projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
liquidio: change manner of detecting whether or not NIC firmware is loaded
[linux-2.6-block.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
cf80e0e4 250#include <crypto/hash.h>
172589cc 251#include <linux/kernel.h>
1da177e4
LT		 252#include <linux/module.h>
253#include <linux/types.h>
254#include <linux/fcntl.h>
255#include <linux/poll.h>
6e9250f5 256#include <linux/inet_diag.h>
1da177e4 257#include <linux/init.h>
1da177e4 258#include <linux/fs.h>
9c55e01c 259#include <linux/skbuff.h>
81b23b4a 260#include <linux/scatterlist.h>
9c55e01c
JA		 261#include <linux/splice.h>
262#include <linux/net.h>
263#include <linux/socket.h>
1da177e4
LT		 264#include <linux/random.h>
265#include <linux/bootmem.h>
57413ebc
MS		 266#include <linux/highmem.h>
267#include <linux/swap.h>
b8059ead 268#include <linux/cache.h>
f4c50d99 269#include <linux/err.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 279
7c0f6ba6 280#include <linux/uaccess.h>
1da177e4 281#include <asm/ioctls.h>
076bb0c8 282#include <net/busy_poll.h>
1da177e4 283
95bd09eb
ED		 284int sysctl_tcp_min_tso_segs __read_mostly = 2;
285
f54b3111
ED		 286int sysctl_tcp_autocorking __read_mostly = 1;
287
dd24c001 288struct percpu_counter tcp_orphan_count;
0a5578cf
ACM		 289EXPORT_SYMBOL_GPL(tcp_orphan_count);
290
a4fe34bf 291long sysctl_tcp_mem[3] __read_mostly;
b8059ead
DM		 292int sysctl_tcp_wmem[3] __read_mostly;
293int sysctl_tcp_rmem[3] __read_mostly;
1da177e4 294
a4fe34bf 295EXPORT_SYMBOL(sysctl_tcp_mem);
1da177e4
LT		 296EXPORT_SYMBOL(sysctl_tcp_rmem);
297EXPORT_SYMBOL(sysctl_tcp_wmem);
298
8d987e5c 299atomic_long_t tcp_memory_allocated; /* Current allocated memory. */
1da177e4 300EXPORT_SYMBOL(tcp_memory_allocated);
1748376b
ED		 301
302/*
303 * Current number of TCP sockets.
304 */
305struct percpu_counter tcp_sockets_allocated;
1da177e4
LT		 306EXPORT_SYMBOL(tcp_sockets_allocated);
307
9c55e01c
JA		 308/*
309 * TCP splice context
310 */
311struct tcp_splice_state {
312 struct pipe_inode_info *pipe;
313 size_t len;
314 unsigned int flags;
315};
316
1da177e4
LT		 317/*
318 * Pressure flag: try to collapse.
319 * Technical note: it is used by multiple contexts non atomically.
3ab224be 320 * All the __sk_mem_schedule() is of this nature: accounting
1da177e4
LT		 321 * is strict, actions are advisory and have some latency.
322 */
06044751
ED		 323unsigned long tcp_memory_pressure __read_mostly;
324EXPORT_SYMBOL_GPL(tcp_memory_pressure);
1da177e4 325
5c52ba17 326void tcp_enter_memory_pressure(struct sock *sk)
1da177e4 327{
06044751
ED		 328 unsigned long val;
329
330 if (tcp_memory_pressure)
331 return;
332 val = jiffies;
333
334 if (!val)
335 val--;
336 if (!cmpxchg(&tcp_memory_pressure, 0, val))
4e673444 337 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES);
1da177e4 338}
06044751
ED		 339EXPORT_SYMBOL_GPL(tcp_enter_memory_pressure);
340
341void tcp_leave_memory_pressure(struct sock *sk)
342{
343 unsigned long val;
344
345 if (!tcp_memory_pressure)
346 return;
347 val = xchg(&tcp_memory_pressure, 0);
348 if (val)
349 NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURESCHRONO,
350 jiffies_to_msecs(jiffies - val));
351}
352EXPORT_SYMBOL_GPL(tcp_leave_memory_pressure);
1da177e4 353
b103cf34
JA		 354/* Convert seconds to retransmits based on initial and max timeout */
355static u8 secs_to_retrans(int seconds, int timeout, int rto_max)
356{
357 u8 res = 0;
358
359 if (seconds > 0) {
360 int period = timeout;
361
362 res = 1;
363 while (seconds > period && res < 255) {
364 res++;
365 timeout <<= 1;
366 if (timeout > rto_max)
367 timeout = rto_max;
368 period += timeout;
369 }
370 }
371 return res;
372}
373
374/* Convert retransmits to seconds based on initial and max timeout */
375static int retrans_to_secs(u8 retrans, int timeout, int rto_max)
376{
377 int period = 0;
378
379 if (retrans > 0) {
380 period = timeout;
381 while (--retrans) {
382 timeout <<= 1;
383 if (timeout > rto_max)
384 timeout = rto_max;
385 period += timeout;
386 }
387 }
388 return period;
389}
390
0263598c
WW		 391static u64 tcp_compute_delivery_rate(const struct tcp_sock *tp)
392{
393 u32 rate = READ_ONCE(tp->rate_delivered);
394 u32 intv = READ_ONCE(tp->rate_interval_us);
395 u64 rate64 = 0;
396
397 if (rate && intv) {
398 rate64 = (u64)rate * tp->mss_cache * USEC_PER_SEC;
399 do_div(rate64, intv);
400 }
401 return rate64;
402}
403
900f65d3
NC		 404/* Address-family independent initialization for a tcp_sock.
405 *
406 * NOTE: A lot of things set to zero explicitly by call to
407 * sk_alloc() so need not be done here.
408 */
409void tcp_init_sock(struct sock *sk)
410{
411 struct inet_connection_sock *icsk = inet_csk(sk);
412 struct tcp_sock *tp = tcp_sk(sk);
413
9f5afeae 414 tp->out_of_order_queue = RB_ROOT;
900f65d3 415 tcp_init_xmit_timers(sk);
46d3ceab 416 INIT_LIST_HEAD(&tp->tsq_node);
900f65d3
NC		 417
418 icsk->icsk_rto = TCP_TIMEOUT_INIT;
740b0f18 419 tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
ac9517fc 420 minmax_reset(&tp->rtt_min, tcp_jiffies32, ~0U);
900f65d3
NC		 421
422 /* So many TCP implementations out there (incorrectly) count the
423 * initial SYN frame in their delayed-ACK and congestion control
424 * algorithms that we must have the following bandaid to talk
425 * efficiently to them. -DaveM
426 */
427 tp->snd_cwnd = TCP_INIT_CWND;
428
d7722e85
SHY		 429 /* There's a bubble in the pipe until at least the first ACK. */
430 tp->app_limited = ~0U;
431
900f65d3
NC		 432 /* See draft-stevens-tcpca-spec-01 for discussion of the
433 * initialization of these values.
434 */
435 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
436 tp->snd_cwnd_clamp = ~0;
437 tp->mss_cache = TCP_MSS_DEFAULT;
438
1043e25f 439 tp->reordering = sock_net(sk)->ipv4.sysctl_tcp_reordering;
55d8694f 440 tcp_assign_congestion_control(sk);
900f65d3 441
ceaa1fef
AV		 442 tp->tsoffset = 0;
443
900f65d3
NC		 444 sk->sk_state = TCP_CLOSE;
445
446 sk->sk_write_space = sk_stream_write_space;
447 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
448
449 icsk->icsk_sync_mss = tcp_sync_mss;
450
900f65d3
NC		 451 sk->sk_sndbuf = sysctl_tcp_wmem[1];
452 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
453
900f65d3 454 sk_sockets_allocated_inc(sk);
900f65d3
NC		 455}
456EXPORT_SYMBOL(tcp_init_sock);
457
c14ac945 458static void tcp_tx_timestamp(struct sock *sk, u16 tsflags, struct sk_buff *skb)
4ed2d765 459{
ad02c4f5 460 if (tsflags && skb) {
f066e2b0 461 struct skb_shared_info *shinfo = skb_shinfo(skb);
6b084928 462 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
4ed2d765 463
c14ac945 464 sock_tx_timestamp(sk, tsflags, &shinfo->tx_flags);
0a2cf20c
SHY		 465 if (tsflags & SOF_TIMESTAMPING_TX_ACK)
466 tcb->txstamp_ack = 1;
467 if (tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
f066e2b0
WB		 468 shinfo->tskey = TCP_SKB_CB(skb)->seq + skb->len - 1;
469 }
4ed2d765
WB		 470}
471
1da177e4
LT		 472/*
473 * Wait for a TCP event.
474 *
475 * Note that we don't need to lock the socket, as the upper poll layers
476 * take care of normal races (between the test and the event) and we don't
477 * go look at any of the socket buffers directly.
478 */
479unsigned int tcp_poll(struct file *file, struct socket *sock, poll_table *wait)
480{
481 unsigned int mask;
482 struct sock *sk = sock->sk;
cf533ea5 483 const struct tcp_sock *tp = tcp_sk(sk);
00fd38d9 484 int state;
1da177e4 485
c3f1dbaf
DM		 486 sock_rps_record_flow(sk);
487
aa395145 488 sock_poll_wait(file, sk_sleep(sk), wait);
00fd38d9
ED		 489
490 state = sk_state_load(sk);
491 if (state == TCP_LISTEN)
dc40c7bc 492 return inet_csk_listen_poll(sk);
1da177e4
LT		 493
494 /* Socket is not locked. We are protected from async events
70efce27
WN		 495 * by poll logic and correct handling of state changes
496 * made by other threads is impossible in any case.
1da177e4
LT		 497 */
498
499 mask = 0;
1da177e4
LT		 500
501 /*
502 * POLLHUP is certainly not done right. But poll() doesn't
503 * have a notion of HUP in just one direction, and for a
504 * socket the read side is more interesting.
505 *
506 * Some poll() documentation says that POLLHUP is incompatible
507 * with the POLLOUT/POLLWR flags, so somebody should check this
508 * all. But careful, it tends to be safer to return too many
509 * bits than too few, and you can easily break real applications
510 * if you don't tell them that something has hung up!
511 *
512 * Check-me.
513 *
514 * Check number 1. POLLHUP is _UNMASKABLE_ event (see UNIX98 and
515 * our fs/select.c). It means that after we received EOF,
516 * poll always returns immediately, making impossible poll() on write()
517 * in state CLOSE_WAIT. One solution is evident --- to set POLLHUP
518 * if and only if shutdown has been made in both directions.
519 * Actually, it is interesting to look how Solaris and DUX
70efce27 520 * solve this dilemma. I would prefer, if POLLHUP were maskable,
1da177e4
LT		 521 * then we could set it on SND_SHUTDOWN. BTW examples given
522 * in Stevens' books assume exactly this behaviour, it explains
70efce27 523 * why POLLHUP is incompatible with POLLOUT. --ANK
1da177e4
LT		 524 *
525 * NOTE. Check for TCP_CLOSE is added. The goal is to prevent
526 * blocking on fresh not-connected or disconnected socket. --ANK
527 */
00fd38d9 528 if (sk->sk_shutdown == SHUTDOWN_MASK || state == TCP_CLOSE)
1da177e4
LT		 529 mask |= POLLHUP;
530 if (sk->sk_shutdown & RCV_SHUTDOWN)
f348d70a 531 mask |= POLLIN | POLLRDNORM | POLLRDHUP;
1da177e4 532
8336886f 533 /* Connected or passive Fast Open socket? */
00fd38d9
ED		 534 if (state != TCP_SYN_SENT &&
535 (state != TCP_SYN_RECV || tp->fastopen_rsk)) {
c7004482
DM		 536 int target = sock_rcvlowat(sk, 0, INT_MAX);
537
538 if (tp->urg_seq == tp->copied_seq &&
539 !sock_flag(sk, SOCK_URGINLINE) &&
540 tp->urg_data)
b634f875 541 target++;
c7004482 542
c7004482 543 if (tp->rcv_nxt - tp->copied_seq >= target)
1da177e4
LT		 544 mask |= POLLIN | POLLRDNORM;
545
546 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
64dc6130 547 if (sk_stream_is_writeable(sk)) {
1da177e4
LT		 548 mask |= POLLOUT | POLLWRNORM;
549 } else { /* send SIGIO later */
9cd3e072 550 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1da177e4
LT		 551 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
552
553 /* Race breaker. If space is freed after
554 * wspace test but before the flags are set,
3c715127 555 * IO signal will be lost. Memory barrier
556 * pairs with the input side.
1da177e4 557 */
3c715127 558 smp_mb__after_atomic();
64dc6130 559 if (sk_stream_is_writeable(sk))
1da177e4
LT		 560 mask |= POLLOUT | POLLWRNORM;
561 }
d84ba638
KM		 562 } else
563 mask |= POLLOUT | POLLWRNORM;
1da177e4
LT		 564
565 if (tp->urg_data & TCP_URG_VALID)
566 mask |= POLLPRI;
d68be71e 567 } else if (state == TCP_SYN_SENT && inet_sk(sk)->defer_connect) {
19f6d3f3
WW		 568 /* Active TCP fastopen socket with defer_connect
569 * Return POLLOUT so application can call write()
570 * in order for kernel to generate SYN+data
571 */
572 mask |= POLLOUT | POLLWRNORM;
1da177e4 573 }
a4d25803
TM		 574 /* This barrier is coupled with smp_wmb() in tcp_reset() */
575 smp_rmb();
4ed2d765 576 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
a4d25803
TM		 577 mask |= POLLERR;
578
1da177e4
LT		 579 return mask;
580}
4bc2f18b 581EXPORT_SYMBOL(tcp_poll);
1da177e4
LT		 582
583int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
584{
585 struct tcp_sock *tp = tcp_sk(sk);
586 int answ;
0e71c55c 587 bool slow;
1da177e4
LT		 588
589 switch (cmd) {
590 case SIOCINQ:
591 if (sk->sk_state == TCP_LISTEN)
592 return -EINVAL;
593
0e71c55c 594 slow = lock_sock_fast(sk);
473bd239 595 answ = tcp_inq(sk);
0e71c55c 596 unlock_sock_fast(sk, slow);
1da177e4
LT		 597 break;
598 case SIOCATMARK:
599 answ = tp->urg_data && tp->urg_seq == tp->copied_seq;
600 break;
601 case SIOCOUTQ:
602 if (sk->sk_state == TCP_LISTEN)
603 return -EINVAL;
604
605 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
606 answ = 0;
607 else
608 answ = tp->write_seq - tp->snd_una;
609 break;
2f4e1b39
MS		 610 case SIOCOUTQNSD:
611 if (sk->sk_state == TCP_LISTEN)
612 return -EINVAL;
613
614 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
615 answ = 0;
616 else
617 answ = tp->write_seq - tp->snd_nxt;
618 break;
1da177e4
LT		 619 default:
620 return -ENOIOCTLCMD;
3ff50b79 621 }
1da177e4
LT		 622
623 return put_user(answ, (int __user *)arg);
624}
4bc2f18b 625EXPORT_SYMBOL(tcp_ioctl);
1da177e4 626
1da177e4
LT		 627static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb)
628{
4de075e0 629 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
1da177e4
LT		 630 tp->pushed_seq = tp->write_seq;
631}
632
a2a385d6 633static inline bool forced_push(const struct tcp_sock *tp)
1da177e4
LT		 634{
635 return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1));
636}
637
f4a775d1 638static void skb_entail(struct sock *sk, struct sk_buff *skb)
1da177e4 639{
9e412ba7 640 struct tcp_sock *tp = tcp_sk(sk);
352d4800
ACM		 641 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
642
643 skb->csum = 0;
644 tcb->seq = tcb->end_seq = tp->write_seq;
4de075e0 645 tcb->tcp_flags = TCPHDR_ACK;
352d4800 646 tcb->sacked = 0;
f4a775d1 647 __skb_header_release(skb);
fe067e8a 648 tcp_add_write_queue_tail(sk, skb);
3ab224be
HA		 649 sk->sk_wmem_queued += skb->truesize;
650 sk_mem_charge(sk, skb->truesize);
89ebd197 651 if (tp->nonagle & TCP_NAGLE_PUSH)
e905a9ed 652 tp->nonagle &= ~TCP_NAGLE_PUSH;
6f021c62
ED		 653
654 tcp_slow_start_after_idle_check(sk);
1da177e4
LT		 655}
656
afeca340 657static inline void tcp_mark_urg(struct tcp_sock *tp, int flags)
1da177e4 658{
33f5f57e 659 if (flags & MSG_OOB)
1da177e4 660 tp->snd_up = tp->write_seq;
1da177e4
LT		 661}
662
f54b3111 663/* If a not yet filled skb is pushed, do not send it if
a181ceb5 664 * we have data packets in Qdisc or NIC queues :
f54b3111
ED		 665 * Because TX completion will happen shortly, it gives a chance
666 * to coalesce future sendmsg() payload into this skb, without
667 * need for a timer, and with no latency trade off.
