_pool_lock
);
2689 spin_unlock_bh(&tcp_md5sig_pool_lock
);
2693 spin_unlock_bh(&tcp_md5sig_pool_lock
);
2696 /* we cannot hold spinlock here because this may sleep. */
2697 struct tcp_md5sig_pool
**p
= __tcp_alloc_md5sig_pool();
2698 spin_lock_bh(&tcp_md5sig_pool_lock
);
2701 spin_unlock_bh(&tcp_md5sig_pool_lock
);
2704 pool
= tcp_md5sig_pool
;
2706 /* oops, it has already been assigned. */
2707 spin_unlock_bh(&tcp_md5sig_pool_lock
);
2708 __tcp_free_md5sig_pool(p
);
2710 tcp_md5sig_pool
= pool
= p
;
2711 spin_unlock_bh(&tcp_md5sig_pool_lock
);
2717 EXPORT_SYMBOL(tcp_alloc_md5sig_pool
);
2719 struct tcp_md5sig_pool
*__tcp_get_md5sig_pool(int cpu
)
2721 struct tcp_md5sig_pool
**p
;
2722 spin_lock_bh(&tcp_md5sig_pool_lock
);
2723 p
= tcp_md5sig_pool
;
2726 spin_unlock_bh(&tcp_md5sig_pool_lock
);
2727 return (p
? *per_cpu_ptr(p
, cpu
) : NULL
);
2730 EXPORT_SYMBOL(__tcp_get_md5sig_pool
);
2732 void __tcp_put_md5sig_pool(void)
2734 tcp_free_md5sig_pool();
2737 EXPORT_SYMBOL(__tcp_put_md5sig_pool
);
2739 int tcp_md5_hash_header(struct tcp_md5sig_pool
*hp
,
2742 struct scatterlist sg
;
2745 __sum16 old_checksum
= th
->check
;
2747 /* options aren't included in the hash */
2748 sg_init_one(&sg
, th
, sizeof(struct tcphdr
));
2749 err
= crypto_hash_update(&hp
->md5_desc
, &sg
, sizeof(struct tcphdr
));
2750 th
->check
= old_checksum
;
2754 EXPORT_SYMBOL(tcp_md5_hash_header
);
2756 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool
*hp
,
2757 struct sk_buff
*skb
, unsigned header_len
)
2759 struct scatterlist sg
;
2760 const struct tcphdr
*tp
= tcp_hdr(skb
);
2761 struct hash_desc
*desc
= &hp
->md5_desc
;
2763 const unsigned head_data_len
= skb_headlen(skb
) > header_len
?
2764 skb_headlen(skb
) - header_len
: 0;
2765 const struct skb_shared_info
*shi
= skb_shinfo(skb
);
2767 sg_init_table(&sg
, 1);
2769 sg_set_buf(&sg
, ((u8
*) tp
) + header_len
, head_data_len
);
2770 if (crypto_hash_update(desc
, &sg
, head_data_len
))
2773 for (i
= 0; i
< shi
->nr_frags
; ++i
) {
2774 const struct skb_frag_struct
*f
= &shi
->frags
[i
];
2775 sg_set_page(&sg
, f
->page
, f
->size
, f
->page_offset
);
2776 if (crypto_hash_update(desc
, &sg
, f
->size
))
2783 EXPORT_SYMBOL(tcp_md5_hash_skb_data
);
2785 int tcp_md5_hash_key(struct tcp_md5sig_pool
*hp
, struct tcp_md5sig_key
*key
)
2787 struct scatterlist sg
;
2789 sg_init_one(&sg
, key
->key
, key
->keylen
);
2790 return crypto_hash_update(&hp
->md5_desc
, &sg
, key
->keylen
);
2793 EXPORT_SYMBOL(tcp_md5_hash_key
);
2797 void tcp_done(struct sock
*sk
)
2799 if (sk
->sk_state
== TCP_SYN_SENT
|| sk
->sk_state
== TCP_SYN_RECV
)
2800 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_ATTEMPTFAILS
);
2802 tcp_set_state(sk
, TCP_CLOSE
);
2803 tcp_clear_xmit_timers(sk
);
2805 sk
->sk_shutdown
= SHUTDOWN_MASK
;
2807 if (!sock_flag(sk
, SOCK_DEAD
))
2808 sk
->sk_state_change(sk
);
2810 inet_csk_destroy_sock(sk
);
2812 EXPORT_SYMBOL_GPL(tcp_done
);
2814 extern struct tcp_congestion_ops tcp_reno
;
2816 static __initdata
unsigned long thash_entries
;
2817 static int __init
set_thash_entries(char *str
)
2821 thash_entries
= simple_strtoul(str
, &str
, 0);
2824 __setup("thash_entries=", set_thash_entries
);
2826 void __init
tcp_init(void)
2828 struct sk_buff
*skb
= NULL
;
2829 unsigned long nr_pages
, limit
;
2830 int order
, i
, max_share
;
2832 BUILD_BUG_ON(sizeof(struct tcp_skb_cb
) > sizeof(skb
->cb
));
2834 percpu_counter_init(&tcp_sockets_allocated
, 0);
2835 percpu_counter_init(&tcp_orphan_count
, 0);
2836 tcp_hashinfo
.bind_bucket_cachep
=
2837 kmem_cache_create("tcp_bind_bucket",
2838 sizeof(struct inet_bind_bucket
), 0,
2839 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
2841 /* Size and allocate the main established and bind bucket
2844 * The methodology is similar to that of the buffer cache.
