]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - net/ipv4/tcp.c
ipv4: Namespaceify tcp_orphan_retries sysctl knob
[mirror_ubuntu-bionic-kernel.git] / net / ipv4 / tcp.c
CommitLineData
1da177e4
LT
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 *
02c30a84 8 * Authors: Ross Biro
1da177e4
LT
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
afd46503
JP
248#define pr_fmt(fmt) "TCP: " fmt
249
172589cc 250#include <linux/kernel.h>
1da177e4
LT
251#include <linux/module.h>
252#include <linux/types.h>
253#include <linux/fcntl.h>
254#include <linux/poll.h>
6e9250f5 255#include <linux/inet_diag.h>
1da177e4 256#include <linux/init.h>
1da177e4 257#include <linux/fs.h>
9c55e01c 258#include <linux/skbuff.h>
81b23b4a 259#include <linux/scatterlist.h>
9c55e01c
JA
260#include <linux/splice.h>
261#include <linux/net.h>
262#include <linux/socket.h>
1da177e4
LT
263#include <linux/random.h>
264#include <linux/bootmem.h>
57413ebc
MS
265#include <linux/highmem.h>
266#include <linux/swap.h>
b8059ead 267#include <linux/cache.h>
f4c50d99 268#include <linux/err.h>
cfb6eeb4 269#include <linux/crypto.h>
da5c78c8 270#include <linux/time.h>
5a0e3ad6 271#include <linux/slab.h>
1da177e4
LT
272
273#include <net/icmp.h>
cf60af03 274#include <net/inet_common.h>
1da177e4
LT
275#include <net/tcp.h>
276#include <net/xfrm.h>
277#include <net/ip.h>
9c55e01c 278#include <net/sock.h>
1da177e4
LT
279
280#include <asm/uaccess.h>
281#include <asm/ioctls.h>
ff5d7497 282#include <asm/unaligned.h>
076bb0c8 283#include <net/busy_poll.h>
1da177e4 284
ab32ea5d 285int sysctl_tcp_fin_timeout __read_mostly = TCP_FIN_TIMEOUT;
1da177e4 286
95bd09eb
ED
287int sysctl_tcp_min_tso_segs __read_mostly = 2;
288
f54b3111
ED
289int sysctl_tcp_autocorking __read_mostly = 1;
290
dd24c001 291struct percpu_counter tcp_orphan_count;
0a5578cf
ACM
292EXPORT_SYMBOL_GPL(tcp_orphan_count);
293
a4fe34bf 294long sysctl_tcp_mem[3] __read_mostly;
b8059ead
DM
295int sysctl_tcp_wmem[3] __read_mostly;
296int sysctl_tcp_rmem[3] __read_mostly;
1da177e4 297
a4fe34bf 298EXPORT_SYMBOL(sysctl_tcp_mem);
1da177e4
LT
299EXPORT_SYMBOL(sysctl_tcp_rmem);
300EXPORT_SYMBOL(sysctl_tcp_wmem);
301
8d987e5c 302atomic_long_t tcp_memory_allocated; /* Current allocated memory. */
1da177e4 303EXPORT_SYMBOL(tcp_memory_allocated);
1748376b
ED
304
305/*
306 * Current number of TCP sockets.
307 */
308struct percpu_counter tcp_sockets_allocated;
1da177e4
LT
309EXPORT_SYMBOL(tcp_sockets_allocated);
310
9c55e01c
JA
311/*
312 * TCP splice context
313 */
314struct tcp_splice_state {
315 struct pipe_inode_info *pipe;
316 size_t len;
317 unsigned int flags;
318};
319
1da177e4
LT
320/*
321 * Pressure flag: try to collapse.
322 * Technical note: it is used by multiple contexts non atomically.
3ab224be 323 * All the __sk_mem_schedule() is of this nature: accounting
1da177e4
LT
324 * is strict, actions are advisory and have some latency.
325 */
4103f8cd 326int tcp_memory_pressure __read_mostly;
1da177e4
LT
327EXPORT_SYMBOL(tcp_memory_pressure);
328
5c52ba17 329void tcp_enter_memory_pressure(struct sock *sk)
1da177e4
LT
330{
331 if (!tcp_memory_pressure) {
4e673444 332 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES);
1da177e4
LT
333 tcp_memory_pressure = 1;
334 }
335}
1da177e4
LT
336EXPORT_SYMBOL(tcp_enter_memory_pressure);
337
b103cf34
JA
338/* Convert seconds to retransmits based on initial and max timeout */
339static u8 secs_to_retrans(int seconds, int timeout, int rto_max)
340{
341 u8 res = 0;
342
343 if (seconds > 0) {
344 int period = timeout;
345
346 res = 1;
347 while (seconds > period && res < 255) {
348 res++;
349 timeout <<= 1;
350 if (timeout > rto_max)
351 timeout = rto_max;
352 period += timeout;
353 }
354 }
355 return res;
356}
357
358/* Convert retransmits to seconds based on initial and max timeout */
359static int retrans_to_secs(u8 retrans, int timeout, int rto_max)
360{
361 int period = 0;
362
363 if (retrans > 0) {
364 period = timeout;
365 while (--retrans) {
366 timeout <<= 1;
367 if (timeout > rto_max)
368 timeout = rto_max;
369 period += timeout;
370 }
371 }
372 return period;
373}
374
900f65d3
NC
375/* Address-family independent initialization for a tcp_sock.
376 *
377 * NOTE: A lot of things set to zero explicitly by call to
378 * sk_alloc() so need not be done here.
379 */
380void tcp_init_sock(struct sock *sk)
381{
382 struct inet_connection_sock *icsk = inet_csk(sk);
383 struct tcp_sock *tp = tcp_sk(sk);
384
996b175e 385 __skb_queue_head_init(&tp->out_of_order_queue);
900f65d3
NC
386 tcp_init_xmit_timers(sk);
387 tcp_prequeue_init(tp);
46d3ceab 388 INIT_LIST_HEAD(&tp->tsq_node);
900f65d3
NC
389
390 icsk->icsk_rto = TCP_TIMEOUT_INIT;
740b0f18 391 tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
f6722583 392 tp->rtt_min[0].rtt = ~0U;
900f65d3
NC
393
394 /* So many TCP implementations out there (incorrectly) count the
395 * initial SYN frame in their delayed-ACK and congestion control
396 * algorithms that we must have the following bandaid to talk
397 * efficiently to them. -DaveM
398 */
399 tp->snd_cwnd = TCP_INIT_CWND;
400
401 /* See draft-stevens-tcpca-spec-01 for discussion of the
402 * initialization of these values.
403 */
404 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
405 tp->snd_cwnd_clamp = ~0;
406 tp->mss_cache = TCP_MSS_DEFAULT;
d654976c 407 u64_stats_init(&tp->syncp);
900f65d3 408
1043e25f 409 tp->reordering = sock_net(sk)->ipv4.sysctl_tcp_reordering;
eed530b6 410 tcp_enable_early_retrans(tp);
55d8694f 411 tcp_assign_congestion_control(sk);
900f65d3 412
ceaa1fef
AV
413 tp->tsoffset = 0;
414
900f65d3
NC
415 sk->sk_state = TCP_CLOSE;
416
417 sk->sk_write_space = sk_stream_write_space;
418 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
419
420 icsk->icsk_sync_mss = tcp_sync_mss;
421
900f65d3
NC
422 sk->sk_sndbuf = sysctl_tcp_wmem[1];
423 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
424
425 local_bh_disable();
3d596f7b
JW
426 if (mem_cgroup_sockets_enabled)
427 sock_update_memcg(sk);
900f65d3
NC
428 sk_sockets_allocated_inc(sk);
429 local_bh_enable();
430}
431EXPORT_SYMBOL(tcp_init_sock);
432
f066e2b0 433static void tcp_tx_timestamp(struct sock *sk, struct sk_buff *skb)
4ed2d765 434{
f066e2b0
WB
435 if (sk->sk_tsflags) {
436 struct skb_shared_info *shinfo = skb_shinfo(skb);
4ed2d765 437
f066e2b0
WB
438 sock_tx_timestamp(sk, &shinfo->tx_flags);
439 if (shinfo->tx_flags & SKBTX_ANY_TSTAMP)
440 shinfo->tskey = TCP_SKB_CB(skb)->seq + skb->len - 1;
441 }
4ed2d765
WB
442}
443
1da177e4
LT
444/*
445 * Wait for a TCP event.
446 *
447 * Note that we don't need to lock the socket, as the upper poll layers
448 * take care of normal races (between the test and the event) and we don't
449 * go look at any of the socket buffers directly.
450 */
451unsigned int tcp_poll(struct file *file, struct socket *sock, poll_table *wait)
452{
453 unsigned int mask;
454 struct sock *sk = sock->sk;
cf533ea5 455 const struct tcp_sock *tp = tcp_sk(sk);
00fd38d9 456 int state;
1da177e4 457
c3f1dbaf
DM
458 sock_rps_record_flow(sk);
459
aa395145 460 sock_poll_wait(file, sk_sleep(sk), wait);
00fd38d9
ED
461
462 state = sk_state_load(sk);
463 if (state == TCP_LISTEN)
dc40c7bc 464 return inet_csk_listen_poll(sk);
1da177e4
LT
465
466 /* Socket is not locked. We are protected from async events
70efce27
WN
467 * by poll logic and correct handling of state changes
468 * made by other threads is impossible in any case.
1da177e4
LT
469 */
470
471 mask = 0;
1da177e4
LT
472
473 /*
474 * POLLHUP is certainly not done right. But poll() doesn't
475 * have a notion of HUP in just one direction, and for a
476 * socket the read side is more interesting.
477 *
478 * Some poll() documentation says that POLLHUP is incompatible
479 * with the POLLOUT/POLLWR flags, so somebody should check this
480 * all. But careful, it tends to be safer to return too many
481 * bits than too few, and you can easily break real applications
482 * if you don't tell them that something has hung up!
483 *
484 * Check-me.
485 *
486 * Check number 1. POLLHUP is _UNMASKABLE_ event (see UNIX98 and
487 * our fs/select.c). It means that after we received EOF,
488 * poll always returns immediately, making impossible poll() on write()
489 * in state CLOSE_WAIT. One solution is evident --- to set POLLHUP
490 * if and only if shutdown has been made in both directions.
491 * Actually, it is interesting to look how Solaris and DUX
70efce27 492 * solve this dilemma. I would prefer, if POLLHUP were maskable,
1da177e4
LT
493 * then we could set it on SND_SHUTDOWN. BTW examples given
494 * in Stevens' books assume exactly this behaviour, it explains
70efce27 495 * why POLLHUP is incompatible with POLLOUT. --ANK
1da177e4
LT
496 *
497 * NOTE. Check for TCP_CLOSE is added. The goal is to prevent
498 * blocking on fresh not-connected or disconnected socket. --ANK
499 */
00fd38d9 500 if (sk->sk_shutdown == SHUTDOWN_MASK || state == TCP_CLOSE)
1da177e4
LT
501 mask |= POLLHUP;
502 if (sk->sk_shutdown & RCV_SHUTDOWN)
f348d70a 503 mask |= POLLIN | POLLRDNORM | POLLRDHUP;
1da177e4 504
8336886f 505 /* Connected or passive Fast Open socket? */
00fd38d9
ED
506 if (state != TCP_SYN_SENT &&
507 (state != TCP_SYN_RECV || tp->fastopen_rsk)) {
c7004482
DM
508 int target = sock_rcvlowat(sk, 0, INT_MAX);
509
510 if (tp->urg_seq == tp->copied_seq &&
511 !sock_flag(sk, SOCK_URGINLINE) &&
512 tp->urg_data)
b634f875 513 target++;
c7004482 514
c7004482 515 if (tp->rcv_nxt - tp->copied_seq >= target)
1da177e4
LT
516 mask |= POLLIN | POLLRDNORM;
517
518 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
64dc6130 519 if (sk_stream_is_writeable(sk)) {
1da177e4
LT
520 mask |= POLLOUT | POLLWRNORM;
521 } else { /* send SIGIO later */
9cd3e072 522 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1da177e4
LT
523 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
524
525 /* Race breaker. If space is freed after
526 * wspace test but before the flags are set,
3c715127 527 * IO signal will be lost. Memory barrier
528 * pairs with the input side.
1da177e4 529 */
3c715127 530 smp_mb__after_atomic();
64dc6130 531 if (sk_stream_is_writeable(sk))
1da177e4
LT
532 mask |= POLLOUT | POLLWRNORM;
533 }
d84ba638
KM
534 } else
535 mask |= POLLOUT | POLLWRNORM;
1da177e4
LT
536
537 if (tp->urg_data & TCP_URG_VALID)
538 mask |= POLLPRI;
539 }
a4d25803
TM
540 /* This barrier is coupled with smp_wmb() in tcp_reset() */
541 smp_rmb();
4ed2d765 542 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
a4d25803
TM
543 mask |= POLLERR;
544
1da177e4
LT
545 return mask;
546}
4bc2f18b 547EXPORT_SYMBOL(tcp_poll);
1da177e4
LT
548
549int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
550{
551 struct tcp_sock *tp = tcp_sk(sk);
552 int answ;
0e71c55c 553 bool slow;
1da177e4
LT
554
555 switch (cmd) {
556 case SIOCINQ:
557 if (sk->sk_state == TCP_LISTEN)
558 return -EINVAL;
559
0e71c55c 560 slow = lock_sock_fast(sk);
1da177e4
LT
561 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
562 answ = 0;
563 else if (sock_flag(sk, SOCK_URGINLINE) ||
564 !tp->urg_data ||
565 before(tp->urg_seq, tp->copied_seq) ||
566 !before(tp->urg_seq, tp->rcv_nxt)) {
91521944 567
1da177e4
LT
568 answ = tp->rcv_nxt - tp->copied_seq;
569
a3374c42
ED
570 /* Subtract 1, if FIN was received */
571 if (answ && sock_flag(sk, SOCK_DONE))
572 answ--;
1da177e4
LT
573 } else
574 answ = tp->urg_seq - tp->copied_seq;
0e71c55c 575 unlock_sock_fast(sk, slow);
1da177e4
LT
576 break;
577 case SIOCATMARK:
578 answ = tp->urg_data && tp->urg_seq == tp->copied_seq;
579 break;
580 case SIOCOUTQ:
581 if (sk->sk_state == TCP_LISTEN)
582 return -EINVAL;
583
584 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
585 answ = 0;
586 else
587 answ = tp->write_seq - tp->snd_una;
588 break;
2f4e1b39
MS
589 case SIOCOUTQNSD:
590 if (sk->sk_state == TCP_LISTEN)
591 return -EINVAL;
592
593 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
594 answ = 0;
595 else
596 answ = tp->write_seq - tp->snd_nxt;
597 break;
1da177e4
LT
598 default:
599 return -ENOIOCTLCMD;
3ff50b79 600 }
1da177e4
LT
601
602 return put_user(answ, (int __user *)arg);
603}
4bc2f18b 604EXPORT_SYMBOL(tcp_ioctl);
1da177e4 605
1da177e4
LT
606static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb)
607{
4de075e0 608 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
1da177e4
LT
609 tp->pushed_seq = tp->write_seq;
610}
611
a2a385d6 612static inline bool forced_push(const struct tcp_sock *tp)
1da177e4
LT
613{
614 return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1));
615}
616
f4a775d1 617static void skb_entail(struct sock *sk, struct sk_buff *skb)
1da177e4 618{
9e412ba7 619 struct tcp_sock *tp = tcp_sk(sk);
352d4800
ACM
620 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
621
622 skb->csum = 0;
623 tcb->seq = tcb->end_seq = tp->write_seq;
4de075e0 624 tcb->tcp_flags = TCPHDR_ACK;
352d4800 625 tcb->sacked = 0;
f4a775d1 626 __skb_header_release(skb);
fe067e8a 627 tcp_add_write_queue_tail(sk, skb);
3ab224be
HA
628 sk->sk_wmem_queued += skb->truesize;
629 sk_mem_charge(sk, skb->truesize);
89ebd197 630 if (tp->nonagle & TCP_NAGLE_PUSH)
e905a9ed 631 tp->nonagle &= ~TCP_NAGLE_PUSH;
6f021c62
ED
632
633 tcp_slow_start_after_idle_check(sk);
1da177e4
LT
634}
635
afeca340 636static inline void tcp_mark_urg(struct tcp_sock *tp, int flags)
1da177e4 637{
33f5f57e 638 if (flags & MSG_OOB)
1da177e4 639 tp->snd_up = tp->write_seq;
1da177e4
LT
640}
641
f54b3111 642/* If a not yet filled skb is pushed, do not send it if
a181ceb5 643 * we have data packets in Qdisc or NIC queues :
f54b3111
ED
644 * Because TX completion will happen shortly, it gives a chance
645 * to coalesce future sendmsg() payload into this skb, without
646 * need for a timer, and with no latency trade off.