668 * As packets containing data payload have a bigger truesize
a181ceb5
ED		 669 * than pure acks (dataless) packets, the last checks prevent
670 * autocorking if we only have an ACK in Qdisc/NIC queues,
671 * or if TX completion was delayed after we processed ACK packet.
f54b3111
ED		 672 */
673static bool tcp_should_autocork(struct sock *sk, struct sk_buff *skb,
674 int size_goal)
1da177e4 675{
f54b3111
ED		 676 return skb->len < size_goal &&
677 sysctl_tcp_autocorking &&
a181ceb5 678 skb != tcp_write_queue_head(sk) &&
14afee4b 679 refcount_read(&sk->sk_wmem_alloc) > skb->truesize;
f54b3111
ED		 680}
681
682static void tcp_push(struct sock *sk, int flags, int mss_now,
683 int nonagle, int size_goal)
684{
685 struct tcp_sock *tp = tcp_sk(sk);
686 struct sk_buff *skb;
afeca340 687
f54b3111
ED		 688 if (!tcp_send_head(sk))
689 return;
afeca340 690
f54b3111
ED		 691 skb = tcp_write_queue_tail(sk);
692 if (!(flags & MSG_MORE) || forced_push(tp))
693 tcp_mark_push(tp, skb);
694
695 tcp_mark_urg(tp, flags);
696
697 if (tcp_should_autocork(sk, skb, size_goal)) {
698
699 /* avoid atomic op if TSQ_THROTTLED bit is already set */
7aa5470c 700 if (!test_bit(TSQ_THROTTLED, &sk->sk_tsq_flags)) {
f54b3111 701 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAUTOCORKING);
7aa5470c 702 set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags);
f54b3111 703 }
a181ceb5
ED		 704 /* It is possible TX completion already happened
705 * before we set TSQ_THROTTLED.
706 */
14afee4b 707 if (refcount_read(&sk->sk_wmem_alloc) > skb->truesize)
a181ceb5 708 return;
1da177e4 709 }
f54b3111
ED		 710
711 if (flags & MSG_MORE)
712 nonagle = TCP_NAGLE_CORK;
713
714 __tcp_push_pending_frames(sk, mss_now, nonagle);
1da177e4
LT		 715}
716
6ff7751d
AB		 717static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
718 unsigned int offset, size_t len)
9c55e01c
JA		 719{
720 struct tcp_splice_state *tss = rd_desc->arg.data;
33966dd0 721 int ret;
9c55e01c 722
a60e3cc7 723 ret = skb_splice_bits(skb, skb->sk, offset, tss->pipe,
25869262 724 min(rd_desc->count, len), tss->flags);
33966dd0
WT		 725 if (ret > 0)
726 rd_desc->count -= ret;
727 return ret;
9c55e01c
JA		 728}
729
730static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss)
731{
732 /* Store TCP splice context information in read_descriptor_t. */
733 read_descriptor_t rd_desc = {
734 .arg.data = tss,
33966dd0 735 .count = tss->len,
9c55e01c
JA		 736 };
737
738 return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv);
739}
740
741/**
742 * tcp_splice_read - splice data from TCP socket to a pipe
743 * @sock: socket to splice from
744 * @ppos: position (not valid)
745 * @pipe: pipe to splice to
746 * @len: number of bytes to splice
747 * @flags: splice modifier flags
748 *
749 * Description:
750 * Will read pages from given socket and fill them into a pipe.
751 *
752 **/
753ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos,
754 struct pipe_inode_info *pipe, size_t len,
755 unsigned int flags)
756{
757 struct sock *sk = sock->sk;
758 struct tcp_splice_state tss = {
759 .pipe = pipe,
760 .len = len,
761 .flags = flags,
762 };
763 long timeo;
764 ssize_t spliced;
765 int ret;
766
3a047bf8 767 sock_rps_record_flow(sk);
9c55e01c
JA		 768 /*
769 * We can't seek on a socket input
770 */
771 if (unlikely(*ppos))
772 return -ESPIPE;
773
774 ret = spliced = 0;
775
776 lock_sock(sk);
777
42324c62 778 timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK);
9c55e01c
JA		 779 while (tss.len) {
780 ret = __tcp_splice_read(sk, &tss);
781 if (ret < 0)
782 break;
783 else if (!ret) {
784 if (spliced)
785 break;
9c55e01c
JA		 786 if (sock_flag(sk, SOCK_DONE))
787 break;
788 if (sk->sk_err) {
789 ret = sock_error(sk);
790 break;
791 }
792 if (sk->sk_shutdown & RCV_SHUTDOWN)
793 break;
794 if (sk->sk_state == TCP_CLOSE) {
795 /*
796 * This occurs when user tries to read
797 * from never connected socket.
798 */
799 if (!sock_flag(sk, SOCK_DONE))
800 ret = -ENOTCONN;
801 break;
802 }
803 if (!timeo) {
804 ret = -EAGAIN;
805 break;
806 }
ccf7abb9
ED		 807 /* if __tcp_splice_read() got nothing while we have
808 * an skb in receive queue, we do not want to loop.
809 * This might happen with URG data.
810 */
811 if (!skb_queue_empty(&sk->sk_receive_queue))
812 break;
dfbafc99 813 sk_wait_data(sk, &timeo, NULL);
9c55e01c
JA		 814 if (signal_pending(current)) {
815 ret = sock_intr_errno(timeo);
816 break;
817 }
818 continue;
819 }
820 tss.len -= ret;
821 spliced += ret;
822
33966dd0
WT		 823 if (!timeo)
824 break;
9c55e01c
JA		 825 release_sock(sk);
826 lock_sock(sk);
827
828 if (sk->sk_err || sk->sk_state == TCP_CLOSE ||
33966dd0 829 (sk->sk_shutdown & RCV_SHUTDOWN) ||
9c55e01c
JA		 830 signal_pending(current))
831 break;
832 }
833
834 release_sock(sk);
835
836 if (spliced)
837 return spliced;
838
839 return ret;
840}
4bc2f18b 841EXPORT_SYMBOL(tcp_splice_read);
9c55e01c 842
eb934478
ED		 843struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
844 bool force_schedule)
f561d0f2
PE		 845{
846 struct sk_buff *skb;
847
848 /* The TCP header must be at least 32-bit aligned. */
849 size = ALIGN(size, 4);
850
8e4d980a
ED		 851 if (unlikely(tcp_under_memory_pressure(sk)))
852 sk_mem_reclaim_partial(sk);
853
f561d0f2 854 skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp);
8e4d980a 855 if (likely(skb)) {
eb934478 856 bool mem_scheduled;
8e4d980a 857
eb934478
ED		 858 if (force_schedule) {
859 mem_scheduled = true;
8e4d980a
ED		 860 sk_forced_mem_schedule(sk, skb->truesize);
861 } else {
eb934478 862 mem_scheduled = sk_wmem_schedule(sk, skb->truesize);
8e4d980a 863 }
eb934478 864 if (likely(mem_scheduled)) {
a21d4572 865 skb_reserve(skb, sk->sk_prot->max_header);
f561d0f2
PE		 866 /*
867 * Make sure that we have exactly size bytes
868 * available to the caller, no more, no less.
869 */
16fad69c 870 skb->reserved_tailroom = skb->end - skb->tail - size;
f561d0f2
PE		 871 return skb;
872 }
873 __kfree_skb(skb);
874 } else {
5c52ba17 875 sk->sk_prot->enter_memory_pressure(sk);
f561d0f2
PE		 876 sk_stream_moderate_sndbuf(sk);
877 }
878 return NULL;
879}
880
0c54b85f
IJ		 881static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now,
882 int large_allowed)
883{
884 struct tcp_sock *tp = tcp_sk(sk);
6c09fa09 885 u32 new_size_goal, size_goal;
605ad7f1
ED		 886
887 if (!large_allowed || !sk_can_gso(sk))
888 return mss_now;
889
6c09fa09
ED		 890 /* Note : tcp_tso_autosize() will eventually split this later */
891 new_size_goal = sk->sk_gso_max_size - 1 - MAX_TCP_HEADER;
605ad7f1
ED		 892 new_size_goal = tcp_bound_to_half_wnd(tp, new_size_goal);
893
894 /* We try hard to avoid divides here */
895 size_goal = tp->gso_segs * mss_now;
896 if (unlikely(new_size_goal < size_goal ||
897 new_size_goal >= size_goal + mss_now)) {
898 tp->gso_segs = min_t(u16, new_size_goal / mss_now,
899 sk->sk_gso_max_segs);
900 size_goal = tp->gso_segs * mss_now;
0c54b85f
IJ		 901 }
902
605ad7f1 903 return max(size_goal, mss_now);
0c54b85f
IJ		 904}
905
906static int tcp_send_mss(struct sock *sk, int *size_goal, int flags)
907{
908 int mss_now;
909
910 mss_now = tcp_current_mss(sk);
911 *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB));
912
913 return mss_now;
914}
915
e3b5616a
DW		 916ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
917 size_t size, int flags)
1da177e4
LT		 918{
919 struct tcp_sock *tp = tcp_sk(sk);
c1b4a7e6 920 int mss_now, size_goal;
1da177e4
LT		 921 int err;
922 ssize_t copied;
923 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
924
8336886f
JC		 925 /* Wait for a connection to finish. One exception is TCP Fast Open
926 * (passive side) where data is allowed to be sent before a connection
927 * is fully established.
928 */
929 if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
930 !tcp_passive_fastopen(sk)) {
686a5624
YM		 931 err = sk_stream_wait_connect(sk, &timeo);
932 if (err != 0)
1da177e4 933 goto out_err;
8336886f 934 }
1da177e4 935
9cd3e072 936 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1da177e4 937
0c54b85f 938 mss_now = tcp_send_mss(sk, &size_goal, flags);
1da177e4
LT		 939 copied = 0;
940
941 err = -EPIPE;
942 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
0d6a775e 943 goto out_err;
1da177e4 944
64022d0b 945 while (size > 0) {
fe067e8a 946 struct sk_buff *skb = tcp_write_queue_tail(sk);
38ba0a65 947 int copy, i;
38ba0a65 948 bool can_coalesce;
1da177e4 949
c134ecb8
MKL		 950 if (!tcp_send_head(sk) || (copy = size_goal - skb->len) <= 0 ||
951 !tcp_skb_can_collapse_to(skb)) {
1da177e4
LT		 952new_segment:
953 if (!sk_stream_memory_free(sk))
954 goto wait_for_sndbuf;
955
eb934478
ED		 956 skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation,
957 skb_queue_empty(&sk->sk_write_queue));
1da177e4
LT		 958 if (!skb)
959 goto wait_for_memory;
960
9e412ba7 961 skb_entail(sk, skb);
c1b4a7e6 962 copy = size_goal;
1da177e4
LT		 963 }
964
965 if (copy > size)
966 copy = size;
967
968 i = skb_shinfo(skb)->nr_frags;
969 can_coalesce = skb_can_coalesce(skb, i, page, offset);
5f74f82e 970 if (!can_coalesce && i >= sysctl_max_skb_frags) {
1da177e4
LT		 971 tcp_mark_push(tp, skb);
972 goto new_segment;
973 }
3ab224be 974 if (!sk_wmem_schedule(sk, copy))
1da177e4 975 goto wait_for_memory;
e905a9ed 976
1da177e4 977 if (can_coalesce) {
9e903e08 978 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1da177e4
LT		 979 } else {
980 get_page(page);
981 skb_fill_page_desc(skb, i, page, offset, copy);
982 }
c9af6db4 983 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
cef401de 984
1da177e4
LT		 985 skb->len += copy;
986 skb->data_len += copy;
987 skb->truesize += copy;
988 sk->sk_wmem_queued += copy;
3ab224be 989 sk_mem_charge(sk, copy);
84fa7933 990 skb->ip_summed = CHECKSUM_PARTIAL;
1da177e4
LT		 991 tp->write_seq += copy;
992 TCP_SKB_CB(skb)->end_seq += copy;
cd7d8498 993 tcp_skb_pcount_set(skb, 0);
1da177e4
LT		 994
995 if (!copied)
4de075e0 996 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1da177e4
LT		 997
998 copied += copy;
64022d0b 999 offset += copy;
686a5624 1000 size -= copy;
ad02c4f5 1001 if (!size)
1da177e4
LT		 1002 goto out;
1003
69d15067 1004 if (skb->len < size_goal || (flags & MSG_OOB))
1da177e4
LT		 1005 continue;
1006
1007 if (forced_push(tp)) {
1008 tcp_mark_push(tp, skb);
9e412ba7 1009 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
fe067e8a 1010 } else if (skb == tcp_send_head(sk))
1da177e4
LT		 1011 tcp_push_one(sk, mss_now);
1012 continue;
1013
1014wait_for_sndbuf:
1015 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1016wait_for_memory:
f54b3111
ED		 1017 tcp_push(sk, flags & ~MSG_MORE, mss_now,
1018 TCP_NAGLE_PUSH, size_goal);
1da177e4 1019
686a5624
YM		 1020 err = sk_stream_wait_memory(sk, &timeo);
1021 if (err != 0)
1da177e4
LT		 1022 goto do_error;
1023
0c54b85f 1024 mss_now = tcp_send_mss(sk, &size_goal, flags);
1da177e4
LT		 1025 }
1026
1027out:
ad02c4f5
SHY		 1028 if (copied) {
1029 tcp_tx_timestamp(sk, sk->sk_tsflags, tcp_write_queue_tail(sk));
1030 if (!(flags & MSG_SENDPAGE_NOTLAST))
1031 tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
1032 }
1da177e4
LT		 1033 return copied;
1034
1035do_error:
1036 if (copied)
1037 goto out;
1038out_err:
ce5ec440 1039 /* make sure we wake any epoll edge trigger waiter */
b0f71bd3
FY		 1040 if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 &&
1041 err == -EAGAIN)) {
ce5ec440 1042 sk->sk_write_space(sk);
b0f71bd3
FY		 1043 tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED);
1044 }
1da177e4
LT		 1045 return sk_stream_error(sk, flags, err);
1046}
e3b5616a 1047EXPORT_SYMBOL_GPL(do_tcp_sendpages);
1da177e4 1048
306b13eb
TH		 1049int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
1050 size_t size, int flags)
1da177e4 1051{
1da177e4 1052 if (!(sk->sk_route_caps & NETIF_F_SG) ||
9a49850d 1053 !sk_check_csum_caps(sk))
7ba42910
CG		 1054 return sock_no_sendpage(sk->sk_socket, page, offset, size,
1055 flags);
1da177e4 1056
d7722e85
SHY		 1057 tcp_rate_check_app_limited(sk); /* is sending application-limited? */
1058
306b13eb
TH		 1059 return do_tcp_sendpages(sk, page, offset, size, flags);
1060}
774c4673 1061EXPORT_SYMBOL_GPL(tcp_sendpage_locked);
306b13eb
TH		 1062
1063int tcp_sendpage(struct sock *sk, struct page *page, int offset,
1064 size_t size, int flags)
1065{
1066 int ret;
1067
1068 lock_sock(sk);
1069 ret = tcp_sendpage_locked(sk, page, offset, size, flags);
1da177e4 1070 release_sock(sk);
306b13eb
TH		 1071
1072 return ret;
1da177e4 1073}
4bc2f18b 1074EXPORT_SYMBOL(tcp_sendpage);
1da177e4 1075
3613b3db
ED		 1076/* Do not bother using a page frag for very small frames.
1077 * But use this heuristic only for the first skb in write queue.
1078 *
1079 * Having no payload in skb->head allows better SACK shifting
1080 * in tcp_shift_skb_data(), reducing sack/rack overhead, because
1081 * write queue has less skbs.
1082 * Each skb can hold up to MAX_SKB_FRAGS * 32Kbytes, or ~0.5 MB.
1083 * This also speeds up tso_fragment(), since it wont fallback
1084 * to tcp_fragment().