2846 tcp_hashinfo
.ehash
=
2847 alloc_large_system_hash("TCP established",
2848 sizeof(struct inet_ehash_bucket
),
2850 (num_physpages
>= 128 * 1024) ?
2853 &tcp_hashinfo
.ehash_size
,
2855 thash_entries
? 0 : 512 * 1024);
2856 tcp_hashinfo
.ehash_size
= 1 << tcp_hashinfo
.ehash_size
;
2857 for (i
= 0; i
< tcp_hashinfo
.ehash_size
; i
++) {
2858 INIT_HLIST_NULLS_HEAD(&tcp_hashinfo
.ehash
[i
].chain
, i
);
2859 INIT_HLIST_NULLS_HEAD(&tcp_hashinfo
.ehash
[i
].twchain
, i
);
2861 if (inet_ehash_locks_alloc(&tcp_hashinfo
))
2862 panic("TCP: failed to alloc ehash_locks");
2863 tcp_hashinfo
.bhash
=
2864 alloc_large_system_hash("TCP bind",
2865 sizeof(struct inet_bind_hashbucket
),
2866 tcp_hashinfo
.ehash_size
,
2867 (num_physpages
>= 128 * 1024) ?
2870 &tcp_hashinfo
.bhash_size
,
2873 tcp_hashinfo
.bhash_size
= 1 << tcp_hashinfo
.bhash_size
;
2874 for (i
= 0; i
< tcp_hashinfo
.bhash_size
; i
++) {
2875 spin_lock_init(&tcp_hashinfo
.bhash
[i
].lock
);
2876 INIT_HLIST_HEAD(&tcp_hashinfo
.bhash
[i
].chain
);
2879 /* Try to be a bit smarter and adjust defaults depending
2880 * on available memory.
2882 for (order
= 0; ((1 << order
) << PAGE_SHIFT
) <
2883 (tcp_hashinfo
.bhash_size
* sizeof(struct inet_bind_hashbucket
));
2887 tcp_death_row
.sysctl_max_tw_buckets
= 180000;
2888 sysctl_tcp_max_orphans
= 4096 << (order
- 4);
2889 sysctl_max_syn_backlog
= 1024;
2890 } else if (order
< 3) {
2891 tcp_death_row
.sysctl_max_tw_buckets
>>= (3 - order
);
2892 sysctl_tcp_max_orphans
>>= (3 - order
);
2893 sysctl_max_syn_backlog
= 128;
2896 /* Set the pressure threshold to be a fraction of global memory that
2897 * is up to 1/2 at 256 MB, decreasing toward zero with the amount of
2898 * memory, with a floor of 128 pages.
2900 nr_pages
= totalram_pages
- totalhigh_pages
;
2901 limit
= min(nr_pages
, 1UL<<(28-PAGE_SHIFT
)) >> (20-PAGE_SHIFT
);
2902 limit
= (limit
* (nr_pages
>> (20-PAGE_SHIFT
))) >> (PAGE_SHIFT
-11);
2903 limit
= max(limit
, 128UL);
2904 sysctl_tcp_mem
[0] = limit
/ 4 * 3;
2905 sysctl_tcp_mem
[1] = limit
;
2906 sysctl_tcp_mem
[2] = sysctl_tcp_mem
[0] * 2;
2908 /* Set per-socket limits to no more than 1/128 the pressure threshold */
2909 limit
= ((unsigned long)sysctl_tcp_mem
[1]) << (PAGE_SHIFT
- 7);
2910 max_share
= min(4UL*1024*1024, limit
);
2912 sysctl_tcp_wmem
[0] = SK_MEM_QUANTUM
;
2913 sysctl_tcp_wmem
[1] = 16*1024;
2914 sysctl_tcp_wmem
[2] = max(64*1024, max_share
);
2916 sysctl_tcp_rmem
[0] = SK_MEM_QUANTUM
;
2917 sysctl_tcp_rmem
[1] = 87380;
2918 sysctl_tcp_rmem
[2] = max(87380, max_share
);
2920 printk(KERN_INFO
"TCP: Hash tables configured "
2921 "(established %d bind %d)\n",
2922 tcp_hashinfo
.ehash_size
, tcp_hashinfo
.bhash_size
);
2924 tcp_register_congestion_control(&tcp_reno
);
2927 EXPORT_SYMBOL(tcp_close
);
2928 EXPORT_SYMBOL(tcp_disconnect
);
2929 EXPORT_SYMBOL(tcp_getsockopt
);
2930 EXPORT_SYMBOL(tcp_ioctl
);
2931 EXPORT_SYMBOL(tcp_poll
);
2932 EXPORT_SYMBOL(tcp_read_sock
);
2933 EXPORT_SYMBOL(tcp_recvmsg
);
2934 EXPORT_SYMBOL(tcp_sendmsg
);
2935 EXPORT_SYMBOL(tcp_splice_read
);
2936 EXPORT_SYMBOL(tcp_sendpage
);
2937 EXPORT_SYMBOL(tcp_setsockopt
);
2938 EXPORT_SYMBOL(tcp_shutdown
);