blob: 61082065b26a068975c411b74eb46739ab0632ca [file] [log] [blame]
Thomas Gleixner2874c5f2019-05-27 08:55:01 +02001// SPDX-License-Identifier: GPL-2.0-or-later
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Implementation of the Transmission Control Protocol(TCP).
8 *
Jesper Juhl02c30a82005-05-05 16:16:16 -07009 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -070010 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Mark Evans, <evansmp@uhura.aston.ac.uk>
12 * Corey Minyard <wf-rch!minyard@relay.EU.net>
13 * Florian La Roche, <flla@stud.uni-sb.de>
14 * Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
15 * Linus Torvalds, <torvalds@cs.helsinki.fi>
16 * Alan Cox, <gw4pts@gw4pts.ampr.org>
17 * Matthew Dillon, <dillon@apollo.west.oic.com>
18 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
19 * Jorge Cwik, <jorge@laser.satlink.net>
20 *
21 * Fixes:
22 * Alan Cox : Numerous verify_area() calls
23 * Alan Cox : Set the ACK bit on a reset
24 * Alan Cox : Stopped it crashing if it closed while
25 * sk->inuse=1 and was trying to connect
26 * (tcp_err()).
27 * Alan Cox : All icmp error handling was broken
28 * pointers passed where wrong and the
29 * socket was looked up backwards. Nobody
30 * tested any icmp error code obviously.
31 * Alan Cox : tcp_err() now handled properly. It
32 * wakes people on errors. poll
33 * behaves and the icmp error race
34 * has gone by moving it into sock.c
35 * Alan Cox : tcp_send_reset() fixed to work for
36 * everything not just packets for
37 * unknown sockets.
38 * Alan Cox : tcp option processing.
39 * Alan Cox : Reset tweaked (still not 100%) [Had
40 * syn rule wrong]
41 * Herp Rosmanith : More reset fixes
42 * Alan Cox : No longer acks invalid rst frames.
43 * Acking any kind of RST is right out.
44 * Alan Cox : Sets an ignore me flag on an rst
45 * receive otherwise odd bits of prattle
46 * escape still
47 * Alan Cox : Fixed another acking RST frame bug.
48 * Should stop LAN workplace lockups.
49 * Alan Cox : Some tidyups using the new skb list
50 * facilities
51 * Alan Cox : sk->keepopen now seems to work
52 * Alan Cox : Pulls options out correctly on accepts
53 * Alan Cox : Fixed assorted sk->rqueue->next errors
54 * Alan Cox : PSH doesn't end a TCP read. Switched a
55 * bit to skb ops.
56 * Alan Cox : Tidied tcp_data to avoid a potential
57 * nasty.
58 * Alan Cox : Added some better commenting, as the
59 * tcp is hard to follow
60 * Alan Cox : Removed incorrect check for 20 * psh
61 * Michael O'Reilly : ack < copied bug fix.
62 * Johannes Stille : Misc tcp fixes (not all in yet).
63 * Alan Cox : FIN with no memory -> CRASH
64 * Alan Cox : Added socket option proto entries.
65 * Also added awareness of them to accept.
66 * Alan Cox : Added TCP options (SOL_TCP)
67 * Alan Cox : Switched wakeup calls to callbacks,
68 * so the kernel can layer network
69 * sockets.
70 * Alan Cox : Use ip_tos/ip_ttl settings.
71 * Alan Cox : Handle FIN (more) properly (we hope).
72 * Alan Cox : RST frames sent on unsynchronised
73 * state ack error.
74 * Alan Cox : Put in missing check for SYN bit.
75 * Alan Cox : Added tcp_select_window() aka NET2E
76 * window non shrink trick.
77 * Alan Cox : Added a couple of small NET2E timer
78 * fixes
79 * Charles Hedrick : TCP fixes
80 * Toomas Tamm : TCP window fixes
81 * Alan Cox : Small URG fix to rlogin ^C ack fight
82 * Charles Hedrick : Rewrote most of it to actually work
83 * Linus : Rewrote tcp_read() and URG handling
84 * completely
85 * Gerhard Koerting: Fixed some missing timer handling
86 * Matthew Dillon : Reworked TCP machine states as per RFC
87 * Gerhard Koerting: PC/TCP workarounds
88 * Adam Caldwell : Assorted timer/timing errors
89 * Matthew Dillon : Fixed another RST bug
90 * Alan Cox : Move to kernel side addressing changes.
91 * Alan Cox : Beginning work on TCP fastpathing
92 * (not yet usable)
93 * Arnt Gulbrandsen: Turbocharged tcp_check() routine.
94 * Alan Cox : TCP fast path debugging
95 * Alan Cox : Window clamping
96 * Michael Riepe : Bug in tcp_check()
97 * Matt Dillon : More TCP improvements and RST bug fixes
98 * Matt Dillon : Yet more small nasties remove from the
99 * TCP code (Be very nice to this man if
100 * tcp finally works 100%) 8)
101 * Alan Cox : BSD accept semantics.
102 * Alan Cox : Reset on closedown bug.
103 * Peter De Schrijver : ENOTCONN check missing in tcp_sendto().
104 * Michael Pall : Handle poll() after URG properly in
105 * all cases.
106 * Michael Pall : Undo the last fix in tcp_read_urg()
107 * (multi URG PUSH broke rlogin).
108 * Michael Pall : Fix the multi URG PUSH problem in
109 * tcp_readable(), poll() after URG
110 * works now.
111 * Michael Pall : recv(...,MSG_OOB) never blocks in the
112 * BSD api.
113 * Alan Cox : Changed the semantics of sk->socket to
114 * fix a race and a signal problem with
115 * accept() and async I/O.
116 * Alan Cox : Relaxed the rules on tcp_sendto().
117 * Yury Shevchuk : Really fixed accept() blocking problem.
118 * Craig I. Hagan : Allow for BSD compatible TIME_WAIT for
119 * clients/servers which listen in on
120 * fixed ports.
121 * Alan Cox : Cleaned the above up and shrank it to
122 * a sensible code size.
123 * Alan Cox : Self connect lockup fix.
124 * Alan Cox : No connect to multicast.
125 * Ross Biro : Close unaccepted children on master
126 * socket close.
127 * Alan Cox : Reset tracing code.
128 * Alan Cox : Spurious resets on shutdown.
129 * Alan Cox : Giant 15 minute/60 second timer error
130 * Alan Cox : Small whoops in polling before an
131 * accept.
132 * Alan Cox : Kept the state trace facility since
133 * it's handy for debugging.
134 * Alan Cox : More reset handler fixes.
135 * Alan Cox : Started rewriting the code based on
136 * the RFC's for other useful protocol
137 * references see: Comer, KA9Q NOS, and
138 * for a reference on the difference
139 * between specifications and how BSD
140 * works see the 4.4lite source.
141 * A.N.Kuznetsov : Don't time wait on completion of tidy
142 * close.
143 * Linus Torvalds : Fin/Shutdown & copied_seq changes.
144 * Linus Torvalds : Fixed BSD port reuse to work first syn
145 * Alan Cox : Reimplemented timers as per the RFC
146 * and using multiple timers for sanity.
147 * Alan Cox : Small bug fixes, and a lot of new
148 * comments.
149 * Alan Cox : Fixed dual reader crash by locking
150 * the buffers (much like datagram.c)
151 * Alan Cox : Fixed stuck sockets in probe. A probe
152 * now gets fed up of retrying without
153 * (even a no space) answer.
154 * Alan Cox : Extracted closing code better
155 * Alan Cox : Fixed the closing state machine to
156 * resemble the RFC.
157 * Alan Cox : More 'per spec' fixes.
158 * Jorge Cwik : Even faster checksumming.
159 * Alan Cox : tcp_data() doesn't ack illegal PSH
160 * only frames. At least one pc tcp stack
161 * generates them.
162 * Alan Cox : Cache last socket.
163 * Alan Cox : Per route irtt.
164 * Matt Day : poll()->select() match BSD precisely on error
165 * Alan Cox : New buffers
166 * Marc Tamsky : Various sk->prot->retransmits and
167 * sk->retransmits misupdating fixed.
168 * Fixed tcp_write_timeout: stuck close,
169 * and TCP syn retries gets used now.
170 * Mark Yarvis : In tcp_read_wakeup(), don't send an
171 * ack if state is TCP_CLOSED.
172 * Alan Cox : Look up device on a retransmit - routes may
173 * change. Doesn't yet cope with MSS shrink right
174 * but it's a start!
175 * Marc Tamsky : Closing in closing fixes.
176 * Mike Shaver : RFC1122 verifications.
177 * Alan Cox : rcv_saddr errors.
178 * Alan Cox : Block double connect().
179 * Alan Cox : Small hooks for enSKIP.
180 * Alexey Kuznetsov: Path MTU discovery.
181 * Alan Cox : Support soft errors.
182 * Alan Cox : Fix MTU discovery pathological case
183 * when the remote claims no mtu!
184 * Marc Tamsky : TCP_CLOSE fix.
185 * Colin (G3TNE) : Send a reset on syn ack replies in
186 * window but wrong (fixes NT lpd problems)
187 * Pedro Roque : Better TCP window handling, delayed ack.
188 * Joerg Reuter : No modification of locked buffers in
189 * tcp_do_retransmit()
190 * Eric Schenk : Changed receiver side silly window
191 * avoidance algorithm to BSD style
192 * algorithm. This doubles throughput
193 * against machines running Solaris,
194 * and seems to result in general
195 * improvement.
196 * Stefan Magdalinski : adjusted tcp_readable() to fix FIONREAD
197 * Willy Konynenberg : Transparent proxying support.
198 * Mike McLagan : Routing by source
199 * Keith Owens : Do proper merging with partial SKB's in
200 * tcp_do_sendmsg to avoid burstiness.
201 * Eric Schenk : Fix fast close down bug with
202 * shutdown() followed by close().
203 * Andi Kleen : Make poll agree with SIGIO
204 * Salvatore Sanfilippo : Support SO_LINGER with linger == 1 and
205 * lingertime == 0 (RFC 793 ABORT Call)
206 * Hirokazu Takahashi : Use copy_from_user() instead of
207 * csum_and_copy_from_user() if possible.
208 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 * Description of States:
210 *
211 * TCP_SYN_SENT sent a connection request, waiting for ack
212 *
213 * TCP_SYN_RECV received a connection request, sent ack,
214 * waiting for final ack in three-way handshake.
215 *
216 * TCP_ESTABLISHED connection established
217 *
218 * TCP_FIN_WAIT1 our side has shutdown, waiting to complete
219 * transmission of remaining buffered data
220 *
221 * TCP_FIN_WAIT2 all buffered data sent, waiting for remote
222 * to shutdown
223 *
224 * TCP_CLOSING both sides have shutdown but we still have
225 * data we have to finish sending
226 *
227 * TCP_TIME_WAIT timeout to catch resent junk before entering
228 * closed, can only be entered from FIN_WAIT2
229 * or CLOSING. Required because the other end
230 * may not have gotten our last ACK causing it
231 * to retransmit the data packet (which we ignore)
232 *
233 * TCP_CLOSE_WAIT remote side has shutdown and is waiting for
234 * us to finish writing our data and to shutdown
235 * (we have to close() to move on to LAST_ACK)
236 *
237 * TCP_LAST_ACK out side has shutdown after remote has
238 * shutdown. There may still be data in our
239 * buffer that we have to finish sending
240 *
241 * TCP_CLOSE socket is finished
242 */
243
Joe Perchesafd465032012-03-12 07:03:32 +0000244#define pr_fmt(fmt) "TCP: " fmt
245
Herbert Xucf80e0e2016-01-24 21:20:23 +0800246#include <crypto/hash.h>
Ilpo Järvinen172589c2007-08-28 15:50:33 -0700247#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248#include <linux/module.h>
249#include <linux/types.h>
250#include <linux/fcntl.h>
251#include <linux/poll.h>
Eric Dumazet6e9250f2015-04-28 16:23:49 -0700252#include <linux/inet_diag.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254#include <linux/fs.h>
Jens Axboe9c55e012007-11-06 23:30:13 -0800255#include <linux/skbuff.h>
Andrew Morton81b23b42008-07-03 03:22:02 -0700256#include <linux/scatterlist.h>
Jens Axboe9c55e012007-11-06 23:30:13 -0800257#include <linux/splice.h>
258#include <linux/net.h>
259#include <linux/socket.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260#include <linux/random.h>
Mike Rapoport57c8a662018-10-30 15:09:49 -0700261#include <linux/memblock.h>
Miquel van Smoorenburg57413eb2008-06-27 17:23:57 -0700262#include <linux/highmem.h>
263#include <linux/swap.h>
David S. Millerb8059ea2006-03-25 01:36:56 -0800264#include <linux/cache.h>
Herbert Xuf4c50d92006-06-22 03:02:40 -0700265#include <linux/err.h>
William Allen Simpsonda5c78c2009-12-02 18:12:09 +0000266#include <linux/time.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +0900267#include <linux/slab.h>
Mike Maloney98aaa912017-08-22 17:08:48 -0400268#include <linux/errqueue.h>
Ursula Braun60e2a772017-10-25 11:01:45 +0200269#include <linux/static_key.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270
271#include <net/icmp.h>
Yuchung Chengcf60af02012-07-19 06:43:09 +0000272#include <net/inet_common.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273#include <net/tcp.h>
274#include <net/xfrm.h>
275#include <net/ip.h>
Jens Axboe9c55e012007-11-06 23:30:13 -0800276#include <net/sock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -0800278#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279#include <asm/ioctls.h>
Eliezer Tamir076bb0c2013-07-10 17:13:17 +0300280#include <net/busy_poll.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281
Eric Dumazetdd24c002008-11-25 21:17:14 -0800282struct percpu_counter tcp_orphan_count;
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -0700283EXPORT_SYMBOL_GPL(tcp_orphan_count);
284
Eric W. Biedermana4fe34b2013-10-19 16:25:36 -0700285long sysctl_tcp_mem[3] __read_mostly;
Eric W. Biedermana4fe34b2013-10-19 16:25:36 -0700286EXPORT_SYMBOL(sysctl_tcp_mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287
Eric Dumazet8d987e52010-11-09 23:24:26 +0000288atomic_long_t tcp_memory_allocated; /* Current allocated memory. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289EXPORT_SYMBOL(tcp_memory_allocated);
Eric Dumazet17483762008-11-25 21:16:35 -0800290
Ursula Braun60e2a772017-10-25 11:01:45 +0200291#if IS_ENABLED(CONFIG_SMC)
292DEFINE_STATIC_KEY_FALSE(tcp_have_smc);
293EXPORT_SYMBOL(tcp_have_smc);
294#endif
295
Eric Dumazet17483762008-11-25 21:16:35 -0800296/*
297 * Current number of TCP sockets.
298 */
299struct percpu_counter tcp_sockets_allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300EXPORT_SYMBOL(tcp_sockets_allocated);
301
302/*
Jens Axboe9c55e012007-11-06 23:30:13 -0800303 * TCP splice context
304 */
305struct tcp_splice_state {
306 struct pipe_inode_info *pipe;
307 size_t len;
308 unsigned int flags;
309};
310
311/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 * Pressure flag: try to collapse.
313 * Technical note: it is used by multiple contexts non atomically.
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800314 * All the __sk_mem_schedule() is of this nature: accounting
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 * is strict, actions are advisory and have some latency.
316 */
Eric Dumazet06044752017-06-07 13:29:12 -0700317unsigned long tcp_memory_pressure __read_mostly;
318EXPORT_SYMBOL_GPL(tcp_memory_pressure);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319
Eric Dumazet2e05fcae2019-06-15 13:19:55 -0700320DEFINE_STATIC_KEY_FALSE(tcp_rx_skb_cache_key);
321EXPORT_SYMBOL(tcp_rx_skb_cache_key);
322
323DEFINE_STATIC_KEY_FALSE(tcp_tx_skb_cache_key);
324
Pavel Emelyanov5c52ba12008-07-16 20:28:10 -0700325void tcp_enter_memory_pressure(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326{
Eric Dumazet06044752017-06-07 13:29:12 -0700327 unsigned long val;
328
329 if (tcp_memory_pressure)
330 return;
331 val = jiffies;
332
333 if (!val)
334 val--;
335 if (!cmpxchg(&tcp_memory_pressure, 0, val))
Pavel Emelyanov4e673442008-07-16 20:30:14 -0700336 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337}
Eric Dumazet06044752017-06-07 13:29:12 -0700338EXPORT_SYMBOL_GPL(tcp_enter_memory_pressure);
339
340void tcp_leave_memory_pressure(struct sock *sk)
341{
342 unsigned long val;
343
344 if (!tcp_memory_pressure)
345 return;
346 val = xchg(&tcp_memory_pressure, 0);
347 if (val)
348 NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURESCHRONO,
349 jiffies_to_msecs(jiffies - val));
350}
351EXPORT_SYMBOL_GPL(tcp_leave_memory_pressure);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352
Julian Anastasovb103cf32009-10-19 10:10:40 +0000353/* Convert seconds to retransmits based on initial and max timeout */
354static u8 secs_to_retrans(int seconds, int timeout, int rto_max)
355{
356 u8 res = 0;
357
358 if (seconds > 0) {
359 int period = timeout;
360
361 res = 1;
362 while (seconds > period && res < 255) {
363 res++;
364 timeout <<= 1;
365 if (timeout > rto_max)
366 timeout = rto_max;
367 period += timeout;
368 }
369 }
370 return res;
371}
372
373/* Convert retransmits to seconds based on initial and max timeout */
374static int retrans_to_secs(u8 retrans, int timeout, int rto_max)
375{
376 int period = 0;
377
378 if (retrans > 0) {
379 period = timeout;
380 while (--retrans) {
381 timeout <<= 1;
382 if (timeout > rto_max)
383 timeout = rto_max;
384 period += timeout;
385 }
386 }
387 return period;
388}
389
Wei Wang02635982017-07-28 10:28:20 -0700390static u64 tcp_compute_delivery_rate(const struct tcp_sock *tp)
391{
392 u32 rate = READ_ONCE(tp->rate_delivered);
393 u32 intv = READ_ONCE(tp->rate_interval_us);
394 u64 rate64 = 0;
395
396 if (rate && intv) {
397 rate64 = (u64)rate * tp->mss_cache * USEC_PER_SEC;
398 do_div(rate64, intv);
399 }
400 return rate64;
401}
402
Neal Cardwell900f65d2012-04-19 09:55:21 +0000403/* Address-family independent initialization for a tcp_sock.