1085 */
1086static int linear_payload_sz(bool first_skb)
1087{
1088 if (first_skb)
1089 return SKB_WITH_OVERHEAD(2048 - MAX_TCP_HEADER);
1090 return 0;
1091}
1092
1093static int select_size(const struct sock *sk, bool sg, bool first_skb)
1da177e4 1094{
cf533ea5 1095 const struct tcp_sock *tp = tcp_sk(sk);
c1b4a7e6 1096 int tmp = tp->mss_cache;
1da177e4 1097
def87cf4 1098 if (sg) {
f07d960d 1099 if (sk_can_gso(sk)) {
3613b3db 1100 tmp = linear_payload_sz(first_skb);
f07d960d 1101 } else {
b4e26f5e
DM		 1102 int pgbreak = SKB_MAX_HEAD(MAX_TCP_HEADER);
1103
1104 if (tmp >= pgbreak &&
1105 tmp <= pgbreak + (MAX_SKB_FRAGS - 1) * PAGE_SIZE)
1106 tmp = pgbreak;
1107 }
1108 }
1da177e4 1109
1da177e4
LT		 1110 return tmp;
1111}
1112
cf60af03
YC		 1113void tcp_free_fastopen_req(struct tcp_sock *tp)
1114{
00db4124 1115 if (tp->fastopen_req) {
cf60af03
YC		 1116 kfree(tp->fastopen_req);
1117 tp->fastopen_req = NULL;
1118 }
1119}
1120
f5ddcbbb
ED		 1121static int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg,
1122 int *copied, size_t size)
cf60af03
YC		 1123{
1124 struct tcp_sock *tp = tcp_sk(sk);
19f6d3f3 1125 struct inet_sock *inet = inet_sk(sk);
ba615f67 1126 struct sockaddr *uaddr = msg->msg_name;
cf60af03
YC		 1127 int err, flags;
1128
ba615f67
WW		 1129 if (!(sysctl_tcp_fastopen & TFO_CLIENT_ENABLE) ||
1130 (uaddr && msg->msg_namelen >= sizeof(uaddr->sa_family) &&
1131 uaddr->sa_family == AF_UNSPEC))
cf60af03 1132 return -EOPNOTSUPP;
00db4124 1133 if (tp->fastopen_req)
cf60af03
YC		 1134 return -EALREADY; /* Another Fast Open is in progress */
1135
1136 tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request),
1137 sk->sk_allocation);
51456b29 1138 if (unlikely(!tp->fastopen_req))
cf60af03
YC		 1139 return -ENOBUFS;
1140 tp->fastopen_req->data = msg;
f5ddcbbb 1141 tp->fastopen_req->size = size;
cf60af03 1142
19f6d3f3
WW		 1143 if (inet->defer_connect) {
1144 err = tcp_connect(sk);
1145 /* Same failure procedure as in tcp_v4/6_connect */
1146 if (err) {
1147 tcp_set_state(sk, TCP_CLOSE);
1148 inet->inet_dport = 0;
1149 sk->sk_route_caps = 0;
1150 }
1151 }
cf60af03 1152 flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0;
ba615f67 1153 err = __inet_stream_connect(sk->sk_socket, uaddr,
3979ad7e 1154 msg->msg_namelen, flags, 1);
7db92362
WW		 1155 /* fastopen_req could already be freed in __inet_stream_connect
1156 * if the connection times out or gets rst
1157 */
1158 if (tp->fastopen_req) {
1159 *copied = tp->fastopen_req->copied;
1160 tcp_free_fastopen_req(tp);
1161 inet->defer_connect = 0;
1162 }
cf60af03
YC		 1163 return err;
1164}
1165
306b13eb 1166int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
1da177e4 1167{
1da177e4 1168 struct tcp_sock *tp = tcp_sk(sk);
f214f915 1169 struct ubuf_info *uarg = NULL;
1da177e4 1170 struct sk_buff *skb;
c14ac945 1171 struct sockcm_cookie sockc;
57be5bda
AV		 1172 int flags, err, copied = 0;
1173 int mss_now = 0, size_goal, copied_syn = 0;
d4011239 1174 bool process_backlog = false;
690e99c4 1175 bool sg;
1da177e4
LT		 1176 long timeo;
1177
1da177e4 1178 flags = msg->msg_flags;
f214f915
WB		 1179
1180 if (flags & MSG_ZEROCOPY && size) {
1181 if (sk->sk_state != TCP_ESTABLISHED) {
1182 err = -EINVAL;
1183 goto out_err;
1184 }
1185
1186 skb = tcp_send_head(sk) ? tcp_write_queue_tail(sk) : NULL;
1187 uarg = sock_zerocopy_realloc(sk, size, skb_zcopy(skb));
1188 if (!uarg) {
1189 err = -ENOBUFS;
1190 goto out_err;
1191 }
1192
1193 /* skb may be freed in main loop, keep extra ref on uarg */
1194 sock_zerocopy_get(uarg);
1195 if (!(sk_check_csum_caps(sk) && sk->sk_route_caps & NETIF_F_SG))
1196 uarg->zerocopy = 0;
1197 }
1198
19f6d3f3 1199 if (unlikely(flags & MSG_FASTOPEN || inet_sk(sk)->defer_connect)) {
f5ddcbbb 1200 err = tcp_sendmsg_fastopen(sk, msg, &copied_syn, size);
cf60af03
YC		 1201 if (err == -EINPROGRESS && copied_syn > 0)
1202 goto out;
1203 else if (err)
1204 goto out_err;
cf60af03
YC		 1205 }
1206
1da177e4
LT		 1207 timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1208
d7722e85
SHY		 1209 tcp_rate_check_app_limited(sk); /* is sending application-limited? */
1210
8336886f
JC		 1211 /* Wait for a connection to finish. One exception is TCP Fast Open
1212 * (passive side) where data is allowed to be sent before a connection
1213 * is fully established.
1214 */
1215 if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
1216 !tcp_passive_fastopen(sk)) {
686a5624
YM		 1217 err = sk_stream_wait_connect(sk, &timeo);
1218 if (err != 0)
cf60af03 1219 goto do_error;
8336886f 1220 }
1da177e4 1221
c0e88ff0
PE		 1222 if (unlikely(tp->repair)) {
1223 if (tp->repair_queue == TCP_RECV_QUEUE) {
1224 copied = tcp_send_rcvq(sk, msg, size);
5924f17a 1225 goto out_nopush;
c0e88ff0
PE		 1226 }
1227
1228 err = -EINVAL;
1229 if (tp->repair_queue == TCP_NO_QUEUE)
1230 goto out_err;
1231
1232 /* 'common' sending to sendq */
1233 }
1234
c14ac945
SHY		 1235 sockc.tsflags = sk->sk_tsflags;
1236 if (msg->msg_controllen) {
1237 err = sock_cmsg_send(sk, msg, &sockc);
1238 if (unlikely(err)) {
1239 err = -EINVAL;
1240 goto out_err;
1241 }
1242 }
1243
1da177e4 1244 /* This should be in poll */
9cd3e072 1245 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1da177e4 1246
1da177e4 1247 /* Ok commence sending. */
1da177e4
LT		 1248 copied = 0;
1249
d41a69f1
ED		 1250restart:
1251 mss_now = tcp_send_mss(sk, &size_goal, flags);
1252
1da177e4
LT		 1253 err = -EPIPE;
1254 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
79d8665b 1255 goto do_error;
1da177e4 1256
690e99c4 1257 sg = !!(sk->sk_route_caps & NETIF_F_SG);
def87cf4 1258
01e97e65 1259 while (msg_data_left(msg)) {
57be5bda
AV		 1260 int copy = 0;
1261 int max = size_goal;
1da177e4 1262
57be5bda
AV		 1263 skb = tcp_write_queue_tail(sk);
1264 if (tcp_send_head(sk)) {
1265 if (skb->ip_summed == CHECKSUM_NONE)
1266 max = mss_now;
1267 copy = max - skb->len;
cf60af03 1268 }
1da177e4 1269
c134ecb8 1270 if (copy <= 0 || !tcp_skb_can_collapse_to(skb)) {
3613b3db
ED		 1271 bool first_skb;
1272
1da177e4 1273new_segment:
57be5bda
AV		 1274 /* Allocate new segment. If the interface is SG,
1275 * allocate skb fitting to single page.
1276 */
1277 if (!sk_stream_memory_free(sk))
1278 goto wait_for_sndbuf;
1da177e4 1279
d4011239
ED		 1280 if (process_backlog && sk_flush_backlog(sk)) {
1281 process_backlog = false;
d41a69f1 1282 goto restart;
d4011239 1283 }
3613b3db 1284 first_skb = skb_queue_empty(&sk->sk_write_queue);
57be5bda 1285 skb = sk_stream_alloc_skb(sk,
3613b3db 1286 select_size(sk, sg, first_skb),
eb934478 1287 sk->sk_allocation,
3613b3db 1288 first_skb);
57be5bda
AV		 1289 if (!skb)
1290 goto wait_for_memory;
1da177e4 1291
d4011239 1292 process_backlog = true;
57be5bda
AV		 1293 /*
1294 * Check whether we can use HW checksum.
1295 */
9a49850d 1296 if (sk_check_csum_caps(sk))
57be5bda 1297 skb->ip_summed = CHECKSUM_PARTIAL;
1da177e4 1298
57be5bda
AV		 1299 skb_entail(sk, skb);
1300 copy = size_goal;
1301 max = size_goal;
9d186cac 1302
57be5bda
AV		 1303 /* All packets are restored as if they have
1304 * already been sent. skb_mstamp isn't set to
1305 * avoid wrong rtt estimation.
1306 */
1307 if (tp->repair)
1308 TCP_SKB_CB(skb)->sacked |= TCPCB_REPAIRED;
1309 }
1da177e4 1310
57be5bda 1311 /* Try to append data to the end of skb. */
01e97e65
AV		 1312 if (copy > msg_data_left(msg))
1313 copy = msg_data_left(msg);
57be5bda
AV		 1314
1315 /* Where to copy to? */
1316 if (skb_availroom(skb) > 0) {
1317 /* We have some space in skb head. Superb! */
1318 copy = min_t(int, copy, skb_availroom(skb));
1319 err = skb_add_data_nocache(sk, skb, &msg->msg_iter, copy);
1320 if (err)
1321 goto do_fault;
f214f915 1322 } else if (!uarg || !uarg->zerocopy) {
57be5bda
AV		 1323 bool merge = true;
1324 int i = skb_shinfo(skb)->nr_frags;
1325 struct page_frag *pfrag = sk_page_frag(sk);
1326
1327 if (!sk_page_frag_refill(sk, pfrag))
1328 goto wait_for_memory;
ef015786 1329
57be5bda
AV		 1330 if (!skb_can_coalesce(skb, i, pfrag->page,
1331 pfrag->offset)) {
ac9e70b1 1332 if (i >= sysctl_max_skb_frags || !sg) {
57be5bda
AV		 1333 tcp_mark_push(tp, skb);
1334 goto new_segment;
1da177e4 1335 }
57be5bda 1336 merge = false;
1da177e4
LT		 1337 }
1338
57be5bda
AV		 1339 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1340
1341 if (!sk_wmem_schedule(sk, copy))
1342 goto wait_for_memory;
1da177e4 1343
57be5bda
AV		 1344 err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
1345 pfrag->page,
1346 pfrag->offset,
1347 copy);
1348 if (err)
1349 goto do_error;
1da177e4 1350
57be5bda
AV		 1351 /* Update the skb. */
1352 if (merge) {
1353 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1354 } else {
1355 skb_fill_page_desc(skb, i, pfrag->page,
1356 pfrag->offset, copy);
4e33e346 1357 page_ref_inc(pfrag->page);
4ed2d765 1358 }
57be5bda 1359 pfrag->offset += copy;
f214f915
WB		 1360 } else {
1361 err = skb_zerocopy_iter_stream(sk, skb, msg, copy, uarg);
1362 if (err == -EMSGSIZE || err == -EEXIST)
1363 goto new_segment;
1364 if (err < 0)
1365 goto do_error;
1366 copy = err;
57be5bda
AV		 1367 }
1368
1369 if (!copied)
1370 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1371
1372 tp->write_seq += copy;
1373 TCP_SKB_CB(skb)->end_seq += copy;
1374 tcp_skb_pcount_set(skb, 0);
1da177e4 1375
57be5bda 1376 copied += copy;
01e97e65 1377 if (!msg_data_left(msg)) {
c134ecb8
MKL		 1378 if (unlikely(flags & MSG_EOR))
1379 TCP_SKB_CB(skb)->eor = 1;
57be5bda
AV		 1380 goto out;
1381 }
1da177e4 1382
57be5bda 1383 if (skb->len < max || (flags & MSG_OOB) || unlikely(tp->repair))
1da177e4
LT		 1384 continue;
1385
57be5bda
AV		 1386 if (forced_push(tp)) {
1387 tcp_mark_push(tp, skb);
1388 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
1389 } else if (skb == tcp_send_head(sk))
1390 tcp_push_one(sk, mss_now);
1391 continue;
1392
1da177e4 1393wait_for_sndbuf:
57be5bda 1394 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1da177e4 1395wait_for_memory:
57be5bda
AV		 1396 if (copied)
1397 tcp_push(sk, flags & ~MSG_MORE, mss_now,
1398 TCP_NAGLE_PUSH, size_goal);
1da177e4 1399
686a5624
YM		 1400 err = sk_stream_wait_memory(sk, &timeo);
1401 if (err != 0)
57be5bda 1402 goto do_error;
1da177e4 1403
57be5bda 1404 mss_now = tcp_send_mss(sk, &size_goal, flags);
1da177e4
LT		 1405 }
1406
1407out:
ad02c4f5
SHY		 1408 if (copied) {
1409 tcp_tx_timestamp(sk, sockc.tsflags, tcp_write_queue_tail(sk));
f54b3111 1410 tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
ad02c4f5 1411 }
5924f17a 1412out_nopush:
f214f915 1413 sock_zerocopy_put(uarg);
cf60af03 1414 return copied + copied_syn;
1da177e4
LT		 1415
1416do_fault:
1417 if (!skb->len) {
fe067e8a
DM		 1418 tcp_unlink_write_queue(skb, sk);
1419 /* It is the one place in all of TCP, except connection
1420 * reset, where we can be unlinking the send_head.
1421 */
1422 tcp_check_send_head(sk, skb);
3ab224be 1423 sk_wmem_free_skb(sk, skb);
1da177e4
LT		 1424 }
1425
1426do_error:
cf60af03 1427 if (copied + copied_syn)
1da177e4
LT		 1428 goto out;
1429out_err:
f214f915 1430 sock_zerocopy_put_abort(uarg);
1da177e4 1431 err = sk_stream_error(sk, flags, err);
ce5ec440 1432 /* make sure we wake any epoll edge trigger waiter */
b0f71bd3
FY		 1433 if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 &&
1434 err == -EAGAIN)) {
ce5ec440 1435 sk->sk_write_space(sk);
b0f71bd3
FY		 1436 tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED);
1437 }
1da177e4
LT		 1438 return err;
1439}
774c4673 1440EXPORT_SYMBOL_GPL(tcp_sendmsg_locked);
306b13eb
TH		 1441
1442int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
1443{
1444 int ret;
1445
1446 lock_sock(sk);
1447 ret = tcp_sendmsg_locked(sk, msg, size);
1448 release_sock(sk);
1449
1450 return ret;
1451}
4bc2f18b 1452EXPORT_SYMBOL(tcp_sendmsg);
1da177e4
LT		 1453
1454/*
1455 * Handle reading urgent data. BSD has very simple semantics for
1456 * this, no blocking and very strange errors 8)
1457 */
1458
377f0a08 1459static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags)
1da177e4
LT		 1460{
1461 struct tcp_sock *tp = tcp_sk(sk);
1462
1463 /* No URG data to read. */
1464 if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data ||
1465 tp->urg_data == TCP_URG_READ)
1466 return -EINVAL; /* Yes this is right ! */
1467
1468 if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE))
1469 return -ENOTCONN;
1470
1471 if (tp->urg_data & TCP_URG_VALID) {
1472 int err = 0;
1473 char c = tp->urg_data;
1474
1475 if (!(flags & MSG_PEEK))
1476 tp->urg_data = TCP_URG_READ;
1477
1478 /* Read urgent data. */
1479 msg->msg_flags |= MSG_OOB;
1480
1481 if (len > 0) {
1482 if (!(flags & MSG_TRUNC))
7eab8d9e 1483 err = memcpy_to_msg(msg, &c, 1);
1da177e4
LT		 1484 len = 1;
1485 } else
1486 msg->msg_flags |= MSG_TRUNC;
1487
1488 return err ? -EFAULT : len;
1489 }
1490
1491 if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
1492 return 0;
1493
1494 /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and
1495 * the available implementations agree in this case:
1496 * this call should never block, independent of the
1497 * blocking state of the socket.
1498 * Mike <pall@rz.uni-karlsruhe.de>
1499 */
1500 return -EAGAIN;
1501}
1502
c0e88ff0
PE		 1503static int tcp_peek_sndq(struct sock *sk, struct msghdr *msg, int len)
1504{
1505 struct sk_buff *skb;
1506 int copied = 0, err = 0;
1507
1508 /* XXX -- need to support SO_PEEK_OFF */
1509
1510 skb_queue_walk(&sk->sk_write_queue, skb) {
51f3d02b 1511 err = skb_copy_datagram_msg(skb, 0, msg, skb->len);
c0e88ff0
PE		 1512 if (err)
1513 break;
1514
1515 copied += skb->len;
1516 }
1517
1518 return err ?: copied;
1519}
1520
1da177e4
LT		 1521/* Clean up the receive buffer for full frames taken by the user,
1522 * then send an ACK if necessary. COPIED is the number of bytes
1523 * tcp_recvmsg has given to the user so far, it speeds up the
1524 * calculation of whether or not we must ACK for the sake of
1525 * a window update.