647 * As packets containing data payload have a bigger truesize
a181ceb5
ED
648 * than pure acks (dataless) packets, the last checks prevent
649 * autocorking if we only have an ACK in Qdisc/NIC queues,
650 * or if TX completion was delayed after we processed ACK packet.
f54b3111
ED
651 */
652static bool tcp_should_autocork(struct sock *sk, struct sk_buff *skb,
653 int size_goal)
1da177e4 654{
f54b3111
ED
655 return skb->len < size_goal &&
656 sysctl_tcp_autocorking &&
a181ceb5 657 skb != tcp_write_queue_head(sk) &&
f54b3111
ED
658 atomic_read(&sk->sk_wmem_alloc) > skb->truesize;
659}
660
661static void tcp_push(struct sock *sk, int flags, int mss_now,
662 int nonagle, int size_goal)
663{
664 struct tcp_sock *tp = tcp_sk(sk);
665 struct sk_buff *skb;
afeca340 666
f54b3111
ED
667 if (!tcp_send_head(sk))
668 return;
afeca340 669
f54b3111
ED
670 skb = tcp_write_queue_tail(sk);
671 if (!(flags & MSG_MORE) || forced_push(tp))
672 tcp_mark_push(tp, skb);
673
674 tcp_mark_urg(tp, flags);
675
676 if (tcp_should_autocork(sk, skb, size_goal)) {
677
678 /* avoid atomic op if TSQ_THROTTLED bit is already set */
679 if (!test_bit(TSQ_THROTTLED, &tp->tsq_flags)) {
680 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAUTOCORKING);
681 set_bit(TSQ_THROTTLED, &tp->tsq_flags);
682 }
a181ceb5
ED
683 /* It is possible TX completion already happened
684 * before we set TSQ_THROTTLED.
685 */
686 if (atomic_read(&sk->sk_wmem_alloc) > skb->truesize)
687 return;
1da177e4 688 }
f54b3111
ED
689
690 if (flags & MSG_MORE)
691 nonagle = TCP_NAGLE_CORK;
692
693 __tcp_push_pending_frames(sk, mss_now, nonagle);
1da177e4
LT
694}
695
6ff7751d
AB
696static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
697 unsigned int offset, size_t len)
9c55e01c
JA
698{
699 struct tcp_splice_state *tss = rd_desc->arg.data;
33966dd0 700 int ret;
9c55e01c 701
a60e3cc7
HFS
702 ret = skb_splice_bits(skb, skb->sk, offset, tss->pipe,
703 min(rd_desc->count, len), tss->flags,
704 skb_socket_splice);
33966dd0
WT
705 if (ret > 0)
706 rd_desc->count -= ret;
707 return ret;
9c55e01c
JA
708}
709
710static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss)
711{
712 /* Store TCP splice context information in read_descriptor_t. */
713 read_descriptor_t rd_desc = {
714 .arg.data = tss,
33966dd0 715 .count = tss->len,
9c55e01c
JA
716 };
717
718 return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv);
719}
720
721/**
722 * tcp_splice_read - splice data from TCP socket to a pipe
723 * @sock: socket to splice from
724 * @ppos: position (not valid)
725 * @pipe: pipe to splice to
726 * @len: number of bytes to splice
727 * @flags: splice modifier flags
728 *
729 * Description:
730 * Will read pages from given socket and fill them into a pipe.
731 *
732 **/
733ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos,
734 struct pipe_inode_info *pipe, size_t len,
735 unsigned int flags)
736{
737 struct sock *sk = sock->sk;
738 struct tcp_splice_state tss = {
739 .pipe = pipe,
740 .len = len,
741 .flags = flags,
742 };
743 long timeo;
744 ssize_t spliced;
745 int ret;
746
3a047bf8 747 sock_rps_record_flow(sk);
9c55e01c
JA
748 /*
749 * We can't seek on a socket input
750 */
751 if (unlikely(*ppos))
752 return -ESPIPE;
753
754 ret = spliced = 0;
755
756 lock_sock(sk);
757
42324c62 758 timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK);
9c55e01c
JA
759 while (tss.len) {
760 ret = __tcp_splice_read(sk, &tss);
761 if (ret < 0)
762 break;
763 else if (!ret) {
764 if (spliced)
765 break;
9c55e01c
JA
766 if (sock_flag(sk, SOCK_DONE))
767 break;
768 if (sk->sk_err) {
769 ret = sock_error(sk);
770 break;
771 }
772 if (sk->sk_shutdown & RCV_SHUTDOWN)
773 break;
774 if (sk->sk_state == TCP_CLOSE) {
775 /*
776 * This occurs when user tries to read
777 * from never connected socket.
778 */
779 if (!sock_flag(sk, SOCK_DONE))
780 ret = -ENOTCONN;
781 break;
782 }
783 if (!timeo) {
784 ret = -EAGAIN;
785 break;
786 }
dfbafc99 787 sk_wait_data(sk, &timeo, NULL);
9c55e01c
JA
788 if (signal_pending(current)) {
789 ret = sock_intr_errno(timeo);
790 break;
791 }
792 continue;
793 }
794 tss.len -= ret;
795 spliced += ret;
796
33966dd0
WT
797 if (!timeo)
798 break;
9c55e01c
JA
799 release_sock(sk);
800 lock_sock(sk);
801
802 if (sk->sk_err || sk->sk_state == TCP_CLOSE ||
33966dd0 803 (sk->sk_shutdown & RCV_SHUTDOWN) ||
9c55e01c
JA
804 signal_pending(current))
805 break;
806 }
807
808 release_sock(sk);
809
810 if (spliced)
811 return spliced;
812
813 return ret;
814}
4bc2f18b 815EXPORT_SYMBOL(tcp_splice_read);
9c55e01c 816
eb934478
ED
817struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
818 bool force_schedule)
f561d0f2
PE
819{
820 struct sk_buff *skb;
821
822 /* The TCP header must be at least 32-bit aligned. */
823 size = ALIGN(size, 4);
824
8e4d980a
ED
825 if (unlikely(tcp_under_memory_pressure(sk)))
826 sk_mem_reclaim_partial(sk);
827
f561d0f2 828 skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp);
8e4d980a 829 if (likely(skb)) {
eb934478 830 bool mem_scheduled;
8e4d980a 831
eb934478
ED
832 if (force_schedule) {
833 mem_scheduled = true;
8e4d980a
ED
834 sk_forced_mem_schedule(sk, skb->truesize);
835 } else {
eb934478 836 mem_scheduled = sk_wmem_schedule(sk, skb->truesize);
8e4d980a 837 }
eb934478 838 if (likely(mem_scheduled)) {
a21d4572 839 skb_reserve(skb, sk->sk_prot->max_header);
f561d0f2
PE
840 /*
841 * Make sure that we have exactly size bytes
842 * available to the caller, no more, no less.
843 */
16fad69c 844 skb->reserved_tailroom = skb->end - skb->tail - size;
f561d0f2
PE
845 return skb;
846 }
847 __kfree_skb(skb);
848 } else {
5c52ba17 849 sk->sk_prot->enter_memory_pressure(sk);
f561d0f2
PE
850 sk_stream_moderate_sndbuf(sk);
851 }
852 return NULL;
853}
854
0c54b85f
IJ
855static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now,
856 int large_allowed)
857{
858 struct tcp_sock *tp = tcp_sk(sk);
6c09fa09 859 u32 new_size_goal, size_goal;
605ad7f1
ED
860
861 if (!large_allowed || !sk_can_gso(sk))
862 return mss_now;
863
6c09fa09
ED
864 /* Note : tcp_tso_autosize() will eventually split this later */
865 new_size_goal = sk->sk_gso_max_size - 1 - MAX_TCP_HEADER;
605ad7f1
ED
866 new_size_goal = tcp_bound_to_half_wnd(tp, new_size_goal);
867
868 /* We try hard to avoid divides here */
869 size_goal = tp->gso_segs * mss_now;
870 if (unlikely(new_size_goal < size_goal ||
871 new_size_goal >= size_goal + mss_now)) {
872 tp->gso_segs = min_t(u16, new_size_goal / mss_now,
873 sk->sk_gso_max_segs);
874 size_goal = tp->gso_segs * mss_now;
0c54b85f
IJ
875 }
876
605ad7f1 877 return max(size_goal, mss_now);
0c54b85f
IJ
878}
879
880static int tcp_send_mss(struct sock *sk, int *size_goal, int flags)
881{
882 int mss_now;
883
884 mss_now = tcp_current_mss(sk);
885 *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB));
886
887 return mss_now;
888}
889
64022d0b
ED
890static ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
891 size_t size, int flags)
1da177e4
LT
892{
893 struct tcp_sock *tp = tcp_sk(sk);
c1b4a7e6 894 int mss_now, size_goal;
1da177e4
LT
895 int err;
896 ssize_t copied;
897 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
898
8336886f
JC
899 /* Wait for a connection to finish. One exception is TCP Fast Open
900 * (passive side) where data is allowed to be sent before a connection
901 * is fully established.
902 */
903 if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
904 !tcp_passive_fastopen(sk)) {
686a5624
YM
905 err = sk_stream_wait_connect(sk, &timeo);
906 if (err != 0)
1da177e4 907 goto out_err;
8336886f 908 }
1da177e4 909
9cd3e072 910 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1da177e4 911
0c54b85f 912 mss_now = tcp_send_mss(sk, &size_goal, flags);
1da177e4
LT
913 copied = 0;
914
915 err = -EPIPE;
916 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
0d6a775e 917 goto out_err;
1da177e4 918
64022d0b 919 while (size > 0) {
fe067e8a 920 struct sk_buff *skb = tcp_write_queue_tail(sk);
38ba0a65 921 int copy, i;
38ba0a65 922 bool can_coalesce;
1da177e4 923
fe067e8a 924 if (!tcp_send_head(sk) || (copy = size_goal - skb->len) <= 0) {
1da177e4
LT
925new_segment:
926 if (!sk_stream_memory_free(sk))
927 goto wait_for_sndbuf;
928
eb934478
ED
929 skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation,
930 skb_queue_empty(&sk->sk_write_queue));
1da177e4
LT
931 if (!skb)
932 goto wait_for_memory;
933
9e412ba7 934 skb_entail(sk, skb);
c1b4a7e6 935 copy = size_goal;
1da177e4
LT
936 }
937
938 if (copy > size)
939 copy = size;
940
941 i = skb_shinfo(skb)->nr_frags;
942 can_coalesce = skb_can_coalesce(skb, i, page, offset);
943 if (!can_coalesce && i >= MAX_SKB_FRAGS) {
944 tcp_mark_push(tp, skb);
945 goto new_segment;
946 }
3ab224be 947 if (!sk_wmem_schedule(sk, copy))
1da177e4 948 goto wait_for_memory;
e905a9ed 949
1da177e4 950 if (can_coalesce) {
9e903e08 951 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1da177e4
LT
952 } else {
953 get_page(page);
954 skb_fill_page_desc(skb, i, page, offset, copy);
955 }
c9af6db4 956 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
cef401de 957
1da177e4
LT
958 skb->len += copy;
959 skb->data_len += copy;
960 skb->truesize += copy;
961 sk->sk_wmem_queued += copy;
3ab224be 962 sk_mem_charge(sk, copy);
84fa7933 963 skb->ip_summed = CHECKSUM_PARTIAL;
1da177e4
LT
964 tp->write_seq += copy;
965 TCP_SKB_CB(skb)->end_seq += copy;
cd7d8498 966 tcp_skb_pcount_set(skb, 0);
1da177e4
LT
967
968 if (!copied)
4de075e0 969 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1da177e4
LT
970
971 copied += copy;
64022d0b 972 offset += copy;
686a5624
YM
973 size -= copy;
974 if (!size) {
4ed2d765 975 tcp_tx_timestamp(sk, skb);
1da177e4 976 goto out;
4ed2d765 977 }
1da177e4 978
69d15067 979 if (skb->len < size_goal || (flags & MSG_OOB))
1da177e4
LT
980 continue;
981
982 if (forced_push(tp)) {
983 tcp_mark_push(tp, skb);
9e412ba7 984 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
fe067e8a 985 } else if (skb == tcp_send_head(sk))
1da177e4
LT
986 tcp_push_one(sk, mss_now);
987 continue;
988
989wait_for_sndbuf:
990 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
991wait_for_memory:
f54b3111
ED
992 tcp_push(sk, flags & ~MSG_MORE, mss_now,
993 TCP_NAGLE_PUSH, size_goal);
1da177e4 994
686a5624
YM
995 err = sk_stream_wait_memory(sk, &timeo);
996 if (err != 0)
1da177e4
LT
997 goto do_error;
998
0c54b85f 999 mss_now = tcp_send_mss(sk, &size_goal, flags);
1da177e4
LT
1000 }
1001
1002out:
35f9c09f 1003 if (copied && !(flags & MSG_SENDPAGE_NOTLAST))
f54b3111 1004 tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
1da177e4
LT
1005 return copied;
1006
1007do_error:
1008 if (copied)
1009 goto out;
1010out_err:
ce5ec440
JB
1011 /* make sure we wake any epoll edge trigger waiter */
1012 if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN))
1013 sk->sk_write_space(sk);
1da177e4
LT
1014 return sk_stream_error(sk, flags, err);
1015}
1016
7ba42910
CG
1017int tcp_sendpage(struct sock *sk, struct page *page, int offset,
1018 size_t size, int flags)
1da177e4
LT
1019{
1020 ssize_t res;
1da177e4 1021
1da177e4 1022 if (!(sk->sk_route_caps & NETIF_F_SG) ||
9a49850d 1023 !sk_check_csum_caps(sk))
7ba42910
CG
1024 return sock_no_sendpage(sk->sk_socket, page, offset, size,
1025 flags);
1da177e4 1026
1da177e4 1027 lock_sock(sk);
64022d0b 1028 res = do_tcp_sendpages(sk, page, offset, size, flags);
1da177e4
LT
1029 release_sock(sk);
1030 return res;
1031}
4bc2f18b 1032EXPORT_SYMBOL(tcp_sendpage);
1da177e4 1033
690e99c4 1034static inline int select_size(const struct sock *sk, bool sg)
1da177e4 1035{
cf533ea5 1036 const struct tcp_sock *tp = tcp_sk(sk);
c1b4a7e6 1037 int tmp = tp->mss_cache;
1da177e4 1038
def87cf4 1039 if (sg) {
f07d960d
ED
1040 if (sk_can_gso(sk)) {
1041 /* Small frames wont use a full page:
1042 * Payload will immediately follow tcp header.