404 *
405 * NOTE: A lot of things set to zero explicitly by call to
406 * sk_alloc() so need not be done here.
407 */
408void tcp_init_sock(struct sock *sk)
409{
410 struct inet_connection_sock *icsk = inet_csk(sk);
411 struct tcp_sock *tp = tcp_sk(sk);
412
Yaogong Wang9f5afea2016-09-07 14:49:28 -0700413 tp->out_of_order_queue = RB_ROOT;
Eric Dumazet75c119a2017-10-05 22:21:27 -0700414 sk->tcp_rtx_queue = RB_ROOT;
Neal Cardwell900f65d2012-04-19 09:55:21 +0000415 tcp_init_xmit_timers(sk);
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000416 INIT_LIST_HEAD(&tp->tsq_node);
Eric Dumazete2080072017-10-04 12:59:58 -0700417 INIT_LIST_HEAD(&tp->tsorted_sent_queue);
Neal Cardwell900f65d2012-04-19 09:55:21 +0000418
419 icsk->icsk_rto = TCP_TIMEOUT_INIT;
Eric Dumazet740b0f12014-02-26 14:02:48 -0800420 tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
Eric Dumazetac9517f2017-05-16 14:00:13 -0700421 minmax_reset(&tp->rtt_min, tcp_jiffies32, ~0U);
Neal Cardwell900f65d2012-04-19 09:55:21 +0000422
423 /* So many TCP implementations out there (incorrectly) count the
424 * initial SYN frame in their delayed-ACK and congestion control
425 * algorithms that we must have the following bandaid to talk
426 * efficiently to them. -DaveM
427 */
428 tp->snd_cwnd = TCP_INIT_CWND;
429
Soheil Hassas Yeganehd7722e82016-09-19 23:39:15 -0400430 /* There's a bubble in the pipe until at least the first ACK. */
431 tp->app_limited = ~0U;
432
Neal Cardwell900f65d2012-04-19 09:55:21 +0000433 /* See draft-stevens-tcpca-spec-01 for discussion of the
434 * initialization of these values.
435 */
436 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
437 tp->snd_cwnd_clamp = ~0;
438 tp->mss_cache = TCP_MSS_DEFAULT;
439
Nikolay Borisov1043e252016-02-03 09:46:52 +0200440 tp->reordering = sock_net(sk)->ipv4.sysctl_tcp_reordering;
Florian Westphal55d86942014-09-26 22:37:32 +0200441 tcp_assign_congestion_control(sk);
Neal Cardwell900f65d2012-04-19 09:55:21 +0000442
Andrey Vaginceaa1fe2013-02-11 05:50:17 +0000443 tp->tsoffset = 0;
Priyaranjan Jha1f255692017-11-03 16:38:48 -0700444 tp->rack.reo_wnd_steps = 1;
Andrey Vaginceaa1fe2013-02-11 05:50:17 +0000445
Neal Cardwell900f65d2012-04-19 09:55:21 +0000446 sk->sk_state = TCP_CLOSE;
447
448 sk->sk_write_space = sk_stream_write_space;
449 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
450
451 icsk->icsk_sync_mss = tcp_sync_mss;
452
Eric Dumazet356d1832017-11-07 00:29:28 -0800453 sk->sk_sndbuf = sock_net(sk)->ipv4.sysctl_tcp_wmem[1];
454 sk->sk_rcvbuf = sock_net(sk)->ipv4.sysctl_tcp_rmem[1];
Neal Cardwell900f65d2012-04-19 09:55:21 +0000455
Neal Cardwell900f65d2012-04-19 09:55:21 +0000456 sk_sockets_allocated_inc(sk);
Eric Dumazet0a6b2a12018-02-19 11:56:47 -0800457 sk->sk_route_forced_caps = NETIF_F_GSO;
Neal Cardwell900f65d2012-04-19 09:55:21 +0000458}
459EXPORT_SYMBOL(tcp_init_sock);
460
Eric Dumazet4e8cc222017-10-05 22:21:23 -0700461static void tcp_tx_timestamp(struct sock *sk, u16 tsflags)
Willem de Bruijn4ed2d762014-08-04 22:11:49 -0400462{
Eric Dumazet4e8cc222017-10-05 22:21:23 -0700463 struct sk_buff *skb = tcp_write_queue_tail(sk);
464
Soheil Hassas Yeganehad02c4f2017-01-04 11:19:34 -0500465 if (tsflags && skb) {
Willem de Bruijnf066e2b2014-08-06 15:09:44 -0400466 struct skb_shared_info *shinfo = skb_shinfo(skb);
Soheil Hassas Yeganeh6b084922016-04-02 23:08:08 -0400467 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
Willem de Bruijn4ed2d762014-08-04 22:11:49 -0400468
Soheil Hassas Yeganehc14ac942016-04-02 23:08:12 -0400469 sock_tx_timestamp(sk, tsflags, &shinfo->tx_flags);
Soheil Hassas Yeganeh0a2cf202016-04-27 23:39:01 -0400470 if (tsflags & SOF_TIMESTAMPING_TX_ACK)
471 tcb->txstamp_ack = 1;
472 if (tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
Willem de Bruijnf066e2b2014-08-06 15:09:44 -0400473 shinfo->tskey = TCP_SKB_CB(skb)->seq + skb->len - 1;
474 }
Willem de Bruijn4ed2d762014-08-04 22:11:49 -0400475}
476
John Fastabend8934ce22018-03-28 12:49:15 -0700477static inline bool tcp_stream_is_readable(const struct tcp_sock *tp,
478 int target, struct sock *sk)
479{
480 return (tp->rcv_nxt - tp->copied_seq >= target) ||
481 (sk->sk_prot->stream_memory_read ?
482 sk->sk_prot->stream_memory_read(sk) : false);
483}
484
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485/*
Linus Torvaldsa11e1d42018-06-28 09:43:44 -0700486 * Wait for a TCP event.
487 *
488 * Note that we don't need to lock the socket, as the upper poll layers
489 * take care of normal races (between the test and the event) and we don't
490 * go look at any of the socket buffers directly.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 */
Linus Torvaldsa11e1d42018-06-28 09:43:44 -0700492__poll_t tcp_poll(struct file *file, struct socket *sock, poll_table *wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493{
Linus Torvaldsa11e1d42018-06-28 09:43:44 -0700494 __poll_t mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 struct sock *sk = sock->sk;
Eric Dumazetcf533ea2011-10-21 05:22:42 -0400496 const struct tcp_sock *tp = tcp_sk(sk);
Eric Dumazet00fd38d2015-11-12 08:43:18 -0800497 int state;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498
Karsten Graul89ab0662018-10-23 13:40:39 +0200499 sock_poll_wait(file, sock, wait);
Linus Torvaldsa11e1d42018-06-28 09:43:44 -0700500
Yafang Shao986ffdf2017-12-20 11:12:52 +0800501 state = inet_sk_state_load(sk);
Eric Dumazet00fd38d2015-11-12 08:43:18 -0800502 if (state == TCP_LISTEN)
Arnaldo Carvalho de Melodc40c7b2005-08-23 21:52:58 -0700503 return inet_csk_listen_poll(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
Linus Torvaldsa11e1d42018-06-28 09:43:44 -0700505 /* Socket is not locked. We are protected from async events
506 * by poll logic and correct handling of state changes
507 * made by other threads is impossible in any case.
508 */
509
510 mask = 0;
511
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 /*
Linus Torvaldsa9a08842018-02-11 14:34:03 -0800513 * EPOLLHUP is certainly not done right. But poll() doesn't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 * have a notion of HUP in just one direction, and for a
515 * socket the read side is more interesting.
516 *
Linus Torvaldsa9a08842018-02-11 14:34:03 -0800517 * Some poll() documentation says that EPOLLHUP is incompatible
518 * with the EPOLLOUT/POLLWR flags, so somebody should check this
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 * all. But careful, it tends to be safer to return too many
520 * bits than too few, and you can easily break real applications
521 * if you don't tell them that something has hung up!
522 *
523 * Check-me.
524 *
Linus Torvaldsa9a08842018-02-11 14:34:03 -0800525 * Check number 1. EPOLLHUP is _UNMASKABLE_ event (see UNIX98 and
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 * our fs/select.c). It means that after we received EOF,
527 * poll always returns immediately, making impossible poll() on write()
Linus Torvaldsa9a08842018-02-11 14:34:03 -0800528 * in state CLOSE_WAIT. One solution is evident --- to set EPOLLHUP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 * if and only if shutdown has been made in both directions.
530 * Actually, it is interesting to look how Solaris and DUX
Linus Torvaldsa9a08842018-02-11 14:34:03 -0800531 * solve this dilemma. I would prefer, if EPOLLHUP were maskable,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 * then we could set it on SND_SHUTDOWN. BTW examples given
533 * in Stevens' books assume exactly this behaviour, it explains
Linus Torvaldsa9a08842018-02-11 14:34:03 -0800534 * why EPOLLHUP is incompatible with EPOLLOUT. --ANK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 *
536 * NOTE. Check for TCP_CLOSE is added. The goal is to prevent
537 * blocking on fresh not-connected or disconnected socket. --ANK
538 */
Eric Dumazet00fd38d2015-11-12 08:43:18 -0800539 if (sk->sk_shutdown == SHUTDOWN_MASK || state == TCP_CLOSE)
Linus Torvaldsa9a08842018-02-11 14:34:03 -0800540 mask |= EPOLLHUP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 if (sk->sk_shutdown & RCV_SHUTDOWN)
Linus Torvaldsa9a08842018-02-11 14:34:03 -0800542 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543
Jerry Chu83368862012-08-31 12:29:12 +0000544 /* Connected or passive Fast Open socket? */
Eric Dumazet00fd38d2015-11-12 08:43:18 -0800545 if (state != TCP_SYN_SENT &&
546 (state != TCP_SYN_RECV || tp->fastopen_rsk)) {
David S. Millerc7004482008-10-06 10:43:54 -0700547 int target = sock_rcvlowat(sk, 0, INT_MAX);
548
549 if (tp->urg_seq == tp->copied_seq &&
550 !sock_flag(sk, SOCK_URGINLINE) &&
551 tp->urg_data)
Alexandra Kossovskyb634f872010-03-18 20:29:24 -0700552 target++;
David S. Millerc7004482008-10-06 10:43:54 -0700553
John Fastabend8934ce22018-03-28 12:49:15 -0700554 if (tcp_stream_is_readable(tp, target, sk))
Linus Torvaldsa9a08842018-02-11 14:34:03 -0800555 mask |= EPOLLIN | EPOLLRDNORM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556
557 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
Eric Dumazet64dc6132013-07-22 20:26:31 -0700558 if (sk_stream_is_writeable(sk)) {
Linus Torvaldsa9a08842018-02-11 14:34:03 -0800559 mask |= EPOLLOUT | EPOLLWRNORM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 } else { /* send SIGIO later */
Eric Dumazet9cd3e072015-11-29 20:03:10 -0800561 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
563
564 /* Race breaker. If space is freed after
565 * wspace test but before the flags are set,
jbaron@akamai.com3c715122015-04-20 20:05:07 +0000566 * IO signal will be lost. Memory barrier
567 * pairs with the input side.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 */
jbaron@akamai.com3c715122015-04-20 20:05:07 +0000569 smp_mb__after_atomic();
Eric Dumazet64dc6132013-07-22 20:26:31 -0700570 if (sk_stream_is_writeable(sk))
Linus Torvaldsa9a08842018-02-11 14:34:03 -0800571 mask |= EPOLLOUT | EPOLLWRNORM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 }
KOSAKI Motohirod84ba632010-08-24 16:05:48 +0000573 } else
Linus Torvaldsa9a08842018-02-11 14:34:03 -0800574 mask |= EPOLLOUT | EPOLLWRNORM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575
576 if (tp->urg_data & TCP_URG_VALID)
Linus Torvaldsa9a08842018-02-11 14:34:03 -0800577 mask |= EPOLLPRI;
Davide Carattid68be712017-04-26 19:07:35 +0200578 } else if (state == TCP_SYN_SENT && inet_sk(sk)->defer_connect) {
Wei Wang19f6d3f2017-01-23 10:59:22 -0800579 /* Active TCP fastopen socket with defer_connect
Linus Torvaldsa9a08842018-02-11 14:34:03 -0800580 * Return EPOLLOUT so application can call write()
Wei Wang19f6d3f2017-01-23 10:59:22 -0800581 * in order for kernel to generate SYN+data
582 */
Linus Torvaldsa9a08842018-02-11 14:34:03 -0800583 mask |= EPOLLOUT | EPOLLWRNORM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 }
Tom Marshalla4d25802010-09-20 15:42:05 -0700585 /* This barrier is coupled with smp_wmb() in tcp_reset() */
586 smp_rmb();
Willem de Bruijn4ed2d762014-08-04 22:11:49 -0400587 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
Linus Torvaldsa9a08842018-02-11 14:34:03 -0800588 mask |= EPOLLERR;
Tom Marshalla4d25802010-09-20 15:42:05 -0700589
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 return mask;
591}
Linus Torvaldsa11e1d42018-06-28 09:43:44 -0700592EXPORT_SYMBOL(tcp_poll);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
594int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
595{
596 struct tcp_sock *tp = tcp_sk(sk);
597 int answ;
Eric Dumazet0e71c552012-10-21 20:06:56 +0000598 bool slow;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599
600 switch (cmd) {
601 case SIOCINQ:
602 if (sk->sk_state == TCP_LISTEN)
603 return -EINVAL;
604
Eric Dumazet0e71c552012-10-21 20:06:56 +0000605 slow = lock_sock_fast(sk);
Tom Herbert473bd232016-03-07 14:11:05 -0800606 answ = tcp_inq(sk);
Eric Dumazet0e71c552012-10-21 20:06:56 +0000607 unlock_sock_fast(sk, slow);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 break;
609 case SIOCATMARK:
610 answ = tp->urg_data && tp->urg_seq == tp->copied_seq;
611 break;
612 case SIOCOUTQ:
613 if (sk->sk_state == TCP_LISTEN)
614 return -EINVAL;
615
616 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
617 answ = 0;
618 else
619 answ = tp->write_seq - tp->snd_una;
620 break;
Mario Schuknecht2f4e1b32011-03-09 14:08:09 -0800621 case SIOCOUTQNSD:
622 if (sk->sk_state == TCP_LISTEN)
623 return -EINVAL;
624
625 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
626 answ = 0;
627 else
628 answ = tp->write_seq - tp->snd_nxt;
629 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 default:
631 return -ENOIOCTLCMD;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700632 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
634 return put_user(answ, (int __user *)arg);
635}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000636EXPORT_SYMBOL(tcp_ioctl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb)
639{
Eric Dumazet4de075e2011-09-27 13:25:05 -0400640 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 tp->pushed_seq = tp->write_seq;
642}
643
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000644static inline bool forced_push(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645{
646 return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1));
647}
648
Eric Dumazetf4a775d2014-09-22 16:29:32 -0700649static void skb_entail(struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700651 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo352d4802006-11-17 19:59:12 -0200652 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
653
654 skb->csum = 0;
655 tcb->seq = tcb->end_seq = tp->write_seq;
Eric Dumazet4de075e2011-09-27 13:25:05 -0400656 tcb->tcp_flags = TCPHDR_ACK;
Arnaldo Carvalho de Melo352d4802006-11-17 19:59:12 -0200657 tcb->sacked = 0;
Eric Dumazetf4a775d2014-09-22 16:29:32 -0700658 __skb_header_release(skb);
David S. Millerfe067e82007-03-07 12:12:44 -0800659 tcp_add_write_queue_tail(sk, skb);
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800660 sk->sk_wmem_queued += skb->truesize;
661 sk_mem_charge(sk, skb->truesize);
David S. Miller89ebd192005-08-23 10:13:06 -0700662 if (tp->nonagle & TCP_NAGLE_PUSH)
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900663 tp->nonagle &= ~TCP_NAGLE_PUSH;
Eric Dumazet6f021c62015-08-21 12:30:00 -0700664
665 tcp_slow_start_after_idle_check(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666}
667
Krishna Kumarafeca342009-12-10 07:16:52 +0000668static inline void tcp_mark_urg(struct tcp_sock *tp, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669{
Ilpo Järvinen33f5f572008-10-07 14:43:06 -0700670 if (flags & MSG_OOB)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 tp->snd_up = tp->write_seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672}
673
Eric Dumazetf54b3112013-12-05 22:36:05 -0800674/* If a not yet filled skb is pushed, do not send it if
Eric Dumazeta181ceb2013-12-17 09:58:30 -0800675 * we have data packets in Qdisc or NIC queues :
Eric Dumazetf54b3112013-12-05 22:36:05 -0800676 * Because TX completion will happen shortly, it gives a chance
677 * to coalesce future sendmsg() payload into this skb, without
678 * need for a timer, and with no latency trade off.
679 * As packets containing data payload have a bigger truesize
Eric Dumazeta181ceb2013-12-17 09:58:30 -0800680 * than pure acks (dataless) packets, the last checks prevent
681 * autocorking if we only have an ACK in Qdisc/NIC queues,
682 * or if TX completion was delayed after we processed ACK packet.