1526 */
3f334078 1527static void tcp_cleanup_rbuf(struct sock *sk, int copied)
1da177e4
LT		 1528{
1529 struct tcp_sock *tp = tcp_sk(sk);
a2a385d6 1530 bool time_to_ack = false;
1da177e4 1531
1da177e4
LT		 1532 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
1533
d792c100 1534 WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq),
2af6fd8b 1535 "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n",
d792c100 1536 tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt);
1da177e4 1537
463c84b9
ACM		 1538 if (inet_csk_ack_scheduled(sk)) {
1539 const struct inet_connection_sock *icsk = inet_csk(sk);
1da177e4
LT		 1540 /* Delayed ACKs frequently hit locked sockets during bulk
1541 * receive. */
463c84b9 1542 if (icsk->icsk_ack.blocked ||
1da177e4 1543 /* Once-per-two-segments ACK was not sent by tcp_input.c */
463c84b9 1544 tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss ||
1da177e4
LT		 1545 /*
1546 * If this read emptied read buffer, we send ACK, if
1547 * connection is not bidirectional, user drained
1548 * receive buffer and there was a small segment
1549 * in queue.
1550 */
1ef9696c
AK		 1551 (copied > 0 &&
1552 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) ||
1553 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) &&
1554 !icsk->icsk_ack.pingpong)) &&
1555 !atomic_read(&sk->sk_rmem_alloc)))
a2a385d6 1556 time_to_ack = true;
1da177e4
LT		 1557 }
1558
1559 /* We send an ACK if we can now advertise a non-zero window
1560 * which has been raised "significantly".
1561 *
1562 * Even if window raised up to infinity, do not send window open ACK
1563 * in states, where we will not receive more. It is useless.
1564 */
1565 if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1566 __u32 rcv_window_now = tcp_receive_window(tp);
1567
1568 /* Optimize, __tcp_select_window() is not cheap. */
1569 if (2*rcv_window_now <= tp->window_clamp) {
1570 __u32 new_window = __tcp_select_window(sk);
1571
1572 /* Send ACK now, if this read freed lots of space
1573 * in our buffer. Certainly, new_window is new window.
1574 * We can advertise it now, if it is not less than current one.
1575 * "Lots" means "at least twice" here.
1576 */
1577 if (new_window && new_window >= 2 * rcv_window_now)
a2a385d6 1578 time_to_ack = true;
1da177e4
LT		 1579 }
1580 }
1581 if (time_to_ack)
1582 tcp_send_ack(sk);
1583}
1584
f26845b4 1585static struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off)
1da177e4
LT		 1586{
1587 struct sk_buff *skb;
1588 u32 offset;
1589
f26845b4 1590 while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) {
1da177e4 1591 offset = seq - TCP_SKB_CB(skb)->seq;
9d691539
ED		 1592 if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
1593 pr_err_once("%s: found a SYN, please report !\n", __func__);
1da177e4 1594 offset--;
9d691539 1595 }
e11ecddf 1596 if (offset < skb->len || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) {
1da177e4
LT		 1597 *off = offset;
1598 return skb;
1599 }
f26845b4
ED		 1600 /* This looks weird, but this can happen if TCP collapsing
1601 * splitted a fat GRO packet, while we released socket lock
1602 * in skb_splice_bits()
1603 */
7bced397 1604 sk_eat_skb(sk, skb);
1da177e4
LT		 1605 }
1606 return NULL;
1607}
1608
1609/*
1610 * This routine provides an alternative to tcp_recvmsg() for routines
1611 * that would like to handle copying from skbuffs directly in 'sendfile'
1612 * fashion.
1613 * Note:
1614 * - It is assumed that the socket was locked by the caller.
1615 * - The routine does not block.
1616 * - At present, there is no support for reading OOB data
1617 * or for 'peeking' the socket using this routine
1618 * (although both would be easy to implement).
1619 */
1620int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
1621 sk_read_actor_t recv_actor)
1622{
1623 struct sk_buff *skb;
1624 struct tcp_sock *tp = tcp_sk(sk);
1625 u32 seq = tp->copied_seq;
1626 u32 offset;
1627 int copied = 0;
1628
1629 if (sk->sk_state == TCP_LISTEN)
1630 return -ENOTCONN;
1631 while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) {
1632 if (offset < skb->len) {
374e7b59
OP		 1633 int used;
1634 size_t len;
1da177e4
LT		 1635
1636 len = skb->len - offset;
1637 /* Stop reading if we hit a patch of urgent data */
1638 if (tp->urg_data) {
1639 u32 urg_offset = tp->urg_seq - seq;
1640 if (urg_offset < len)
1641 len = urg_offset;
1642 if (!len)
1643 break;
1644 }
1645 used = recv_actor(desc, skb, offset, len);
ff905b1e 1646 if (used <= 0) {
ddb61a57
JA		 1647 if (!copied)
1648 copied = used;
1649 break;
1650 } else if (used <= len) {
1da177e4
LT		 1651 seq += used;
1652 copied += used;
1653 offset += used;
1654 }
02275a2e 1655 /* If recv_actor drops the lock (e.g. TCP splice
293ad604
OP		 1656 * receive) the skb pointer might be invalid when
1657 * getting here: tcp_collapse might have deleted it
1658 * while aggregating skbs from the socket queue.
1659 */
02275a2e
WT		 1660 skb = tcp_recv_skb(sk, seq - 1, &offset);
1661 if (!skb)
1da177e4 1662 break;
02275a2e
WT		 1663 /* TCP coalescing might have appended data to the skb.
1664 * Try to splice more frags
1665 */
1666 if (offset + 1 != skb->len)
1667 continue;
1da177e4 1668 }
e11ecddf 1669 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) {
7bced397 1670 sk_eat_skb(sk, skb);
1da177e4
LT		 1671 ++seq;
1672 break;
1673 }
7bced397 1674 sk_eat_skb(sk, skb);
1da177e4
LT		 1675 if (!desc->count)
1676 break;
baff42ab 1677 tp->copied_seq = seq;
1da177e4
LT		 1678 }
1679 tp->copied_seq = seq;
1680
1681 tcp_rcv_space_adjust(sk);
1682
1683 /* Clean up data we have read: This will do ACK frames. */
f26845b4
ED		 1684 if (copied > 0) {
1685 tcp_recv_skb(sk, seq, &offset);
0e4b4992 1686 tcp_cleanup_rbuf(sk, copied);
f26845b4 1687 }
1da177e4
LT		 1688 return copied;
1689}
4bc2f18b 1690EXPORT_SYMBOL(tcp_read_sock);
1da177e4 1691
32035585
TH		 1692int tcp_peek_len(struct socket *sock)
1693{
1694 return tcp_inq(sock->sk);
1695}
1696EXPORT_SYMBOL(tcp_peek_len);
1697
1da177e4
LT		 1698/*
1699 * This routine copies from a sock struct into the user buffer.
1700 *
1701 * Technical note: in 2.3 we work on _locked_ socket, so that
1702 * tricks with *seq access order and skb->users are not required.
1703 * Probably, code can be easily improved even more.
1704 */
1705
1b784140
YX		 1706int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
1707 int flags, int *addr_len)
1da177e4
LT		 1708{
1709 struct tcp_sock *tp = tcp_sk(sk);
1710 int copied = 0;
1711 u32 peek_seq;
1712 u32 *seq;
1713 unsigned long used;
1714 int err;
1715 int target; /* Read at least this many bytes */
1716 long timeo;
dfbafc99 1717 struct sk_buff *skb, *last;
77527313 1718 u32 urg_hole = 0;
1da177e4 1719
4ed2d765 1720 if (unlikely(flags & MSG_ERRQUEUE))
f4713a3d 1721 return inet_recv_error(sk, msg, len, addr_len);
4ed2d765 1722
cbf55001
ET		 1723 if (sk_can_busy_loop(sk) && skb_queue_empty(&sk->sk_receive_queue) &&
1724 (sk->sk_state == TCP_ESTABLISHED))
1725 sk_busy_loop(sk, nonblock);
d30e383b 1726
1da177e4
LT		 1727 lock_sock(sk);
1728
1da177e4
LT		 1729 err = -ENOTCONN;
1730 if (sk->sk_state == TCP_LISTEN)
1731 goto out;
1732
1733 timeo = sock_rcvtimeo(sk, nonblock);
1734
1735 /* Urgent data needs to be handled specially. */
1736 if (flags & MSG_OOB)
1737 goto recv_urg;
1738
c0e88ff0
PE		 1739 if (unlikely(tp->repair)) {
1740 err = -EPERM;
1741 if (!(flags & MSG_PEEK))
1742 goto out;
1743
1744 if (tp->repair_queue == TCP_SEND_QUEUE)
1745 goto recv_sndq;
1746
1747 err = -EINVAL;
1748 if (tp->repair_queue == TCP_NO_QUEUE)
1749 goto out;
1750
1751 /* 'common' recv queue MSG_PEEK-ing */
1752 }
1753
1da177e4
LT		 1754 seq = &tp->copied_seq;
1755 if (flags & MSG_PEEK) {
1756 peek_seq = tp->copied_seq;
1757 seq = &peek_seq;
1758 }
1759
1760 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1761
1762 do {
1da177e4
LT		 1763 u32 offset;
1764
1765 /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */
1766 if (tp->urg_data && tp->urg_seq == *seq) {
1767 if (copied)
1768 break;
1769 if (signal_pending(current)) {
1770 copied = timeo ? sock_intr_errno(timeo) : -EAGAIN;
1771 break;
1772 }
1773 }
1774
1775 /* Next get a buffer. */
1776
dfbafc99 1777 last = skb_peek_tail(&sk->sk_receive_queue);
91521944 1778 skb_queue_walk(&sk->sk_receive_queue, skb) {
dfbafc99 1779 last = skb;
1da177e4
LT		 1780 /* Now that we have two receive queues this
1781 * shouldn't happen.
1782 */
d792c100 1783 if (WARN(before(*seq, TCP_SKB_CB(skb)->seq),
2af6fd8b
JP		 1784 "recvmsg bug: copied %X seq %X rcvnxt %X fl %X\n",
1785 *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt,
1786 flags))
1da177e4 1787 break;
d792c100 1788
1da177e4 1789 offset = *seq - TCP_SKB_CB(skb)->seq;
9d691539
ED		 1790 if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
1791 pr_err_once("%s: found a SYN, please report !\n", __func__);
1da177e4 1792 offset--;
9d691539 1793 }
1da177e4
LT		 1794 if (offset < skb->len)
1795 goto found_ok_skb;
e11ecddf 1796 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
1da177e4 1797 goto found_fin_ok;
2af6fd8b
JP		 1798 WARN(!(flags & MSG_PEEK),
1799 "recvmsg bug 2: copied %X seq %X rcvnxt %X fl %X\n",
1800 *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags);
91521944 1801 }
1da177e4
LT		 1802
1803 /* Well, if we have backlog, try to process it now yet. */
1804
1805 if (copied >= target && !sk->sk_backlog.tail)
1806 break;
1807
1808 if (copied) {
1809 if (sk->sk_err ||
1810 sk->sk_state == TCP_CLOSE ||
1811 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1812 !timeo ||
518a09ef 1813 signal_pending(current))
1da177e4
LT		 1814 break;
1815 } else {
1816 if (sock_flag(sk, SOCK_DONE))
1817 break;
1818
1819 if (sk->sk_err) {
1820 copied = sock_error(sk);
1821 break;
1822 }
1823
1824 if (sk->sk_shutdown & RCV_SHUTDOWN)
1825 break;
1826
1827 if (sk->sk_state == TCP_CLOSE) {
1828 if (!sock_flag(sk, SOCK_DONE)) {
1829 /* This occurs when user tries to read
1830 * from never connected socket.
1831 */
1832 copied = -ENOTCONN;
1833 break;
1834 }
1835 break;
1836 }
1837
1838 if (!timeo) {
1839 copied = -EAGAIN;
1840 break;
1841 }
1842
1843 if (signal_pending(current)) {
1844 copied = sock_intr_errno(timeo);
1845 break;
1846 }
1847 }
1848
0e4b4992 1849 tcp_cleanup_rbuf(sk, copied);
1da177e4 1850
1da177e4
LT		 1851 if (copied >= target) {
1852 /* Do not sleep, just process backlog. */
1853 release_sock(sk);
1854 lock_sock(sk);
dfbafc99
SD		 1855 } else {
1856 sk_wait_data(sk, &timeo, last);
1857 }
1da177e4 1858
77527313
IJ		 1859 if ((flags & MSG_PEEK) &&
1860 (peek_seq - copied - urg_hole != tp->copied_seq)) {
e87cc472
JP		 1861 net_dbg_ratelimited("TCP(%s:%d): Application bug, race in MSG_PEEK\n",
1862 current->comm,
1863 task_pid_nr(current));
1da177e4
LT		 1864 peek_seq = tp->copied_seq;
1865 }
1866 continue;
1867
1868 found_ok_skb:
1869 /* Ok so how much can we use? */
1870 used = skb->len - offset;
1871 if (len < used)
1872 used = len;
1873
1874 /* Do we have urgent data here? */
1875 if (tp->urg_data) {
1876 u32 urg_offset = tp->urg_seq - *seq;
1877 if (urg_offset < used) {
1878 if (!urg_offset) {
1879 if (!sock_flag(sk, SOCK_URGINLINE)) {
1880 ++*seq;
77527313 1881 urg_hole++;
1da177e4
LT		 1882 offset++;
1883 used--;
1884 if (!used)
1885 goto skip_copy;
1886 }
1887 } else
1888 used = urg_offset;
1889 }
1890 }
1891
1892 if (!(flags & MSG_TRUNC)) {
51f3d02b 1893 err = skb_copy_datagram_msg(skb, offset, msg, used);
7bced397
DW		 1894 if (err) {
1895 /* Exception. Bailout! */
1896 if (!copied)
1897 copied = -EFAULT;
1898 break;
1da177e4
LT		 1899 }
1900 }
1901
1902 *seq += used;
1903 copied += used;
1904 len -= used;
1905
1906 tcp_rcv_space_adjust(sk);
1907
1908skip_copy:
45f119bf 1909 if (tp->urg_data && after(tp->copied_seq, tp->urg_seq))
1da177e4 1910 tp->urg_data = 0;
1da177e4
LT		 1911 if (used + offset < skb->len)
1912 continue;
1913
e11ecddf 1914 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
1da177e4 1915 goto found_fin_ok;
7bced397
DW		 1916 if (!(flags & MSG_PEEK))
1917 sk_eat_skb(sk, skb);
1da177e4
LT		 1918 continue;
1919
1920 found_fin_ok:
1921 /* Process the FIN. */
1922 ++*seq;
7bced397
DW		 1923 if (!(flags & MSG_PEEK))
1924 sk_eat_skb(sk, skb);
1da177e4
LT		 1925 break;
1926 } while (len > 0);
1927
1da177e4
LT		 1928 /* According to UNIX98, msg_name/msg_namelen are ignored
1929 * on connected socket. I was just happy when found this 8) --ANK
1930 */
1931
1932 /* Clean up data we have read: This will do ACK frames. */
0e4b4992 1933 tcp_cleanup_rbuf(sk, copied);
1da177e4 1934
1da177e4
LT		 1935 release_sock(sk);
1936 return copied;
1937
1938out:
1da177e4
LT		 1939 release_sock(sk);
1940 return err;
1941
1942recv_urg:
377f0a08 1943 err = tcp_recv_urg(sk, msg, len, flags);
1da177e4 1944 goto out;
c0e88ff0
PE		 1945
1946recv_sndq:
1947 err = tcp_peek_sndq(sk, msg, len);
1948 goto out;
1da177e4 1949}
4bc2f18b 1950EXPORT_SYMBOL(tcp_recvmsg);
1da177e4 1951
490d5046
IJ		 1952void tcp_set_state(struct sock *sk, int state)
1953{
1954 int oldstate = sk->sk_state;
1955
1956 switch (state) {
1957 case TCP_ESTABLISHED:
1958 if (oldstate != TCP_ESTABLISHED)
81cc8a75 1959 TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
490d5046
IJ		 1960 break;
1961
1962 case TCP_CLOSE:
1963 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
81cc8a75 1964 TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS);
490d5046
IJ		 1965
1966 sk->sk_prot->unhash(sk);
1967 if (inet_csk(sk)->icsk_bind_hash &&
1968 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
ab1e0a13 1969 inet_put_port(sk);
490d5046
IJ		 1970 /* fall through */
1971 default:
5a5f3a8d 1972 if (oldstate == TCP_ESTABLISHED)
74688e48 1973 TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
490d5046
IJ		 1974 }
1975
1976 /* Change state AFTER socket is unhashed to avoid closed
1977 * socket sitting in hash tables.