1043 */
1044 tmp = SKB_WITH_OVERHEAD(2048 - MAX_TCP_HEADER);
1045 } else {
b4e26f5e
DM
1046 int pgbreak = SKB_MAX_HEAD(MAX_TCP_HEADER);
1047
1048 if (tmp >= pgbreak &&
1049 tmp <= pgbreak + (MAX_SKB_FRAGS - 1) * PAGE_SIZE)
1050 tmp = pgbreak;
1051 }
1052 }
1da177e4 1053
1da177e4
LT
1054 return tmp;
1055}
1056
cf60af03
YC
1057void tcp_free_fastopen_req(struct tcp_sock *tp)
1058{
00db4124 1059 if (tp->fastopen_req) {
cf60af03
YC
1060 kfree(tp->fastopen_req);
1061 tp->fastopen_req = NULL;
1062 }
1063}
1064
f5ddcbbb
ED
1065static int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg,
1066 int *copied, size_t size)
cf60af03
YC
1067{
1068 struct tcp_sock *tp = tcp_sk(sk);
1069 int err, flags;
1070
1071 if (!(sysctl_tcp_fastopen & TFO_CLIENT_ENABLE))
1072 return -EOPNOTSUPP;
00db4124 1073 if (tp->fastopen_req)
cf60af03
YC
1074 return -EALREADY; /* Another Fast Open is in progress */
1075
1076 tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request),
1077 sk->sk_allocation);
51456b29 1078 if (unlikely(!tp->fastopen_req))
cf60af03
YC
1079 return -ENOBUFS;
1080 tp->fastopen_req->data = msg;
f5ddcbbb 1081 tp->fastopen_req->size = size;
cf60af03
YC
1082
1083 flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0;
1084 err = __inet_stream_connect(sk->sk_socket, msg->msg_name,
1085 msg->msg_namelen, flags);
f5ddcbbb 1086 *copied = tp->fastopen_req->copied;
cf60af03
YC
1087 tcp_free_fastopen_req(tp);
1088 return err;
1089}
1090
1b784140 1091int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
1da177e4 1092{
1da177e4
LT
1093 struct tcp_sock *tp = tcp_sk(sk);
1094 struct sk_buff *skb;
57be5bda
AV
1095 int flags, err, copied = 0;
1096 int mss_now = 0, size_goal, copied_syn = 0;
690e99c4 1097 bool sg;
1da177e4
LT
1098 long timeo;
1099
1100 lock_sock(sk);
1da177e4
LT
1101
1102 flags = msg->msg_flags;
cf60af03 1103 if (flags & MSG_FASTOPEN) {
f5ddcbbb 1104 err = tcp_sendmsg_fastopen(sk, msg, &copied_syn, size);
cf60af03
YC
1105 if (err == -EINPROGRESS && copied_syn > 0)
1106 goto out;
1107 else if (err)
1108 goto out_err;
cf60af03
YC
1109 }
1110
1da177e4
LT
1111 timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1112
8336886f
JC
1113 /* Wait for a connection to finish. One exception is TCP Fast Open
1114 * (passive side) where data is allowed to be sent before a connection
1115 * is fully established.
1116 */
1117 if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
1118 !tcp_passive_fastopen(sk)) {
686a5624
YM
1119 err = sk_stream_wait_connect(sk, &timeo);
1120 if (err != 0)
cf60af03 1121 goto do_error;
8336886f 1122 }
1da177e4 1123
c0e88ff0
PE
1124 if (unlikely(tp->repair)) {
1125 if (tp->repair_queue == TCP_RECV_QUEUE) {
1126 copied = tcp_send_rcvq(sk, msg, size);
5924f17a 1127 goto out_nopush;
c0e88ff0
PE
1128 }
1129
1130 err = -EINVAL;
1131 if (tp->repair_queue == TCP_NO_QUEUE)
1132 goto out_err;
1133
1134 /* 'common' sending to sendq */
1135 }
1136
1da177e4 1137 /* This should be in poll */
9cd3e072 1138 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1da177e4 1139
0c54b85f 1140 mss_now = tcp_send_mss(sk, &size_goal, flags);
1da177e4
LT
1141
1142 /* Ok commence sending. */
1da177e4
LT
1143 copied = 0;
1144
1145 err = -EPIPE;
1146 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
0d6a775e 1147 goto out_err;
1da177e4 1148
690e99c4 1149 sg = !!(sk->sk_route_caps & NETIF_F_SG);
def87cf4 1150
01e97e65 1151 while (msg_data_left(msg)) {
57be5bda
AV
1152 int copy = 0;
1153 int max = size_goal;
1da177e4 1154
57be5bda
AV
1155 skb = tcp_write_queue_tail(sk);
1156 if (tcp_send_head(sk)) {
1157 if (skb->ip_summed == CHECKSUM_NONE)
1158 max = mss_now;
1159 copy = max - skb->len;
cf60af03 1160 }
1da177e4 1161
57be5bda 1162 if (copy <= 0) {
1da177e4 1163new_segment:
57be5bda
AV
1164 /* Allocate new segment. If the interface is SG,
1165 * allocate skb fitting to single page.
1166 */
1167 if (!sk_stream_memory_free(sk))
1168 goto wait_for_sndbuf;
1da177e4 1169
57be5bda
AV
1170 skb = sk_stream_alloc_skb(sk,
1171 select_size(sk, sg),
eb934478
ED
1172 sk->sk_allocation,
1173 skb_queue_empty(&sk->sk_write_queue));
57be5bda
AV
1174 if (!skb)
1175 goto wait_for_memory;
1da177e4 1176
57be5bda
AV
1177 /*
1178 * Check whether we can use HW checksum.
1179 */
9a49850d 1180 if (sk_check_csum_caps(sk))
57be5bda 1181 skb->ip_summed = CHECKSUM_PARTIAL;
1da177e4 1182
57be5bda
AV
1183 skb_entail(sk, skb);
1184 copy = size_goal;
1185 max = size_goal;
9d186cac 1186
57be5bda
AV
1187 /* All packets are restored as if they have
1188 * already been sent. skb_mstamp isn't set to
1189 * avoid wrong rtt estimation.
1190 */
1191 if (tp->repair)
1192 TCP_SKB_CB(skb)->sacked |= TCPCB_REPAIRED;
1193 }
1da177e4 1194
57be5bda 1195 /* Try to append data to the end of skb. */
01e97e65
AV
1196 if (copy > msg_data_left(msg))
1197 copy = msg_data_left(msg);
57be5bda
AV
1198
1199 /* Where to copy to? */
1200 if (skb_availroom(skb) > 0) {
1201 /* We have some space in skb head. Superb! */
1202 copy = min_t(int, copy, skb_availroom(skb));
1203 err = skb_add_data_nocache(sk, skb, &msg->msg_iter, copy);
1204 if (err)
1205 goto do_fault;
1206 } else {
1207 bool merge = true;
1208 int i = skb_shinfo(skb)->nr_frags;
1209 struct page_frag *pfrag = sk_page_frag(sk);
1210
1211 if (!sk_page_frag_refill(sk, pfrag))
1212 goto wait_for_memory;
ef015786 1213
57be5bda
AV
1214 if (!skb_can_coalesce(skb, i, pfrag->page,
1215 pfrag->offset)) {
1216 if (i == MAX_SKB_FRAGS || !sg) {
1217 tcp_mark_push(tp, skb);
1218 goto new_segment;
1da177e4 1219 }
57be5bda 1220 merge = false;
1da177e4
LT
1221 }
1222
57be5bda
AV
1223 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1224
1225 if (!sk_wmem_schedule(sk, copy))
1226 goto wait_for_memory;
1da177e4 1227
57be5bda
AV
1228 err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
1229 pfrag->page,
1230 pfrag->offset,
1231 copy);
1232 if (err)
1233 goto do_error;
1da177e4 1234
57be5bda
AV
1235 /* Update the skb. */
1236 if (merge) {
1237 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1238 } else {
1239 skb_fill_page_desc(skb, i, pfrag->page,
1240 pfrag->offset, copy);
1241 get_page(pfrag->page);
4ed2d765 1242 }
57be5bda
AV
1243 pfrag->offset += copy;
1244 }
1245
1246 if (!copied)
1247 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1248
1249 tp->write_seq += copy;
1250 TCP_SKB_CB(skb)->end_seq += copy;
1251 tcp_skb_pcount_set(skb, 0);
1da177e4 1252
57be5bda 1253 copied += copy;
01e97e65 1254 if (!msg_data_left(msg)) {
57be5bda
AV
1255 tcp_tx_timestamp(sk, skb);
1256 goto out;
1257 }
1da177e4 1258
57be5bda 1259 if (skb->len < max || (flags & MSG_OOB) || unlikely(tp->repair))
1da177e4
LT
1260 continue;
1261
57be5bda
AV
1262 if (forced_push(tp)) {
1263 tcp_mark_push(tp, skb);
1264 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
1265 } else if (skb == tcp_send_head(sk))
1266 tcp_push_one(sk, mss_now);
1267 continue;
1268
1da177e4 1269wait_for_sndbuf:
57be5bda 1270 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1da177e4 1271wait_for_memory:
57be5bda
AV
1272 if (copied)
1273 tcp_push(sk, flags & ~MSG_MORE, mss_now,
1274 TCP_NAGLE_PUSH, size_goal);
1da177e4 1275
686a5624
YM
1276 err = sk_stream_wait_memory(sk, &timeo);
1277 if (err != 0)
57be5bda 1278 goto do_error;
1da177e4 1279
57be5bda 1280 mss_now = tcp_send_mss(sk, &size_goal, flags);
1da177e4
LT
1281 }
1282
1283out:
ec342325 1284 if (copied)
f54b3111 1285 tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
5924f17a 1286out_nopush:
1da177e4 1287 release_sock(sk);
cf60af03 1288 return copied + copied_syn;
1da177e4
LT
1289
1290do_fault:
1291 if (!skb->len) {
fe067e8a
DM
1292 tcp_unlink_write_queue(skb, sk);
1293 /* It is the one place in all of TCP, except connection
1294 * reset, where we can be unlinking the send_head.
1295 */
1296 tcp_check_send_head(sk, skb);
3ab224be 1297 sk_wmem_free_skb(sk, skb);
1da177e4
LT
1298 }
1299
1300do_error:
cf60af03 1301 if (copied + copied_syn)
1da177e4
LT
1302 goto out;
1303out_err:
1304 err = sk_stream_error(sk, flags, err);
ce5ec440
JB
1305 /* make sure we wake any epoll edge trigger waiter */
1306 if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN))
1307 sk->sk_write_space(sk);
1da177e4
LT
1308 release_sock(sk);
1309 return err;
1310}
4bc2f18b 1311EXPORT_SYMBOL(tcp_sendmsg);
1da177e4
LT
1312
1313/*
1314 * Handle reading urgent data. BSD has very simple semantics for
1315 * this, no blocking and very strange errors 8)
1316 */
1317
377f0a08 1318static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags)
1da177e4
LT
1319{
1320 struct tcp_sock *tp = tcp_sk(sk);
1321
1322 /* No URG data to read. */
1323 if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data ||
1324 tp->urg_data == TCP_URG_READ)
1325 return -EINVAL; /* Yes this is right ! */
1326
1327 if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE))
1328 return -ENOTCONN;
1329
1330 if (tp->urg_data & TCP_URG_VALID) {
1331 int err = 0;
1332 char c = tp->urg_data;
1333
1334 if (!(flags & MSG_PEEK))
1335 tp->urg_data = TCP_URG_READ;
1336
1337 /* Read urgent data. */
1338 msg->msg_flags |= MSG_OOB;
1339
1340 if (len > 0) {
1341 if (!(flags & MSG_TRUNC))
7eab8d9e 1342 err = memcpy_to_msg(msg, &c, 1);
1da177e4
LT
1343 len = 1;
1344 } else
1345 msg->msg_flags |= MSG_TRUNC;
1346
1347 return err ? -EFAULT : len;
1348 }
1349
1350 if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
1351 return 0;
1352
1353 /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and
1354 * the available implementations agree in this case:
1355 * this call should never block, independent of the
1356 * blocking state of the socket.
1357 * Mike <pall@rz.uni-karlsruhe.de>
1358 */
1359 return -EAGAIN;
1360}
1361
c0e88ff0
PE
1362static int tcp_peek_sndq(struct sock *sk, struct msghdr *msg, int len)
1363{
1364 struct sk_buff *skb;
1365 int copied = 0, err = 0;
1366
1367 /* XXX -- need to support SO_PEEK_OFF */
1368
1369 skb_queue_walk(&sk->sk_write_queue, skb) {
51f3d02b 1370 err = skb_copy_datagram_msg(skb, 0, msg, skb->len);
c0e88ff0
PE
1371 if (err)
1372 break;
1373
1374 copied += skb->len;
1375 }
1376
1377 return err ?: copied;
1378}
1379
1da177e4
LT
1380/* Clean up the receive buffer for full frames taken by the user,
1381 * then send an ACK if necessary. COPIED is the number of bytes
1382 * tcp_recvmsg has given to the user so far, it speeds up the
1383 * calculation of whether or not we must ACK for the sake of
1384 * a window update.
1385 */
3f334078 1386static void tcp_cleanup_rbuf(struct sock *sk, int copied)
1da177e4
LT
1387{
1388 struct tcp_sock *tp = tcp_sk(sk);
a2a385d6 1389 bool time_to_ack = false;
1da177e4 1390
1da177e4
LT
1391 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
1392
d792c100 1393 WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq),
2af6fd8b 1394 "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n",
d792c100 1395 tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt);
1da177e4 1396
463c84b9
ACM
1397 if (inet_csk_ack_scheduled(sk)) {
1398 const struct inet_connection_sock *icsk = inet_csk(sk);
1da177e4
LT
1399 /* Delayed ACKs frequently hit locked sockets during bulk
1400 * receive. */
463c84b9 1401 if (icsk->icsk_ack.blocked ||
1da177e4 1402 /* Once-per-two-segments ACK was not sent by tcp_input.c */
463c84b9 1403 tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss ||
1da177e4
LT
1404 /*
1405 * If this read emptied read buffer, we send ACK, if
1406 * connection is not bidirectional, user drained
1407 * receive buffer and there was a small segment
1408 * in queue.