Eric Dumazetf54b3112013-12-05 22:36:05 -0800683 */
684static bool tcp_should_autocork(struct sock *sk, struct sk_buff *skb,
685 int size_goal)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686{
Eric Dumazetf54b3112013-12-05 22:36:05 -0800687 return skb->len < size_goal &&
Eric Dumazet790f00e2017-10-27 07:47:29 -0700688 sock_net(sk)->ipv4.sysctl_tcp_autocorking &&
Eric Dumazet114f39f2018-05-02 20:25:13 -0700689 !tcp_rtx_queue_empty(sk) &&
Reshetova, Elena14afee42017-06-30 13:08:00 +0300690 refcount_read(&sk->sk_wmem_alloc) > skb->truesize;
Eric Dumazetf54b3112013-12-05 22:36:05 -0800691}
Krishna Kumarafeca342009-12-10 07:16:52 +0000692
Eric Dumazetf54b3112013-12-05 22:36:05 -0800693static void tcp_push(struct sock *sk, int flags, int mss_now,
694 int nonagle, int size_goal)
695{
696 struct tcp_sock *tp = tcp_sk(sk);
697 struct sk_buff *skb;
Krishna Kumarafeca342009-12-10 07:16:52 +0000698
Eric Dumazetf54b3112013-12-05 22:36:05 -0800699 skb = tcp_write_queue_tail(sk);
Eric Dumazet75c119a2017-10-05 22:21:27 -0700700 if (!skb)
701 return;
Eric Dumazetf54b3112013-12-05 22:36:05 -0800702 if (!(flags & MSG_MORE) || forced_push(tp))
703 tcp_mark_push(tp, skb);
704
705 tcp_mark_urg(tp, flags);
706
707 if (tcp_should_autocork(sk, skb, size_goal)) {
708
709 /* avoid atomic op if TSQ_THROTTLED bit is already set */
Eric Dumazet7aa54702016-12-03 11:14:57 -0800710 if (!test_bit(TSQ_THROTTLED, &sk->sk_tsq_flags)) {
Eric Dumazetf54b3112013-12-05 22:36:05 -0800711 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAUTOCORKING);
Eric Dumazet7aa54702016-12-03 11:14:57 -0800712 set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags);
Eric Dumazetf54b3112013-12-05 22:36:05 -0800713 }
Eric Dumazeta181ceb2013-12-17 09:58:30 -0800714 /* It is possible TX completion already happened
715 * before we set TSQ_THROTTLED.
716 */
Reshetova, Elena14afee42017-06-30 13:08:00 +0300717 if (refcount_read(&sk->sk_wmem_alloc) > skb->truesize)
Eric Dumazeta181ceb2013-12-17 09:58:30 -0800718 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 }
Eric Dumazetf54b3112013-12-05 22:36:05 -0800720
721 if (flags & MSG_MORE)
722 nonagle = TCP_NAGLE_CORK;
723
724 __tcp_push_pending_frames(sk, mss_now, nonagle);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725}
726
Adrian Bunk6ff77512007-11-06 23:32:26 -0800727static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
728 unsigned int offset, size_t len)
Jens Axboe9c55e012007-11-06 23:30:13 -0800729{
730 struct tcp_splice_state *tss = rd_desc->arg.data;
Willy Tarreau33966dd02009-01-13 16:04:36 -0800731 int ret;
Jens Axboe9c55e012007-11-06 23:30:13 -0800732
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +0200733 ret = skb_splice_bits(skb, skb->sk, offset, tss->pipe,
Al Viro25869262016-09-17 21:02:10 -0400734 min(rd_desc->count, len), tss->flags);
Willy Tarreau33966dd02009-01-13 16:04:36 -0800735 if (ret > 0)
736 rd_desc->count -= ret;
737 return ret;
Jens Axboe9c55e012007-11-06 23:30:13 -0800738}
739
740static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss)
741{
742 /* Store TCP splice context information in read_descriptor_t. */
743 read_descriptor_t rd_desc = {
744 .arg.data = tss,
Willy Tarreau33966dd02009-01-13 16:04:36 -0800745 .count = tss->len,
Jens Axboe9c55e012007-11-06 23:30:13 -0800746 };
747
748 return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv);
749}
750
751/**
752 * tcp_splice_read - splice data from TCP socket to a pipe
753 * @sock: socket to splice from
754 * @ppos: position (not valid)
755 * @pipe: pipe to splice to
756 * @len: number of bytes to splice
757 * @flags: splice modifier flags
758 *
759 * Description:
760 * Will read pages from given socket and fill them into a pipe.
761 *
762 **/
763ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos,
764 struct pipe_inode_info *pipe, size_t len,
765 unsigned int flags)
766{
767 struct sock *sk = sock->sk;
768 struct tcp_splice_state tss = {
769 .pipe = pipe,
770 .len = len,
771 .flags = flags,
772 };
773 long timeo;
774 ssize_t spliced;
775 int ret;
776
Changli Gao3a047bf2010-07-12 21:00:12 +0000777 sock_rps_record_flow(sk);
Jens Axboe9c55e012007-11-06 23:30:13 -0800778 /*
779 * We can't seek on a socket input
780 */
781 if (unlikely(*ppos))
782 return -ESPIPE;
783
784 ret = spliced = 0;
785
786 lock_sock(sk);
787
Eric Dumazet42324c62009-10-01 15:26:00 -0700788 timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK);
Jens Axboe9c55e012007-11-06 23:30:13 -0800789 while (tss.len) {
790 ret = __tcp_splice_read(sk, &tss);
791 if (ret < 0)
792 break;
793 else if (!ret) {
794 if (spliced)
795 break;
Jens Axboe9c55e012007-11-06 23:30:13 -0800796 if (sock_flag(sk, SOCK_DONE))
797 break;
798 if (sk->sk_err) {
799 ret = sock_error(sk);
800 break;
801 }
802 if (sk->sk_shutdown & RCV_SHUTDOWN)
803 break;
804 if (sk->sk_state == TCP_CLOSE) {
805 /*
806 * This occurs when user tries to read
807 * from never connected socket.
808 */
Eric Dumazetc47078d2018-07-07 23:15:56 -0700809 ret = -ENOTCONN;
Jens Axboe9c55e012007-11-06 23:30:13 -0800810 break;
811 }
812 if (!timeo) {
813 ret = -EAGAIN;
814 break;
815 }
Eric Dumazetccf7abb2017-02-03 14:59:38 -0800816 /* if __tcp_splice_read() got nothing while we have
817 * an skb in receive queue, we do not want to loop.
818 * This might happen with URG data.
819 */
820 if (!skb_queue_empty(&sk->sk_receive_queue))
821 break;
Sabrina Dubrocadfbafc92015-07-24 18:19:25 +0200822 sk_wait_data(sk, &timeo, NULL);
Jens Axboe9c55e012007-11-06 23:30:13 -0800823 if (signal_pending(current)) {
824 ret = sock_intr_errno(timeo);
825 break;
826 }
827 continue;
828 }
829 tss.len -= ret;
830 spliced += ret;
831
Willy Tarreau33966dd02009-01-13 16:04:36 -0800832 if (!timeo)
833 break;
Jens Axboe9c55e012007-11-06 23:30:13 -0800834 release_sock(sk);
835 lock_sock(sk);
836
837 if (sk->sk_err || sk->sk_state == TCP_CLOSE ||
Willy Tarreau33966dd02009-01-13 16:04:36 -0800838 (sk->sk_shutdown & RCV_SHUTDOWN) ||
Jens Axboe9c55e012007-11-06 23:30:13 -0800839 signal_pending(current))
840 break;
841 }
842
843 release_sock(sk);
844
845 if (spliced)
846 return spliced;
847
848 return ret;
849}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000850EXPORT_SYMBOL(tcp_splice_read);
Jens Axboe9c55e012007-11-06 23:30:13 -0800851
Eric Dumazeteb934472015-05-19 13:26:55 -0700852struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
853 bool force_schedule)
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100854{
855 struct sk_buff *skb;
856
Eric Dumazet4f661542019-03-26 08:34:55 -0700857 if (likely(!size)) {
858 skb = sk->sk_tx_skb_cache;
Eric Dumazet858f5012019-05-15 09:10:15 -0700859 if (skb) {
Eric Dumazetd7cc399e2019-04-19 16:02:03 -0700860 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
Eric Dumazet472c2e02019-03-22 08:56:39 -0700861 sk->sk_tx_skb_cache = NULL;
862 pskb_trim(skb, 0);
863 INIT_LIST_HEAD(&skb->tcp_tsorted_anchor);
Eric Dumazeteb70a1a2019-03-29 12:46:17 -0700864 skb_shinfo(skb)->tx_flags = 0;
865 memset(TCP_SKB_CB(skb), 0, sizeof(struct tcp_skb_cb));
Eric Dumazet472c2e02019-03-22 08:56:39 -0700866 return skb;
867 }
868 }
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100869 /* The TCP header must be at least 32-bit aligned. */
870 size = ALIGN(size, 4);
871
Eric Dumazet8e4d9802015-05-15 12:39:28 -0700872 if (unlikely(tcp_under_memory_pressure(sk)))
873 sk_mem_reclaim_partial(sk);
874
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100875 skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp);
Eric Dumazet8e4d9802015-05-15 12:39:28 -0700876 if (likely(skb)) {
Eric Dumazeteb934472015-05-19 13:26:55 -0700877 bool mem_scheduled;
Eric Dumazet8e4d9802015-05-15 12:39:28 -0700878
Eric Dumazeteb934472015-05-19 13:26:55 -0700879 if (force_schedule) {
880 mem_scheduled = true;
Eric Dumazet8e4d9802015-05-15 12:39:28 -0700881 sk_forced_mem_schedule(sk, skb->truesize);
882 } else {
Eric Dumazeteb934472015-05-19 13:26:55 -0700883 mem_scheduled = sk_wmem_schedule(sk, skb->truesize);
Eric Dumazet8e4d9802015-05-15 12:39:28 -0700884 }
Eric Dumazeteb934472015-05-19 13:26:55 -0700885 if (likely(mem_scheduled)) {
Eric Dumazeta21d4572012-04-10 20:30:48 +0000886 skb_reserve(skb, sk->sk_prot->max_header);
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100887 /*
888 * Make sure that we have exactly size bytes
889 * available to the caller, no more, no less.
890 */
Eric Dumazet16fad692013-03-14 05:40:32 +0000891 skb->reserved_tailroom = skb->end - skb->tail - size;
Eric Dumazete2080072017-10-04 12:59:58 -0700892 INIT_LIST_HEAD(&skb->tcp_tsorted_anchor);
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100893 return skb;
894 }
895 __kfree_skb(skb);
896 } else {
Pavel Emelyanov5c52ba12008-07-16 20:28:10 -0700897 sk->sk_prot->enter_memory_pressure(sk);
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100898 sk_stream_moderate_sndbuf(sk);
899 }
900 return NULL;
901}
902
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000903static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now,
904 int large_allowed)
905{
906 struct tcp_sock *tp = tcp_sk(sk);
Eric Dumazet6c09fa02015-03-05 08:03:06 -0800907 u32 new_size_goal, size_goal;
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000908
Eric Dumazet74d4a8f2018-02-19 11:56:48 -0800909 if (!large_allowed)
Eric Dumazet605ad7f2014-12-07 12:22:18 -0800910 return mss_now;
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000911
Eric Dumazet6c09fa02015-03-05 08:03:06 -0800912 /* Note : tcp_tso_autosize() will eventually split this later */
913 new_size_goal = sk->sk_gso_max_size - 1 - MAX_TCP_HEADER;
Eric Dumazet605ad7f2014-12-07 12:22:18 -0800914 new_size_goal = tcp_bound_to_half_wnd(tp, new_size_goal);
Eric Dumazet95bd09e2013-08-27 05:46:32 -0700915
Eric Dumazet605ad7f2014-12-07 12:22:18 -0800916 /* We try hard to avoid divides here */
917 size_goal = tp->gso_segs * mss_now;
918 if (unlikely(new_size_goal < size_goal ||
919 new_size_goal >= size_goal + mss_now)) {
920 tp->gso_segs = min_t(u16, new_size_goal / mss_now,
921 sk->sk_gso_max_segs);
922 size_goal = tp->gso_segs * mss_now;
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000923 }
924
Eric Dumazet605ad7f2014-12-07 12:22:18 -0800925 return max(size_goal, mss_now);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000926}
927
928static int tcp_send_mss(struct sock *sk, int *size_goal, int flags)
929{
930 int mss_now;
931
932 mss_now = tcp_current_mss(sk);
933 *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB));
934
935 return mss_now;
936}
937
Eric Dumazetfdfc5c82019-08-26 09:19:15 -0700938/* In some cases, both sendpage() and sendmsg() could have added
939 * an skb to the write queue, but failed adding payload on it.
940 * We need to remove it to consume less memory, but more
941 * importantly be able to generate EPOLLOUT for Edge Trigger epoll()
942 * users.
943 */
944static void tcp_remove_empty_skb(struct sock *sk, struct sk_buff *skb)
945{
946 if (skb && !skb->len) {
947 tcp_unlink_write_queue(skb, sk);
948 if (tcp_write_queue_empty(sk))
949 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
950 sk_wmem_free_skb(sk, skb);
951 }
952}
953
Dave Watsone3b56162017-06-14 11:37:26 -0700954ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
955 size_t size, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956{
957 struct tcp_sock *tp = tcp_sk(sk);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700958 int mss_now, size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 int err;
960 ssize_t copied;
961 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
962
Vasily Averina10674b2019-03-06 14:10:22 +0300963 if (IS_ENABLED(CONFIG_DEBUG_VM) &&
964 WARN_ONCE(PageSlab(page), "page must not be a Slab one"))
965 return -EINVAL;
966
Jerry Chu83368862012-08-31 12:29:12 +0000967 /* Wait for a connection to finish. One exception is TCP Fast Open
968 * (passive side) where data is allowed to be sent before a connection
969 * is fully established.