1978 */
00fd38d9 1979 sk_state_store(sk, state);
490d5046
IJ		 1980
1981#ifdef STATE_TRACE
5a5f3a8d 1982 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n", sk, statename[oldstate], statename[state]);
490d5046
IJ		 1983#endif
1984}
1985EXPORT_SYMBOL_GPL(tcp_set_state);
1986
1da177e4
LT		 1987/*
1988 * State processing on a close. This implements the state shift for
1989 * sending our FIN frame. Note that we only send a FIN for some
1990 * states. A shutdown() may have already sent the FIN, or we may be
1991 * closed.
1992 */
1993
9b5b5cff 1994static const unsigned char new_state[16] = {
1da177e4 1995 /* current state: new state: action: */
0980c1e3
ED		 1996 [0 /* (Invalid) */] = TCP_CLOSE,
1997 [TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
1998 [TCP_SYN_SENT] = TCP_CLOSE,
1999 [TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2000 [TCP_FIN_WAIT1] = TCP_FIN_WAIT1,
2001 [TCP_FIN_WAIT2] = TCP_FIN_WAIT2,
2002 [TCP_TIME_WAIT] = TCP_CLOSE,
2003 [TCP_CLOSE] = TCP_CLOSE,
2004 [TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN,
2005 [TCP_LAST_ACK] = TCP_LAST_ACK,
2006 [TCP_LISTEN] = TCP_CLOSE,
2007 [TCP_CLOSING] = TCP_CLOSING,
2008 [TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */
1da177e4
LT		 2009};
2010
2011static int tcp_close_state(struct sock *sk)
2012{
2013 int next = (int)new_state[sk->sk_state];
2014 int ns = next & TCP_STATE_MASK;
2015
2016 tcp_set_state(sk, ns);
2017
2018 return next & TCP_ACTION_FIN;
2019}
2020
2021/*
2022 * Shutdown the sending side of a connection. Much like close except
1f29b058 2023 * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD).
1da177e4
LT		 2024 */
2025
2026void tcp_shutdown(struct sock *sk, int how)
2027{
2028 /* We need to grab some memory, and put together a FIN,
2029 * and then put it into the queue to be sent.
2030 * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92.
2031 */
2032 if (!(how & SEND_SHUTDOWN))
2033 return;
2034
2035 /* If we've already sent a FIN, or it's a closed state, skip this. */
2036 if ((1 << sk->sk_state) &
2037 (TCPF_ESTABLISHED | TCPF_SYN_SENT |
2038 TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) {
2039 /* Clear out any half completed packets. FIN if needed. */
2040 if (tcp_close_state(sk))
2041 tcp_send_fin(sk);
2042 }
2043}
4bc2f18b 2044EXPORT_SYMBOL(tcp_shutdown);
1da177e4 2045
efcdbf24
AS		 2046bool tcp_check_oom(struct sock *sk, int shift)
2047{
2048 bool too_many_orphans, out_of_socket_memory;
2049
2050 too_many_orphans = tcp_too_many_orphans(sk, shift);
2051 out_of_socket_memory = tcp_out_of_memory(sk);
2052
e87cc472
JP		 2053 if (too_many_orphans)
2054 net_info_ratelimited("too many orphaned sockets\n");
2055 if (out_of_socket_memory)
2056 net_info_ratelimited("out of memory -- consider tuning tcp_mem\n");
efcdbf24
AS		 2057 return too_many_orphans || out_of_socket_memory;
2058}
2059
1da177e4
LT		 2060void tcp_close(struct sock *sk, long timeout)
2061{
2062 struct sk_buff *skb;
2063 int data_was_unread = 0;
75c2d907 2064 int state;
1da177e4
LT		 2065
2066 lock_sock(sk);
2067 sk->sk_shutdown = SHUTDOWN_MASK;
2068
2069 if (sk->sk_state == TCP_LISTEN) {
2070 tcp_set_state(sk, TCP_CLOSE);
2071
2072 /* Special case. */
0a5578cf 2073 inet_csk_listen_stop(sk);
1da177e4
LT		 2074
2075 goto adjudge_to_death;
2076 }
2077
2078 /* We need to flush the recv. buffs. We do this only on the
2079 * descriptor close, not protocol-sourced closes, because the
2080 * reader process may not have drained the data yet!
2081 */
2082 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
e11ecddf
ED		 2083 u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq;
2084
2085 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
2086 len--;
1da177e4
LT		 2087 data_was_unread += len;
2088 __kfree_skb(skb);
2089 }
2090
3ab224be 2091 sk_mem_reclaim(sk);
1da177e4 2092
565b7b2d
KK		 2093 /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */
2094 if (sk->sk_state == TCP_CLOSE)
2095 goto adjudge_to_death;
2096
65bb723c
GR		 2097 /* As outlined in RFC 2525, section 2.17, we send a RST here because
2098 * data was lost. To witness the awful effects of the old behavior of
2099 * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk
2100 * GET in an FTP client, suspend the process, wait for the client to
2101 * advertise a zero window, then kill -9 the FTP client, wheee...
2102 * Note: timeout is always zero in such a case.
1da177e4 2103 */
ee995283
PE		 2104 if (unlikely(tcp_sk(sk)->repair)) {
2105 sk->sk_prot->disconnect(sk, 0);
2106 } else if (data_was_unread) {
1da177e4 2107 /* Unread data was tossed, zap the connection. */
6aef70a8 2108 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE);
1da177e4 2109 tcp_set_state(sk, TCP_CLOSE);
aa133076 2110 tcp_send_active_reset(sk, sk->sk_allocation);
1da177e4
LT		 2111 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
2112 /* Check zero linger _after_ checking for unread data. */
2113 sk->sk_prot->disconnect(sk, 0);
6aef70a8 2114 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONDATA);
1da177e4
LT		 2115 } else if (tcp_close_state(sk)) {
2116 /* We FIN if the application ate all the data before
2117 * zapping the connection.
2118 */
2119
2120 /* RED-PEN. Formally speaking, we have broken TCP state
2121 * machine. State transitions:
2122 *
2123 * TCP_ESTABLISHED -> TCP_FIN_WAIT1
2124 * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible)
2125 * TCP_CLOSE_WAIT -> TCP_LAST_ACK
2126 *
2127 * are legal only when FIN has been sent (i.e. in window),
2128 * rather than queued out of window. Purists blame.
2129 *
2130 * F.e. "RFC state" is ESTABLISHED,
2131 * if Linux state is FIN-WAIT-1, but FIN is still not sent.
2132 *
2133 * The visible declinations are that sometimes
2134 * we enter time-wait state, when it is not required really
2135 * (harmless), do not send active resets, when they are
2136 * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when
2137 * they look as CLOSING or LAST_ACK for Linux)
2138 * Probably, I missed some more holelets.
2139 * --ANK
8336886f
JC		 2140 * XXX (TFO) - To start off we don't support SYN+ACK+FIN
2141 * in a single packet! (May consider it later but will
2142 * probably need API support or TCP_CORK SYN-ACK until
2143 * data is written and socket is closed.)
1da177e4
LT		 2144 */
2145 tcp_send_fin(sk);
2146 }
2147
2148 sk_stream_wait_close(sk, timeout);
2149
2150adjudge_to_death:
75c2d907
HX		 2151 state = sk->sk_state;
2152 sock_hold(sk);
2153 sock_orphan(sk);
75c2d907 2154
1da177e4
LT		 2155 /* It is the last release_sock in its life. It will remove backlog. */
2156 release_sock(sk);
2157
2158
2159 /* Now socket is owned by kernel and we acquire BH lock
a777f715 2160 * to finish close. No need to check for user refs.
1da177e4
LT		 2161 */
2162 local_bh_disable();
2163 bh_lock_sock(sk);
547b792c 2164 WARN_ON(sock_owned_by_user(sk));
1da177e4 2165
eb4dea58
HX		 2166 percpu_counter_inc(sk->sk_prot->orphan_count);
2167
75c2d907
HX		 2168 /* Have we already been destroyed by a softirq or backlog? */
2169 if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE)
2170 goto out;
1da177e4
LT		 2171
2172 /* This is a (useful) BSD violating of the RFC. There is a
2173 * problem with TCP as specified in that the other end could
2174 * keep a socket open forever with no application left this end.
b10bd54c 2175 * We use a 1 minute timeout (about the same as BSD) then kill
1da177e4
LT		 2176 * our end. If they send after that then tough - BUT: long enough
2177 * that we won't make the old 4*rto = almost no time - whoops
2178 * reset mistake.
2179 *
2180 * Nope, it was not mistake. It is really desired behaviour
2181 * f.e. on http servers, when such sockets are useless, but
2182 * consume significant resources. Let's do it with special
2183 * linger2 option. --ANK
2184 */
2185
2186 if (sk->sk_state == TCP_FIN_WAIT2) {
2187 struct tcp_sock *tp = tcp_sk(sk);
2188 if (tp->linger2 < 0) {
2189 tcp_set_state(sk, TCP_CLOSE);
2190 tcp_send_active_reset(sk, GFP_ATOMIC);
02a1d6e7 2191 __NET_INC_STATS(sock_net(sk),
de0744af 2192 LINUX_MIB_TCPABORTONLINGER);
1da177e4 2193 } else {
463c84b9 2194 const int tmo = tcp_fin_time(sk);
1da177e4
LT		 2195
2196 if (tmo > TCP_TIMEWAIT_LEN) {
52499afe
DM		 2197 inet_csk_reset_keepalive_timer(sk,
2198 tmo - TCP_TIMEWAIT_LEN);
1da177e4 2199 } else {
1da177e4
LT		 2200 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
2201 goto out;
2202 }
2203 }
2204 }
2205 if (sk->sk_state != TCP_CLOSE) {
3ab224be 2206 sk_mem_reclaim(sk);
efcdbf24 2207 if (tcp_check_oom(sk, 0)) {
1da177e4
LT		 2208 tcp_set_state(sk, TCP_CLOSE);
2209 tcp_send_active_reset(sk, GFP_ATOMIC);
02a1d6e7 2210 __NET_INC_STATS(sock_net(sk),
de0744af 2211 LINUX_MIB_TCPABORTONMEMORY);
1da177e4
LT		 2212 }
2213 }
1da177e4 2214
8336886f
JC		 2215 if (sk->sk_state == TCP_CLOSE) {
2216 struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
2217 /* We could get here with a non-NULL req if the socket is
2218 * aborted (e.g., closed with unread data) before 3WHS
2219 * finishes.
2220 */
00db4124 2221 if (req)
8336886f 2222 reqsk_fastopen_remove(sk, req, false);
0a5578cf 2223 inet_csk_destroy_sock(sk);
8336886f 2224 }
1da177e4
LT		 2225 /* Otherwise, socket is reprieved until protocol close. */
2226
2227out:
2228 bh_unlock_sock(sk);
2229 local_bh_enable();
2230 sock_put(sk);
2231}
4bc2f18b 2232EXPORT_SYMBOL(tcp_close);
1da177e4
LT		 2233
2234/* These states need RST on ABORT according to RFC793 */
2235
a2a385d6 2236static inline bool tcp_need_reset(int state)
1da177e4
LT		 2237{
2238 return (1 << state) &
2239 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 |
2240 TCPF_FIN_WAIT2 | TCPF_SYN_RECV);
2241}
2242
2243int tcp_disconnect(struct sock *sk, int flags)
2244{
2245 struct inet_sock *inet = inet_sk(sk);
463c84b9 2246 struct inet_connection_sock *icsk = inet_csk(sk);
1da177e4
LT		 2247 struct tcp_sock *tp = tcp_sk(sk);
2248 int err = 0;
2249 int old_state = sk->sk_state;
2250
2251 if (old_state != TCP_CLOSE)
2252 tcp_set_state(sk, TCP_CLOSE);
2253
2254 /* ABORT function of RFC793 */
2255 if (old_state == TCP_LISTEN) {
0a5578cf 2256 inet_csk_listen_stop(sk);
ee995283
PE		 2257 } else if (unlikely(tp->repair)) {
2258 sk->sk_err = ECONNABORTED;
1da177e4
LT		 2259 } else if (tcp_need_reset(old_state) ||
2260 (tp->snd_nxt != tp->write_seq &&
2261 (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) {
caa20d9a 2262 /* The last check adjusts for discrepancy of Linux wrt. RFC
1da177e4
LT		 2263 * states
2264 */
2265 tcp_send_active_reset(sk, gfp_any());
2266 sk->sk_err = ECONNRESET;
2267 } else if (old_state == TCP_SYN_SENT)
2268 sk->sk_err = ECONNRESET;
2269
2270 tcp_clear_xmit_timers(sk);
2271 __skb_queue_purge(&sk->sk_receive_queue);
fe067e8a 2272 tcp_write_queue_purge(sk);
cf1ef3f0 2273 tcp_fastopen_active_disable_ofo_check(sk);
9f5afeae 2274 skb_rbtree_purge(&tp->out_of_order_queue);
1da177e4 2275
c720c7e8 2276 inet->inet_dport = 0;
1da177e4
LT		 2277
2278 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
2279 inet_reset_saddr(sk);
2280
2281 sk->sk_shutdown = 0;
2282 sock_reset_flag(sk, SOCK_DONE);
740b0f18 2283 tp->srtt_us = 0;
686a5624
YM		 2284 tp->write_seq += tp->max_window + 2;
2285 if (tp->write_seq == 0)
1da177e4 2286 tp->write_seq = 1;
463c84b9 2287 icsk->icsk_backoff = 0;
1da177e4 2288 tp->snd_cwnd = 2;
6687e988 2289 icsk->icsk_probes_out = 0;
1da177e4 2290 tp->packets_out = 0;
0b6a05c1 2291 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
1da177e4 2292 tp->snd_cwnd_cnt = 0;
1fdf475a 2293 tp->window_clamp = 0;
6687e988 2294 tcp_set_ca_state(sk, TCP_CA_Open);
1da177e4 2295 tcp_clear_retrans(tp);
463c84b9 2296 inet_csk_delack_init(sk);
499350a5
WW		 2297 /* Initialize rcv_mss to TCP_MIN_MSS to avoid division by 0
2298 * issue in __tcp_select_window()
2299 */
2300 icsk->icsk_ack.rcv_mss = TCP_MIN_MSS;
fe067e8a 2301 tcp_init_send_head(sk);
b40b4f79 2302 memset(&tp->rx_opt, 0, sizeof(tp->rx_opt));
1da177e4 2303 __sk_dst_reset(sk);
d747a7a5
WC		 2304 dst_release(sk->sk_rx_dst);
2305 sk->sk_rx_dst = NULL;
17c3060b 2306 tcp_saved_syn_free(tp);
1da177e4 2307
7db92362
WW		 2308 /* Clean up fastopen related fields */
2309 tcp_free_fastopen_req(tp);
2310 inet->defer_connect = 0;
2311
c720c7e8 2312 WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
1da177e4
LT		 2313
2314 sk->sk_error_report(sk);
2315 return err;
2316}
4bc2f18b 2317EXPORT_SYMBOL(tcp_disconnect);
1da177e4 2318
a2a385d6 2319static inline bool tcp_can_repair_sock(const struct sock *sk)
ee995283 2320{
52e804c6 2321 return ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN) &&
319b0534 2322 (sk->sk_state != TCP_LISTEN);
ee995283
PE		 2323}
2324
b1ed4c4f
AV		 2325static int tcp_repair_set_window(struct tcp_sock *tp, char __user *optbuf, int len)
2326{
2327 struct tcp_repair_window opt;
2328
2329 if (!tp->repair)
2330 return -EPERM;
2331
2332 if (len != sizeof(opt))
2333 return -EINVAL;
2334
2335 if (copy_from_user(&opt, optbuf, sizeof(opt)))
2336 return -EFAULT;
2337
2338 if (opt.max_window < opt.snd_wnd)
2339 return -EINVAL;
2340
2341 if (after(opt.snd_wl1, tp->rcv_nxt + opt.rcv_wnd))
2342 return -EINVAL;
2343
2344 if (after(opt.rcv_wup, tp->rcv_nxt))
2345 return -EINVAL;
2346
2347 tp->snd_wl1 = opt.snd_wl1;
2348 tp->snd_wnd = opt.snd_wnd;
2349 tp->max_window = opt.max_window;
2350
2351 tp->rcv_wnd = opt.rcv_wnd;
2352 tp->rcv_wup = opt.rcv_wup;
2353
2354 return 0;
2355}
2356
15e56515 2357static int tcp_repair_options_est(struct sock *sk,
de248a75 2358 struct tcp_repair_opt __user *optbuf, unsigned int len)
b139ba4e 2359{
15e56515 2360 struct tcp_sock *tp = tcp_sk(sk);
de248a75 2361 struct tcp_repair_opt opt;
b139ba4e 2362
de248a75
PE		 2363 while (len >= sizeof(opt)) {
2364 if (copy_from_user(&opt, optbuf, sizeof(opt)))
b139ba4e
PE		 2365 return -EFAULT;
2366
2367 optbuf++;
de248a75 2368 len -= sizeof(opt);
b139ba4e 2369
de248a75
PE		 2370 switch (opt.opt_code) {
2371 case TCPOPT_MSS:
2372 tp->rx_opt.mss_clamp = opt.opt_val;
15e56515 2373 tcp_mtup_init(sk);
b139ba4e 2374 break;
de248a75 2375 case TCPOPT_WINDOW:
bc26ccd8
AV		 2376 {
2377 u16 snd_wscale = opt.opt_val & 0xFFFF;
2378 u16 rcv_wscale = opt.opt_val >> 16;
2379
589c49cb 2380 if (snd_wscale > TCP_MAX_WSCALE || rcv_wscale > TCP_MAX_WSCALE)
bc26ccd8 2381 return -EFBIG;
b139ba4e 2382
bc26ccd8
AV		 2383 tp->rx_opt.snd_wscale = snd_wscale;
2384 tp->rx_opt.rcv_wscale = rcv_wscale;
2385 tp->rx_opt.wscale_ok = 1;
2386 }
b139ba4e 2387 break;
b139ba4e 2388 case TCPOPT_SACK_PERM:
de248a75
PE		 2389 if (opt.opt_val != 0)
2390 return -EINVAL;
2391
b139ba4e
PE		 2392 tp->rx_opt.sack_ok |= TCP_SACK_SEEN;
2393 if (sysctl_tcp_fack)
2394 tcp_enable_fack(tp);
2395 break;
2396 case TCPOPT_TIMESTAMP:
de248a75
PE		 2397 if (opt.opt_val != 0)
2398 return -EINVAL;
2399
b139ba4e
PE		 2400 tp->rx_opt.tstamp_ok = 1;
2401 break;
2402 }
2403 }
2404
2405 return 0;
2406}
2407
1da177e4
LT		 2408/*
2409 * Socket option code for TCP.