1409 */
1ef9696c
AK
1410 (copied > 0 &&
1411 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) ||
1412 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) &&
1413 !icsk->icsk_ack.pingpong)) &&
1414 !atomic_read(&sk->sk_rmem_alloc)))
a2a385d6 1415 time_to_ack = true;
1da177e4
LT
1416 }
1417
1418 /* We send an ACK if we can now advertise a non-zero window
1419 * which has been raised "significantly".
1420 *
1421 * Even if window raised up to infinity, do not send window open ACK
1422 * in states, where we will not receive more. It is useless.
1423 */
1424 if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1425 __u32 rcv_window_now = tcp_receive_window(tp);
1426
1427 /* Optimize, __tcp_select_window() is not cheap. */
1428 if (2*rcv_window_now <= tp->window_clamp) {
1429 __u32 new_window = __tcp_select_window(sk);
1430
1431 /* Send ACK now, if this read freed lots of space
1432 * in our buffer. Certainly, new_window is new window.
1433 * We can advertise it now, if it is not less than current one.
1434 * "Lots" means "at least twice" here.
1435 */
1436 if (new_window && new_window >= 2 * rcv_window_now)
a2a385d6 1437 time_to_ack = true;
1da177e4
LT
1438 }
1439 }
1440 if (time_to_ack)
1441 tcp_send_ack(sk);
1442}
1443
1444static void tcp_prequeue_process(struct sock *sk)
1445{
1446 struct sk_buff *skb;
1447 struct tcp_sock *tp = tcp_sk(sk);
1448
6f67c817 1449 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPPREQUEUED);
1da177e4
LT
1450
1451 /* RX process wants to run with disabled BHs, though it is not
1452 * necessary */
1453 local_bh_disable();
1454 while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
c57943a1 1455 sk_backlog_rcv(sk, skb);
1da177e4
LT
1456 local_bh_enable();
1457
1458 /* Clear memory counter. */
1459 tp->ucopy.memory = 0;
1460}
1461
f26845b4 1462static struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off)
1da177e4
LT
1463{
1464 struct sk_buff *skb;
1465 u32 offset;
1466
f26845b4 1467 while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) {
1da177e4 1468 offset = seq - TCP_SKB_CB(skb)->seq;
9d691539
ED
1469 if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
1470 pr_err_once("%s: found a SYN, please report !\n", __func__);
1da177e4 1471 offset--;
9d691539 1472 }
e11ecddf 1473 if (offset < skb->len || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) {
1da177e4
LT
1474 *off = offset;
1475 return skb;
1476 }
f26845b4
ED
1477 /* This looks weird, but this can happen if TCP collapsing
1478 * splitted a fat GRO packet, while we released socket lock
1479 * in skb_splice_bits()
1480 */
7bced397 1481 sk_eat_skb(sk, skb);
1da177e4
LT
1482 }
1483 return NULL;
1484}
1485
1486/*
1487 * This routine provides an alternative to tcp_recvmsg() for routines
1488 * that would like to handle copying from skbuffs directly in 'sendfile'
1489 * fashion.
1490 * Note:
1491 * - It is assumed that the socket was locked by the caller.
1492 * - The routine does not block.
1493 * - At present, there is no support for reading OOB data
1494 * or for 'peeking' the socket using this routine
1495 * (although both would be easy to implement).
1496 */
1497int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
1498 sk_read_actor_t recv_actor)
1499{
1500 struct sk_buff *skb;
1501 struct tcp_sock *tp = tcp_sk(sk);
1502 u32 seq = tp->copied_seq;
1503 u32 offset;
1504 int copied = 0;
1505
1506 if (sk->sk_state == TCP_LISTEN)
1507 return -ENOTCONN;
1508 while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) {
1509 if (offset < skb->len) {
374e7b59
OP
1510 int used;
1511 size_t len;
1da177e4
LT
1512
1513 len = skb->len - offset;
1514 /* Stop reading if we hit a patch of urgent data */
1515 if (tp->urg_data) {
1516 u32 urg_offset = tp->urg_seq - seq;
1517 if (urg_offset < len)
1518 len = urg_offset;
1519 if (!len)
1520 break;
1521 }
1522 used = recv_actor(desc, skb, offset, len);
ff905b1e 1523 if (used <= 0) {
ddb61a57
JA
1524 if (!copied)
1525 copied = used;
1526 break;
1527 } else if (used <= len) {
1da177e4
LT
1528 seq += used;
1529 copied += used;
1530 offset += used;
1531 }
02275a2e 1532 /* If recv_actor drops the lock (e.g. TCP splice
293ad604
OP
1533 * receive) the skb pointer might be invalid when
1534 * getting here: tcp_collapse might have deleted it
1535 * while aggregating skbs from the socket queue.
1536 */
02275a2e
WT
1537 skb = tcp_recv_skb(sk, seq - 1, &offset);
1538 if (!skb)
1da177e4 1539 break;
02275a2e
WT
1540 /* TCP coalescing might have appended data to the skb.
1541 * Try to splice more frags
1542 */
1543 if (offset + 1 != skb->len)
1544 continue;
1da177e4 1545 }
e11ecddf 1546 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) {
7bced397 1547 sk_eat_skb(sk, skb);
1da177e4
LT
1548 ++seq;
1549 break;
1550 }
7bced397 1551 sk_eat_skb(sk, skb);
1da177e4
LT
1552 if (!desc->count)
1553 break;
baff42ab 1554 tp->copied_seq = seq;
1da177e4
LT
1555 }
1556 tp->copied_seq = seq;
1557
1558 tcp_rcv_space_adjust(sk);
1559
1560 /* Clean up data we have read: This will do ACK frames. */
f26845b4
ED
1561 if (copied > 0) {
1562 tcp_recv_skb(sk, seq, &offset);
0e4b4992 1563 tcp_cleanup_rbuf(sk, copied);
f26845b4 1564 }
1da177e4
LT
1565 return copied;
1566}
4bc2f18b 1567EXPORT_SYMBOL(tcp_read_sock);
1da177e4
LT
1568
1569/*
1570 * This routine copies from a sock struct into the user buffer.
1571 *
1572 * Technical note: in 2.3 we work on _locked_ socket, so that
1573 * tricks with *seq access order and skb->users are not required.
1574 * Probably, code can be easily improved even more.
1575 */
1576
1b784140
YX
1577int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
1578 int flags, int *addr_len)
1da177e4
LT
1579{
1580 struct tcp_sock *tp = tcp_sk(sk);
1581 int copied = 0;
1582 u32 peek_seq;
1583 u32 *seq;
1584 unsigned long used;
1585 int err;
1586 int target; /* Read at least this many bytes */
1587 long timeo;
1588 struct task_struct *user_recv = NULL;
dfbafc99 1589 struct sk_buff *skb, *last;
77527313 1590 u32 urg_hole = 0;
1da177e4 1591
4ed2d765 1592 if (unlikely(flags & MSG_ERRQUEUE))
f4713a3d 1593 return inet_recv_error(sk, msg, len, addr_len);
4ed2d765 1594
cbf55001
ET
1595 if (sk_can_busy_loop(sk) && skb_queue_empty(&sk->sk_receive_queue) &&
1596 (sk->sk_state == TCP_ESTABLISHED))
1597 sk_busy_loop(sk, nonblock);
d30e383b 1598
1da177e4
LT
1599 lock_sock(sk);
1600
1da177e4
LT
1601 err = -ENOTCONN;
1602 if (sk->sk_state == TCP_LISTEN)
1603 goto out;
1604
1605 timeo = sock_rcvtimeo(sk, nonblock);
1606
1607 /* Urgent data needs to be handled specially. */
1608 if (flags & MSG_OOB)
1609 goto recv_urg;
1610
c0e88ff0
PE
1611 if (unlikely(tp->repair)) {
1612 err = -EPERM;
1613 if (!(flags & MSG_PEEK))
1614 goto out;
1615
1616 if (tp->repair_queue == TCP_SEND_QUEUE)
1617 goto recv_sndq;
1618
1619 err = -EINVAL;
1620 if (tp->repair_queue == TCP_NO_QUEUE)
1621 goto out;
1622
1623 /* 'common' recv queue MSG_PEEK-ing */
1624 }
1625
1da177e4
LT
1626 seq = &tp->copied_seq;
1627 if (flags & MSG_PEEK) {
1628 peek_seq = tp->copied_seq;
1629 seq = &peek_seq;
1630 }
1631
1632 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1633
1634 do {
1da177e4
LT
1635 u32 offset;
1636
1637 /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */
1638 if (tp->urg_data && tp->urg_seq == *seq) {
1639 if (copied)
1640 break;
1641 if (signal_pending(current)) {
1642 copied = timeo ? sock_intr_errno(timeo) : -EAGAIN;
1643 break;
1644 }
1645 }
1646
1647 /* Next get a buffer. */
1648
dfbafc99 1649 last = skb_peek_tail(&sk->sk_receive_queue);
91521944 1650 skb_queue_walk(&sk->sk_receive_queue, skb) {
dfbafc99 1651 last = skb;
1da177e4
LT
1652 /* Now that we have two receive queues this
1653 * shouldn't happen.
1654 */
d792c100 1655 if (WARN(before(*seq, TCP_SKB_CB(skb)->seq),
2af6fd8b
JP
1656 "recvmsg bug: copied %X seq %X rcvnxt %X fl %X\n",
1657 *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt,
1658 flags))
1da177e4 1659 break;
d792c100 1660
1da177e4 1661 offset = *seq - TCP_SKB_CB(skb)->seq;
9d691539
ED
1662 if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
1663 pr_err_once("%s: found a SYN, please report !\n", __func__);
1da177e4 1664 offset--;
9d691539 1665 }
1da177e4
LT
1666 if (offset < skb->len)
1667 goto found_ok_skb;
e11ecddf 1668 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
1da177e4 1669 goto found_fin_ok;
2af6fd8b
JP
1670 WARN(!(flags & MSG_PEEK),
1671 "recvmsg bug 2: copied %X seq %X rcvnxt %X fl %X\n",
1672 *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags);
91521944 1673 }
1da177e4
LT
1674
1675 /* Well, if we have backlog, try to process it now yet. */
1676
1677 if (copied >= target && !sk->sk_backlog.tail)
1678 break;
1679
1680 if (copied) {
1681 if (sk->sk_err ||
1682 sk->sk_state == TCP_CLOSE ||
1683 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1684 !timeo ||
518a09ef 1685 signal_pending(current))
1da177e4
LT
1686 break;
1687 } else {
1688 if (sock_flag(sk, SOCK_DONE))
1689 break;
1690
1691 if (sk->sk_err) {
1692 copied = sock_error(sk);
1693 break;
1694 }
1695
1696 if (sk->sk_shutdown & RCV_SHUTDOWN)
1697 break;
1698
1699 if (sk->sk_state == TCP_CLOSE) {
1700 if (!sock_flag(sk, SOCK_DONE)) {
1701 /* This occurs when user tries to read
1702 * from never connected socket.
1703 */
1704 copied = -ENOTCONN;
1705 break;
1706 }
1707 break;
1708 }
1709
1710 if (!timeo) {
1711 copied = -EAGAIN;
1712 break;
1713 }
1714
1715 if (signal_pending(current)) {
1716 copied = sock_intr_errno(timeo);
1717 break;
1718 }
1719 }
1720
0e4b4992 1721 tcp_cleanup_rbuf(sk, copied);
1da177e4 1722
7df55125 1723 if (!sysctl_tcp_low_latency && tp->ucopy.task == user_recv) {
1da177e4
LT
1724 /* Install new reader */
1725 if (!user_recv && !(flags & (MSG_TRUNC | MSG_PEEK))) {
1726 user_recv = current;
1727 tp->ucopy.task = user_recv;
f4362a2c 1728 tp->ucopy.msg = msg;
1da177e4
LT
1729 }
1730
1731 tp->ucopy.len = len;
1732
547b792c
IJ
1733 WARN_ON(tp->copied_seq != tp->rcv_nxt &&
1734 !(flags & (MSG_PEEK | MSG_TRUNC)));
1da177e4
LT
1735
1736 /* Ugly... If prequeue is not empty, we have to
1737 * process it before releasing socket, otherwise
1738 * order will be broken at second iteration.
1739 * More elegant solution is required!!!
1740 *
1741 * Look: we have the following (pseudo)queues:
1742 *
1743 * 1. packets in flight
1744 * 2. backlog
1745 * 3. prequeue
1746 * 4. receive_queue
1747 *
1748 * Each queue can be processed only if the next ones
1749 * are empty. At this point we have empty receive_queue.
1750 * But prequeue _can_ be not empty after 2nd iteration,
1751 * when we jumped to start of loop because backlog
1752 * processing added something to receive_queue.
1753 * We cannot release_sock(), because backlog contains
1754 * packets arrived _after_ prequeued ones.
1755 *
1756 * Shortly, algorithm is clear --- to process all
1757 * the queues in order. We could make it more directly,
1758 * requeueing packets from backlog to prequeue, if
1759 * is not empty. It is more elegant, but eats cycles,
1760 * unfortunately.