970 */
971 if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
972 !tcp_passive_fastopen(sk)) {
Yuvaraja Mariappan686a5622015-10-06 10:53:29 -0700973 err = sk_stream_wait_connect(sk, &timeo);
974 if (err != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 goto out_err;
Jerry Chu83368862012-08-31 12:29:12 +0000976 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977
Eric Dumazet9cd3e072015-11-29 20:03:10 -0800978 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000980 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 copied = 0;
982
983 err = -EPIPE;
984 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
Ilpo Järvinen0d6a7752009-02-28 04:44:41 +0000985 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986
Eric Dumazet64022d02012-12-01 13:07:02 +0000987 while (size > 0) {
David S. Millerfe067e82007-03-07 12:12:44 -0800988 struct sk_buff *skb = tcp_write_queue_tail(sk);
Eric Dumazet38ba0a62012-04-23 17:48:27 +0000989 int copy, i;
Eric Dumazet38ba0a62012-04-23 17:48:27 +0000990 bool can_coalesce;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991
Eric Dumazet75c119a2017-10-05 22:21:27 -0700992 if (!skb || (copy = size_goal - skb->len) <= 0 ||
Martin KaFai Lauc134ecb2016-04-25 14:44:48 -0700993 !tcp_skb_can_collapse_to(skb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994new_segment:
995 if (!sk_stream_memory_free(sk))
996 goto wait_for_sndbuf;
997
Eric Dumazeteb934472015-05-19 13:26:55 -0700998 skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation,
Eric Dumazet75c119a2017-10-05 22:21:27 -0700999 tcp_rtx_and_write_queues_empty(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 if (!skb)
1001 goto wait_for_memory;
1002
Jakub Kicinski41477662019-08-07 17:03:59 -07001003#ifdef CONFIG_TLS_DEVICE
1004 skb->decrypted = !!(flags & MSG_SENDPAGE_DECRYPTED);
1005#endif
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001006 skb_entail(sk, skb);
David S. Millerc1b4a7e2005-07-05 15:24:38 -07001007 copy = size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 }
1009
1010 if (copy > size)
1011 copy = size;
1012
1013 i = skb_shinfo(skb)->nr_frags;
1014 can_coalesce = skb_can_coalesce(skb, i, page, offset);
Hans Westgaard Ry5f74f82e2016-02-03 09:26:57 +01001015 if (!can_coalesce && i >= sysctl_max_skb_frags) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 tcp_mark_push(tp, skb);
1017 goto new_segment;
1018 }
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001019 if (!sk_wmem_schedule(sk, copy))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 goto wait_for_memory;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001021
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 if (can_coalesce) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00001023 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 } else {
1025 get_page(page);
1026 skb_fill_page_desc(skb, i, page, offset, copy);
1027 }
John Fastabend312fc2b2018-03-18 12:57:00 -07001028
1029 if (!(flags & MSG_NO_SHARED_FRAGS))
1030 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
Eric Dumazetcef401d2013-01-25 20:34:37 +00001031
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 skb->len += copy;
1033 skb->data_len += copy;
1034 skb->truesize += copy;
1035 sk->sk_wmem_queued += copy;
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001036 sk_mem_charge(sk, copy);
Patrick McHardy84fa7932006-08-29 16:44:56 -07001037 skb->ip_summed = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 tp->write_seq += copy;
1039 TCP_SKB_CB(skb)->end_seq += copy;
Eric Dumazetcd7d8492014-09-24 04:11:22 -07001040 tcp_skb_pcount_set(skb, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041
1042 if (!copied)
Eric Dumazet4de075e2011-09-27 13:25:05 -04001043 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
1045 copied += copy;
Eric Dumazet64022d02012-12-01 13:07:02 +00001046 offset += copy;
Yuvaraja Mariappan686a5622015-10-06 10:53:29 -07001047 size -= copy;
Soheil Hassas Yeganehad02c4f2017-01-04 11:19:34 -05001048 if (!size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 goto out;
1050
Herbert Xu69d15062008-03-22 15:47:05 -07001051 if (skb->len < size_goal || (flags & MSG_OOB))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 continue;
1053
1054 if (forced_push(tp)) {
1055 tcp_mark_push(tp, skb);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001056 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
David S. Millerfe067e82007-03-07 12:12:44 -08001057 } else if (skb == tcp_send_head(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 tcp_push_one(sk, mss_now);
1059 continue;
1060
1061wait_for_sndbuf:
1062 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1063wait_for_memory:
Eric Dumazetf54b3112013-12-05 22:36:05 -08001064 tcp_push(sk, flags & ~MSG_MORE, mss_now,
1065 TCP_NAGLE_PUSH, size_goal);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066
Yuvaraja Mariappan686a5622015-10-06 10:53:29 -07001067 err = sk_stream_wait_memory(sk, &timeo);
1068 if (err != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 goto do_error;
1070
Ilpo Järvinen0c54b852009-03-14 14:23:05 +00001071 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 }
1073
1074out:
Soheil Hassas Yeganehad02c4f2017-01-04 11:19:34 -05001075 if (copied) {
Eric Dumazet4e8cc222017-10-05 22:21:23 -07001076 tcp_tx_timestamp(sk, sk->sk_tsflags);
Soheil Hassas Yeganehad02c4f2017-01-04 11:19:34 -05001077 if (!(flags & MSG_SENDPAGE_NOTLAST))
1078 tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
1079 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 return copied;
1081
1082do_error:
Eric Dumazetfdfc5c82019-08-26 09:19:15 -07001083 tcp_remove_empty_skb(sk, tcp_write_queue_tail(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 if (copied)
1085 goto out;
1086out_err:
Jason Baronce5ec442015-05-20 15:52:53 +00001087 /* make sure we wake any epoll edge trigger waiter */
Francis Yanb0f71bd2016-11-27 23:07:16 -08001088 if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 &&
1089 err == -EAGAIN)) {
Jason Baronce5ec442015-05-20 15:52:53 +00001090 sk->sk_write_space(sk);
Francis Yanb0f71bd2016-11-27 23:07:16 -08001091 tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED);
1092 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 return sk_stream_error(sk, flags, err);
1094}
Dave Watsone3b56162017-06-14 11:37:26 -07001095EXPORT_SYMBOL_GPL(do_tcp_sendpages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096
Tom Herbert306b13e2017-07-28 16:22:41 -07001097int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
1098 size_t size, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099{
Eric Dumazetdead7cd2018-02-19 11:56:49 -08001100 if (!(sk->sk_route_caps & NETIF_F_SG))
Eric Dumazetbd9dfc52017-08-25 06:27:05 -07001101 return sock_no_sendpage_locked(sk, page, offset, size, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102
Soheil Hassas Yeganehd7722e82016-09-19 23:39:15 -04001103 tcp_rate_check_app_limited(sk); /* is sending application-limited? */
1104
Tom Herbert306b13e2017-07-28 16:22:41 -07001105 return do_tcp_sendpages(sk, page, offset, size, flags);
1106}
David S. Miller774c4672017-08-16 15:40:44 -07001107EXPORT_SYMBOL_GPL(tcp_sendpage_locked);
Tom Herbert306b13e2017-07-28 16:22:41 -07001108
1109int tcp_sendpage(struct sock *sk, struct page *page, int offset,
1110 size_t size, int flags)
1111{
1112 int ret;
1113
1114 lock_sock(sk);
1115 ret = tcp_sendpage_locked(sk, page, offset, size, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 release_sock(sk);
Tom Herbert306b13e2017-07-28 16:22:41 -07001117
1118 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001120EXPORT_SYMBOL(tcp_sendpage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121
Yuchung Chengcf60af02012-07-19 06:43:09 +00001122void tcp_free_fastopen_req(struct tcp_sock *tp)
1123{
Ian Morris00db4122015-04-03 09:17:27 +01001124 if (tp->fastopen_req) {
Yuchung Chengcf60af02012-07-19 06:43:09 +00001125 kfree(tp->fastopen_req);
1126 tp->fastopen_req = NULL;
1127 }
1128}
1129
Eric Dumazetf5ddcbb2014-02-20 10:09:18 -08001130static int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg,
Willem de Bruijnf859a442019-01-25 11:17:23 -05001131 int *copied, size_t size,
1132 struct ubuf_info *uarg)
Yuchung Chengcf60af02012-07-19 06:43:09 +00001133{
1134 struct tcp_sock *tp = tcp_sk(sk);
Wei Wang19f6d3f2017-01-23 10:59:22 -08001135 struct inet_sock *inet = inet_sk(sk);
Wei Wangba615f62017-05-24 09:59:31 -07001136 struct sockaddr *uaddr = msg->msg_name;
Yuchung Chengcf60af02012-07-19 06:43:09 +00001137 int err, flags;
1138
Haishuang Yane1cfcbe2017-09-27 11:35:40 +08001139 if (!(sock_net(sk)->ipv4.sysctl_tcp_fastopen & TFO_CLIENT_ENABLE) ||
Wei Wangba615f62017-05-24 09:59:31 -07001140 (uaddr && msg->msg_namelen >= sizeof(uaddr->sa_family) &&
1141 uaddr->sa_family == AF_UNSPEC))
Yuchung Chengcf60af02012-07-19 06:43:09 +00001142 return -EOPNOTSUPP;
Ian Morris00db4122015-04-03 09:17:27 +01001143 if (tp->fastopen_req)
Yuchung Chengcf60af02012-07-19 06:43:09 +00001144 return -EALREADY; /* Another Fast Open is in progress */
1145
1146 tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request),
1147 sk->sk_allocation);
Ian Morris51456b22015-04-03 09:17:26 +01001148 if (unlikely(!tp->fastopen_req))
Yuchung Chengcf60af02012-07-19 06:43:09 +00001149 return -ENOBUFS;
1150 tp->fastopen_req->data = msg;
Eric Dumazetf5ddcbb2014-02-20 10:09:18 -08001151 tp->fastopen_req->size = size;
Willem de Bruijnf859a442019-01-25 11:17:23 -05001152 tp->fastopen_req->uarg = uarg;
Yuchung Chengcf60af02012-07-19 06:43:09 +00001153
Wei Wang19f6d3f2017-01-23 10:59:22 -08001154 if (inet->defer_connect) {
1155 err = tcp_connect(sk);
1156 /* Same failure procedure as in tcp_v4/6_connect */
1157 if (err) {
1158 tcp_set_state(sk, TCP_CLOSE);
1159 inet->inet_dport = 0;
1160 sk->sk_route_caps = 0;
1161 }
1162 }
Yuchung Chengcf60af02012-07-19 06:43:09 +00001163 flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0;
Wei Wangba615f62017-05-24 09:59:31 -07001164 err = __inet_stream_connect(sk->sk_socket, uaddr,
Willy Tarreau3979ad72017-01-25 14:42:46 +01001165 msg->msg_namelen, flags, 1);
Wei Wang7db92362017-03-01 13:29:48 -08001166 /* fastopen_req could already be freed in __inet_stream_connect
1167 * if the connection times out or gets rst
1168 */
1169 if (tp->fastopen_req) {
1170 *copied = tp->fastopen_req->copied;
1171 tcp_free_fastopen_req(tp);
1172 inet->defer_connect = 0;
1173 }
Yuchung Chengcf60af02012-07-19 06:43:09 +00001174 return err;
1175}
1176
Tom Herbert306b13e2017-07-28 16:22:41 -07001177int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 struct tcp_sock *tp = tcp_sk(sk);
Willem de Bruijnf214f912017-08-03 16:29:44 -04001180 struct ubuf_info *uarg = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 struct sk_buff *skb;
Soheil Hassas Yeganehc14ac942016-04-02 23:08:12 -04001182 struct sockcm_cookie sockc;
Al Viro57be5bd2014-11-28 13:40:20 -05001183 int flags, err, copied = 0;
1184 int mss_now = 0, size_goal, copied_syn = 0;
Eric Dumazetd4011232016-05-02 21:49:25 -07001185 bool process_backlog = false;
Eric Dumazet74d4a8f2018-02-19 11:56:48 -08001186 bool zc = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 long timeo;
1188
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 flags = msg->msg_flags;
Willem de Bruijnf214f912017-08-03 16:29:44 -04001190
Vincent Whitchurch5cf4a852018-09-06 15:54:59 +02001191 if (flags & MSG_ZEROCOPY && size && sock_flag(sk, SOCK_ZEROCOPY)) {
Eric Dumazet75c119a2017-10-05 22:21:27 -07001192 skb = tcp_write_queue_tail(sk);
Willem de Bruijnf214f912017-08-03 16:29:44 -04001193 uarg = sock_zerocopy_realloc(sk, size, skb_zcopy(skb));
1194 if (!uarg) {
1195 err = -ENOBUFS;
1196 goto out_err;
1197 }
1198
Eric Dumazetdead7cd2018-02-19 11:56:49 -08001199 zc = sk->sk_route_caps & NETIF_F_SG;
Willem de Bruijn02583ad2017-12-22 19:00:19 -05001200 if (!zc)
Willem de Bruijnf214f912017-08-03 16:29:44 -04001201 uarg->zerocopy = 0;
1202 }
1203
Yuchung Cheng16ae6aa2018-04-25 11:33:08 -07001204 if (unlikely(flags & MSG_FASTOPEN || inet_sk(sk)->defer_connect) &&
1205 !tp->repair) {
Willem de Bruijnf859a442019-01-25 11:17:23 -05001206 err = tcp_sendmsg_fastopen(sk, msg, &copied_syn, size, uarg);
Yuchung Chengcf60af02012-07-19 06:43:09 +00001207 if (err == -EINPROGRESS && copied_syn > 0)
1208 goto out;
1209 else if (err)
1210 goto out_err;
Yuchung Chengcf60af02012-07-19 06:43:09 +00001211 }
1212
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1214
Soheil Hassas Yeganehd7722e82016-09-19 23:39:15 -04001215 tcp_rate_check_app_limited(sk); /* is sending application-limited? */
1216
Jerry Chu83368862012-08-31 12:29:12 +00001217 /* Wait for a connection to finish. One exception is TCP Fast Open
1218 * (passive side) where data is allowed to be sent before a connection
1219 * is fully established.
1220 */
1221 if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
1222 !tcp_passive_fastopen(sk)) {
Yuvaraja Mariappan686a5622015-10-06 10:53:29 -07001223 err = sk_stream_wait_connect(sk, &timeo);
1224 if (err != 0)
Yuchung Chengcf60af02012-07-19 06:43:09 +00001225 goto do_error;
Jerry Chu83368862012-08-31 12:29:12 +00001226 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00001228 if (unlikely(tp->repair)) {
1229 if (tp->repair_queue == TCP_RECV_QUEUE) {
1230 copied = tcp_send_rcvq(sk, msg, size);
Christoph Paasch5924f172014-06-28 18:26:37 +02001231 goto out_nopush;
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00001232 }
1233
1234 err = -EINVAL;
1235 if (tp->repair_queue == TCP_NO_QUEUE)
1236 goto out_err;
1237
1238 /* 'common' sending to sendq */
1239 }
1240
Willem de Bruijn657a0662018-07-06 10:12:56 -04001241 sockcm_init(&sockc, sk);
Soheil Hassas Yeganehc14ac942016-04-02 23:08:12 -04001242 if (msg->msg_controllen) {
1243 err = sock_cmsg_send(sk, msg, &sockc);
1244 if (unlikely(err)) {
1245 err = -EINVAL;
1246 goto out_err;
1247 }
1248 }
1249
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 /* This should be in poll */
Eric Dumazet9cd3e072015-11-29 20:03:10 -08001251 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 /* Ok commence sending. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 copied = 0;
1255
Eric Dumazetd41a69f2016-04-29 14:16:53 -07001256restart:
1257 mss_now = tcp_send_mss(sk, &size_goal, flags);
1258
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 err = -EPIPE;
1260 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
Eric Dumazet79d86652016-11-02 14:41:50 -07001261 goto do_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262
Al Viro01e97e62014-12-15 21:39:31 -05001263 while (msg_data_left(msg)) {
Al Viro57be5bd2014-11-28 13:40:20 -05001264 int copy = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265
Al Viro57be5bd2014-11-28 13:40:20 -05001266 skb = tcp_write_queue_tail(sk);
Eric Dumazet65ec6092018-02-19 11:56:50 -08001267 if (skb)
1268 copy = size_goal - skb->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269
Martin KaFai Lauc134ecb2016-04-25 14:44:48 -07001270 if (copy <= 0 || !tcp_skb_can_collapse_to(skb)) {
Eric Dumazet3613b3d2016-09-15 09:33:02 -07001271 bool first_skb;
1272
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273new_segment:
Al Viro57be5bd2014-11-28 13:40:20 -05001274 if (!sk_stream_memory_free(sk))
1275 goto wait_for_sndbuf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276
Eric Dumazetd4011232016-05-02 21:49:25 -07001277 if (process_backlog && sk_flush_backlog(sk)) {
1278 process_backlog = false;
Eric Dumazetd41a69f2016-04-29 14:16:53 -07001279 goto restart;
Eric Dumazetd4011232016-05-02 21:49:25 -07001280 }
Eric Dumazet75c119a2017-10-05 22:21:27 -07001281 first_skb = tcp_rtx_and_write_queues_empty(sk);
Eric Dumazet472c2e02019-03-22 08:56:39 -07001282 skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation,
Eric Dumazet3613b3d2016-09-15 09:33:02 -07001283 first_skb);
Al Viro57be5bd2014-11-28 13:40:20 -05001284 if (!skb)
1285 goto wait_for_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286
Eric Dumazetd4011232016-05-02 21:49:25 -07001287 process_backlog = true;
Eric Dumazetdead7cd2018-02-19 11:56:49 -08001288 skb->ip_summed = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289
Al Viro57be5bd2014-11-28 13:40:20 -05001290 skb_entail(sk, skb);
1291 copy = size_goal;
Andrey Vagin9d186ca2014-08-13 16:03:10 +04001292
Al Viro57be5bd2014-11-28 13:40:20 -05001293 /* All packets are restored as if they have
Eric Dumazetd3edd062018-09-21 08:51:50 -07001294 * already been sent. skb_mstamp_ns isn't set to
Al Viro57be5bd2014-11-28 13:40:20 -05001295 * avoid wrong rtt estimation.