2410 */
3fdadf7d 2411static int do_tcp_setsockopt(struct sock *sk, int level,
b7058842 2412 int optname, char __user *optval, unsigned int optlen)
1da177e4
LT		 2413{
2414 struct tcp_sock *tp = tcp_sk(sk);
463c84b9 2415 struct inet_connection_sock *icsk = inet_csk(sk);
1e579caa 2416 struct net *net = sock_net(sk);
1da177e4
LT		 2417 int val;
2418 int err = 0;
2419
e56fb50f
WAS		 2420 /* These are data/string values, all the others are ints */
2421 switch (optname) {
2422 case TCP_CONGESTION: {
5f8ef48d
SH		 2423 char name[TCP_CA_NAME_MAX];
2424
2425 if (optlen < 1)
2426 return -EINVAL;
2427
2428 val = strncpy_from_user(name, optval,
4fdb78d3 2429 min_t(long, TCP_CA_NAME_MAX-1, optlen));
5f8ef48d
SH		 2430 if (val < 0)
2431 return -EFAULT;
2432 name[val] = 0;
2433
2434 lock_sock(sk);
91b5b21c 2435 err = tcp_set_congestion_control(sk, name, true);
5f8ef48d
SH		 2436 release_sock(sk);
2437 return err;
2438 }
734942cc
DW		 2439 case TCP_ULP: {
2440 char name[TCP_ULP_NAME_MAX];
2441
2442 if (optlen < 1)
2443 return -EINVAL;
2444
2445 val = strncpy_from_user(name, optval,
2446 min_t(long, TCP_ULP_NAME_MAX - 1,
2447 optlen));
2448 if (val < 0)
2449 return -EFAULT;
2450 name[val] = 0;
2451
2452 lock_sock(sk);
2453 err = tcp_set_ulp(sk, name);
2454 release_sock(sk);
2455 return err;
2456 }
e56fb50f
WAS		 2457 default:
2458 /* fallthru */
2459 break;
ccbd6a5a 2460 }
5f8ef48d 2461
1da177e4
LT		 2462 if (optlen < sizeof(int))
2463 return -EINVAL;
2464
2465 if (get_user(val, (int __user *)optval))
2466 return -EFAULT;
2467
2468 lock_sock(sk);
2469
2470 switch (optname) {
2471 case TCP_MAXSEG:
2472 /* Values greater than interface MTU won't take effect. However
2473 * at the point when this call is done we typically don't yet
a777f715
RC		 2474 * know which interface is going to be used
2475 */
cfc62d87 2476 if (val && (val < TCP_MIN_MSS || val > MAX_TCP_WINDOW)) {
1da177e4
LT		 2477 err = -EINVAL;
2478 break;
2479 }
2480 tp->rx_opt.user_mss = val;
2481 break;
2482
2483 case TCP_NODELAY:
2484 if (val) {
2485 /* TCP_NODELAY is weaker than TCP_CORK, so that
2486 * this option on corked socket is remembered, but
2487 * it is not activated until cork is cleared.
2488 *
2489 * However, when TCP_NODELAY is set we make
2490 * an explicit push, which overrides even TCP_CORK
2491 * for currently queued segments.
2492 */
2493 tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH;
9e412ba7 2494 tcp_push_pending_frames(sk);
1da177e4
LT		 2495 } else {
2496 tp->nonagle &= ~TCP_NAGLE_OFF;
2497 }
2498 break;
2499
36e31b0a
AP		 2500 case TCP_THIN_LINEAR_TIMEOUTS:
2501 if (val < 0 || val > 1)
2502 err = -EINVAL;
2503 else
2504 tp->thin_lto = val;
2505 break;
2506
7e380175
AP		 2507 case TCP_THIN_DUPACK:
2508 if (val < 0 || val > 1)
2509 err = -EINVAL;
7e380175
AP		 2510 break;
2511
ee995283
PE		 2512 case TCP_REPAIR:
2513 if (!tcp_can_repair_sock(sk))
2514 err = -EPERM;
2515 else if (val == 1) {
2516 tp->repair = 1;
2517 sk->sk_reuse = SK_FORCE_REUSE;
2518 tp->repair_queue = TCP_NO_QUEUE;
2519 } else if (val == 0) {
2520 tp->repair = 0;
2521 sk->sk_reuse = SK_NO_REUSE;
2522 tcp_send_window_probe(sk);
2523 } else
2524 err = -EINVAL;
2525
2526 break;
2527
2528 case TCP_REPAIR_QUEUE:
2529 if (!tp->repair)
2530 err = -EPERM;
2531 else if (val < TCP_QUEUES_NR)
2532 tp->repair_queue = val;
2533 else
2534 err = -EINVAL;
2535 break;
2536
2537 case TCP_QUEUE_SEQ:
2538 if (sk->sk_state != TCP_CLOSE)
2539 err = -EPERM;
2540 else if (tp->repair_queue == TCP_SEND_QUEUE)
2541 tp->write_seq = val;
2542 else if (tp->repair_queue == TCP_RECV_QUEUE)
2543 tp->rcv_nxt = val;
2544 else
2545 err = -EINVAL;
2546 break;
2547
b139ba4e
PE		 2548 case TCP_REPAIR_OPTIONS:
2549 if (!tp->repair)
2550 err = -EINVAL;
2551 else if (sk->sk_state == TCP_ESTABLISHED)
15e56515 2552 err = tcp_repair_options_est(sk,
de248a75
PE		 2553 (struct tcp_repair_opt __user *)optval,
2554 optlen);
b139ba4e
PE		 2555 else
2556 err = -EPERM;
2557 break;
2558
1da177e4
LT		 2559 case TCP_CORK:
2560 /* When set indicates to always queue non-full frames.
2561 * Later the user clears this option and we transmit
2562 * any pending partial frames in the queue. This is
2563 * meant to be used alongside sendfile() to get properly
2564 * filled frames when the user (for example) must write
2565 * out headers with a write() call first and then use
2566 * sendfile to send out the data parts.
2567 *
2568 * TCP_CORK can be set together with TCP_NODELAY and it is
2569 * stronger than TCP_NODELAY.
2570 */
2571 if (val) {
2572 tp->nonagle |= TCP_NAGLE_CORK;
2573 } else {
2574 tp->nonagle &= ~TCP_NAGLE_CORK;
2575 if (tp->nonagle&TCP_NAGLE_OFF)
2576 tp->nonagle |= TCP_NAGLE_PUSH;
9e412ba7 2577 tcp_push_pending_frames(sk);
1da177e4
LT		 2578 }
2579 break;
2580
2581 case TCP_KEEPIDLE:
2582 if (val < 1 || val > MAX_TCP_KEEPIDLE)
2583 err = -EINVAL;
2584 else {
2585 tp->keepalive_time = val * HZ;
2586 if (sock_flag(sk, SOCK_KEEPOPEN) &&
2587 !((1 << sk->sk_state) &
2588 (TCPF_CLOSE | TCPF_LISTEN))) {
6c37e5de 2589 u32 elapsed = keepalive_time_elapsed(tp);
1da177e4
LT		 2590 if (tp->keepalive_time > elapsed)
2591 elapsed = tp->keepalive_time - elapsed;
2592 else
2593 elapsed = 0;
463c84b9 2594 inet_csk_reset_keepalive_timer(sk, elapsed);
1da177e4
LT		 2595 }
2596 }
2597 break;
2598 case TCP_KEEPINTVL:
2599 if (val < 1 || val > MAX_TCP_KEEPINTVL)
2600 err = -EINVAL;
2601 else
2602 tp->keepalive_intvl = val * HZ;
2603 break;
2604 case TCP_KEEPCNT:
2605 if (val < 1 || val > MAX_TCP_KEEPCNT)
2606 err = -EINVAL;
2607 else
2608 tp->keepalive_probes = val;
2609 break;
2610 case TCP_SYNCNT:
2611 if (val < 1 || val > MAX_TCP_SYNCNT)
2612 err = -EINVAL;
2613 else
463c84b9 2614 icsk->icsk_syn_retries = val;
1da177e4
LT		 2615 break;
2616
cd8ae852
ED		 2617 case TCP_SAVE_SYN:
2618 if (val < 0 || val > 1)
2619 err = -EINVAL;
2620 else
2621 tp->save_syn = val;
2622 break;
2623
1da177e4
LT		 2624 case TCP_LINGER2:
2625 if (val < 0)
2626 tp->linger2 = -1;
1e579caa 2627 else if (val > net->ipv4.sysctl_tcp_fin_timeout / HZ)
1da177e4
LT		 2628 tp->linger2 = 0;
2629 else
2630 tp->linger2 = val * HZ;
2631 break;
2632
2633 case TCP_DEFER_ACCEPT:
b103cf34
JA		 2634 /* Translate value in seconds to number of retransmits */
2635 icsk->icsk_accept_queue.rskq_defer_accept =
2636 secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ,
2637 TCP_RTO_MAX / HZ);
1da177e4
LT		 2638 break;
2639
2640 case TCP_WINDOW_CLAMP:
2641 if (!val) {
2642 if (sk->sk_state != TCP_CLOSE) {
2643 err = -EINVAL;
2644 break;
2645 }
2646 tp->window_clamp = 0;
2647 } else
2648 tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ?
2649 SOCK_MIN_RCVBUF / 2 : val;
2650 break;
2651
2652 case TCP_QUICKACK:
2653 if (!val) {
463c84b9 2654 icsk->icsk_ack.pingpong = 1;
1da177e4 2655 } else {
463c84b9 2656 icsk->icsk_ack.pingpong = 0;
1da177e4
LT		 2657 if ((1 << sk->sk_state) &
2658 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) &&
463c84b9
ACM		 2659 inet_csk_ack_scheduled(sk)) {
2660 icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;
0e4b4992 2661 tcp_cleanup_rbuf(sk, 1);
1da177e4 2662 if (!(val & 1))
463c84b9 2663 icsk->icsk_ack.pingpong = 1;
1da177e4
LT		 2664 }
2665 }
2666 break;
2667
cfb6eeb4
YH		 2668#ifdef CONFIG_TCP_MD5SIG
2669 case TCP_MD5SIG:
8917a777 2670 case TCP_MD5SIG_EXT:
cfb6eeb4 2671 /* Read the IP->Key mappings from userspace */
8917a777 2672 err = tp->af_specific->md5_parse(sk, optname, optval, optlen);
cfb6eeb4
YH		 2673 break;
2674#endif
dca43c75 2675 case TCP_USER_TIMEOUT:
b248230c 2676 /* Cap the max time in ms TCP will retry or probe the window
dca43c75
JC		 2677 * before giving up and aborting (ETIMEDOUT) a connection.