1761 */
b03efcfb 1762 if (!skb_queue_empty(&tp->ucopy.prequeue))
1da177e4
LT
1763 goto do_prequeue;
1764
1765 /* __ Set realtime policy in scheduler __ */
1766 }
1767
1768 if (copied >= target) {
1769 /* Do not sleep, just process backlog. */
1770 release_sock(sk);
1771 lock_sock(sk);
dfbafc99
SD
1772 } else {
1773 sk_wait_data(sk, &timeo, last);
1774 }
1da177e4
LT
1775
1776 if (user_recv) {
1777 int chunk;
1778
1779 /* __ Restore normal policy in scheduler __ */
1780
686a5624
YM
1781 chunk = len - tp->ucopy.len;
1782 if (chunk != 0) {
ed88098e 1783 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMBACKLOG, chunk);
1da177e4
LT
1784 len -= chunk;
1785 copied += chunk;
1786 }
1787
1788 if (tp->rcv_nxt == tp->copied_seq &&
b03efcfb 1789 !skb_queue_empty(&tp->ucopy.prequeue)) {
1da177e4
LT
1790do_prequeue:
1791 tcp_prequeue_process(sk);
1792
686a5624
YM
1793 chunk = len - tp->ucopy.len;
1794 if (chunk != 0) {
ed88098e 1795 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
1da177e4
LT
1796 len -= chunk;
1797 copied += chunk;
1798 }
1799 }
1800 }
77527313
IJ
1801 if ((flags & MSG_PEEK) &&
1802 (peek_seq - copied - urg_hole != tp->copied_seq)) {
e87cc472
JP
1803 net_dbg_ratelimited("TCP(%s:%d): Application bug, race in MSG_PEEK\n",
1804 current->comm,
1805 task_pid_nr(current));
1da177e4
LT
1806 peek_seq = tp->copied_seq;
1807 }
1808 continue;
1809
1810 found_ok_skb:
1811 /* Ok so how much can we use? */
1812 used = skb->len - offset;
1813 if (len < used)
1814 used = len;
1815
1816 /* Do we have urgent data here? */
1817 if (tp->urg_data) {
1818 u32 urg_offset = tp->urg_seq - *seq;
1819 if (urg_offset < used) {
1820 if (!urg_offset) {
1821 if (!sock_flag(sk, SOCK_URGINLINE)) {
1822 ++*seq;
77527313 1823 urg_hole++;
1da177e4
LT
1824 offset++;
1825 used--;
1826 if (!used)
1827 goto skip_copy;
1828 }
1829 } else
1830 used = urg_offset;
1831 }
1832 }
1833
1834 if (!(flags & MSG_TRUNC)) {
51f3d02b 1835 err = skb_copy_datagram_msg(skb, offset, msg, used);
7bced397
DW
1836 if (err) {
1837 /* Exception. Bailout! */
1838 if (!copied)
1839 copied = -EFAULT;
1840 break;
1da177e4
LT
1841 }
1842 }
1843
1844 *seq += used;
1845 copied += used;
1846 len -= used;
1847
1848 tcp_rcv_space_adjust(sk);
1849
1850skip_copy:
1851 if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) {
1852 tp->urg_data = 0;
9e412ba7 1853 tcp_fast_path_check(sk);
1da177e4
LT
1854 }
1855 if (used + offset < skb->len)
1856 continue;
1857
e11ecddf 1858 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
1da177e4 1859 goto found_fin_ok;
7bced397
DW
1860 if (!(flags & MSG_PEEK))
1861 sk_eat_skb(sk, skb);
1da177e4
LT
1862 continue;
1863
1864 found_fin_ok:
1865 /* Process the FIN. */
1866 ++*seq;
7bced397
DW
1867 if (!(flags & MSG_PEEK))
1868 sk_eat_skb(sk, skb);
1da177e4
LT
1869 break;
1870 } while (len > 0);
1871
1872 if (user_recv) {
b03efcfb 1873 if (!skb_queue_empty(&tp->ucopy.prequeue)) {
1da177e4
LT
1874 int chunk;
1875
1876 tp->ucopy.len = copied > 0 ? len : 0;
1877
1878 tcp_prequeue_process(sk);
1879
1880 if (copied > 0 && (chunk = len - tp->ucopy.len) != 0) {
ed88098e 1881 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
1da177e4
LT
1882 len -= chunk;
1883 copied += chunk;
1884 }
1885 }
1886
1887 tp->ucopy.task = NULL;
1888 tp->ucopy.len = 0;
1889 }
1890
1891 /* According to UNIX98, msg_name/msg_namelen are ignored
1892 * on connected socket. I was just happy when found this 8) --ANK
1893 */
1894
1895 /* Clean up data we have read: This will do ACK frames. */
0e4b4992 1896 tcp_cleanup_rbuf(sk, copied);
1da177e4 1897
1da177e4
LT
1898 release_sock(sk);
1899 return copied;
1900
1901out:
1da177e4
LT
1902 release_sock(sk);
1903 return err;
1904
1905recv_urg:
377f0a08 1906 err = tcp_recv_urg(sk, msg, len, flags);
1da177e4 1907 goto out;
c0e88ff0
PE
1908
1909recv_sndq:
1910 err = tcp_peek_sndq(sk, msg, len);
1911 goto out;
1da177e4 1912}
4bc2f18b 1913EXPORT_SYMBOL(tcp_recvmsg);
1da177e4 1914
490d5046
IJ
1915void tcp_set_state(struct sock *sk, int state)
1916{
1917 int oldstate = sk->sk_state;
1918
1919 switch (state) {
1920 case TCP_ESTABLISHED:
1921 if (oldstate != TCP_ESTABLISHED)
81cc8a75 1922 TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
490d5046
IJ
1923 break;
1924
1925 case TCP_CLOSE:
1926 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
81cc8a75 1927 TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS);
490d5046
IJ
1928
1929 sk->sk_prot->unhash(sk);
1930 if (inet_csk(sk)->icsk_bind_hash &&
1931 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
ab1e0a13 1932 inet_put_port(sk);
490d5046
IJ
1933 /* fall through */
1934 default:
5a5f3a8d 1935 if (oldstate == TCP_ESTABLISHED)
74688e48 1936 TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
490d5046
IJ
1937 }
1938
1939 /* Change state AFTER socket is unhashed to avoid closed
1940 * socket sitting in hash tables.
1941 */
00fd38d9 1942 sk_state_store(sk, state);
490d5046
IJ
1943
1944#ifdef STATE_TRACE
5a5f3a8d 1945 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n", sk, statename[oldstate], statename[state]);
490d5046
IJ
1946#endif
1947}
1948EXPORT_SYMBOL_GPL(tcp_set_state);
1949
1da177e4
LT
1950/*
1951 * State processing on a close. This implements the state shift for
1952 * sending our FIN frame. Note that we only send a FIN for some
1953 * states. A shutdown() may have already sent the FIN, or we may be
1954 * closed.
1955 */
1956
9b5b5cff 1957static const unsigned char new_state[16] = {
1da177e4 1958 /* current state: new state: action: */
0980c1e3
ED
1959 [0 /* (Invalid) */] = TCP_CLOSE,
1960 [TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
1961 [TCP_SYN_SENT] = TCP_CLOSE,
1962 [TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
1963 [TCP_FIN_WAIT1] = TCP_FIN_WAIT1,
1964 [TCP_FIN_WAIT2] = TCP_FIN_WAIT2,
1965 [TCP_TIME_WAIT] = TCP_CLOSE,
1966 [TCP_CLOSE] = TCP_CLOSE,
1967 [TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN,
1968 [TCP_LAST_ACK] = TCP_LAST_ACK,
1969 [TCP_LISTEN] = TCP_CLOSE,
1970 [TCP_CLOSING] = TCP_CLOSING,
1971 [TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */
1da177e4
LT
1972};
1973
1974static int tcp_close_state(struct sock *sk)
1975{
1976 int next = (int)new_state[sk->sk_state];
1977 int ns = next & TCP_STATE_MASK;
1978
1979 tcp_set_state(sk, ns);
1980
1981 return next & TCP_ACTION_FIN;
1982}
1983
1984/*
1985 * Shutdown the sending side of a connection. Much like close except
1f29b058 1986 * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD).
1da177e4
LT
1987 */
1988
1989void tcp_shutdown(struct sock *sk, int how)
1990{
1991 /* We need to grab some memory, and put together a FIN,
1992 * and then put it into the queue to be sent.
1993 * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92.
1994 */
1995 if (!(how & SEND_SHUTDOWN))
1996 return;
1997
1998 /* If we've already sent a FIN, or it's a closed state, skip this. */
1999 if ((1 << sk->sk_state) &
2000 (TCPF_ESTABLISHED | TCPF_SYN_SENT |
2001 TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) {
2002 /* Clear out any half completed packets. FIN if needed. */
2003 if (tcp_close_state(sk))
2004 tcp_send_fin(sk);
2005 }
2006}
4bc2f18b 2007EXPORT_SYMBOL(tcp_shutdown);
1da177e4 2008
efcdbf24
AS
2009bool tcp_check_oom(struct sock *sk, int shift)
2010{
2011 bool too_many_orphans, out_of_socket_memory;
2012
2013 too_many_orphans = tcp_too_many_orphans(sk, shift);
2014 out_of_socket_memory = tcp_out_of_memory(sk);
2015
e87cc472
JP
2016 if (too_many_orphans)
2017 net_info_ratelimited("too many orphaned sockets\n");
2018 if (out_of_socket_memory)
2019 net_info_ratelimited("out of memory -- consider tuning tcp_mem\n");
efcdbf24
AS
2020 return too_many_orphans || out_of_socket_memory;
2021}
2022
1da177e4
LT
2023void tcp_close(struct sock *sk, long timeout)
2024{
2025 struct sk_buff *skb;
2026 int data_was_unread = 0;
75c2d907 2027 int state;
1da177e4
LT
2028
2029 lock_sock(sk);
2030 sk->sk_shutdown = SHUTDOWN_MASK;
2031
2032 if (sk->sk_state == TCP_LISTEN) {
2033 tcp_set_state(sk, TCP_CLOSE);
2034
2035 /* Special case. */
0a5578cf 2036 inet_csk_listen_stop(sk);
1da177e4
LT
2037
2038 goto adjudge_to_death;
2039 }
2040
2041 /* We need to flush the recv. buffs. We do this only on the
2042 * descriptor close, not protocol-sourced closes, because the
2043 * reader process may not have drained the data yet!
2044 */
2045 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
e11ecddf
ED
2046 u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq;
2047
2048 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
2049 len--;
1da177e4
LT
2050 data_was_unread += len;
2051 __kfree_skb(skb);
2052 }
2053
3ab224be 2054 sk_mem_reclaim(sk);
1da177e4 2055
565b7b2d
KK
2056 /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */
2057 if (sk->sk_state == TCP_CLOSE)
2058 goto adjudge_to_death;
2059
65bb723c
GR
2060 /* As outlined in RFC 2525, section 2.17, we send a RST here because
2061 * data was lost. To witness the awful effects of the old behavior of
2062 * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk
2063 * GET in an FTP client, suspend the process, wait for the client to
2064 * advertise a zero window, then kill -9 the FTP client, wheee...
2065 * Note: timeout is always zero in such a case.
1da177e4 2066 */
ee995283
PE
2067 if (unlikely(tcp_sk(sk)->repair)) {
2068 sk->sk_prot->disconnect(sk, 0);
2069 } else if (data_was_unread) {
1da177e4 2070 /* Unread data was tossed, zap the connection. */
6f67c817 2071 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE);
1da177e4 2072 tcp_set_state(sk, TCP_CLOSE);
aa133076 2073 tcp_send_active_reset(sk, sk->sk_allocation);
1da177e4
LT
2074 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
2075 /* Check zero linger _after_ checking for unread data. */
2076 sk->sk_prot->disconnect(sk, 0);
6f67c817 2077 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONDATA);
1da177e4
LT
2078 } else if (tcp_close_state(sk)) {
2079 /* We FIN if the application ate all the data before
2080 * zapping the connection.
2081 */
2082
2083 /* RED-PEN. Formally speaking, we have broken TCP state
2084 * machine. State transitions:
2085 *
2086 * TCP_ESTABLISHED -> TCP_FIN_WAIT1
2087 * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible)
2088 * TCP_CLOSE_WAIT -> TCP_LAST_ACK
2089 *
2090 * are legal only when FIN has been sent (i.e. in window),
2091 * rather than queued out of window. Purists blame.
2092 *
2093 * F.e. "RFC state" is ESTABLISHED,
2094 * if Linux state is FIN-WAIT-1, but FIN is still not sent.
2095 *
2096 * The visible declinations are that sometimes
2097 * we enter time-wait state, when it is not required really
2098 * (harmless), do not send active resets, when they are
2099 * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when
2100 * they look as CLOSING or LAST_ACK for Linux)
2101 * Probably, I missed some more holelets.
2102 * --ANK
8336886f
JC
2103 * XXX (TFO) - To start off we don't support SYN+ACK+FIN
2104 * in a single packet! (May consider it later but will
2105 * probably need API support or TCP_CORK SYN-ACK until
2106 * data is written and socket is closed.)
1da177e4
LT
2107 */
2108 tcp_send_fin(sk);
2109 }
2110
2111 sk_stream_wait_close(sk, timeout);
2112
2113adjudge_to_death:
75c2d907
HX
2114 state = sk->sk_state;
2115 sock_hold(sk);
2116 sock_orphan(sk);
75c2d907 2117
1da177e4
LT
2118 /* It is the last release_sock in its life. It will remove backlog. */
2119 release_sock(sk);
2120
2121
2122 /* Now socket is owned by kernel and we acquire BH lock
2123 to finish close. No need to check for user refs.
2124 */
2125 local_bh_disable();
2126 bh_lock_sock(sk);
547b792c 2127 WARN_ON(sock_owned_by_user(sk));
1da177e4 2128
eb4dea58
HX
2129 percpu_counter_inc(sk->sk_prot->orphan_count);
2130
75c2d907
HX
2131 /* Have we already been destroyed by a softirq or backlog? */
2132 if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE)
2133 goto out;
1da177e4
LT
2134
2135 /* This is a (useful) BSD violating of the RFC. There is a
2136 * problem with TCP as specified in that the other end could
2137 * keep a socket open forever with no application left this end.
b10bd54c 2138 * We use a 1 minute timeout (about the same as BSD) then kill
1da177e4
LT
2139 * our end. If they send after that then tough - BUT: long enough
2140 * that we won't make the old 4*rto = almost no time - whoops
2141 * reset mistake.
2142 *
2143 * Nope, it was not mistake. It is really desired behaviour
2144 * f.e. on http servers, when such sockets are useless, but
2145 * consume significant resources. Let's do it with special
2146 * linger2 option. --ANK
2147 */
2148
2149 if (sk->sk_state == TCP_FIN_WAIT2) {
2150 struct tcp_sock *tp = tcp_sk(sk);
2151 if (tp->linger2 < 0) {
2152 tcp_set_state(sk, TCP_CLOSE);
2153 tcp_send_active_reset(sk, GFP_ATOMIC);
de0744af
PE
2154 NET_INC_STATS_BH(sock_net(sk),
2155 LINUX_MIB_TCPABORTONLINGER);
1da177e4 2156 } else {
463c84b9 2157 const int tmo = tcp_fin_time(sk);
1da177e4
LT
2158
2159 if (tmo > TCP_TIMEWAIT_LEN) {
52499afe
DM
2160 inet_csk_reset_keepalive_timer(sk,
2161 tmo - TCP_TIMEWAIT_LEN);
1da177e4 2162 } else {
1da177e4
LT
2163 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
2164 goto out;
2165 }
2166 }
2167 }
2168 if (sk->sk_state != TCP_CLOSE) {
3ab224be 2169 sk_mem_reclaim(sk);
efcdbf24 2170 if (tcp_check_oom(sk, 0)) {
1da177e4
LT
2171 tcp_set_state(sk, TCP_CLOSE);
2172 tcp_send_active_reset(sk, GFP_ATOMIC);
de0744af
PE
2173 NET_INC_STATS_BH(sock_net(sk),
2174 LINUX_MIB_TCPABORTONMEMORY);
1da177e4
LT
2175 }
2176 }
1da177e4 2177
8336886f
JC
2178 if (sk->sk_state == TCP_CLOSE) {
2179 struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
2180 /* We could get here with a non-NULL req if the socket is
2181 * aborted (e.g., closed with unread data) before 3WHS
2182 * finishes.