1296 */
1297 if (tp->repair)
1298 TCP_SKB_CB(skb)->sacked |= TCPCB_REPAIRED;
1299 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300
Al Viro57be5bd2014-11-28 13:40:20 -05001301 /* Try to append data to the end of skb. */
Al Viro01e97e62014-12-15 21:39:31 -05001302 if (copy > msg_data_left(msg))
1303 copy = msg_data_left(msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
Al Viro57be5bd2014-11-28 13:40:20 -05001305 /* Where to copy to? */
Willem de Bruijn02583ad2017-12-22 19:00:19 -05001306 if (skb_availroom(skb) > 0 && !zc) {
Al Viro57be5bd2014-11-28 13:40:20 -05001307 /* We have some space in skb head. Superb! */
1308 copy = min_t(int, copy, skb_availroom(skb));
1309 err = skb_add_data_nocache(sk, skb, &msg->msg_iter, copy);
1310 if (err)
1311 goto do_fault;
Willem de Bruijn02583ad2017-12-22 19:00:19 -05001312 } else if (!zc) {
Al Viro57be5bd2014-11-28 13:40:20 -05001313 bool merge = true;
1314 int i = skb_shinfo(skb)->nr_frags;
1315 struct page_frag *pfrag = sk_page_frag(sk);
Eric Dumazet761965e2011-12-04 07:05:17 +00001316
Al Viro57be5bd2014-11-28 13:40:20 -05001317 if (!sk_page_frag_refill(sk, pfrag))
1318 goto wait_for_memory;
Eric Dumazet761965e2011-12-04 07:05:17 +00001319
Al Viro57be5bd2014-11-28 13:40:20 -05001320 if (!skb_can_coalesce(skb, i, pfrag->page,
1321 pfrag->offset)) {
Eric Dumazet74d4a8f2018-02-19 11:56:48 -08001322 if (i >= sysctl_max_skb_frags) {
Al Viro57be5bd2014-11-28 13:40:20 -05001323 tcp_mark_push(tp, skb);
1324 goto new_segment;
Eric Dumazet5640f762012-09-23 23:04:42 +00001325 }
Al Viro57be5bd2014-11-28 13:40:20 -05001326 merge = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 }
1328
Al Viro57be5bd2014-11-28 13:40:20 -05001329 copy = min_t(int, copy, pfrag->size - pfrag->offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330
Al Viro57be5bd2014-11-28 13:40:20 -05001331 if (!sk_wmem_schedule(sk, copy))
1332 goto wait_for_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
Al Viro57be5bd2014-11-28 13:40:20 -05001334 err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
1335 pfrag->page,
1336 pfrag->offset,
1337 copy);
1338 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 goto do_error;
1340
Al Viro57be5bd2014-11-28 13:40:20 -05001341 /* Update the skb. */
1342 if (merge) {
1343 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1344 } else {
1345 skb_fill_page_desc(skb, i, pfrag->page,
1346 pfrag->offset, copy);
Eric Dumazet4e33e342017-02-17 09:11:42 -08001347 page_ref_inc(pfrag->page);
Al Viro57be5bd2014-11-28 13:40:20 -05001348 }
1349 pfrag->offset += copy;
Willem de Bruijnf214f912017-08-03 16:29:44 -04001350 } else {
1351 err = skb_zerocopy_iter_stream(sk, skb, msg, copy, uarg);
Willem de Bruijn111856c2017-12-22 19:00:18 -05001352 if (err == -EMSGSIZE || err == -EEXIST) {
1353 tcp_mark_push(tp, skb);
Willem de Bruijnf214f912017-08-03 16:29:44 -04001354 goto new_segment;
Willem de Bruijn111856c2017-12-22 19:00:18 -05001355 }
Willem de Bruijnf214f912017-08-03 16:29:44 -04001356 if (err < 0)
1357 goto do_error;
1358 copy = err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 }
Al Viro57be5bd2014-11-28 13:40:20 -05001360
1361 if (!copied)
1362 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1363
1364 tp->write_seq += copy;
1365 TCP_SKB_CB(skb)->end_seq += copy;
1366 tcp_skb_pcount_set(skb, 0);
1367
1368 copied += copy;
Al Viro01e97e62014-12-15 21:39:31 -05001369 if (!msg_data_left(msg)) {
Martin KaFai Lauc134ecb2016-04-25 14:44:48 -07001370 if (unlikely(flags & MSG_EOR))
1371 TCP_SKB_CB(skb)->eor = 1;
Al Viro57be5bd2014-11-28 13:40:20 -05001372 goto out;
1373 }
1374
Eric Dumazet65ec6092018-02-19 11:56:50 -08001375 if (skb->len < size_goal || (flags & MSG_OOB) || unlikely(tp->repair))
Al Viro57be5bd2014-11-28 13:40:20 -05001376 continue;
1377
1378 if (forced_push(tp)) {
1379 tcp_mark_push(tp, skb);
1380 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
1381 } else if (skb == tcp_send_head(sk))
1382 tcp_push_one(sk, mss_now);
1383 continue;
1384
1385wait_for_sndbuf:
1386 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1387wait_for_memory:
1388 if (copied)
1389 tcp_push(sk, flags & ~MSG_MORE, mss_now,
1390 TCP_NAGLE_PUSH, size_goal);
1391
Yuvaraja Mariappan686a5622015-10-06 10:53:29 -07001392 err = sk_stream_wait_memory(sk, &timeo);
1393 if (err != 0)
Al Viro57be5bd2014-11-28 13:40:20 -05001394 goto do_error;
1395
1396 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 }
1398
1399out:
Soheil Hassas Yeganehad02c4f2017-01-04 11:19:34 -05001400 if (copied) {
Eric Dumazet4e8cc222017-10-05 22:21:23 -07001401 tcp_tx_timestamp(sk, sockc.tsflags);
Eric Dumazetf54b3112013-12-05 22:36:05 -08001402 tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
Soheil Hassas Yeganehad02c4f2017-01-04 11:19:34 -05001403 }
Christoph Paasch5924f172014-06-28 18:26:37 +02001404out_nopush:
Willem de Bruijnf214f912017-08-03 16:29:44 -04001405 sock_zerocopy_put(uarg);
Yuchung Chengcf60af02012-07-19 06:43:09 +00001406 return copied + copied_syn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408do_error:
Eric Dumazetfdfc5c82019-08-26 09:19:15 -07001409 skb = tcp_write_queue_tail(sk);
1410do_fault:
1411 tcp_remove_empty_skb(sk, skb);
1412
Yuchung Chengcf60af02012-07-19 06:43:09 +00001413 if (copied + copied_syn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 goto out;
1415out_err:
Willem de Bruijn52900d22018-11-30 15:32:40 -05001416 sock_zerocopy_put_abort(uarg, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 err = sk_stream_error(sk, flags, err);
Jason Baronce5ec442015-05-20 15:52:53 +00001418 /* make sure we wake any epoll edge trigger waiter */
Francis Yanb0f71bd2016-11-27 23:07:16 -08001419 if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 &&
1420 err == -EAGAIN)) {
Jason Baronce5ec442015-05-20 15:52:53 +00001421 sk->sk_write_space(sk);
Francis Yanb0f71bd2016-11-27 23:07:16 -08001422 tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED);
1423 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 return err;
1425}
David S. Miller774c4672017-08-16 15:40:44 -07001426EXPORT_SYMBOL_GPL(tcp_sendmsg_locked);
Tom Herbert306b13e2017-07-28 16:22:41 -07001427
1428int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
1429{
1430 int ret;
1431
1432 lock_sock(sk);
1433 ret = tcp_sendmsg_locked(sk, msg, size);
1434 release_sock(sk);
1435
1436 return ret;
1437}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001438EXPORT_SYMBOL(tcp_sendmsg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439
1440/*
1441 * Handle reading urgent data. BSD has very simple semantics for
1442 * this, no blocking and very strange errors 8)
1443 */
1444
Rami Rosen377f0a02009-03-31 14:43:17 -07001445static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446{
1447 struct tcp_sock *tp = tcp_sk(sk);
1448
1449 /* No URG data to read. */
1450 if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data ||
1451 tp->urg_data == TCP_URG_READ)
1452 return -EINVAL; /* Yes this is right ! */
1453
1454 if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE))
1455 return -ENOTCONN;
1456
1457 if (tp->urg_data & TCP_URG_VALID) {
1458 int err = 0;
1459 char c = tp->urg_data;
1460
1461 if (!(flags & MSG_PEEK))
1462 tp->urg_data = TCP_URG_READ;
1463
1464 /* Read urgent data. */
1465 msg->msg_flags |= MSG_OOB;
1466
1467 if (len > 0) {
1468 if (!(flags & MSG_TRUNC))
Al Viro7eab8d92014-04-06 21:51:23 -04001469 err = memcpy_to_msg(msg, &c, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 len = 1;
1471 } else
1472 msg->msg_flags |= MSG_TRUNC;
1473
1474 return err ? -EFAULT : len;
1475 }
1476
1477 if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
1478 return 0;
1479
1480 /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and
1481 * the available implementations agree in this case:
1482 * this call should never block, independent of the
1483 * blocking state of the socket.
1484 * Mike <pall@rz.uni-karlsruhe.de>
1485 */
1486 return -EAGAIN;
1487}
1488
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00001489static int tcp_peek_sndq(struct sock *sk, struct msghdr *msg, int len)
1490{
1491 struct sk_buff *skb;
1492 int copied = 0, err = 0;
1493
1494 /* XXX -- need to support SO_PEEK_OFF */
1495
Eric Dumazet75c119a2017-10-05 22:21:27 -07001496 skb_rbtree_walk(skb, &sk->tcp_rtx_queue) {
1497 err = skb_copy_datagram_msg(skb, 0, msg, skb->len);
1498 if (err)
1499 return err;
1500 copied += skb->len;
1501 }
1502
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00001503 skb_queue_walk(&sk->sk_write_queue, skb) {
David S. Miller51f3d022014-11-05 16:46:40 -05001504 err = skb_copy_datagram_msg(skb, 0, msg, skb->len);
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00001505 if (err)
1506 break;
1507
1508 copied += skb->len;
1509 }
1510
1511 return err ?: copied;
1512}
1513
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514/* Clean up the receive buffer for full frames taken by the user,
1515 * then send an ACK if necessary. COPIED is the number of bytes
1516 * tcp_recvmsg has given to the user so far, it speeds up the
1517 * calculation of whether or not we must ACK for the sake of
1518 * a window update.
1519 */
Dan Williams3f334072013-12-30 17:12:07 -08001520static void tcp_cleanup_rbuf(struct sock *sk, int copied)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521{
1522 struct tcp_sock *tp = tcp_sk(sk);
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001523 bool time_to_ack = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
1526
Ilpo Järvinend792c102009-11-13 13:56:33 -08001527 WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq),
Joe Perches2af6fd82010-10-30 11:08:53 +00001528 "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n",
Ilpo Järvinend792c102009-11-13 13:56:33 -08001529 tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001531 if (inet_csk_ack_scheduled(sk)) {
1532 const struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 /* Delayed ACKs frequently hit locked sockets during bulk
1534 * receive. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001535 if (icsk->icsk_ack.blocked ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 /* Once-per-two-segments ACK was not sent by tcp_input.c */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001537 tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 /*
1539 * If this read emptied read buffer, we send ACK, if
1540 * connection is not bidirectional, user drained
1541 * receive buffer and there was a small segment
1542 * in queue.
1543 */
Alexey Kuznetsov1ef96962006-09-19 12:52:50 -07001544 (copied > 0 &&
1545 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) ||
1546 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) &&
Wei Wang31954cd2019-01-25 10:53:19 -08001547 !inet_csk_in_pingpong_mode(sk))) &&
Alexey Kuznetsov1ef96962006-09-19 12:52:50 -07001548 !atomic_read(&sk->sk_rmem_alloc)))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001549 time_to_ack = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 }
1551
1552 /* We send an ACK if we can now advertise a non-zero window
1553 * which has been raised "significantly".
1554 *
1555 * Even if window raised up to infinity, do not send window open ACK
1556 * in states, where we will not receive more. It is useless.
1557 */
1558 if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1559 __u32 rcv_window_now = tcp_receive_window(tp);
1560
1561 /* Optimize, __tcp_select_window() is not cheap. */
1562 if (2*rcv_window_now <= tp->window_clamp) {
1563 __u32 new_window = __tcp_select_window(sk);
1564
1565 /* Send ACK now, if this read freed lots of space
1566 * in our buffer. Certainly, new_window is new window.
1567 * We can advertise it now, if it is not less than current one.
1568 * "Lots" means "at least twice" here.
1569 */
1570 if (new_window && new_window >= 2 * rcv_window_now)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001571 time_to_ack = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 }
1573 }
1574 if (time_to_ack)
1575 tcp_send_ack(sk);
1576}
1577
Eric Dumazetf26845b2013-01-09 20:59:09 +00001578static struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579{
1580 struct sk_buff *skb;
1581 u32 offset;
1582
Eric Dumazetf26845b2013-01-09 20:59:09 +00001583 while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 offset = seq - TCP_SKB_CB(skb)->seq;
Eric Dumazet9d691532016-02-01 21:03:08 -08001585 if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
1586 pr_err_once("%s: found a SYN, please report !\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 offset--;
Eric Dumazet9d691532016-02-01 21:03:08 -08001588 }
Eric Dumazete11ecdd2014-09-15 04:19:51 -07001589 if (offset < skb->len || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 *off = offset;
1591 return skb;
1592 }
Eric Dumazetf26845b2013-01-09 20:59:09 +00001593 /* This looks weird, but this can happen if TCP collapsing
1594 * splitted a fat GRO packet, while we released socket lock
1595 * in skb_splice_bits()
1596 */
Dan Williams7bced392013-12-30 12:37:29 -08001597 sk_eat_skb(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 }
1599 return NULL;
1600}
1601
1602/*
1603 * This routine provides an alternative to tcp_recvmsg() for routines
1604 * that would like to handle copying from skbuffs directly in 'sendfile'
1605 * fashion.
1606 * Note:
1607 * - It is assumed that the socket was locked by the caller.
1608 * - The routine does not block.
1609 * - At present, there is no support for reading OOB data
1610 * or for 'peeking' the socket using this routine
1611 * (although both would be easy to implement).
1612 */
1613int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
1614 sk_read_actor_t recv_actor)
1615{
1616 struct sk_buff *skb;
1617 struct tcp_sock *tp = tcp_sk(sk);
1618 u32 seq = tp->copied_seq;
1619 u32 offset;
1620 int copied = 0;
1621
1622 if (sk->sk_state == TCP_LISTEN)
1623 return -ENOTCONN;
1624 while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) {
1625 if (offset < skb->len) {
Octavian Purdila374e7b52008-07-03 03:31:21 -07001626 int used;
1627 size_t len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628
1629 len = skb->len - offset;
1630 /* Stop reading if we hit a patch of urgent data */
1631 if (tp->urg_data) {
1632 u32 urg_offset = tp->urg_seq - seq;
1633 if (urg_offset < len)
1634 len = urg_offset;
1635 if (!len)
1636 break;
1637 }
1638 used = recv_actor(desc, skb, offset, len);
Eric Dumazetff905b12013-01-10 07:06:10 +00001639 if (used <= 0) {
Jens Axboeddb61a52007-06-23 23:07:50 -07001640 if (!copied)
1641 copied = used;
1642 break;
1643 } else if (used <= len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 seq += used;
1645 copied += used;
1646 offset += used;
1647 }
Willy Tarreau02275a22012-12-02 11:49:27 +00001648 /* If recv_actor drops the lock (e.g. TCP splice
Octavian Purdila293ad602008-06-04 15:45:58 -07001649 * receive) the skb pointer might be invalid when
1650 * getting here: tcp_collapse might have deleted it
1651 * while aggregating skbs from the socket queue.
1652 */
Willy Tarreau02275a22012-12-02 11:49:27 +00001653 skb = tcp_recv_skb(sk, seq - 1, &offset);
1654 if (!skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 break;
Willy Tarreau02275a22012-12-02 11:49:27 +00001656 /* TCP coalescing might have appended data to the skb.