2678 */
42493570
HL		 2679 if (val < 0)
2680 err = -EINVAL;
2681 else
2682 icsk->icsk_user_timeout = msecs_to_jiffies(val);
dca43c75 2683 break;
8336886f
JC		 2684
2685 case TCP_FASTOPEN:
2686 if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE |
dfea2aa6
CP		 2687 TCPF_LISTEN))) {
2688 tcp_fastopen_init_key_once(true);
2689
0536fcc0 2690 fastopen_queue_tune(sk, val);
dfea2aa6 2691 } else {
8336886f 2692 err = -EINVAL;
dfea2aa6 2693 }
8336886f 2694 break;
19f6d3f3
WW		 2695 case TCP_FASTOPEN_CONNECT:
2696 if (val > 1 || val < 0) {
2697 err = -EINVAL;
2698 } else if (sysctl_tcp_fastopen & TFO_CLIENT_ENABLE) {
2699 if (sk->sk_state == TCP_CLOSE)
2700 tp->fastopen_connect = val;
2701 else
2702 err = -EINVAL;
2703 } else {
2704 err = -EOPNOTSUPP;
2705 }
2706 break;
93be6ce0
AV		 2707 case TCP_TIMESTAMP:
2708 if (!tp->repair)
2709 err = -EPERM;
2710 else
9a568de4 2711 tp->tsoffset = val - tcp_time_stamp_raw();
93be6ce0 2712 break;
b1ed4c4f
AV		 2713 case TCP_REPAIR_WINDOW:
2714 err = tcp_repair_set_window(tp, optval, optlen);
2715 break;
c9bee3b7
ED		 2716 case TCP_NOTSENT_LOWAT:
2717 tp->notsent_lowat = val;
2718 sk->sk_write_space(sk);
2719 break;
1da177e4
LT		 2720 default:
2721 err = -ENOPROTOOPT;
2722 break;
3ff50b79
SH		 2723 }
2724
1da177e4
LT		 2725 release_sock(sk);
2726 return err;
2727}
2728
3fdadf7d 2729int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval,
b7058842 2730 unsigned int optlen)
3fdadf7d 2731{
cf533ea5 2732 const struct inet_connection_sock *icsk = inet_csk(sk);
3fdadf7d
DM		 2733
2734 if (level != SOL_TCP)
2735 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
2736 optval, optlen);
2737 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2738}
4bc2f18b 2739EXPORT_SYMBOL(tcp_setsockopt);
3fdadf7d
DM		 2740
2741#ifdef CONFIG_COMPAT
543d9cfe 2742int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
b7058842 2743 char __user *optval, unsigned int optlen)
3fdadf7d 2744{
dec73ff0
ACM		 2745 if (level != SOL_TCP)
2746 return inet_csk_compat_setsockopt(sk, level, optname,
2747 optval, optlen);
3fdadf7d
DM		 2748 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2749}
543d9cfe 2750EXPORT_SYMBOL(compat_tcp_setsockopt);
3fdadf7d
DM		 2751#endif
2752
efd90174
FY		 2753static void tcp_get_info_chrono_stats(const struct tcp_sock *tp,
2754 struct tcp_info *info)
2755{
2756 u64 stats[__TCP_CHRONO_MAX], total = 0;
2757 enum tcp_chrono i;
2758
2759 for (i = TCP_CHRONO_BUSY; i < __TCP_CHRONO_MAX; ++i) {
2760 stats[i] = tp->chrono_stat[i - 1];
2761 if (i == tp->chrono_type)
628174cc 2762 stats[i] += tcp_jiffies32 - tp->chrono_start;
efd90174
FY		 2763 stats[i] *= USEC_PER_SEC / HZ;
2764 total += stats[i];
2765 }
2766
2767 info->tcpi_busy_time = total;
2768 info->tcpi_rwnd_limited = stats[TCP_CHRONO_RWND_LIMITED];
2769 info->tcpi_sndbuf_limited = stats[TCP_CHRONO_SNDBUF_LIMITED];
2770}
2771
1da177e4 2772/* Return information about state of tcp endpoint in API format. */
0df48c26 2773void tcp_get_info(struct sock *sk, struct tcp_info *info)
1da177e4 2774{
35ac838a 2775 const struct tcp_sock *tp = tcp_sk(sk); /* iff sk_type == SOCK_STREAM */
463c84b9 2776 const struct inet_connection_sock *icsk = inet_csk(sk);
0263598c 2777 u32 now;
ff5d7497 2778 u64 rate64;
67db3e4b 2779 bool slow;
fad9dfef 2780 u32 rate;
1da177e4
LT		 2781
2782 memset(info, 0, sizeof(*info));
35ac838a
CG		 2783 if (sk->sk_type != SOCK_STREAM)
2784 return;
1da177e4 2785
00fd38d9
ED		 2786 info->tcpi_state = sk_state_load(sk);
2787
ccbf3bfa
ED		 2788 /* Report meaningful fields for all TCP states, including listeners */
2789 rate = READ_ONCE(sk->sk_pacing_rate);
2790 rate64 = rate != ~0U ? rate : ~0ULL;
f522a5fc 2791 info->tcpi_pacing_rate = rate64;
ccbf3bfa
ED		 2792
2793 rate = READ_ONCE(sk->sk_max_pacing_rate);
2794 rate64 = rate != ~0U ? rate : ~0ULL;
f522a5fc 2795 info->tcpi_max_pacing_rate = rate64;
ccbf3bfa
ED		 2796
2797 info->tcpi_reordering = tp->reordering;
2798 info->tcpi_snd_cwnd = tp->snd_cwnd;
2799
2800 if (info->tcpi_state == TCP_LISTEN) {
2801 /* listeners aliased fields :
2802 * tcpi_unacked -> Number of children ready for accept()
2803 * tcpi_sacked -> max backlog
2804 */
2805 info->tcpi_unacked = sk->sk_ack_backlog;
2806 info->tcpi_sacked = sk->sk_max_ack_backlog;
2807 return;
2808 }
b369e7fd
ED		 2809
2810 slow = lock_sock_fast(sk);
2811
6687e988 2812 info->tcpi_ca_state = icsk->icsk_ca_state;
463c84b9 2813 info->tcpi_retransmits = icsk->icsk_retransmits;
6687e988 2814 info->tcpi_probes = icsk->icsk_probes_out;
463c84b9 2815 info->tcpi_backoff = icsk->icsk_backoff;
1da177e4
LT		 2816
2817 if (tp->rx_opt.tstamp_ok)
2818 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
e60402d0 2819 if (tcp_is_sack(tp))
1da177e4
LT		 2820 info->tcpi_options |= TCPI_OPT_SACK;
2821 if (tp->rx_opt.wscale_ok) {
2822 info->tcpi_options |= TCPI_OPT_WSCALE;
2823 info->tcpi_snd_wscale = tp->rx_opt.snd_wscale;
2824 info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale;
e905a9ed 2825 }
1da177e4 2826
b5c5693b 2827 if (tp->ecn_flags & TCP_ECN_OK)
1da177e4 2828 info->tcpi_options |= TCPI_OPT_ECN;
b5c5693b
ED		 2829 if (tp->ecn_flags & TCP_ECN_SEEN)
2830 info->tcpi_options |= TCPI_OPT_ECN_SEEN;
6f73601e
YC		 2831 if (tp->syn_data_acked)
2832 info->tcpi_options |= TCPI_OPT_SYN_DATA;
1da177e4 2833
463c84b9
ACM		 2834 info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto);
2835 info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato);
c1b4a7e6 2836 info->tcpi_snd_mss = tp->mss_cache;
463c84b9 2837 info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss;
1da177e4 2838
ccbf3bfa
ED		 2839 info->tcpi_unacked = tp->packets_out;
2840 info->tcpi_sacked = tp->sacked_out;
2841
1da177e4
LT		 2842 info->tcpi_lost = tp->lost_out;
2843 info->tcpi_retrans = tp->retrans_out;
2844 info->tcpi_fackets = tp->fackets_out;
2845
d635fbe2 2846 now = tcp_jiffies32;
1da177e4 2847 info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime);
463c84b9 2848 info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime);
1da177e4
LT		 2849 info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp);
2850
d83d8461 2851 info->tcpi_pmtu = icsk->icsk_pmtu_cookie;
1da177e4 2852 info->tcpi_rcv_ssthresh = tp->rcv_ssthresh;
740b0f18
ED		 2853 info->tcpi_rtt = tp->srtt_us >> 3;
2854 info->tcpi_rttvar = tp->mdev_us >> 2;
1da177e4 2855 info->tcpi_snd_ssthresh = tp->snd_ssthresh;
1da177e4 2856 info->tcpi_advmss = tp->advmss;
1da177e4 2857
645f4c6f 2858 info->tcpi_rcv_rtt = tp->rcv_rtt_est.rtt_us >> 3;
1da177e4
LT		 2859 info->tcpi_rcv_space = tp->rcvq_space.space;
2860
2861 info->tcpi_total_retrans = tp->total_retrans;
977cb0ec 2862
f522a5fc
ED		 2863 info->tcpi_bytes_acked = tp->bytes_acked;
2864 info->tcpi_bytes_received = tp->bytes_received;
67db3e4b 2865 info->tcpi_notsent_bytes = max_t(int, 0, tp->write_seq - tp->snd_nxt);
efd90174 2866 tcp_get_info_chrono_stats(tp, info);
67db3e4b 2867
2efd055c
MRL		 2868 info->tcpi_segs_out = tp->segs_out;
2869 info->tcpi_segs_in = tp->segs_in;
cd9b2660 2870
cd9b2660 2871 info->tcpi_min_rtt = tcp_min_rtt(tp);
a44d6eac
MKL		 2872 info->tcpi_data_segs_in = tp->data_segs_in;
2873 info->tcpi_data_segs_out = tp->data_segs_out;
eb8329e0
YC		 2874
2875 info->tcpi_delivery_rate_app_limited = tp->rate_app_limited ? 1 : 0;
0263598c
WW		 2876 rate64 = tcp_compute_delivery_rate(tp);
2877 if (rate64)
f522a5fc 2878 info->tcpi_delivery_rate = rate64;
b369e7fd 2879 unlock_sock_fast(sk, slow);
1da177e4 2880}
1da177e4
LT		 2881EXPORT_SYMBOL_GPL(tcp_get_info);
2882
1c885808
FY		 2883struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk)
2884{
2885 const struct tcp_sock *tp = tcp_sk(sk);
2886 struct sk_buff *stats;
2887 struct tcp_info info;
bb7c19f9
WW		 2888 u64 rate64;
2889 u32 rate;
1c885808 2890
bb7c19f9
WW		 2891 stats = alloc_skb(7 * nla_total_size_64bit(sizeof(u64)) +
2892 3 * nla_total_size(sizeof(u32)) +
2893 2 * nla_total_size(sizeof(u8)), GFP_ATOMIC);
1c885808
FY		 2894 if (!stats)
2895 return NULL;
2896
2897 tcp_get_info_chrono_stats(tp, &info);
2898 nla_put_u64_64bit(stats, TCP_NLA_BUSY,
2899 info.tcpi_busy_time, TCP_NLA_PAD);
2900 nla_put_u64_64bit(stats, TCP_NLA_RWND_LIMITED,
2901 info.tcpi_rwnd_limited, TCP_NLA_PAD);
2902 nla_put_u64_64bit(stats, TCP_NLA_SNDBUF_LIMITED,
2903 info.tcpi_sndbuf_limited, TCP_NLA_PAD);
7e98102f
YC		 2904 nla_put_u64_64bit(stats, TCP_NLA_DATA_SEGS_OUT,
2905 tp->data_segs_out, TCP_NLA_PAD);
2906 nla_put_u64_64bit(stats, TCP_NLA_TOTAL_RETRANS,
2907 tp->total_retrans, TCP_NLA_PAD);
bb7c19f9
WW		 2908
2909 rate = READ_ONCE(sk->sk_pacing_rate);
2910 rate64 = rate != ~0U ? rate : ~0ULL;
2911 nla_put_u64_64bit(stats, TCP_NLA_PACING_RATE, rate64, TCP_NLA_PAD);
2912
2913 rate64 = tcp_compute_delivery_rate(tp);
2914 nla_put_u64_64bit(stats, TCP_NLA_DELIVERY_RATE, rate64, TCP_NLA_PAD);
2915
2916 nla_put_u32(stats, TCP_NLA_SND_CWND, tp->snd_cwnd);
2917 nla_put_u32(stats, TCP_NLA_REORDERING, tp->reordering);
2918 nla_put_u32(stats, TCP_NLA_MIN_RTT, tcp_min_rtt(tp));
2919
2920 nla_put_u8(stats, TCP_NLA_RECUR_RETRANS, inet_csk(sk)->icsk_retransmits);
2921 nla_put_u8(stats, TCP_NLA_DELIVERY_RATE_APP_LMT, !!tp->rate_app_limited);
1c885808
FY		 2922 return stats;
2923}
2924
3fdadf7d
DM		 2925static int do_tcp_getsockopt(struct sock *sk, int level,
2926 int optname, char __user *optval, int __user *optlen)
1da177e4 2927{
295f7324 2928 struct inet_connection_sock *icsk = inet_csk(sk);
1da177e4 2929 struct tcp_sock *tp = tcp_sk(sk);
6fa25166 2930 struct net *net = sock_net(sk);
1da177e4
LT		 2931 int val, len;
2932
1da177e4
LT		 2933 if (get_user(len, optlen))
2934 return -EFAULT;
2935
2936 len = min_t(unsigned int, len, sizeof(int));
2937
2938 if (len < 0)
2939 return -EINVAL;
2940
2941 switch (optname) {
2942 case TCP_MAXSEG:
c1b4a7e6 2943 val = tp->mss_cache;
1da177e4
LT		 2944 if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
2945 val = tp->rx_opt.user_mss;
5e6a3ce6
PE		 2946 if (tp->repair)
2947 val = tp->rx_opt.mss_clamp;
1da177e4
LT		 2948 break;
2949 case TCP_NODELAY:
2950 val = !!(tp->nonagle&TCP_NAGLE_OFF);
2951 break;
2952 case TCP_CORK:
2953 val = !!(tp->nonagle&TCP_NAGLE_CORK);
2954 break;
2955 case TCP_KEEPIDLE:
df19a626 2956 val = keepalive_time_when(tp) / HZ;
1da177e4
LT		 2957 break;
2958 case TCP_KEEPINTVL:
df19a626 2959 val = keepalive_intvl_when(tp) / HZ;
1da177e4
LT		 2960 break;
2961 case TCP_KEEPCNT:
df19a626 2962 val = keepalive_probes(tp);
1da177e4
LT		 2963 break;
2964 case TCP_SYNCNT:
6fa25166 2965 val = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries;
1da177e4
LT		 2966 break;
2967 case TCP_LINGER2:
2968 val = tp->linger2;
2969 if (val >= 0)
1e579caa 2970 val = (val ? : net->ipv4.sysctl_tcp_fin_timeout) / HZ;
1da177e4
LT		 2971 break;
2972 case TCP_DEFER_ACCEPT:
b103cf34
JA		 2973 val = retrans_to_secs(icsk->icsk_accept_queue.rskq_defer_accept,
2974 TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ);
1da177e4
LT		 2975 break;
2976 case TCP_WINDOW_CLAMP:
2977 val = tp->window_clamp;
2978 break;
2979 case TCP_INFO: {
2980 struct tcp_info info;
2981
2982 if (get_user(len, optlen))
2983 return -EFAULT;
2984
2985 tcp_get_info(sk, &info);
2986
2987 len = min_t(unsigned int, len, sizeof(info));
2988 if (put_user(len, optlen))
2989 return -EFAULT;
2990 if (copy_to_user(optval, &info, len))
2991 return -EFAULT;
2992 return 0;
2993 }
6e9250f5
ED		 2994 case TCP_CC_INFO: {
2995 const struct tcp_congestion_ops *ca_ops;
2996 union tcp_cc_info info;
2997 size_t sz = 0;
2998 int attr;
2999
3000 if (get_user(len, optlen))
3001 return -EFAULT;
3002
3003 ca_ops = icsk->icsk_ca_ops;
3004 if (ca_ops && ca_ops->get_info)
3005 sz = ca_ops->get_info(sk, ~0U, &attr, &info);
3006
3007 len = min_t(unsigned int, len, sz);
3008 if (put_user(len, optlen))
3009 return -EFAULT;
3010 if (copy_to_user(optval, &info, len))
3011 return -EFAULT;
3012 return 0;
3013 }
1da177e4 3014 case TCP_QUICKACK:
295f7324 3015 val = !icsk->icsk_ack.pingpong;
1da177e4 3016 break;
5f8ef48d
SH		 3017
3018 case TCP_CONGESTION:
3019 if (get_user(len, optlen))
3020 return -EFAULT;
3021 len = min_t(unsigned int, len, TCP_CA_NAME_MAX);
3022 if (put_user(len, optlen))
3023 return -EFAULT;
6687e988 3024 if (copy_to_user(optval, icsk->icsk_ca_ops->name, len))
5f8ef48d
SH		 3025 return -EFAULT;
3026 return 0;
e56fb50f 3027
734942cc
DW		 3028 case TCP_ULP:
3029 if (get_user(len, optlen))
3030 return -EFAULT;
3031 len = min_t(unsigned int, len, TCP_ULP_NAME_MAX);
d97af30f
DW		 3032 if (!icsk->icsk_ulp_ops) {
3033 if (put_user(0, optlen))
3034 return -EFAULT;
3035 return 0;
3036 }
734942cc
DW		 3037 if (put_user(len, optlen))
3038 return -EFAULT;
3039 if (copy_to_user(optval, icsk->icsk_ulp_ops->name, len))
3040 return -EFAULT;
3041 return 0;
3042
3c0fef0b
JH		 3043 case TCP_THIN_LINEAR_TIMEOUTS:
3044 val = tp->thin_lto;
3045 break;
4a7f6009 3046
3c0fef0b 3047 case TCP_THIN_DUPACK:
4a7f6009 3048 val = 0;
3c0fef0b 3049 break;
dca43c75 3050
ee995283
PE		 3051 case TCP_REPAIR:
3052 val = tp->repair;
3053 break;
3054
3055 case TCP_REPAIR_QUEUE:
3056 if (tp->repair)
3057 val = tp->repair_queue;
3058 else
3059 return -EINVAL;
3060 break;
3061
b1ed4c4f
AV		 3062 case TCP_REPAIR_WINDOW: {
3063 struct tcp_repair_window opt;
3064
3065 if (get_user(len, optlen))
3066 return -EFAULT;
3067
3068 if (len != sizeof(opt))
3069 return -EINVAL;
3070
3071 if (!tp->repair)
3072 return -EPERM;
3073
3074 opt.snd_wl1 = tp->snd_wl1;
3075 opt.snd_wnd = tp->snd_wnd;