2183 */
00db4124 2184 if (req)
8336886f 2185 reqsk_fastopen_remove(sk, req, false);
0a5578cf 2186 inet_csk_destroy_sock(sk);
8336886f 2187 }
1da177e4
LT
2188 /* Otherwise, socket is reprieved until protocol close. */
2189
2190out:
2191 bh_unlock_sock(sk);
2192 local_bh_enable();
2193 sock_put(sk);
2194}
4bc2f18b 2195EXPORT_SYMBOL(tcp_close);
1da177e4
LT
2196
2197/* These states need RST on ABORT according to RFC793 */
2198
a2a385d6 2199static inline bool tcp_need_reset(int state)
1da177e4
LT
2200{
2201 return (1 << state) &
2202 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 |
2203 TCPF_FIN_WAIT2 | TCPF_SYN_RECV);
2204}
2205
2206int tcp_disconnect(struct sock *sk, int flags)
2207{
2208 struct inet_sock *inet = inet_sk(sk);
463c84b9 2209 struct inet_connection_sock *icsk = inet_csk(sk);
1da177e4
LT
2210 struct tcp_sock *tp = tcp_sk(sk);
2211 int err = 0;
2212 int old_state = sk->sk_state;
2213
2214 if (old_state != TCP_CLOSE)
2215 tcp_set_state(sk, TCP_CLOSE);
2216
2217 /* ABORT function of RFC793 */
2218 if (old_state == TCP_LISTEN) {
0a5578cf 2219 inet_csk_listen_stop(sk);
ee995283
PE
2220 } else if (unlikely(tp->repair)) {
2221 sk->sk_err = ECONNABORTED;
1da177e4
LT
2222 } else if (tcp_need_reset(old_state) ||
2223 (tp->snd_nxt != tp->write_seq &&
2224 (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) {
caa20d9a 2225 /* The last check adjusts for discrepancy of Linux wrt. RFC
1da177e4
LT
2226 * states
2227 */
2228 tcp_send_active_reset(sk, gfp_any());
2229 sk->sk_err = ECONNRESET;
2230 } else if (old_state == TCP_SYN_SENT)
2231 sk->sk_err = ECONNRESET;
2232
2233 tcp_clear_xmit_timers(sk);
2234 __skb_queue_purge(&sk->sk_receive_queue);
fe067e8a 2235 tcp_write_queue_purge(sk);
1da177e4
LT
2236 __skb_queue_purge(&tp->out_of_order_queue);
2237
c720c7e8 2238 inet->inet_dport = 0;
1da177e4
LT
2239
2240 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
2241 inet_reset_saddr(sk);
2242
2243 sk->sk_shutdown = 0;
2244 sock_reset_flag(sk, SOCK_DONE);
740b0f18 2245 tp->srtt_us = 0;
686a5624
YM
2246 tp->write_seq += tp->max_window + 2;
2247 if (tp->write_seq == 0)
1da177e4 2248 tp->write_seq = 1;
463c84b9 2249 icsk->icsk_backoff = 0;
1da177e4 2250 tp->snd_cwnd = 2;
6687e988 2251 icsk->icsk_probes_out = 0;
1da177e4 2252 tp->packets_out = 0;
0b6a05c1 2253 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
1da177e4 2254 tp->snd_cwnd_cnt = 0;
1fdf475a 2255 tp->window_clamp = 0;
6687e988 2256 tcp_set_ca_state(sk, TCP_CA_Open);
1da177e4 2257 tcp_clear_retrans(tp);
463c84b9 2258 inet_csk_delack_init(sk);
fe067e8a 2259 tcp_init_send_head(sk);
b40b4f79 2260 memset(&tp->rx_opt, 0, sizeof(tp->rx_opt));
1da177e4
LT
2261 __sk_dst_reset(sk);
2262
c720c7e8 2263 WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
1da177e4
LT
2264
2265 sk->sk_error_report(sk);
2266 return err;
2267}
4bc2f18b 2268EXPORT_SYMBOL(tcp_disconnect);
1da177e4 2269
a2a385d6 2270static inline bool tcp_can_repair_sock(const struct sock *sk)
ee995283 2271{
52e804c6 2272 return ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN) &&
ee995283
PE
2273 ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_ESTABLISHED));
2274}
2275
de248a75
PE
2276static int tcp_repair_options_est(struct tcp_sock *tp,
2277 struct tcp_repair_opt __user *optbuf, unsigned int len)
b139ba4e 2278{
de248a75 2279 struct tcp_repair_opt opt;
b139ba4e 2280
de248a75
PE
2281 while (len >= sizeof(opt)) {
2282 if (copy_from_user(&opt, optbuf, sizeof(opt)))
b139ba4e
PE
2283 return -EFAULT;
2284
2285 optbuf++;
de248a75 2286 len -= sizeof(opt);
b139ba4e 2287
de248a75
PE
2288 switch (opt.opt_code) {
2289 case TCPOPT_MSS:
2290 tp->rx_opt.mss_clamp = opt.opt_val;
b139ba4e 2291 break;
de248a75 2292 case TCPOPT_WINDOW:
bc26ccd8
AV
2293 {
2294 u16 snd_wscale = opt.opt_val & 0xFFFF;
2295 u16 rcv_wscale = opt.opt_val >> 16;
2296
2297 if (snd_wscale > 14 || rcv_wscale > 14)
2298 return -EFBIG;
b139ba4e 2299
bc26ccd8
AV
2300 tp->rx_opt.snd_wscale = snd_wscale;
2301 tp->rx_opt.rcv_wscale = rcv_wscale;
2302 tp->rx_opt.wscale_ok = 1;
2303 }
b139ba4e 2304 break;
b139ba4e 2305 case TCPOPT_SACK_PERM:
de248a75
PE
2306 if (opt.opt_val != 0)
2307 return -EINVAL;
2308
b139ba4e
PE
2309 tp->rx_opt.sack_ok |= TCP_SACK_SEEN;
2310 if (sysctl_tcp_fack)
2311 tcp_enable_fack(tp);
2312 break;
2313 case TCPOPT_TIMESTAMP:
de248a75
PE
2314 if (opt.opt_val != 0)
2315 return -EINVAL;
2316
b139ba4e
PE
2317 tp->rx_opt.tstamp_ok = 1;
2318 break;
2319 }
2320 }
2321
2322 return 0;
2323}
2324
1da177e4
LT
2325/*
2326 * Socket option code for TCP.
2327 */
3fdadf7d 2328static int do_tcp_setsockopt(struct sock *sk, int level,
b7058842 2329 int optname, char __user *optval, unsigned int optlen)
1da177e4
LT
2330{
2331 struct tcp_sock *tp = tcp_sk(sk);
463c84b9 2332 struct inet_connection_sock *icsk = inet_csk(sk);
1da177e4
LT
2333 int val;
2334 int err = 0;
2335
e56fb50f
WAS
2336 /* These are data/string values, all the others are ints */
2337 switch (optname) {
2338 case TCP_CONGESTION: {
5f8ef48d
SH
2339 char name[TCP_CA_NAME_MAX];
2340
2341 if (optlen < 1)
2342 return -EINVAL;
2343
2344 val = strncpy_from_user(name, optval,
4fdb78d3 2345 min_t(long, TCP_CA_NAME_MAX-1, optlen));
5f8ef48d
SH
2346 if (val < 0)
2347 return -EFAULT;
2348 name[val] = 0;
2349
2350 lock_sock(sk);
6687e988 2351 err = tcp_set_congestion_control(sk, name);
5f8ef48d
SH
2352 release_sock(sk);
2353 return err;
2354 }
e56fb50f
WAS
2355 default:
2356 /* fallthru */
2357 break;
ccbd6a5a 2358 }
5f8ef48d 2359
1da177e4
LT
2360 if (optlen < sizeof(int))
2361 return -EINVAL;
2362
2363 if (get_user(val, (int __user *)optval))
2364 return -EFAULT;
2365
2366 lock_sock(sk);
2367
2368 switch (optname) {
2369 case TCP_MAXSEG:
2370 /* Values greater than interface MTU won't take effect. However
2371 * at the point when this call is done we typically don't yet
2372 * know which interface is going to be used */
c39508d6 2373 if (val < TCP_MIN_MSS || val > MAX_TCP_WINDOW) {
1da177e4
LT
2374 err = -EINVAL;
2375 break;
2376 }
2377 tp->rx_opt.user_mss = val;
2378 break;
2379
2380 case TCP_NODELAY:
2381 if (val) {
2382 /* TCP_NODELAY is weaker than TCP_CORK, so that
2383 * this option on corked socket is remembered, but
2384 * it is not activated until cork is cleared.
2385 *
2386 * However, when TCP_NODELAY is set we make
2387 * an explicit push, which overrides even TCP_CORK
2388 * for currently queued segments.
2389 */
2390 tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH;
9e412ba7 2391 tcp_push_pending_frames(sk);
1da177e4
LT
2392 } else {
2393 tp->nonagle &= ~TCP_NAGLE_OFF;
2394 }
2395 break;
2396
36e31b0a
AP
2397 case TCP_THIN_LINEAR_TIMEOUTS:
2398 if (val < 0 || val > 1)
2399 err = -EINVAL;
2400 else
2401 tp->thin_lto = val;
2402 break;
2403
7e380175
AP
2404 case TCP_THIN_DUPACK:
2405 if (val < 0 || val > 1)
2406 err = -EINVAL;
e2e5c4c0 2407 else {
7e380175 2408 tp->thin_dupack = val;
eed530b6
YC
2409 if (tp->thin_dupack)
2410 tcp_disable_early_retrans(tp);
e2e5c4c0 2411 }
7e380175
AP
2412 break;
2413
ee995283
PE
2414 case TCP_REPAIR:
2415 if (!tcp_can_repair_sock(sk))
2416 err = -EPERM;
2417 else if (val == 1) {
2418 tp->repair = 1;
2419 sk->sk_reuse = SK_FORCE_REUSE;
2420 tp->repair_queue = TCP_NO_QUEUE;
2421 } else if (val == 0) {
2422 tp->repair = 0;
2423 sk->sk_reuse = SK_NO_REUSE;
2424 tcp_send_window_probe(sk);
2425 } else
2426 err = -EINVAL;
2427
2428 break;
2429
2430 case TCP_REPAIR_QUEUE:
2431 if (!tp->repair)
2432 err = -EPERM;
2433 else if (val < TCP_QUEUES_NR)
2434 tp->repair_queue = val;
2435 else
2436 err = -EINVAL;
2437 break;
2438
2439 case TCP_QUEUE_SEQ:
2440 if (sk->sk_state != TCP_CLOSE)
2441 err = -EPERM;
2442 else if (tp->repair_queue == TCP_SEND_QUEUE)
2443 tp->write_seq = val;
2444 else if (tp->repair_queue == TCP_RECV_QUEUE)
2445 tp->rcv_nxt = val;
2446 else
2447 err = -EINVAL;
2448 break;
2449
b139ba4e
PE
2450 case TCP_REPAIR_OPTIONS:
2451 if (!tp->repair)
2452 err = -EINVAL;
2453 else if (sk->sk_state == TCP_ESTABLISHED)
de248a75
PE
2454 err = tcp_repair_options_est(tp,
2455 (struct tcp_repair_opt __user *)optval,
2456 optlen);
b139ba4e
PE
2457 else
2458 err = -EPERM;
2459 break;
2460
1da177e4
LT
2461 case TCP_CORK:
2462 /* When set indicates to always queue non-full frames.
2463 * Later the user clears this option and we transmit
2464 * any pending partial frames in the queue. This is
2465 * meant to be used alongside sendfile() to get properly
2466 * filled frames when the user (for example) must write
2467 * out headers with a write() call first and then use
2468 * sendfile to send out the data parts.
2469 *
2470 * TCP_CORK can be set together with TCP_NODELAY and it is
2471 * stronger than TCP_NODELAY.
2472 */
2473 if (val) {
2474 tp->nonagle |= TCP_NAGLE_CORK;
2475 } else {
2476 tp->nonagle &= ~TCP_NAGLE_CORK;
2477 if (tp->nonagle&TCP_NAGLE_OFF)
2478 tp->nonagle |= TCP_NAGLE_PUSH;
9e412ba7 2479 tcp_push_pending_frames(sk);
1da177e4
LT
2480 }
2481 break;
2482
2483 case TCP_KEEPIDLE:
2484 if (val < 1 || val > MAX_TCP_KEEPIDLE)
2485 err = -EINVAL;
2486 else {
2487 tp->keepalive_time = val * HZ;
2488 if (sock_flag(sk, SOCK_KEEPOPEN) &&
2489 !((1 << sk->sk_state) &
2490 (TCPF_CLOSE | TCPF_LISTEN))) {
6c37e5de 2491 u32 elapsed = keepalive_time_elapsed(tp);
1da177e4
LT
2492 if (tp->keepalive_time > elapsed)
2493 elapsed = tp->keepalive_time - elapsed;
2494 else
2495 elapsed = 0;
463c84b9 2496 inet_csk_reset_keepalive_timer(sk, elapsed);
1da177e4
LT
2497 }
2498 }
2499 break;
2500 case TCP_KEEPINTVL:
2501 if (val < 1 || val > MAX_TCP_KEEPINTVL)
2502 err = -EINVAL;
2503 else
2504 tp->keepalive_intvl = val * HZ;
2505 break;
2506 case TCP_KEEPCNT:
2507 if (val < 1 || val > MAX_TCP_KEEPCNT)
2508 err = -EINVAL;
2509 else
2510 tp->keepalive_probes = val;
2511 break;
2512 case TCP_SYNCNT:
2513 if (val < 1 || val > MAX_TCP_SYNCNT)
2514 err = -EINVAL;
2515 else
463c84b9 2516 icsk->icsk_syn_retries = val;
1da177e4
LT
2517 break;
2518
cd8ae852
ED
2519 case TCP_SAVE_SYN:
2520 if (val < 0 || val > 1)
2521 err = -EINVAL;
2522 else
2523 tp->save_syn = val;
2524 break;
2525
1da177e4
LT
2526 case TCP_LINGER2:
2527 if (val < 0)
2528 tp->linger2 = -1;
2529 else if (val > sysctl_tcp_fin_timeout / HZ)
2530 tp->linger2 = 0;
2531 else
2532 tp->linger2 = val * HZ;
2533 break;
2534
2535 case TCP_DEFER_ACCEPT:
b103cf34
JA
2536 /* Translate value in seconds to number of retransmits */
2537 icsk->icsk_accept_queue.rskq_defer_accept =
2538 secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ,
2539 TCP_RTO_MAX / HZ);
1da177e4
LT
2540 break;
2541
2542 case TCP_WINDOW_CLAMP:
2543 if (!val) {
2544 if (sk->sk_state != TCP_CLOSE) {
2545 err = -EINVAL;
2546 break;
2547 }
2548 tp->window_clamp = 0;
2549 } else
2550 tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ?
2551 SOCK_MIN_RCVBUF / 2 : val;
2552 break;
2553
2554 case TCP_QUICKACK:
2555 if (!val) {
463c84b9 2556 icsk->icsk_ack.pingpong = 1;
1da177e4 2557 } else {
463c84b9 2558 icsk->icsk_ack.pingpong = 0;
1da177e4
LT
2559 if ((1 << sk->sk_state) &
2560 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) &&
463c84b9
ACM
2561 inet_csk_ack_scheduled(sk)) {
2562 icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;
0e4b4992 2563 tcp_cleanup_rbuf(sk, 1);
1da177e4 2564 if (!(val & 1))
463c84b9 2565 icsk->icsk_ack.pingpong = 1;
1da177e4
LT
2566 }
2567 }
2568 break;
2569
cfb6eeb4
YH
2570#ifdef CONFIG_TCP_MD5SIG
2571 case TCP_MD5SIG:
2572 /* Read the IP->Key mappings from userspace */
2573 err = tp->af_specific->md5_parse(sk, optval, optlen);
2574 break;
2575#endif
dca43c75 2576 case TCP_USER_TIMEOUT:
b248230c 2577 /* Cap the max time in ms TCP will retry or probe the window
dca43c75
JC
2578 * before giving up and aborting (ETIMEDOUT) a connection.