1657 * Try to splice more frags
1658 */
1659 if (offset + 1 != skb->len)
1660 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 }
Eric Dumazete11ecdd2014-09-15 04:19:51 -07001662 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) {
Dan Williams7bced392013-12-30 12:37:29 -08001663 sk_eat_skb(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 ++seq;
1665 break;
1666 }
Dan Williams7bced392013-12-30 12:37:29 -08001667 sk_eat_skb(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 if (!desc->count)
1669 break;
Steven J. Magnanibaff42a2010-03-30 13:56:01 -07001670 tp->copied_seq = seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 }
1672 tp->copied_seq = seq;
1673
1674 tcp_rcv_space_adjust(sk);
1675
1676 /* Clean up data we have read: This will do ACK frames. */
Eric Dumazetf26845b2013-01-09 20:59:09 +00001677 if (copied > 0) {
1678 tcp_recv_skb(sk, seq, &offset);
Chris Leech0e4b4992006-05-23 18:00:16 -07001679 tcp_cleanup_rbuf(sk, copied);
Eric Dumazetf26845b2013-01-09 20:59:09 +00001680 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 return copied;
1682}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001683EXPORT_SYMBOL(tcp_read_sock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684
Tom Herbert32035582016-08-28 14:43:18 -07001685int tcp_peek_len(struct socket *sock)
1686{
1687 return tcp_inq(sock->sk);
1688}
1689EXPORT_SYMBOL(tcp_peek_len);
1690
Eric Dumazetd1361842018-04-16 10:33:35 -07001691/* Make sure sk_rcvbuf is big enough to satisfy SO_RCVLOWAT hint */
1692int tcp_set_rcvlowat(struct sock *sk, int val)
1693{
Soheil Hassas Yeganeh867f8162018-06-08 22:47:10 -04001694 int cap;
1695
1696 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK)
1697 cap = sk->sk_rcvbuf >> 1;
1698 else
1699 cap = sock_net(sk)->ipv4.sysctl_tcp_rmem[2] >> 1;
1700 val = min(val, cap);
Eric Dumazetd1361842018-04-16 10:33:35 -07001701 sk->sk_rcvlowat = val ? : 1;
Eric Dumazet03f45c82018-04-16 10:33:37 -07001702
1703 /* Check if we need to signal EPOLLIN right now */
1704 tcp_data_ready(sk);
1705
Eric Dumazetd1361842018-04-16 10:33:35 -07001706 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK)
1707 return 0;
1708
Eric Dumazetd1361842018-04-16 10:33:35 -07001709 val <<= 1;
Eric Dumazetd1361842018-04-16 10:33:35 -07001710 if (val > sk->sk_rcvbuf) {
1711 sk->sk_rcvbuf = val;
1712 tcp_sk(sk)->window_clamp = tcp_win_from_space(sk, val);
1713 }
1714 return 0;
1715}
1716EXPORT_SYMBOL(tcp_set_rcvlowat);
1717
Eric Dumazet05255b82018-04-27 08:58:08 -07001718#ifdef CONFIG_MMU
1719static const struct vm_operations_struct tcp_vm_ops = {
1720};
1721
Eric Dumazet93ab6cc2018-04-16 10:33:38 -07001722int tcp_mmap(struct file *file, struct socket *sock,
1723 struct vm_area_struct *vma)
1724{
Eric Dumazet05255b82018-04-27 08:58:08 -07001725 if (vma->vm_flags & (VM_WRITE | VM_EXEC))
1726 return -EPERM;
1727 vma->vm_flags &= ~(VM_MAYWRITE | VM_MAYEXEC);
1728
1729 /* Instruct vm_insert_page() to not down_read(mmap_sem) */
1730 vma->vm_flags |= VM_MIXEDMAP;
1731
1732 vma->vm_ops = &tcp_vm_ops;
1733 return 0;
1734}
1735EXPORT_SYMBOL(tcp_mmap);
1736
1737static int tcp_zerocopy_receive(struct sock *sk,
1738 struct tcp_zerocopy_receive *zc)
1739{
1740 unsigned long address = (unsigned long)zc->address;
1741 const skb_frag_t *frags = NULL;
1742 u32 length = 0, seq, offset;
1743 struct vm_area_struct *vma;
1744 struct sk_buff *skb = NULL;
Eric Dumazet93ab6cc2018-04-16 10:33:38 -07001745 struct tcp_sock *tp;
Soheil Hassas Yeganeh8f2b0292018-09-26 16:57:03 -04001746 int inq;
Eric Dumazet93ab6cc2018-04-16 10:33:38 -07001747 int ret;
1748
Eric Dumazet05255b82018-04-27 08:58:08 -07001749 if (address & (PAGE_SIZE - 1) || address != zc->address)
Eric Dumazet93ab6cc2018-04-16 10:33:38 -07001750 return -EINVAL;
1751
Eric Dumazet93ab6cc2018-04-16 10:33:38 -07001752 if (sk->sk_state == TCP_LISTEN)
Eric Dumazet05255b82018-04-27 08:58:08 -07001753 return -ENOTCONN;
Eric Dumazet93ab6cc2018-04-16 10:33:38 -07001754
1755 sock_rps_record_flow(sk);
1756
Eric Dumazet05255b82018-04-27 08:58:08 -07001757 down_read(&current->mm->mmap_sem);
1758
1759 ret = -EINVAL;
1760 vma = find_vma(current->mm, address);
1761 if (!vma || vma->vm_start > address || vma->vm_ops != &tcp_vm_ops)
Eric Dumazet93ab6cc2018-04-16 10:33:38 -07001762 goto out;
Eric Dumazet05255b82018-04-27 08:58:08 -07001763 zc->length = min_t(unsigned long, zc->length, vma->vm_end - address);
1764
Eric Dumazet93ab6cc2018-04-16 10:33:38 -07001765 tp = tcp_sk(sk);
1766 seq = tp->copied_seq;
Soheil Hassas Yeganeh8f2b0292018-09-26 16:57:03 -04001767 inq = tcp_inq(sk);
1768 zc->length = min_t(u32, zc->length, inq);
Eric Dumazet05255b82018-04-27 08:58:08 -07001769 zc->length &= ~(PAGE_SIZE - 1);
Soheil Hassas Yeganeh8f2b0292018-09-26 16:57:03 -04001770 if (zc->length) {
1771 zap_page_range(vma, address, zc->length);
1772 zc->recv_skip_hint = 0;
1773 } else {
1774 zc->recv_skip_hint = inq;
1775 }
Eric Dumazet05255b82018-04-27 08:58:08 -07001776 ret = 0;
1777 while (length + PAGE_SIZE <= zc->length) {
1778 if (zc->recv_skip_hint < PAGE_SIZE) {
1779 if (skb) {
1780 skb = skb->next;
1781 offset = seq - TCP_SKB_CB(skb)->seq;
1782 } else {
1783 skb = tcp_recv_skb(sk, seq, &offset);
1784 }
1785
1786 zc->recv_skip_hint = skb->len - offset;
1787 offset -= skb_headlen(skb);
1788 if ((int)offset < 0 || skb_has_frag_list(skb))
1789 break;
1790 frags = skb_shinfo(skb)->frags;
1791 while (offset) {
1792 if (frags->size > offset)
1793 goto out;
1794 offset -= frags->size;
1795 frags++;
1796 }
1797 }
Soheil Hassas Yeganeh789762c2018-09-26 16:57:04 -04001798 if (frags->size != PAGE_SIZE || frags->page_offset) {
1799 int remaining = zc->recv_skip_hint;
1800
1801 while (remaining && (frags->size != PAGE_SIZE ||
1802 frags->page_offset)) {
1803 remaining -= frags->size;
1804 frags++;
1805 }
1806 zc->recv_skip_hint -= remaining;
Eric Dumazet05255b82018-04-27 08:58:08 -07001807 break;
Soheil Hassas Yeganeh789762c2018-09-26 16:57:04 -04001808 }
Eric Dumazet05255b82018-04-27 08:58:08 -07001809 ret = vm_insert_page(vma, address + length,
1810 skb_frag_page(frags));
1811 if (ret)
1812 break;
1813 length += PAGE_SIZE;
1814 seq += PAGE_SIZE;
1815 zc->recv_skip_hint -= PAGE_SIZE;
Eric Dumazet93ab6cc2018-04-16 10:33:38 -07001816 frags++;
1817 }
Eric Dumazet93ab6cc2018-04-16 10:33:38 -07001818out:
Eric Dumazet05255b82018-04-27 08:58:08 -07001819 up_read(&current->mm->mmap_sem);
1820 if (length) {
1821 tp->copied_seq = seq;
1822 tcp_rcv_space_adjust(sk);
1823
1824 /* Clean up data we have read: This will do ACK frames. */
1825 tcp_recv_skb(sk, seq, &offset);
1826 tcp_cleanup_rbuf(sk, length);
1827 ret = 0;
1828 if (length == zc->length)
1829 zc->recv_skip_hint = 0;
1830 } else {
1831 if (!zc->recv_skip_hint && sock_flag(sk, SOCK_DONE))
1832 ret = -EIO;
1833 }
1834 zc->length = length;
Eric Dumazet93ab6cc2018-04-16 10:33:38 -07001835 return ret;
1836}
Eric Dumazet05255b82018-04-27 08:58:08 -07001837#endif
Eric Dumazet93ab6cc2018-04-16 10:33:38 -07001838
Mike Maloney98aaa912017-08-22 17:08:48 -04001839static void tcp_update_recv_tstamps(struct sk_buff *skb,
Deepa Dinamani97184752019-02-02 07:34:51 -08001840 struct scm_timestamping_internal *tss)
Mike Maloney98aaa912017-08-22 17:08:48 -04001841{
1842 if (skb->tstamp)
Deepa Dinamani97184752019-02-02 07:34:51 -08001843 tss->ts[0] = ktime_to_timespec64(skb->tstamp);
Mike Maloney98aaa912017-08-22 17:08:48 -04001844 else
Deepa Dinamani97184752019-02-02 07:34:51 -08001845 tss->ts[0] = (struct timespec64) {0};
Mike Maloney98aaa912017-08-22 17:08:48 -04001846
1847 if (skb_hwtstamps(skb)->hwtstamp)
Deepa Dinamani97184752019-02-02 07:34:51 -08001848 tss->ts[2] = ktime_to_timespec64(skb_hwtstamps(skb)->hwtstamp);
Mike Maloney98aaa912017-08-22 17:08:48 -04001849 else
Deepa Dinamani97184752019-02-02 07:34:51 -08001850 tss->ts[2] = (struct timespec64) {0};
Mike Maloney98aaa912017-08-22 17:08:48 -04001851}
1852
1853/* Similar to __sock_recv_timestamp, but does not require an skb */
Wei Yongjun809a79e2018-01-10 07:43:15 +00001854static void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk,
Deepa Dinamani97184752019-02-02 07:34:51 -08001855 struct scm_timestamping_internal *tss)
Mike Maloney98aaa912017-08-22 17:08:48 -04001856{
Deepa Dinamani887feae2019-02-02 07:34:50 -08001857 int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW);
Mike Maloney98aaa912017-08-22 17:08:48 -04001858 bool has_timestamping = false;
1859
1860 if (tss->ts[0].tv_sec || tss->ts[0].tv_nsec) {
1861 if (sock_flag(sk, SOCK_RCVTSTAMP)) {
1862 if (sock_flag(sk, SOCK_RCVTSTAMPNS)) {
Deepa Dinamani887feae2019-02-02 07:34:50 -08001863 if (new_tstamp) {
1864 struct __kernel_timespec kts = {tss->ts[0].tv_sec, tss->ts[0].tv_nsec};
Mike Maloney98aaa912017-08-22 17:08:48 -04001865
Deepa Dinamani887feae2019-02-02 07:34:50 -08001866 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW,
1867 sizeof(kts), &kts);
1868 } else {
Deepa Dinamani97184752019-02-02 07:34:51 -08001869 struct timespec ts_old = timespec64_to_timespec(tss->ts[0]);
1870
Deepa Dinamani887feae2019-02-02 07:34:50 -08001871 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD,
Deepa Dinamani97184752019-02-02 07:34:51 -08001872 sizeof(ts_old), &ts_old);
Deepa Dinamani887feae2019-02-02 07:34:50 -08001873 }
1874 } else {
1875 if (new_tstamp) {
1876 struct __kernel_sock_timeval stv;
1877
1878 stv.tv_sec = tss->ts[0].tv_sec;
1879 stv.tv_usec = tss->ts[0].tv_nsec / 1000;
1880 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW,
1881 sizeof(stv), &stv);
1882 } else {
1883 struct __kernel_old_timeval tv;
1884
1885 tv.tv_sec = tss->ts[0].tv_sec;
1886 tv.tv_usec = tss->ts[0].tv_nsec / 1000;
1887 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD,
1888 sizeof(tv), &tv);
1889 }
Mike Maloney98aaa912017-08-22 17:08:48 -04001890 }
1891 }
1892
1893 if (sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE)
1894 has_timestamping = true;
1895 else
Deepa Dinamani97184752019-02-02 07:34:51 -08001896 tss->ts[0] = (struct timespec64) {0};
Mike Maloney98aaa912017-08-22 17:08:48 -04001897 }
1898
1899 if (tss->ts[2].tv_sec || tss->ts[2].tv_nsec) {
1900 if (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)
1901 has_timestamping = true;
1902 else
Deepa Dinamani97184752019-02-02 07:34:51 -08001903 tss->ts[2] = (struct timespec64) {0};
Mike Maloney98aaa912017-08-22 17:08:48 -04001904 }
1905
1906 if (has_timestamping) {
Deepa Dinamani97184752019-02-02 07:34:51 -08001907 tss->ts[1] = (struct timespec64) {0};
1908 if (sock_flag(sk, SOCK_TSTAMP_NEW))
1909 put_cmsg_scm_timestamping64(msg, tss);
1910 else
1911 put_cmsg_scm_timestamping(msg, tss);
Mike Maloney98aaa912017-08-22 17:08:48 -04001912 }
1913}
1914
Soheil Hassas Yeganehb75eba72018-05-01 15:39:15 -04001915static int tcp_inq_hint(struct sock *sk)
1916{
1917 const struct tcp_sock *tp = tcp_sk(sk);
1918 u32 copied_seq = READ_ONCE(tp->copied_seq);
1919 u32 rcv_nxt = READ_ONCE(tp->rcv_nxt);
1920 int inq;
1921
1922 inq = rcv_nxt - copied_seq;
1923 if (unlikely(inq < 0 || copied_seq != READ_ONCE(tp->copied_seq))) {
1924 lock_sock(sk);
1925 inq = tp->rcv_nxt - tp->copied_seq;
1926 release_sock(sk);
1927 }
Soheil Hassas Yeganeh6466e712019-03-06 13:01:36 -05001928 /* After receiving a FIN, tell the user-space to continue reading
1929 * by returning a non-zero inq.
1930 */
1931 if (inq == 0 && sock_flag(sk, SOCK_DONE))
1932 inq = 1;
Soheil Hassas Yeganehb75eba72018-05-01 15:39:15 -04001933 return inq;
1934}
1935
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936/*
1937 * This routine copies from a sock struct into the user buffer.
1938 *
1939 * Technical note: in 2.3 we work on _locked_ socket, so that
1940 * tricks with *seq access order and skb->users are not required.
1941 * Probably, code can be easily improved even more.
1942 */
1943
Ying Xue1b784142015-03-02 15:37:48 +08001944int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
1945 int flags, int *addr_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946{
1947 struct tcp_sock *tp = tcp_sk(sk);
1948 int copied = 0;
1949 u32 peek_seq;
1950 u32 *seq;
1951 unsigned long used;
Soheil Hassas Yeganehb75eba72018-05-01 15:39:15 -04001952 int err, inq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 int target; /* Read at least this many bytes */
1954 long timeo;
Sabrina Dubrocadfbafc92015-07-24 18:19:25 +02001955 struct sk_buff *skb, *last;
Ilpo Järvinen77527312009-05-10 20:32:34 +00001956 u32 urg_hole = 0;
Deepa Dinamani97184752019-02-02 07:34:51 -08001957 struct scm_timestamping_internal tss;
Mike Maloney98aaa912017-08-22 17:08:48 -04001958 bool has_tss = false;
Soheil Hassas Yeganehb75eba72018-05-01 15:39:15 -04001959 bool has_cmsg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960
Willem de Bruijn4ed2d762014-08-04 22:11:49 -04001961 if (unlikely(flags & MSG_ERRQUEUE))
Willem de Bruijnf4713a32014-11-26 14:53:02 -05001962 return inet_recv_error(sk, msg, len, addr_len);
Willem de Bruijn4ed2d762014-08-04 22:11:49 -04001963
Eliezer Tamircbf55002013-07-08 16:20:34 +03001964 if (sk_can_busy_loop(sk) && skb_queue_empty(&sk->sk_receive_queue) &&
1965 (sk->sk_state == TCP_ESTABLISHED))
1966 sk_busy_loop(sk, nonblock);
Eliezer Tamird30e3832013-06-10 11:40:10 +03001967
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 lock_sock(sk);
1969
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 err = -ENOTCONN;
1971 if (sk->sk_state == TCP_LISTEN)
1972 goto out;
1973
Soheil Hassas Yeganehb75eba72018-05-01 15:39:15 -04001974 has_cmsg = tp->recvmsg_inq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 timeo = sock_rcvtimeo(sk, nonblock);
1976
1977 /* Urgent data needs to be handled specially. */
1978 if (flags & MSG_OOB)
1979 goto recv_urg;
1980
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00001981 if (unlikely(tp->repair)) {
1982 err = -EPERM;
1983 if (!(flags & MSG_PEEK))
1984 goto out;
1985
1986 if (tp->repair_queue == TCP_SEND_QUEUE)
1987 goto recv_sndq;
1988
1989 err = -EINVAL;
1990 if (tp->repair_queue == TCP_NO_QUEUE)
1991 goto out;
1992
1993 /* 'common' recv queue MSG_PEEK-ing */
1994 }
1995
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 seq = &tp->copied_seq;
1997 if (flags & MSG_PEEK) {
1998 peek_seq = tp->copied_seq;
1999 seq = &peek_seq;
2000 }
2001
2002 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
2003
2004 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 u32 offset;
2006
2007 /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */
2008 if (tp->urg_data && tp->urg_seq == *seq) {
2009 if (copied)
2010 break;
2011 if (signal_pending(current)) {
2012 copied = timeo ? sock_intr_errno(timeo) : -EAGAIN;
2013 break;
2014 }
2015 }
2016
2017 /* Next get a buffer. */
2018
Sabrina Dubrocadfbafc92015-07-24 18:19:25 +02002019 last = skb_peek_tail(&sk->sk_receive_queue);
David S. Miller91521942009-05-28 21:35:47 -07002020 skb_queue_walk(&sk->sk_receive_queue, skb) {
Sabrina Dubrocadfbafc92015-07-24 18:19:25 +02002021 last = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 /* Now that we have two receive queues this
2023 * shouldn't happen.
2024 */
Ilpo Järvinend792c102009-11-13 13:56:33 -08002025 if (WARN(before(*seq, TCP_SKB_CB(skb)->seq),
Randy Dunlape56b8ce2018-07-17 18:27:45 -07002026 "TCP recvmsg seq # bug: copied %X, seq %X, rcvnxt %X, fl %X\n",
Joe Perches2af6fd82010-10-30 11:08:53 +00002027 *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt,
2028 flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 break;
Ilpo Järvinend792c102009-11-13 13:56:33 -08002030
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 offset = *seq - TCP_SKB_CB(skb)->seq;
Eric Dumazet9d691532016-02-01 21:03:08 -08002032 if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
2033 pr_err_once("%s: found a SYN, please report !\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 offset--;
Eric Dumazet9d691532016-02-01 21:03:08 -08002035 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 if (offset < skb->len)
2037 goto found_ok_skb;
Eric Dumazete11ecdd2014-09-15 04:19:51 -07002038 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 goto found_fin_ok;
Joe Perches2af6fd82010-10-30 11:08:53 +00002040 WARN(!(flags & MSG_PEEK),
Randy Dunlape56b8ce2018-07-17 18:27:45 -07002041 "TCP recvmsg seq # bug 2: copied %X, seq %X, rcvnxt %X, fl %X\n",
Joe Perches2af6fd82010-10-30 11:08:53 +00002042 *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags);
David S. Miller91521942009-05-28 21:35:47 -07002043 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044
2045 /* Well, if we have backlog, try to process it now yet. */
2046
2047 if (copied >= target && !sk->sk_backlog.tail)
2048 break;
2049
2050 if (copied) {
2051 if (sk->sk_err ||
2052 sk->sk_state == TCP_CLOSE ||
2053 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2054 !timeo ||
David S. Miller518a09e2008-11-05 03:36:01 -08002055 signal_pending(current))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 break;
2057 } else {
2058 if (sock_flag(sk, SOCK_DONE))
2059 break;
2060
2061 if (sk->sk_err) {
2062 copied = sock_error(sk);
2063 break;
2064 }
2065
2066 if (sk->sk_shutdown & RCV_SHUTDOWN)
2067 break;
2068
2069 if (sk->sk_state == TCP_CLOSE) {
Eric Dumazetc47078d2018-07-07 23:15:56 -07002070 /* This occurs when user tries to read
2071 * from never connected socket.