3076 opt.max_window = tp->max_window;
3077 opt.rcv_wnd = tp->rcv_wnd;
3078 opt.rcv_wup = tp->rcv_wup;
3079
3080 if (copy_to_user(optval, &opt, len))
3081 return -EFAULT;
3082 return 0;
3083 }
ee995283
PE		 3084 case TCP_QUEUE_SEQ:
3085 if (tp->repair_queue == TCP_SEND_QUEUE)
3086 val = tp->write_seq;
3087 else if (tp->repair_queue == TCP_RECV_QUEUE)
3088 val = tp->rcv_nxt;
3089 else
3090 return -EINVAL;
3091 break;
3092
dca43c75
JC		 3093 case TCP_USER_TIMEOUT:
3094 val = jiffies_to_msecs(icsk->icsk_user_timeout);
3095 break;
1536e285
KN		 3096
3097 case TCP_FASTOPEN:
0536fcc0 3098 val = icsk->icsk_accept_queue.fastopenq.max_qlen;
1536e285
KN		 3099 break;
3100
19f6d3f3
WW		 3101 case TCP_FASTOPEN_CONNECT:
3102 val = tp->fastopen_connect;
3103 break;
3104
93be6ce0 3105 case TCP_TIMESTAMP:
9a568de4 3106 val = tcp_time_stamp_raw() + tp->tsoffset;
93be6ce0 3107 break;
c9bee3b7
ED		 3108 case TCP_NOTSENT_LOWAT:
3109 val = tp->notsent_lowat;
3110 break;
cd8ae852
ED		 3111 case TCP_SAVE_SYN:
3112 val = tp->save_syn;
3113 break;
3114 case TCP_SAVED_SYN: {
3115 if (get_user(len, optlen))
3116 return -EFAULT;
3117
3118 lock_sock(sk);
3119 if (tp->saved_syn) {
aea0929e
EM		 3120 if (len < tp->saved_syn[0]) {
3121 if (put_user(tp->saved_syn[0], optlen)) {
3122 release_sock(sk);
3123 return -EFAULT;
3124 }
3125 release_sock(sk);
3126 return -EINVAL;
3127 }
3128 len = tp->saved_syn[0];
cd8ae852
ED		 3129 if (put_user(len, optlen)) {
3130 release_sock(sk);
3131 return -EFAULT;
3132 }
3133 if (copy_to_user(optval, tp->saved_syn + 1, len)) {
3134 release_sock(sk);
3135 return -EFAULT;
3136 }
3137 tcp_saved_syn_free(tp);
3138 release_sock(sk);
3139 } else {
3140 release_sock(sk);
3141 len = 0;
3142 if (put_user(len, optlen))
3143 return -EFAULT;
3144 }
3145 return 0;
3146 }
1da177e4
LT		 3147 default:
3148 return -ENOPROTOOPT;
3ff50b79 3149 }
1da177e4
LT		 3150
3151 if (put_user(len, optlen))
3152 return -EFAULT;
3153 if (copy_to_user(optval, &val, len))
3154 return -EFAULT;
3155 return 0;
3156}
3157
3fdadf7d
DM		 3158int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval,
3159 int __user *optlen)
3160{
3161 struct inet_connection_sock *icsk = inet_csk(sk);
3162
3163 if (level != SOL_TCP)
3164 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
3165 optval, optlen);
3166 return do_tcp_getsockopt(sk, level, optname, optval, optlen);
3167}
4bc2f18b 3168EXPORT_SYMBOL(tcp_getsockopt);
3fdadf7d
DM		 3169
3170#ifdef CONFIG_COMPAT
543d9cfe
ACM		 3171int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
3172 char __user *optval, int __user *optlen)
3fdadf7d 3173{
dec73ff0
ACM		 3174 if (level != SOL_TCP)
3175 return inet_csk_compat_getsockopt(sk, level, optname,
3176 optval, optlen);
3fdadf7d
DM		 3177 return do_tcp_getsockopt(sk, level, optname, optval, optlen);
3178}
543d9cfe 3179EXPORT_SYMBOL(compat_tcp_getsockopt);
3fdadf7d 3180#endif
1da177e4 3181
cfb6eeb4 3182#ifdef CONFIG_TCP_MD5SIG
349ce993 3183static DEFINE_PER_CPU(struct tcp_md5sig_pool, tcp_md5sig_pool);
71cea17e 3184static DEFINE_MUTEX(tcp_md5sig_mutex);
349ce993 3185static bool tcp_md5sig_pool_populated = false;
cfb6eeb4 3186
71cea17e 3187static void __tcp_alloc_md5sig_pool(void)
cfb6eeb4 3188{
cf80e0e4 3189 struct crypto_ahash *hash;
cfb6eeb4 3190 int cpu;
cfb6eeb4 3191
cf80e0e4 3192 hash = crypto_alloc_ahash("md5", 0, CRYPTO_ALG_ASYNC);
70477371 3193 if (IS_ERR(hash))
cf80e0e4
HX		 3194 return;
3195
cfb6eeb4 3196 for_each_possible_cpu(cpu) {
19689e38 3197 void *scratch = per_cpu(tcp_md5sig_pool, cpu).scratch;
cf80e0e4 3198 struct ahash_request *req;
cfb6eeb4 3199
19689e38
ED		 3200 if (!scratch) {
3201 scratch = kmalloc_node(sizeof(union tcp_md5sum_block) +
3202 sizeof(struct tcphdr),
3203 GFP_KERNEL,
3204 cpu_to_node(cpu));
3205 if (!scratch)
3206 return;
3207 per_cpu(tcp_md5sig_pool, cpu).scratch = scratch;
3208 }
cf80e0e4
HX		 3209 if (per_cpu(tcp_md5sig_pool, cpu).md5_req)
3210 continue;
3211
3212 req = ahash_request_alloc(hash, GFP_KERNEL);
3213 if (!req)
3214 return;
3215
3216 ahash_request_set_callback(req, 0, NULL, NULL);
3217
3218 per_cpu(tcp_md5sig_pool, cpu).md5_req = req;
cfb6eeb4 3219 }
349ce993
ED		 3220 /* before setting tcp_md5sig_pool_populated, we must commit all writes
3221 * to memory. See smp_rmb() in tcp_get_md5sig_pool()
71cea17e
ED		 3222 */
3223 smp_wmb();
349ce993 3224 tcp_md5sig_pool_populated = true;
cfb6eeb4
YH		 3225}
3226
71cea17e 3227bool tcp_alloc_md5sig_pool(void)
cfb6eeb4 3228{
349ce993 3229 if (unlikely(!tcp_md5sig_pool_populated)) {
71cea17e
ED		 3230 mutex_lock(&tcp_md5sig_mutex);
3231
349ce993 3232 if (!tcp_md5sig_pool_populated)
71cea17e
ED		 3233 __tcp_alloc_md5sig_pool();
3234
3235 mutex_unlock(&tcp_md5sig_mutex);
cfb6eeb4 3236 }
349ce993 3237 return tcp_md5sig_pool_populated;
cfb6eeb4 3238}
cfb6eeb4
YH		 3239EXPORT_SYMBOL(tcp_alloc_md5sig_pool);
3240
35790c04
ED		 3241
3242/**
3243 * tcp_get_md5sig_pool - get md5sig_pool for this user
3244 *
3245 * We use percpu structure, so if we succeed, we exit with preemption
3246 * and BH disabled, to make sure another thread or softirq handling
3247 * wont try to get same context.
3248 */
3249struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
cfb6eeb4 3250{
35790c04 3251 local_bh_disable();
cfb6eeb4 3252
349ce993
ED		 3253 if (tcp_md5sig_pool_populated) {
3254 /* coupled with smp_wmb() in __tcp_alloc_md5sig_pool() */
3255 smp_rmb();
3256 return this_cpu_ptr(&tcp_md5sig_pool);
3257 }
35790c04
ED		 3258 local_bh_enable();
3259 return NULL;
3260}
3261EXPORT_SYMBOL(tcp_get_md5sig_pool);
cfb6eeb4 3262
49a72dfb 3263int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp,
cf533ea5 3264 const struct sk_buff *skb, unsigned int header_len)
49a72dfb
AL		 3265{
3266 struct scatterlist sg;
3267 const struct tcphdr *tp = tcp_hdr(skb);
cf80e0e4 3268 struct ahash_request *req = hp->md5_req;
95c96174
ED		 3269 unsigned int i;
3270 const unsigned int head_data_len = skb_headlen(skb) > header_len ?
3271 skb_headlen(skb) - header_len : 0;
49a72dfb 3272 const struct skb_shared_info *shi = skb_shinfo(skb);
d7fd1b57 3273 struct sk_buff *frag_iter;
49a72dfb
AL		 3274
3275 sg_init_table(&sg, 1);
3276
3277 sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len);
cf80e0e4
HX		 3278 ahash_request_set_crypt(req, &sg, NULL, head_data_len);
3279 if (crypto_ahash_update(req))
49a72dfb
AL		 3280 return 1;
3281
3282 for (i = 0; i < shi->nr_frags; ++i) {
3283 const struct skb_frag_struct *f = &shi->frags[i];
54d27fcb
ED		 3284 unsigned int offset = f->page_offset;
3285 struct page *page = skb_frag_page(f) + (offset >> PAGE_SHIFT);
3286
3287 sg_set_page(&sg, page, skb_frag_size(f),
3288 offset_in_page(offset));
cf80e0e4
HX		 3289 ahash_request_set_crypt(req, &sg, NULL, skb_frag_size(f));
3290 if (crypto_ahash_update(req))
49a72dfb
AL		 3291 return 1;
3292 }
3293
d7fd1b57
ED		 3294 skb_walk_frags(skb, frag_iter)
3295 if (tcp_md5_hash_skb_data(hp, frag_iter, 0))
3296 return 1;
3297
49a72dfb
AL		 3298 return 0;
3299}
49a72dfb
AL		 3300EXPORT_SYMBOL(tcp_md5_hash_skb_data);
3301
cf533ea5 3302int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, const struct tcp_md5sig_key *key)
49a72dfb
AL		 3303{
3304 struct scatterlist sg;
3305
3306 sg_init_one(&sg, key->key, key->keylen);
cf80e0e4
HX		 3307 ahash_request_set_crypt(hp->md5_req, &sg, NULL, key->keylen);
3308 return crypto_ahash_update(hp->md5_req);
49a72dfb 3309}
49a72dfb
AL		 3310EXPORT_SYMBOL(tcp_md5_hash_key);
3311
cfb6eeb4
YH		 3312#endif
3313
4ac02bab
AK		 3314void tcp_done(struct sock *sk)
3315{
8336886f
JC		 3316 struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
3317
5a5f3a8d 3318 if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
c10d9310 3319 TCP_INC_STATS(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
4ac02bab
AK		 3320
3321 tcp_set_state(sk, TCP_CLOSE);
3322 tcp_clear_xmit_timers(sk);
00db4124 3323 if (req)
8336886f 3324 reqsk_fastopen_remove(sk, req, false);
4ac02bab
AK		 3325
3326 sk->sk_shutdown = SHUTDOWN_MASK;
3327
3328 if (!sock_flag(sk, SOCK_DEAD))
3329 sk->sk_state_change(sk);
3330 else
3331 inet_csk_destroy_sock(sk);
3332}
3333EXPORT_SYMBOL_GPL(tcp_done);
3334
c1e64e29
LC		 3335int tcp_abort(struct sock *sk, int err)
3336{
3337 if (!sk_fullsock(sk)) {
07f6f4a3
ED		 3338 if (sk->sk_state == TCP_NEW_SYN_RECV) {
3339 struct request_sock *req = inet_reqsk(sk);
3340
3341 local_bh_disable();
3342 inet_csk_reqsk_queue_drop_and_put(req->rsk_listener,
3343 req);
3344 local_bh_enable();
3345 return 0;
3346 }
c1e64e29
LC		 3347 return -EOPNOTSUPP;
3348 }
3349
3350 /* Don't race with userspace socket closes such as tcp_close. */
3351 lock_sock(sk);
3352
2010b93e
LC		 3353 if (sk->sk_state == TCP_LISTEN) {
3354 tcp_set_state(sk, TCP_CLOSE);
3355 inet_csk_listen_stop(sk);
3356 }
3357
c1e64e29
LC		 3358 /* Don't race with BH socket closes such as inet_csk_listen_stop. */
3359 local_bh_disable();
3360 bh_lock_sock(sk);
3361
3362 if (!sock_flag(sk, SOCK_DEAD)) {
3363 sk->sk_err = err;
3364 /* This barrier is coupled with smp_rmb() in tcp_poll() */
3365 smp_wmb();
3366 sk->sk_error_report(sk);
3367 if (tcp_need_reset(sk->sk_state))
3368 tcp_send_active_reset(sk, GFP_ATOMIC);
3369 tcp_done(sk);
3370 }
3371
3372 bh_unlock_sock(sk);
3373 local_bh_enable();
3374 release_sock(sk);
c1e64e29
LC		 3375 return 0;
3376}
3377EXPORT_SYMBOL_GPL(tcp_abort);
3378
5f8ef48d 3379extern struct tcp_congestion_ops tcp_reno;
1da177e4
LT		 3380
3381static __initdata unsigned long thash_entries;
3382static int __init set_thash_entries(char *str)
3383{
413c27d8
EZ		 3384 ssize_t ret;
3385
1da177e4
LT		 3386 if (!str)
3387 return 0;
413c27d8
EZ		 3388
3389 ret = kstrtoul(str, 0, &thash_entries);
3390 if (ret)
3391 return 0;
3392
1da177e4
LT		 3393 return 1;
3394}
3395__setup("thash_entries=", set_thash_entries);
3396
47d7a88c 3397static void __init tcp_init_mem(void)
4acb4190 3398{
b66e91cc
ED		 3399 unsigned long limit = nr_free_buffer_pages() / 16;
3400
4acb4190 3401 limit = max(limit, 128UL);
b66e91cc
ED		 3402 sysctl_tcp_mem[0] = limit / 4 * 3; /* 4.68 % */
3403 sysctl_tcp_mem[1] = limit; /* 6.25 % */
3404 sysctl_tcp_mem[2] = sysctl_tcp_mem[0] * 2; /* 9.37 % */
4acb4190
GC		 3405}
3406
1da177e4
LT		 3407void __init tcp_init(void)
3408{
b49960a0 3409 int max_rshare, max_wshare, cnt;
b2d3ea4a 3410 unsigned long limit;
074b8517 3411 unsigned int i;
1da177e4 3412
b2d3ea4a
ED		 3413 BUILD_BUG_ON(sizeof(struct tcp_skb_cb) >
3414 FIELD_SIZEOF(struct sk_buff, cb));
1da177e4 3415
908c7f19
TH		 3416 percpu_counter_init(&tcp_sockets_allocated, 0, GFP_KERNEL);
3417 percpu_counter_init(&tcp_orphan_count, 0, GFP_KERNEL);
1946e672 3418 inet_hashinfo_init(&tcp_hashinfo);
6e04e021
ACM		 3419 tcp_hashinfo.bind_bucket_cachep =
3420 kmem_cache_create("tcp_bind_bucket",
3421 sizeof(struct inet_bind_bucket), 0,
20c2df83 3422 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1da177e4 3423
1da177e4
LT		 3424 /* Size and allocate the main established and bind bucket
3425 * hash tables.
3426 *
3427 * The methodology is similar to that of the buffer cache.
3428 */
6e04e021 3429 tcp_hashinfo.ehash =
1da177e4 3430 alloc_large_system_hash("TCP established",
0f7ff927 3431 sizeof(struct inet_ehash_bucket),
1da177e4 3432 thash_entries,
fd90b29d 3433 17, /* one slot per 128 KB of memory */
9e950efa 3434 0,
1da177e4 3435 NULL,
f373b53b 3436 &tcp_hashinfo.ehash_mask,
31fe62b9 3437 0,
0ccfe618 3438 thash_entries ? 0 : 512 * 1024);
05dbc7b5 3439 for (i = 0; i <= tcp_hashinfo.ehash_mask; i++)
3ab5aee7 3440 INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i);
05dbc7b5 3441
230140cf
ED		 3442 if (inet_ehash_locks_alloc(&tcp_hashinfo))
3443 panic("TCP: failed to alloc ehash_locks");
6e04e021 3444 tcp_hashinfo.bhash =
1da177e4 3445 alloc_large_system_hash("TCP bind",
0f7ff927 3446 sizeof(struct inet_bind_hashbucket),
f373b53b 3447 tcp_hashinfo.ehash_mask + 1,
fd90b29d 3448 17, /* one slot per 128 KB of memory */
9e950efa 3449 0,
6e04e021 3450 &tcp_hashinfo.bhash_size,
1da177e4 3451 NULL,
31fe62b9 3452 0,
1da177e4 3453 64 * 1024);
074b8517 3454 tcp_hashinfo.bhash_size = 1U << tcp_hashinfo.bhash_size;
6e04e021
ACM		 3455 for (i = 0; i < tcp_hashinfo.bhash_size; i++) {
3456 spin_lock_init(&tcp_hashinfo.bhash[i].lock);
3457 INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain);
1da177e4
LT		 3458 }
3459
c5ed63d6
ED		 3460
3461 cnt = tcp_hashinfo.ehash_mask + 1;
c5ed63d6 3462 sysctl_tcp_max_orphans = cnt / 2;
1da177e4 3463
a4fe34bf 3464 tcp_init_mem();
c43b874d 3465 /* Set per-socket limits to no more than 1/128 the pressure threshold */
5fb84b14 3466 limit = nr_free_buffer_pages() << (PAGE_SHIFT - 7);
b49960a0
ED		 3467 max_wshare = min(4UL*1024*1024, limit);
3468 max_rshare = min(6UL*1024*1024, limit);
7b4f4b5e 3469
3ab224be 3470 sysctl_tcp_wmem[0] = SK_MEM_QUANTUM;
7b4f4b5e 3471 sysctl_tcp_wmem[1] = 16*1024;
b49960a0 3472 sysctl_tcp_wmem[2] = max(64*1024, max_wshare);
7b4f4b5e 3473
3ab224be 3474 sysctl_tcp_rmem[0] = SK_MEM_QUANTUM;
7b4f4b5e 3475 sysctl_tcp_rmem[1] = 87380;
b49960a0 3476 sysctl_tcp_rmem[2] = max(87380, max_rshare);
1da177e4 3477
afd46503 3478 pr_info("Hash tables configured (established %u bind %u)\n",
058bd4d2 3479 tcp_hashinfo.ehash_mask + 1, tcp_hashinfo.bhash_size);
317a76f9 3480
1946e672 3481 tcp_v4_init();
51c5d0c4 3482 tcp_metrics_init();
55d8694f 3483 BUG_ON(tcp_register_congestion_control(&tcp_reno) != 0);
46d3ceab 3484 tcp_tasklet_init();
1da177e4 3485}
Linux 6.x block layer and io_uring tree(s)