2579 */
42493570
HL
2580 if (val < 0)
2581 err = -EINVAL;
2582 else
2583 icsk->icsk_user_timeout = msecs_to_jiffies(val);
dca43c75 2584 break;
8336886f
JC
2585
2586 case TCP_FASTOPEN:
2587 if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE |
dfea2aa6
CP
2588 TCPF_LISTEN))) {
2589 tcp_fastopen_init_key_once(true);
2590
0536fcc0 2591 fastopen_queue_tune(sk, val);
dfea2aa6 2592 } else {
8336886f 2593 err = -EINVAL;
dfea2aa6 2594 }
8336886f 2595 break;
93be6ce0
AV
2596 case TCP_TIMESTAMP:
2597 if (!tp->repair)
2598 err = -EPERM;
2599 else
2600 tp->tsoffset = val - tcp_time_stamp;
2601 break;
c9bee3b7
ED
2602 case TCP_NOTSENT_LOWAT:
2603 tp->notsent_lowat = val;
2604 sk->sk_write_space(sk);
2605 break;
1da177e4
LT
2606 default:
2607 err = -ENOPROTOOPT;
2608 break;
3ff50b79
SH
2609 }
2610
1da177e4
LT
2611 release_sock(sk);
2612 return err;
2613}
2614
3fdadf7d 2615int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval,
b7058842 2616 unsigned int optlen)
3fdadf7d 2617{
cf533ea5 2618 const struct inet_connection_sock *icsk = inet_csk(sk);
3fdadf7d
DM
2619
2620 if (level != SOL_TCP)
2621 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
2622 optval, optlen);
2623 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2624}
4bc2f18b 2625EXPORT_SYMBOL(tcp_setsockopt);
3fdadf7d
DM
2626
2627#ifdef CONFIG_COMPAT
543d9cfe 2628int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
b7058842 2629 char __user *optval, unsigned int optlen)
3fdadf7d 2630{
dec73ff0
ACM
2631 if (level != SOL_TCP)
2632 return inet_csk_compat_setsockopt(sk, level, optname,
2633 optval, optlen);
3fdadf7d
DM
2634 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2635}
543d9cfe 2636EXPORT_SYMBOL(compat_tcp_setsockopt);
3fdadf7d
DM
2637#endif
2638
1da177e4 2639/* Return information about state of tcp endpoint in API format. */
0df48c26 2640void tcp_get_info(struct sock *sk, struct tcp_info *info)
1da177e4 2641{
35ac838a 2642 const struct tcp_sock *tp = tcp_sk(sk); /* iff sk_type == SOCK_STREAM */
463c84b9 2643 const struct inet_connection_sock *icsk = inet_csk(sk);
1da177e4 2644 u32 now = tcp_time_stamp;
d654976c 2645 unsigned int start;
ff5d7497 2646 u64 rate64;
fad9dfef 2647 u32 rate;
1da177e4
LT
2648
2649 memset(info, 0, sizeof(*info));
35ac838a
CG
2650 if (sk->sk_type != SOCK_STREAM)
2651 return;
1da177e4 2652
00fd38d9
ED
2653 info->tcpi_state = sk_state_load(sk);
2654
6687e988 2655 info->tcpi_ca_state = icsk->icsk_ca_state;
463c84b9 2656 info->tcpi_retransmits = icsk->icsk_retransmits;
6687e988 2657 info->tcpi_probes = icsk->icsk_probes_out;
463c84b9 2658 info->tcpi_backoff = icsk->icsk_backoff;
1da177e4
LT
2659
2660 if (tp->rx_opt.tstamp_ok)
2661 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
e60402d0 2662 if (tcp_is_sack(tp))
1da177e4
LT
2663 info->tcpi_options |= TCPI_OPT_SACK;
2664 if (tp->rx_opt.wscale_ok) {
2665 info->tcpi_options |= TCPI_OPT_WSCALE;
2666 info->tcpi_snd_wscale = tp->rx_opt.snd_wscale;
2667 info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale;
e905a9ed 2668 }
1da177e4 2669
b5c5693b 2670 if (tp->ecn_flags & TCP_ECN_OK)
1da177e4 2671 info->tcpi_options |= TCPI_OPT_ECN;
b5c5693b
ED
2672 if (tp->ecn_flags & TCP_ECN_SEEN)
2673 info->tcpi_options |= TCPI_OPT_ECN_SEEN;
6f73601e
YC
2674 if (tp->syn_data_acked)
2675 info->tcpi_options |= TCPI_OPT_SYN_DATA;
1da177e4 2676
463c84b9
ACM
2677 info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto);
2678 info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato);
c1b4a7e6 2679 info->tcpi_snd_mss = tp->mss_cache;
463c84b9 2680 info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss;
1da177e4 2681
00fd38d9 2682 if (info->tcpi_state == TCP_LISTEN) {
5ee3afba
RJ
2683 info->tcpi_unacked = sk->sk_ack_backlog;
2684 info->tcpi_sacked = sk->sk_max_ack_backlog;
2685 } else {
2686 info->tcpi_unacked = tp->packets_out;
2687 info->tcpi_sacked = tp->sacked_out;
2688 }
1da177e4
LT
2689 info->tcpi_lost = tp->lost_out;
2690 info->tcpi_retrans = tp->retrans_out;
2691 info->tcpi_fackets = tp->fackets_out;
2692
2693 info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime);
463c84b9 2694 info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime);
1da177e4
LT
2695 info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp);
2696
d83d8461 2697 info->tcpi_pmtu = icsk->icsk_pmtu_cookie;
1da177e4 2698 info->tcpi_rcv_ssthresh = tp->rcv_ssthresh;
740b0f18
ED
2699 info->tcpi_rtt = tp->srtt_us >> 3;
2700 info->tcpi_rttvar = tp->mdev_us >> 2;
1da177e4
LT
2701 info->tcpi_snd_ssthresh = tp->snd_ssthresh;
2702 info->tcpi_snd_cwnd = tp->snd_cwnd;
2703 info->tcpi_advmss = tp->advmss;
2704 info->tcpi_reordering = tp->reordering;
2705
2706 info->tcpi_rcv_rtt = jiffies_to_usecs(tp->rcv_rtt_est.rtt)>>3;
2707 info->tcpi_rcv_space = tp->rcvq_space.space;
2708
2709 info->tcpi_total_retrans = tp->total_retrans;
977cb0ec 2710
fad9dfef 2711 rate = READ_ONCE(sk->sk_pacing_rate);
ff5d7497
ED
2712 rate64 = rate != ~0U ? rate : ~0ULL;
2713 put_unaligned(rate64, &info->tcpi_pacing_rate);
fad9dfef
ED
2714
2715 rate = READ_ONCE(sk->sk_max_pacing_rate);
ff5d7497
ED
2716 rate64 = rate != ~0U ? rate : ~0ULL;
2717 put_unaligned(rate64, &info->tcpi_max_pacing_rate);
0df48c26 2718
d654976c
ED
2719 do {
2720 start = u64_stats_fetch_begin_irq(&tp->syncp);
ff5d7497
ED
2721 put_unaligned(tp->bytes_acked, &info->tcpi_bytes_acked);
2722 put_unaligned(tp->bytes_received, &info->tcpi_bytes_received);
d654976c 2723 } while (u64_stats_fetch_retry_irq(&tp->syncp, start));
2efd055c
MRL
2724 info->tcpi_segs_out = tp->segs_out;
2725 info->tcpi_segs_in = tp->segs_in;
1da177e4 2726}
1da177e4
LT
2727EXPORT_SYMBOL_GPL(tcp_get_info);
2728
3fdadf7d
DM
2729static int do_tcp_getsockopt(struct sock *sk, int level,
2730 int optname, char __user *optval, int __user *optlen)
1da177e4 2731{
295f7324 2732 struct inet_connection_sock *icsk = inet_csk(sk);
1da177e4 2733 struct tcp_sock *tp = tcp_sk(sk);
6fa25166 2734 struct net *net = sock_net(sk);
1da177e4
LT
2735 int val, len;
2736
1da177e4
LT
2737 if (get_user(len, optlen))
2738 return -EFAULT;
2739
2740 len = min_t(unsigned int, len, sizeof(int));
2741
2742 if (len < 0)
2743 return -EINVAL;
2744
2745 switch (optname) {
2746 case TCP_MAXSEG:
c1b4a7e6 2747 val = tp->mss_cache;
1da177e4
LT
2748 if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
2749 val = tp->rx_opt.user_mss;
5e6a3ce6
PE
2750 if (tp->repair)
2751 val = tp->rx_opt.mss_clamp;
1da177e4
LT
2752 break;
2753 case TCP_NODELAY:
2754 val = !!(tp->nonagle&TCP_NAGLE_OFF);
2755 break;
2756 case TCP_CORK:
2757 val = !!(tp->nonagle&TCP_NAGLE_CORK);
2758 break;
2759 case TCP_KEEPIDLE:
df19a626 2760 val = keepalive_time_when(tp) / HZ;
1da177e4
LT
2761 break;
2762 case TCP_KEEPINTVL:
df19a626 2763 val = keepalive_intvl_when(tp) / HZ;
1da177e4
LT
2764 break;
2765 case TCP_KEEPCNT:
df19a626 2766 val = keepalive_probes(tp);
1da177e4
LT
2767 break;
2768 case TCP_SYNCNT:
6fa25166 2769 val = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries;
1da177e4
LT
2770 break;
2771 case TCP_LINGER2:
2772 val = tp->linger2;
2773 if (val >= 0)
2774 val = (val ? : sysctl_tcp_fin_timeout) / HZ;
2775 break;
2776 case TCP_DEFER_ACCEPT:
b103cf34
JA
2777 val = retrans_to_secs(icsk->icsk_accept_queue.rskq_defer_accept,
2778 TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ);
1da177e4
LT
2779 break;
2780 case TCP_WINDOW_CLAMP:
2781 val = tp->window_clamp;
2782 break;
2783 case TCP_INFO: {
2784 struct tcp_info info;
2785
2786 if (get_user(len, optlen))
2787 return -EFAULT;
2788
2789 tcp_get_info(sk, &info);
2790
2791 len = min_t(unsigned int, len, sizeof(info));
2792 if (put_user(len, optlen))
2793 return -EFAULT;
2794 if (copy_to_user(optval, &info, len))
2795 return -EFAULT;
2796 return 0;
2797 }
6e9250f5
ED
2798 case TCP_CC_INFO: {
2799 const struct tcp_congestion_ops *ca_ops;
2800 union tcp_cc_info info;
2801 size_t sz = 0;
2802 int attr;
2803
2804 if (get_user(len, optlen))
2805 return -EFAULT;
2806
2807 ca_ops = icsk->icsk_ca_ops;
2808 if (ca_ops && ca_ops->get_info)
2809 sz = ca_ops->get_info(sk, ~0U, &attr, &info);
2810
2811 len = min_t(unsigned int, len, sz);
2812 if (put_user(len, optlen))
2813 return -EFAULT;
2814 if (copy_to_user(optval, &info, len))
2815 return -EFAULT;
2816 return 0;
2817 }
1da177e4 2818 case TCP_QUICKACK:
295f7324 2819 val = !icsk->icsk_ack.pingpong;
1da177e4 2820 break;
5f8ef48d
SH
2821
2822 case TCP_CONGESTION:
2823 if (get_user(len, optlen))
2824 return -EFAULT;
2825 len = min_t(unsigned int, len, TCP_CA_NAME_MAX);
2826 if (put_user(len, optlen))
2827 return -EFAULT;
6687e988 2828 if (copy_to_user(optval, icsk->icsk_ca_ops->name, len))
5f8ef48d
SH
2829 return -EFAULT;
2830 return 0;
e56fb50f 2831
3c0fef0b
JH
2832 case TCP_THIN_LINEAR_TIMEOUTS:
2833 val = tp->thin_lto;
2834 break;
2835 case TCP_THIN_DUPACK:
2836 val = tp->thin_dupack;
2837 break;
dca43c75 2838
ee995283
PE
2839 case TCP_REPAIR:
2840 val = tp->repair;
2841 break;
2842
2843 case TCP_REPAIR_QUEUE:
2844 if (tp->repair)
2845 val = tp->repair_queue;
2846 else
2847 return -EINVAL;
2848 break;
2849
2850 case TCP_QUEUE_SEQ:
2851 if (tp->repair_queue == TCP_SEND_QUEUE)
2852 val = tp->write_seq;
2853 else if (tp->repair_queue == TCP_RECV_QUEUE)
2854 val = tp->rcv_nxt;
2855 else
2856 return -EINVAL;
2857 break;
2858
dca43c75
JC
2859 case TCP_USER_TIMEOUT:
2860 val = jiffies_to_msecs(icsk->icsk_user_timeout);
2861 break;
1536e285
KN
2862
2863 case TCP_FASTOPEN:
0536fcc0 2864 val = icsk->icsk_accept_queue.fastopenq.max_qlen;
1536e285
KN
2865 break;
2866
93be6ce0
AV
2867 case TCP_TIMESTAMP:
2868 val = tcp_time_stamp + tp->tsoffset;
2869 break;
c9bee3b7
ED
2870 case TCP_NOTSENT_LOWAT:
2871 val = tp->notsent_lowat;
2872 break;
cd8ae852
ED
2873 case TCP_SAVE_SYN:
2874 val = tp->save_syn;
2875 break;
2876 case TCP_SAVED_SYN: {
2877 if (get_user(len, optlen))
2878 return -EFAULT;
2879
2880 lock_sock(sk);
2881 if (tp->saved_syn) {
aea0929e
EM
2882 if (len < tp->saved_syn[0]) {
2883 if (put_user(tp->saved_syn[0], optlen)) {
2884 release_sock(sk);
2885 return -EFAULT;
2886 }
2887 release_sock(sk);
2888 return -EINVAL;
2889 }
2890 len = tp->saved_syn[0];
cd8ae852
ED
2891 if (put_user(len, optlen)) {
2892 release_sock(sk);
2893 return -EFAULT;
2894 }
2895 if (copy_to_user(optval, tp->saved_syn + 1, len)) {
2896 release_sock(sk);
2897 return -EFAULT;
2898 }
2899 tcp_saved_syn_free(tp);
2900 release_sock(sk);
2901 } else {
2902 release_sock(sk);
2903 len = 0;
2904 if (put_user(len, optlen))
2905 return -EFAULT;
2906 }
2907 return 0;
2908 }
1da177e4
LT
2909 default:
2910 return -ENOPROTOOPT;
3ff50b79 2911 }
1da177e4
LT
2912
2913 if (put_user(len, optlen))
2914 return -EFAULT;
2915 if (copy_to_user(optval, &val, len))
2916 return -EFAULT;
2917 return 0;
2918}
2919
3fdadf7d
DM
2920int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval,
2921 int __user *optlen)
2922{
2923 struct inet_connection_sock *icsk = inet_csk(sk);
2924
2925 if (level != SOL_TCP)
2926 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
2927 optval, optlen);
2928 return do_tcp_getsockopt(sk, level, optname, optval, optlen);
2929}
4bc2f18b 2930EXPORT_SYMBOL(tcp_getsockopt);
3fdadf7d
DM
2931
2932#ifdef CONFIG_COMPAT
543d9cfe
ACM
2933int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
2934 char __user *optval, int __user *optlen)
3fdadf7d 2935{
dec73ff0
ACM
2936 if (level != SOL_TCP)
2937 return inet_csk_compat_getsockopt(sk, level, optname,
2938 optval, optlen);
3fdadf7d
DM
2939 return do_tcp_getsockopt(sk, level, optname, optval, optlen);
2940}
543d9cfe 2941EXPORT_SYMBOL(compat_tcp_getsockopt);
3fdadf7d 2942#endif
1da177e4 2943
cfb6eeb4 2944#ifdef CONFIG_TCP_MD5SIG
349ce993 2945static DEFINE_PER_CPU(struct tcp_md5sig_pool, tcp_md5sig_pool);
71cea17e 2946static DEFINE_MUTEX(tcp_md5sig_mutex);
349ce993 2947static bool tcp_md5sig_pool_populated = false;
cfb6eeb4 2948
71cea17e 2949static void __tcp_alloc_md5sig_pool(void)
cfb6eeb4
YH
2950{
2951 int cpu;
cfb6eeb4
YH
2952
2953 for_each_possible_cpu(cpu) {
349ce993
ED
2954 if (!per_cpu(tcp_md5sig_pool, cpu).md5_desc.tfm) {
2955 struct crypto_hash *hash;
cfb6eeb4 2956
349ce993
ED
2957 hash = crypto_alloc_hash("md5", 0, CRYPTO_ALG_ASYNC);
2958 if (IS_ERR_OR_NULL(hash))
2959 return;
2960 per_cpu(tcp_md5sig_pool, cpu).md5_desc.tfm = hash;
2961 }
cfb6eeb4 2962 }
349ce993
ED
2963 /* before setting tcp_md5sig_pool_populated, we must commit all writes
2964 * to memory. See smp_rmb() in tcp_get_md5sig_pool()
71cea17e
ED
2965 */
2966 smp_wmb();
349ce993 2967 tcp_md5sig_pool_populated = true;
cfb6eeb4
YH
2968}
2969
71cea17e 2970bool tcp_alloc_md5sig_pool(void)
cfb6eeb4 2971{
349ce993 2972 if (unlikely(!tcp_md5sig_pool_populated)) {
71cea17e
ED
2973 mutex_lock(&tcp_md5sig_mutex);
2974
349ce993 2975 if (!tcp_md5sig_pool_populated)
71cea17e
ED
2976 __tcp_alloc_md5sig_pool();
2977
2978 mutex_unlock(&tcp_md5sig_mutex);
cfb6eeb4 2979 }
349ce993 2980 return tcp_md5sig_pool_populated;
cfb6eeb4 2981}
cfb6eeb4
YH
2982EXPORT_SYMBOL(tcp_alloc_md5sig_pool);
2983
35790c04
ED
2984
2985/**
2986 * tcp_get_md5sig_pool - get md5sig_pool for this user
2987 *
2988 * We use percpu structure, so if we succeed, we exit with preemption
2989 * and BH disabled, to make sure another thread or softirq handling
2990 * wont try to get same context.