2072 */
2073 copied = -ENOTCONN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 break;
2075 }
2076
2077 if (!timeo) {
2078 copied = -EAGAIN;
2079 break;
2080 }
2081
2082 if (signal_pending(current)) {
2083 copied = sock_intr_errno(timeo);
2084 break;
2085 }
2086 }
2087
Chris Leech0e4b4992006-05-23 18:00:16 -07002088 tcp_cleanup_rbuf(sk, copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 if (copied >= target) {
2091 /* Do not sleep, just process backlog. */
2092 release_sock(sk);
2093 lock_sock(sk);
Sabrina Dubrocadfbafc92015-07-24 18:19:25 +02002094 } else {
2095 sk_wait_data(sk, &timeo, last);
2096 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097
Ilpo Järvinen77527312009-05-10 20:32:34 +00002098 if ((flags & MSG_PEEK) &&
2099 (peek_seq - copied - urg_hole != tp->copied_seq)) {
Joe Perchese87cc472012-05-13 21:56:26 +00002100 net_dbg_ratelimited("TCP(%s:%d): Application bug, race in MSG_PEEK\n",
2101 current->comm,
2102 task_pid_nr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 peek_seq = tp->copied_seq;
2104 }
2105 continue;
2106
Pedro Tammelafdb8b292018-12-06 10:45:28 -02002107found_ok_skb:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 /* Ok so how much can we use? */
2109 used = skb->len - offset;
2110 if (len < used)
2111 used = len;
2112
2113 /* Do we have urgent data here? */
2114 if (tp->urg_data) {
2115 u32 urg_offset = tp->urg_seq - *seq;
2116 if (urg_offset < used) {
2117 if (!urg_offset) {
2118 if (!sock_flag(sk, SOCK_URGINLINE)) {
2119 ++*seq;
Ilpo Järvinen77527312009-05-10 20:32:34 +00002120 urg_hole++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 offset++;
2122 used--;
2123 if (!used)
2124 goto skip_copy;
2125 }
2126 } else
2127 used = urg_offset;
2128 }
2129 }
2130
2131 if (!(flags & MSG_TRUNC)) {
David S. Miller51f3d022014-11-05 16:46:40 -05002132 err = skb_copy_datagram_msg(skb, offset, msg, used);
Dan Williams7bced392013-12-30 12:37:29 -08002133 if (err) {
2134 /* Exception. Bailout! */
2135 if (!copied)
2136 copied = -EFAULT;
2137 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 }
2139 }
2140
2141 *seq += used;
2142 copied += used;
2143 len -= used;
2144
2145 tcp_rcv_space_adjust(sk);
2146
2147skip_copy:
Florian Westphal31770e32017-08-30 19:24:58 +02002148 if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 tp->urg_data = 0;
Florian Westphal31770e32017-08-30 19:24:58 +02002150 tcp_fast_path_check(sk);
2151 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 if (used + offset < skb->len)
2153 continue;
2154
Mike Maloney98aaa912017-08-22 17:08:48 -04002155 if (TCP_SKB_CB(skb)->has_rxtstamp) {
2156 tcp_update_recv_tstamps(skb, &tss);
2157 has_tss = true;
Soheil Hassas Yeganehb75eba72018-05-01 15:39:15 -04002158 has_cmsg = true;
Mike Maloney98aaa912017-08-22 17:08:48 -04002159 }
Eric Dumazete11ecdd2014-09-15 04:19:51 -07002160 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 goto found_fin_ok;
Dan Williams7bced392013-12-30 12:37:29 -08002162 if (!(flags & MSG_PEEK))
2163 sk_eat_skb(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 continue;
2165
Pedro Tammelafdb8b292018-12-06 10:45:28 -02002166found_fin_ok:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 /* Process the FIN. */
2168 ++*seq;
Dan Williams7bced392013-12-30 12:37:29 -08002169 if (!(flags & MSG_PEEK))
2170 sk_eat_skb(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 break;
2172 } while (len > 0);
2173
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 /* According to UNIX98, msg_name/msg_namelen are ignored
2175 * on connected socket. I was just happy when found this 8) --ANK
2176 */
2177
2178 /* Clean up data we have read: This will do ACK frames. */
Chris Leech0e4b4992006-05-23 18:00:16 -07002179 tcp_cleanup_rbuf(sk, copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 release_sock(sk);
Soheil Hassas Yeganehb75eba72018-05-01 15:39:15 -04002182
2183 if (has_cmsg) {
2184 if (has_tss)
2185 tcp_recv_timestamp(msg, sk, &tss);
2186 if (tp->recvmsg_inq) {
2187 inq = tcp_inq_hint(sk);
2188 put_cmsg(msg, SOL_TCP, TCP_CM_INQ, sizeof(inq), &inq);
2189 }
2190 }
2191
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 return copied;
2193
2194out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 release_sock(sk);
2196 return err;
2197
2198recv_urg:
Rami Rosen377f0a02009-03-31 14:43:17 -07002199 err = tcp_recv_urg(sk, msg, len, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 goto out;
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00002201
2202recv_sndq:
2203 err = tcp_peek_sndq(sk, msg, len);
2204 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002206EXPORT_SYMBOL(tcp_recvmsg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207
Ilpo Järvinen490d5042008-01-12 03:17:20 -08002208void tcp_set_state(struct sock *sk, int state)
2209{
2210 int oldstate = sk->sk_state;
2211
Lawrence Brakmod4487492018-01-25 16:14:15 -08002212 /* We defined a new enum for TCP states that are exported in BPF
2213 * so as not force the internal TCP states to be frozen. The
2214 * following checks will detect if an internal state value ever
2215 * differs from the BPF value. If this ever happens, then we will
2216 * need to remap the internal value to the BPF value before calling
2217 * tcp_call_bpf_2arg.
2218 */
2219 BUILD_BUG_ON((int)BPF_TCP_ESTABLISHED != (int)TCP_ESTABLISHED);
2220 BUILD_BUG_ON((int)BPF_TCP_SYN_SENT != (int)TCP_SYN_SENT);
2221 BUILD_BUG_ON((int)BPF_TCP_SYN_RECV != (int)TCP_SYN_RECV);
2222 BUILD_BUG_ON((int)BPF_TCP_FIN_WAIT1 != (int)TCP_FIN_WAIT1);
2223 BUILD_BUG_ON((int)BPF_TCP_FIN_WAIT2 != (int)TCP_FIN_WAIT2);
2224 BUILD_BUG_ON((int)BPF_TCP_TIME_WAIT != (int)TCP_TIME_WAIT);
2225 BUILD_BUG_ON((int)BPF_TCP_CLOSE != (int)TCP_CLOSE);
2226 BUILD_BUG_ON((int)BPF_TCP_CLOSE_WAIT != (int)TCP_CLOSE_WAIT);
2227 BUILD_BUG_ON((int)BPF_TCP_LAST_ACK != (int)TCP_LAST_ACK);
2228 BUILD_BUG_ON((int)BPF_TCP_LISTEN != (int)TCP_LISTEN);
2229 BUILD_BUG_ON((int)BPF_TCP_CLOSING != (int)TCP_CLOSING);
2230 BUILD_BUG_ON((int)BPF_TCP_NEW_SYN_RECV != (int)TCP_NEW_SYN_RECV);
2231 BUILD_BUG_ON((int)BPF_TCP_MAX_STATES != (int)TCP_MAX_STATES);
2232
2233 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_STATE_CB_FLAG))
2234 tcp_call_bpf_2arg(sk, BPF_SOCK_OPS_STATE_CB, oldstate, state);
Song Liue8fce232017-10-23 09:20:27 -07002235
Ilpo Järvinen490d5042008-01-12 03:17:20 -08002236 switch (state) {
2237 case TCP_ESTABLISHED:
2238 if (oldstate != TCP_ESTABLISHED)
Pavel Emelyanov81cc8a72008-07-16 20:22:04 -07002239 TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08002240 break;
2241
2242 case TCP_CLOSE:
2243 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
Pavel Emelyanov81cc8a72008-07-16 20:22:04 -07002244 TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08002245
2246 sk->sk_prot->unhash(sk);
2247 if (inet_csk(sk)->icsk_bind_hash &&
2248 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -08002249 inet_put_port(sk);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08002250 /* fall through */
2251 default:
Jianjun Kong5a5f3a82008-11-03 00:24:34 -08002252 if (oldstate == TCP_ESTABLISHED)
Pavel Emelyanov74688e42008-07-16 20:22:46 -07002253 TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08002254 }
2255
2256 /* Change state AFTER socket is unhashed to avoid closed
2257 * socket sitting in hash tables.
2258 */
Yafang Shao563e0bb2017-12-20 11:12:51 +08002259 inet_sk_state_store(sk, state);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08002260}
2261EXPORT_SYMBOL_GPL(tcp_set_state);
2262
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263/*
2264 * State processing on a close. This implements the state shift for
2265 * sending our FIN frame. Note that we only send a FIN for some
2266 * states. A shutdown() may have already sent the FIN, or we may be
2267 * closed.
2268 */
2269
Arjan van de Ven9b5b5cf2005-11-29 16:21:38 -08002270static const unsigned char new_state[16] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 /* current state: new state: action: */
Eric Dumazet0980c1e2015-03-24 15:58:53 -07002272 [0 /* (Invalid) */] = TCP_CLOSE,
2273 [TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2274 [TCP_SYN_SENT] = TCP_CLOSE,
2275 [TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2276 [TCP_FIN_WAIT1] = TCP_FIN_WAIT1,
2277 [TCP_FIN_WAIT2] = TCP_FIN_WAIT2,
2278 [TCP_TIME_WAIT] = TCP_CLOSE,
2279 [TCP_CLOSE] = TCP_CLOSE,
2280 [TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN,
2281 [TCP_LAST_ACK] = TCP_LAST_ACK,
2282 [TCP_LISTEN] = TCP_CLOSE,
2283 [TCP_CLOSING] = TCP_CLOSING,
2284 [TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285};
2286
2287static int tcp_close_state(struct sock *sk)
2288{
2289 int next = (int)new_state[sk->sk_state];
2290 int ns = next & TCP_STATE_MASK;
2291
2292 tcp_set_state(sk, ns);
2293
2294 return next & TCP_ACTION_FIN;
2295}
2296
2297/*
2298 * Shutdown the sending side of a connection. Much like close except
Satoru SATOH1f29b052008-04-21 02:27:58 -07002299 * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD).
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 */
2301
2302void tcp_shutdown(struct sock *sk, int how)
2303{
2304 /* We need to grab some memory, and put together a FIN,
2305 * and then put it into the queue to be sent.
2306 * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92.
2307 */
2308 if (!(how & SEND_SHUTDOWN))
2309 return;
2310
2311 /* If we've already sent a FIN, or it's a closed state, skip this. */
2312 if ((1 << sk->sk_state) &
2313 (TCPF_ESTABLISHED | TCPF_SYN_SENT |
2314 TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) {
2315 /* Clear out any half completed packets. FIN if needed. */
2316 if (tcp_close_state(sk))
2317 tcp_send_fin(sk);
2318 }
2319}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002320EXPORT_SYMBOL(tcp_shutdown);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321
Arun Sharmaefcdbf22012-01-30 14:16:06 -08002322bool tcp_check_oom(struct sock *sk, int shift)
2323{
2324 bool too_many_orphans, out_of_socket_memory;
2325
2326 too_many_orphans = tcp_too_many_orphans(sk, shift);
2327 out_of_socket_memory = tcp_out_of_memory(sk);
2328
Joe Perchese87cc472012-05-13 21:56:26 +00002329 if (too_many_orphans)
2330 net_info_ratelimited("too many orphaned sockets\n");
2331 if (out_of_socket_memory)
2332 net_info_ratelimited("out of memory -- consider tuning tcp_mem\n");
Arun Sharmaefcdbf22012-01-30 14:16:06 -08002333 return too_many_orphans || out_of_socket_memory;
2334}
2335
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336void tcp_close(struct sock *sk, long timeout)
2337{
2338 struct sk_buff *skb;
2339 int data_was_unread = 0;
Herbert Xu75c2d9072006-05-03 23:31:35 -07002340 int state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341
2342 lock_sock(sk);
2343 sk->sk_shutdown = SHUTDOWN_MASK;
2344
2345 if (sk->sk_state == TCP_LISTEN) {
2346 tcp_set_state(sk, TCP_CLOSE);
2347
2348 /* Special case. */
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07002349 inet_csk_listen_stop(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350
2351 goto adjudge_to_death;
2352 }
2353
2354 /* We need to flush the recv. buffs. We do this only on the
2355 * descriptor close, not protocol-sourced closes, because the
2356 * reader process may not have drained the data yet!
2357 */
2358 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
Eric Dumazete11ecdd2014-09-15 04:19:51 -07002359 u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq;
2360
2361 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
2362 len--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 data_was_unread += len;
2364 __kfree_skb(skb);
2365 }
2366
Hideo Aoki3ab224b2007-12-31 00:11:19 -08002367 sk_mem_reclaim(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368
Konstantin Khorenko565b7b22010-06-24 21:54:58 -07002369 /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */
2370 if (sk->sk_state == TCP_CLOSE)
2371 goto adjudge_to_death;
2372
Gerrit Renker65bb7232007-04-28 21:21:46 -07002373 /* As outlined in RFC 2525, section 2.17, we send a RST here because
2374 * data was lost. To witness the awful effects of the old behavior of
2375 * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk
2376 * GET in an FTP client, suspend the process, wait for the client to
2377 * advertise a zero window, then kill -9 the FTP client, wheee...
2378 * Note: timeout is always zero in such a case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 */
Pavel Emelyanovee995282012-04-19 03:40:39 +00002380 if (unlikely(tcp_sk(sk)->repair)) {
2381 sk->sk_prot->disconnect(sk, 0);
2382 } else if (data_was_unread) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383 /* Unread data was tossed, zap the connection. */
Eric Dumazet6aef70a2016-04-27 16:44:27 -07002384 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 tcp_set_state(sk, TCP_CLOSE);
Wu Fengguangaa133072009-09-02 23:45:45 -07002386 tcp_send_active_reset(sk, sk->sk_allocation);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
2388 /* Check zero linger _after_ checking for unread data. */
2389 sk->sk_prot->disconnect(sk, 0);
Eric Dumazet6aef70a2016-04-27 16:44:27 -07002390 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONDATA);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 } else if (tcp_close_state(sk)) {
2392 /* We FIN if the application ate all the data before
2393 * zapping the connection.
2394 */
2395
2396 /* RED-PEN. Formally speaking, we have broken TCP state
2397 * machine. State transitions:
2398 *
2399 * TCP_ESTABLISHED -> TCP_FIN_WAIT1
2400 * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible)
2401 * TCP_CLOSE_WAIT -> TCP_LAST_ACK
2402 *
2403 * are legal only when FIN has been sent (i.e. in window),
2404 * rather than queued out of window. Purists blame.
2405 *
2406 * F.e. "RFC state" is ESTABLISHED,
2407 * if Linux state is FIN-WAIT-1, but FIN is still not sent.
2408 *
2409 * The visible declinations are that sometimes
2410 * we enter time-wait state, when it is not required really
2411 * (harmless), do not send active resets, when they are
2412 * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when
2413 * they look as CLOSING or LAST_ACK for Linux)
2414 * Probably, I missed some more holelets.
2415 * --ANK
Jerry Chu83368862012-08-31 12:29:12 +00002416 * XXX (TFO) - To start off we don't support SYN+ACK+FIN
2417 * in a single packet! (May consider it later but will
2418 * probably need API support or TCP_CORK SYN-ACK until
2419 * data is written and socket is closed.)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 */
2421 tcp_send_fin(sk);
2422 }
2423
2424 sk_stream_wait_close(sk, timeout);
2425
2426adjudge_to_death:
Herbert Xu75c2d9072006-05-03 23:31:35 -07002427 state = sk->sk_state;
2428 sock_hold(sk);
2429 sock_orphan(sk);
Herbert Xu75c2d9072006-05-03 23:31:35 -07002430
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 local_bh_disable();
2432 bh_lock_sock(sk);
Eric Dumazet8873c062018-10-01 23:24:26 -07002433 /* remove backlog if any, without releasing ownership. */
2434 __release_sock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435
Herbert Xueb4dea52008-12-29 23:04:08 -08002436 percpu_counter_inc(sk->sk_prot->orphan_count);
2437
Herbert Xu75c2d9072006-05-03 23:31:35 -07002438 /* Have we already been destroyed by a softirq or backlog? */
2439 if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE)
2440 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441
2442 /* This is a (useful) BSD violating of the RFC. There is a
2443 * problem with TCP as specified in that the other end could
2444 * keep a socket open forever with no application left this end.
Jesper Juhlb10bd542014-02-09 23:30:32 +01002445 * We use a 1 minute timeout (about the same as BSD) then kill
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 * our end. If they send after that then tough - BUT: long enough
2447 * that we won't make the old 4*rto = almost no time - whoops
2448 * reset mistake.
2449 *
2450 * Nope, it was not mistake. It is really desired behaviour
2451 * f.e. on http servers, when such sockets are useless, but
2452 * consume significant resources. Let's do it with special
2453 * linger2 option. --ANK
2454 */
2455
2456 if (sk->sk_state == TCP_FIN_WAIT2) {
2457 struct tcp_sock *tp = tcp_sk(sk);
2458 if (tp->linger2 < 0) {
2459 tcp_set_state(sk, TCP_CLOSE);
2460 tcp_send_active_reset(sk, GFP_ATOMIC);
Eric Dumazet02a1d6e2016-04-27 16:44:39 -07002461 __NET_INC_STATS(sock_net(sk),
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07002462 LINUX_MIB_TCPABORTONLINGER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 } else {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002464 const int tmo = tcp_fin_time(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465
2466 if (tmo > TCP_TIMEWAIT_LEN) {
David S. Miller52499af2006-07-31 22:32:09 -07002467 inet_csk_reset_keepalive_timer(sk,
2468 tmo - TCP_TIMEWAIT_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
2471 goto out;
2472 }
2473 }
2474 }
2475 if (sk->sk_state != TCP_CLOSE) {
Hideo Aoki3ab224b2007-12-31 00:11:19 -08002476 sk_mem_reclaim(sk);
Arun Sharmaefcdbf22012-01-30 14:16:06 -08002477 if (tcp_check_oom(sk, 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 tcp_set_state(sk, TCP_CLOSE);
2479 tcp_send_active_reset(sk, GFP_ATOMIC);
Eric Dumazet02a1d6e2016-04-27 16:44:39 -07002480 __NET_INC_STATS(sock_net(sk),
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07002481 LINUX_MIB_TCPABORTONMEMORY);
Dan Streetman4ee806d2018-01-18 16:14:26 -05002482 } else if (!check_net(sock_net(sk))) {
2483 /* Not possible to send reset; just close */
2484 tcp_set_state(sk, TCP_CLOSE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 }
2486 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487
Jerry Chu83368862012-08-31 12:29:12 +00002488 if (sk->sk_state == TCP_CLOSE) {
2489 struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
2490 /* We could get here with a non-NULL req if the socket is
2491 * aborted (e.g., closed with unread data) before 3WHS
2492 * finishes.