2991 */
2992struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
cfb6eeb4 2993{
35790c04 2994 local_bh_disable();
cfb6eeb4 2995
349ce993
ED
2996 if (tcp_md5sig_pool_populated) {
2997 /* coupled with smp_wmb() in __tcp_alloc_md5sig_pool() */
2998 smp_rmb();
2999 return this_cpu_ptr(&tcp_md5sig_pool);
3000 }
35790c04
ED
3001 local_bh_enable();
3002 return NULL;
3003}
3004EXPORT_SYMBOL(tcp_get_md5sig_pool);
cfb6eeb4 3005
49a72dfb 3006int tcp_md5_hash_header(struct tcp_md5sig_pool *hp,
ca35a0ef 3007 const struct tcphdr *th)
49a72dfb
AL
3008{
3009 struct scatterlist sg;
ca35a0ef 3010 struct tcphdr hdr;
49a72dfb
AL
3011 int err;
3012
ca35a0ef
ED
3013 /* We are not allowed to change tcphdr, make a local copy */
3014 memcpy(&hdr, th, sizeof(hdr));
3015 hdr.check = 0;
3016
49a72dfb 3017 /* options aren't included in the hash */
ca35a0ef
ED
3018 sg_init_one(&sg, &hdr, sizeof(hdr));
3019 err = crypto_hash_update(&hp->md5_desc, &sg, sizeof(hdr));
49a72dfb
AL
3020 return err;
3021}
49a72dfb
AL
3022EXPORT_SYMBOL(tcp_md5_hash_header);
3023
3024int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp,
cf533ea5 3025 const struct sk_buff *skb, unsigned int header_len)
49a72dfb
AL
3026{
3027 struct scatterlist sg;
3028 const struct tcphdr *tp = tcp_hdr(skb);
3029 struct hash_desc *desc = &hp->md5_desc;
95c96174
ED
3030 unsigned int i;
3031 const unsigned int head_data_len = skb_headlen(skb) > header_len ?
3032 skb_headlen(skb) - header_len : 0;
49a72dfb 3033 const struct skb_shared_info *shi = skb_shinfo(skb);
d7fd1b57 3034 struct sk_buff *frag_iter;
49a72dfb
AL
3035
3036 sg_init_table(&sg, 1);
3037
3038 sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len);
3039 if (crypto_hash_update(desc, &sg, head_data_len))
3040 return 1;
3041
3042 for (i = 0; i < shi->nr_frags; ++i) {
3043 const struct skb_frag_struct *f = &shi->frags[i];
54d27fcb
ED
3044 unsigned int offset = f->page_offset;
3045 struct page *page = skb_frag_page(f) + (offset >> PAGE_SHIFT);
3046
3047 sg_set_page(&sg, page, skb_frag_size(f),
3048 offset_in_page(offset));
9e903e08 3049 if (crypto_hash_update(desc, &sg, skb_frag_size(f)))
49a72dfb
AL
3050 return 1;
3051 }
3052
d7fd1b57
ED
3053 skb_walk_frags(skb, frag_iter)
3054 if (tcp_md5_hash_skb_data(hp, frag_iter, 0))
3055 return 1;
3056
49a72dfb
AL
3057 return 0;
3058}
49a72dfb
AL
3059EXPORT_SYMBOL(tcp_md5_hash_skb_data);
3060
cf533ea5 3061int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, const struct tcp_md5sig_key *key)
49a72dfb
AL
3062{
3063 struct scatterlist sg;
3064
3065 sg_init_one(&sg, key->key, key->keylen);
3066 return crypto_hash_update(&hp->md5_desc, &sg, key->keylen);
3067}
49a72dfb
AL
3068EXPORT_SYMBOL(tcp_md5_hash_key);
3069
cfb6eeb4
YH
3070#endif
3071
4ac02bab
AK
3072void tcp_done(struct sock *sk)
3073{
8336886f
JC
3074 struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
3075
5a5f3a8d 3076 if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
63231bdd 3077 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
4ac02bab
AK
3078
3079 tcp_set_state(sk, TCP_CLOSE);
3080 tcp_clear_xmit_timers(sk);
00db4124 3081 if (req)
8336886f 3082 reqsk_fastopen_remove(sk, req, false);
4ac02bab
AK
3083
3084 sk->sk_shutdown = SHUTDOWN_MASK;
3085
3086 if (!sock_flag(sk, SOCK_DEAD))
3087 sk->sk_state_change(sk);
3088 else
3089 inet_csk_destroy_sock(sk);
3090}
3091EXPORT_SYMBOL_GPL(tcp_done);
3092
c1e64e29
LC
3093int tcp_abort(struct sock *sk, int err)
3094{
3095 if (!sk_fullsock(sk)) {
07f6f4a3
ED
3096 if (sk->sk_state == TCP_NEW_SYN_RECV) {
3097 struct request_sock *req = inet_reqsk(sk);
3098
3099 local_bh_disable();
3100 inet_csk_reqsk_queue_drop_and_put(req->rsk_listener,
3101 req);
3102 local_bh_enable();
3103 return 0;
3104 }
c1e64e29
LC
3105 sock_gen_put(sk);
3106 return -EOPNOTSUPP;
3107 }
3108
3109 /* Don't race with userspace socket closes such as tcp_close. */
3110 lock_sock(sk);
3111
2010b93e
LC
3112 if (sk->sk_state == TCP_LISTEN) {
3113 tcp_set_state(sk, TCP_CLOSE);
3114 inet_csk_listen_stop(sk);
3115 }
3116
c1e64e29
LC
3117 /* Don't race with BH socket closes such as inet_csk_listen_stop. */
3118 local_bh_disable();
3119 bh_lock_sock(sk);
3120
3121 if (!sock_flag(sk, SOCK_DEAD)) {
3122 sk->sk_err = err;
3123 /* This barrier is coupled with smp_rmb() in tcp_poll() */
3124 smp_wmb();
3125 sk->sk_error_report(sk);
3126 if (tcp_need_reset(sk->sk_state))
3127 tcp_send_active_reset(sk, GFP_ATOMIC);
3128 tcp_done(sk);
3129 }
3130
3131 bh_unlock_sock(sk);
3132 local_bh_enable();
3133 release_sock(sk);
3134 sock_put(sk);
3135 return 0;
3136}
3137EXPORT_SYMBOL_GPL(tcp_abort);
3138
5f8ef48d 3139extern struct tcp_congestion_ops tcp_reno;
1da177e4
LT
3140
3141static __initdata unsigned long thash_entries;
3142static int __init set_thash_entries(char *str)
3143{
413c27d8
EZ
3144 ssize_t ret;
3145
1da177e4
LT
3146 if (!str)
3147 return 0;
413c27d8
EZ
3148
3149 ret = kstrtoul(str, 0, &thash_entries);
3150 if (ret)
3151 return 0;
3152
1da177e4
LT
3153 return 1;
3154}
3155__setup("thash_entries=", set_thash_entries);
3156
47d7a88c 3157static void __init tcp_init_mem(void)
4acb4190 3158{
b66e91cc
ED
3159 unsigned long limit = nr_free_buffer_pages() / 16;
3160
4acb4190 3161 limit = max(limit, 128UL);
b66e91cc
ED
3162 sysctl_tcp_mem[0] = limit / 4 * 3; /* 4.68 % */
3163 sysctl_tcp_mem[1] = limit; /* 6.25 % */
3164 sysctl_tcp_mem[2] = sysctl_tcp_mem[0] * 2; /* 9.37 % */
4acb4190
GC
3165}
3166
1da177e4
LT
3167void __init tcp_init(void)
3168{
f03d78db 3169 unsigned long limit;
b49960a0 3170 int max_rshare, max_wshare, cnt;
074b8517 3171 unsigned int i;
1da177e4 3172
b4772ef8 3173 sock_skb_cb_check_size(sizeof(struct tcp_skb_cb));
1da177e4 3174
908c7f19
TH
3175 percpu_counter_init(&tcp_sockets_allocated, 0, GFP_KERNEL);
3176 percpu_counter_init(&tcp_orphan_count, 0, GFP_KERNEL);
6e04e021
ACM
3177 tcp_hashinfo.bind_bucket_cachep =
3178 kmem_cache_create("tcp_bind_bucket",
3179 sizeof(struct inet_bind_bucket), 0,
20c2df83 3180 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1da177e4 3181
1da177e4
LT
3182 /* Size and allocate the main established and bind bucket
3183 * hash tables.
3184 *
3185 * The methodology is similar to that of the buffer cache.
3186 */
6e04e021 3187 tcp_hashinfo.ehash =
1da177e4 3188 alloc_large_system_hash("TCP established",
0f7ff927 3189 sizeof(struct inet_ehash_bucket),
1da177e4 3190 thash_entries,
fd90b29d 3191 17, /* one slot per 128 KB of memory */
9e950efa 3192 0,
1da177e4 3193 NULL,
f373b53b 3194 &tcp_hashinfo.ehash_mask,
31fe62b9 3195 0,
0ccfe618 3196 thash_entries ? 0 : 512 * 1024);
05dbc7b5 3197 for (i = 0; i <= tcp_hashinfo.ehash_mask; i++)
3ab5aee7 3198 INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i);
05dbc7b5 3199
230140cf
ED
3200 if (inet_ehash_locks_alloc(&tcp_hashinfo))
3201 panic("TCP: failed to alloc ehash_locks");
6e04e021 3202 tcp_hashinfo.bhash =
1da177e4 3203 alloc_large_system_hash("TCP bind",
0f7ff927 3204 sizeof(struct inet_bind_hashbucket),
f373b53b 3205 tcp_hashinfo.ehash_mask + 1,
fd90b29d 3206 17, /* one slot per 128 KB of memory */
9e950efa 3207 0,
6e04e021 3208 &tcp_hashinfo.bhash_size,
1da177e4 3209 NULL,
31fe62b9 3210 0,
1da177e4 3211 64 * 1024);
074b8517 3212 tcp_hashinfo.bhash_size = 1U << tcp_hashinfo.bhash_size;
6e04e021
ACM
3213 for (i = 0; i < tcp_hashinfo.bhash_size; i++) {
3214 spin_lock_init(&tcp_hashinfo.bhash[i].lock);
3215 INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain);
1da177e4
LT
3216 }
3217
c5ed63d6
ED
3218
3219 cnt = tcp_hashinfo.ehash_mask + 1;
3220
3221 tcp_death_row.sysctl_max_tw_buckets = cnt / 2;
3222 sysctl_tcp_max_orphans = cnt / 2;
3223 sysctl_max_syn_backlog = max(128, cnt / 256);
1da177e4 3224
a4fe34bf 3225 tcp_init_mem();
c43b874d 3226 /* Set per-socket limits to no more than 1/128 the pressure threshold */
5fb84b14 3227 limit = nr_free_buffer_pages() << (PAGE_SHIFT - 7);
b49960a0
ED
3228 max_wshare = min(4UL*1024*1024, limit);
3229 max_rshare = min(6UL*1024*1024, limit);
7b4f4b5e 3230
3ab224be 3231 sysctl_tcp_wmem[0] = SK_MEM_QUANTUM;
7b4f4b5e 3232 sysctl_tcp_wmem[1] = 16*1024;
b49960a0 3233 sysctl_tcp_wmem[2] = max(64*1024, max_wshare);
7b4f4b5e 3234
3ab224be 3235 sysctl_tcp_rmem[0] = SK_MEM_QUANTUM;
7b4f4b5e 3236 sysctl_tcp_rmem[1] = 87380;
b49960a0 3237 sysctl_tcp_rmem[2] = max(87380, max_rshare);
1da177e4 3238
afd46503 3239 pr_info("Hash tables configured (established %u bind %u)\n",
058bd4d2 3240 tcp_hashinfo.ehash_mask + 1, tcp_hashinfo.bhash_size);
317a76f9 3241
51c5d0c4 3242 tcp_metrics_init();
55d8694f 3243 BUG_ON(tcp_register_congestion_control(&tcp_reno) != 0);
46d3ceab 3244 tcp_tasklet_init();
1da177e4 3245}