2493 */
Ian Morris00db4122015-04-03 09:17:27 +01002494 if (req)
Jerry Chu83368862012-08-31 12:29:12 +00002495 reqsk_fastopen_remove(sk, req, false);
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07002496 inet_csk_destroy_sock(sk);
Jerry Chu83368862012-08-31 12:29:12 +00002497 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 /* Otherwise, socket is reprieved until protocol close. */
2499
2500out:
2501 bh_unlock_sock(sk);
2502 local_bh_enable();
Eric Dumazet8873c062018-10-01 23:24:26 -07002503 release_sock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 sock_put(sk);
2505}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002506EXPORT_SYMBOL(tcp_close);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507
2508/* These states need RST on ABORT according to RFC793 */
2509
Eric Dumazeta2a385d2012-05-16 23:15:34 +00002510static inline bool tcp_need_reset(int state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511{
2512 return (1 << state) &
2513 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 |
2514 TCPF_FIN_WAIT2 | TCPF_SYN_RECV);
2515}
2516
Eric Dumazet75c119a2017-10-05 22:21:27 -07002517static void tcp_rtx_queue_purge(struct sock *sk)
2518{
2519 struct rb_node *p = rb_first(&sk->tcp_rtx_queue);
2520
2521 while (p) {
2522 struct sk_buff *skb = rb_to_skb(p);
2523
2524 p = rb_next(p);
2525 /* Since we are deleting whole queue, no need to
2526 * list_del(&skb->tcp_tsorted_anchor)
2527 */
2528 tcp_rtx_queue_unlink(skb, sk);
2529 sk_wmem_free_skb(sk, skb);
2530 }
2531}
2532
Eric Dumazetac3f09b2017-10-05 22:21:22 -07002533void tcp_write_queue_purge(struct sock *sk)
2534{
2535 struct sk_buff *skb;
2536
2537 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
2538 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
2539 tcp_skb_tsorted_anchor_cleanup(skb);
2540 sk_wmem_free_skb(sk, skb);
2541 }
Eric Dumazet75c119a2017-10-05 22:21:27 -07002542 tcp_rtx_queue_purge(sk);
Eric Dumazet472c2e02019-03-22 08:56:39 -07002543 skb = sk->sk_tx_skb_cache;
2544 if (skb) {
Eric Dumazet472c2e02019-03-22 08:56:39 -07002545 __kfree_skb(skb);
2546 sk->sk_tx_skb_cache = NULL;
2547 }
Eric Dumazetac3f09b2017-10-05 22:21:22 -07002548 INIT_LIST_HEAD(&tcp_sk(sk)->tsorted_sent_queue);
2549 sk_mem_reclaim(sk);
2550 tcp_clear_all_retrans_hints(tcp_sk(sk));
Soheil Hassas Yeganehbffd1682018-04-14 20:44:46 -04002551 tcp_sk(sk)->packets_out = 0;
Eric Dumazet04c03112019-02-15 13:36:20 -08002552 inet_csk(sk)->icsk_backoff = 0;
Eric Dumazetac3f09b2017-10-05 22:21:22 -07002553}
2554
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555int tcp_disconnect(struct sock *sk, int flags)
2556{
2557 struct inet_sock *inet = inet_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002558 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 struct tcp_sock *tp = tcp_sk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 int old_state = sk->sk_state;
2561
2562 if (old_state != TCP_CLOSE)
2563 tcp_set_state(sk, TCP_CLOSE);
2564
2565 /* ABORT function of RFC793 */
2566 if (old_state == TCP_LISTEN) {
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07002567 inet_csk_listen_stop(sk);
Pavel Emelyanovee995282012-04-19 03:40:39 +00002568 } else if (unlikely(tp->repair)) {
2569 sk->sk_err = ECONNABORTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 } else if (tcp_need_reset(old_state) ||
2571 (tp->snd_nxt != tp->write_seq &&
2572 (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) {
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -08002573 /* The last check adjusts for discrepancy of Linux wrt. RFC
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574 * states
2575 */
2576 tcp_send_active_reset(sk, gfp_any());
2577 sk->sk_err = ECONNRESET;
2578 } else if (old_state == TCP_SYN_SENT)
2579 sk->sk_err = ECONNRESET;
2580
2581 tcp_clear_xmit_timers(sk);
2582 __skb_queue_purge(&sk->sk_receive_queue);
Eric Dumazet8b27dae2019-03-22 08:56:40 -07002583 if (sk->sk_rx_skb_cache) {
2584 __kfree_skb(sk->sk_rx_skb_cache);
2585 sk->sk_rx_skb_cache = NULL;
2586 }
Eric Dumazet6508b672018-07-07 23:00:01 -07002587 tp->copied_seq = tp->rcv_nxt;
2588 tp->urg_data = 0;
David S. Millerfe067e82007-03-07 12:12:44 -08002589 tcp_write_queue_purge(sk);
Wei Wangcf1ef3f2017-04-20 14:45:46 -07002590 tcp_fastopen_active_disable_ofo_check(sk);
Yaogong Wang9f5afea2016-09-07 14:49:28 -07002591 skb_rbtree_purge(&tp->out_of_order_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592
Eric Dumazetc720c7e82009-10-15 06:30:45 +00002593 inet->inet_dport = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594
2595 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
2596 inet_reset_saddr(sk);
2597
2598 sk->sk_shutdown = 0;
2599 sock_reset_flag(sk, SOCK_DONE);
Eric Dumazet740b0f12014-02-26 14:02:48 -08002600 tp->srtt_us = 0;
Eric Dumazetb9e2e682019-01-17 11:23:36 -08002601 tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
Wei Wang3f6c65d2018-06-19 21:42:50 -07002602 tp->rcv_rtt_last_tsecr = 0;
Yuvaraja Mariappan686a5622015-10-06 10:53:29 -07002603 tp->write_seq += tp->max_window + 2;
2604 if (tp->write_seq == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605 tp->write_seq = 1;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002606 icsk->icsk_backoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 tp->snd_cwnd = 2;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002608 icsk->icsk_probes_out = 0;
Eric Dumazet6a408142019-01-17 11:23:33 -08002609 icsk->icsk_rto = TCP_TIMEOUT_INIT;
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -07002610 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
Eric Dumazet3a9a57f2019-01-17 11:23:37 -08002611 tp->snd_cwnd = TCP_INIT_CWND;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 tp->snd_cwnd_cnt = 0;
Eric Dumazet1fdf4752009-11-30 12:53:30 -08002613 tp->window_clamp = 0;
Yuchung Chenge21db6f2018-04-17 23:18:48 -07002614 tp->delivered_ce = 0;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002615 tcp_set_ca_state(sk, TCP_CA_Open);
Yousuk Seungd4761752017-12-07 13:41:34 -08002616 tp->is_sack_reneg = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617 tcp_clear_retrans(tp);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002618 inet_csk_delack_init(sk);
Wei Wang499350a2017-05-18 11:22:33 -07002619 /* Initialize rcv_mss to TCP_MIN_MSS to avoid division by 0
2620 * issue in __tcp_select_window()
2621 */
2622 icsk->icsk_ack.rcv_mss = TCP_MIN_MSS;
Srinivas Ajib40b4f72007-05-03 17:32:28 -07002623 memset(&tp->rx_opt, 0, sizeof(tp->rx_opt));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624 __sk_dst_reset(sk);
WANG Congd747a7a2017-06-24 23:50:30 -07002625 dst_release(sk->sk_rx_dst);
2626 sk->sk_rx_dst = NULL;
Eric Dumazet17c30602017-04-08 08:07:33 -07002627 tcp_saved_syn_free(tp);
Eric Dumazet5d9f4262018-05-17 14:47:26 -07002628 tp->compressed_ack = 0;
Wei Wangba113c32018-07-31 17:46:21 -07002629 tp->bytes_sent = 0;
Christoph Paasche858faf2019-07-06 16:13:07 -07002630 tp->bytes_acked = 0;
2631 tp->bytes_received = 0;
Wei Wangfb31c9b2018-07-31 17:46:22 -07002632 tp->bytes_retrans = 0;
Yuchung Cheng77881742018-08-29 14:53:56 -07002633 tp->duplicate_sack[0].start_seq = 0;
2634 tp->duplicate_sack[0].end_seq = 0;
Wei Wang7e10b652018-07-31 17:46:23 -07002635 tp->dsack_dups = 0;
Wei Wang7ec65372018-07-31 17:46:24 -07002636 tp->reord_seen = 0;
Eric Dumazet5c701542019-01-17 11:23:39 -08002637 tp->retrans_out = 0;
2638 tp->sacked_out = 0;
2639 tp->tlp_high_seq = 0;
2640 tp->last_oow_ack_time = 0;
Eric Dumazet6cda8b72019-01-17 11:23:40 -08002641 /* There's a bubble in the pipe until at least the first ACK. */
2642 tp->app_limited = ~0U;
Eric Dumazet792c4352019-01-17 11:23:41 -08002643 tp->rack.mstamp = 0;
2644 tp->rack.advanced = 0;
2645 tp->rack.reo_wnd_steps = 1;
2646 tp->rack.last_delivered = 0;
2647 tp->rack.reo_wnd_persist = 0;
2648 tp->rack.dsack_seen = 0;
Eric Dumazet6bcdc402019-01-17 11:23:42 -08002649 tp->syn_data_acked = 0;
2650 tp->rx_opt.saw_tstamp = 0;
2651 tp->rx_opt.dsack = 0;
2652 tp->rx_opt.num_sacks = 0;
Eric Dumazet6cda8b72019-01-17 11:23:40 -08002653
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654
Wei Wang7db92362017-03-01 13:29:48 -08002655 /* Clean up fastopen related fields */
2656 tcp_free_fastopen_req(tp);
2657 inet->defer_connect = 0;
2658
Eric Dumazetc720c7e82009-10-15 06:30:45 +00002659 WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660
Li RongQing9b42d552018-01-26 16:40:41 +08002661 if (sk->sk_frag.page) {
2662 put_page(sk->sk_frag.page);
2663 sk->sk_frag.page = NULL;
2664 sk->sk_frag.offset = 0;
2665 }
2666
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 sk->sk_error_report(sk);
YueHaibinga01512b2018-08-03 16:28:48 +08002668 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002670EXPORT_SYMBOL(tcp_disconnect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671
Eric Dumazeta2a385d2012-05-16 23:15:34 +00002672static inline bool tcp_can_repair_sock(const struct sock *sk)
Pavel Emelyanovee995282012-04-19 03:40:39 +00002673{
Eric W. Biederman52e804c2012-11-16 03:03:05 +00002674 return ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN) &&
Andrey Vagin319b0532016-11-14 18:15:14 -08002675 (sk->sk_state != TCP_LISTEN);
Pavel Emelyanovee995282012-04-19 03:40:39 +00002676}
2677
Andrey Vaginb1ed4c42016-06-27 15:33:56 -07002678static int tcp_repair_set_window(struct tcp_sock *tp, char __user *optbuf, int len)
2679{
2680 struct tcp_repair_window opt;
2681
2682 if (!tp->repair)
2683 return -EPERM;
2684
2685 if (len != sizeof(opt))
2686 return -EINVAL;
2687
2688 if (copy_from_user(&opt, optbuf, sizeof(opt)))
2689 return -EFAULT;
2690
2691 if (opt.max_window < opt.snd_wnd)
2692 return -EINVAL;
2693
2694 if (after(opt.snd_wl1, tp->rcv_nxt + opt.rcv_wnd))
2695 return -EINVAL;
2696
2697 if (after(opt.rcv_wup, tp->rcv_nxt))
2698 return -EINVAL;
2699
2700 tp->snd_wl1 = opt.snd_wl1;
2701 tp->snd_wnd = opt.snd_wnd;
2702 tp->max_window = opt.max_window;
2703
2704 tp->rcv_wnd = opt.rcv_wnd;
2705 tp->rcv_wup = opt.rcv_wup;
2706
2707 return 0;
2708}
2709
Douglas Caetano dos Santos15e56512017-05-26 14:28:00 -03002710static int tcp_repair_options_est(struct sock *sk,
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002711 struct tcp_repair_opt __user *optbuf, unsigned int len)
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002712{
Douglas Caetano dos Santos15e56512017-05-26 14:28:00 -03002713 struct tcp_sock *tp = tcp_sk(sk);
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002714 struct tcp_repair_opt opt;
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002715
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002716 while (len >= sizeof(opt)) {
2717 if (copy_from_user(&opt, optbuf, sizeof(opt)))
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002718 return -EFAULT;
2719
2720 optbuf++;
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002721 len -= sizeof(opt);
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002722
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002723 switch (opt.opt_code) {
2724 case TCPOPT_MSS:
2725 tp->rx_opt.mss_clamp = opt.opt_val;
Douglas Caetano dos Santos15e56512017-05-26 14:28:00 -03002726 tcp_mtup_init(sk);
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002727 break;
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002728 case TCPOPT_WINDOW:
Andrey Vaginbc26ccd2012-09-19 09:40:00 +00002729 {
2730 u16 snd_wscale = opt.opt_val & 0xFFFF;
2731 u16 rcv_wscale = opt.opt_val >> 16;
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002732
Gao Feng589c49c2017-04-04 21:09:48 +08002733 if (snd_wscale > TCP_MAX_WSCALE || rcv_wscale > TCP_MAX_WSCALE)
Andrey Vaginbc26ccd2012-09-19 09:40:00 +00002734 return -EFBIG;
2735
2736 tp->rx_opt.snd_wscale = snd_wscale;
2737 tp->rx_opt.rcv_wscale = rcv_wscale;
2738 tp->rx_opt.wscale_ok = 1;
2739 }
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002740 break;
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002741 case TCPOPT_SACK_PERM:
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002742 if (opt.opt_val != 0)
2743 return -EINVAL;
2744
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002745 tp->rx_opt.sack_ok |= TCP_SACK_SEEN;
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002746 break;
2747 case TCPOPT_TIMESTAMP:
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002748 if (opt.opt_val != 0)
2749 return -EINVAL;
2750
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002751 tp->rx_opt.tstamp_ok = 1;
2752 break;
2753 }
2754 }
2755
2756 return 0;
2757}
2758
Eric Dumazeta842fe12019-06-12 11:57:25 -07002759DEFINE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2760EXPORT_SYMBOL(tcp_tx_delay_enabled);
2761
2762static void tcp_enable_tx_delay(void)
2763{
2764 if (!static_branch_unlikely(&tcp_tx_delay_enabled)) {
2765 static int __tcp_tx_delay_enabled = 0;
2766
2767 if (cmpxchg(&__tcp_tx_delay_enabled, 0, 1) == 0) {
2768 static_branch_enable(&tcp_tx_delay_enabled);
2769 pr_info("TCP_TX_DELAY enabled\n");
2770 }
2771 }
2772}
2773
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775 * Socket option code for TCP.
2776 */
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002777static int do_tcp_setsockopt(struct sock *sk, int level,
David S. Millerb7058842009-09-30 16:12:20 -07002778 int optname, char __user *optval, unsigned int optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779{
2780 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002781 struct inet_connection_sock *icsk = inet_csk(sk);
Nikolay Borisov1e579ca2016-02-03 09:46:56 +02002782 struct net *net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783 int val;
2784 int err = 0;
2785
William Allen Simpsone56fb502009-12-02 18:19:30 +00002786 /* These are data/string values, all the others are ints */
2787 switch (optname) {
2788 case TCP_CONGESTION: {
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002789 char name[TCP_CA_NAME_MAX];
2790
2791 if (optlen < 1)
2792 return -EINVAL;
2793
2794 val = strncpy_from_user(name, optval,
Andrew Morton4fdb78d2009-10-01 15:02:20 -07002795 min_t(long, TCP_CA_NAME_MAX-1, optlen));
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002796 if (val < 0)
2797 return -EFAULT;
2798 name[val] = 0;
2799
2800 lock_sock(sk);
Eric Dumazet8d650cd2019-07-18 19:28:14 -07002801 err = tcp_set_congestion_control(sk, name, true, true,
2802 ns_capable(sock_net(sk)->user_ns,
2803 CAP_NET_ADMIN));
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002804 release_sock(sk);
2805 return err;
2806 }
Dave Watson734942c2017-06-14 11:37:14 -07002807 case TCP_ULP: {
2808 char name[TCP_ULP_NAME_MAX];
2809
2810 if (optlen < 1)
2811 return -EINVAL;
2812
2813 val = strncpy_from_user(name, optval,
2814 min_t(long, TCP_ULP_NAME_MAX - 1,
2815 optlen));
2816 if (val < 0)
2817 return -EFAULT;
2818 name[val] = 0;
2819
2820 lock_sock(sk);
2821 err = tcp_set_ulp(sk, name);
2822 release_sock(sk);
2823 return err;
2824 }
Yuchung Cheng1fba70e2017-10-18 11:22:51 -07002825 case TCP_FASTOPEN_KEY: {
Jason Baron0f1ce022019-05-29 12:33:58 -04002826 __u8 key[TCP_FASTOPEN_KEY_BUF_LENGTH];
2827 __u8 *backup_key = NULL;
Yuchung Cheng1fba70e2017-10-18 11:22:51 -07002828
Jason Baron0f1ce022019-05-29 12:33:58 -04002829 /* Allow a backup key as well to facilitate key rotation
2830 * First key is the active one.
2831 */
2832 if (optlen != TCP_FASTOPEN_KEY_LENGTH &&
2833 optlen != TCP_FASTOPEN_KEY_BUF_LENGTH)
Yuchung Cheng1fba70e2017-10-18 11:22:51 -07002834 return -EINVAL;
2835
2836 if (copy_from_user(key, optval, optlen))
2837 return -EFAULT;
2838
Jason Baron0f1ce022019-05-29 12:33:58 -04002839 if (optlen == TCP_FASTOPEN_KEY_BUF_LENGTH)
2840 backup_key = key + TCP_FASTOPEN_KEY_LENGTH;
2841
Ard Biesheuvel438ac882019-06-19 23:46:28 +02002842 return tcp_fastopen_reset_cipher(net, sk, key, backup_key);
Yuchung Cheng1fba70e2017-10-18 11:22:51 -07002843 }
William Allen Simpsone56fb502009-12-02 18:19:30 +00002844 default:
2845 /* fallthru */
2846 break;
Joe Perchesccbd6a52010-05-14 10:58:26 +00002847 }
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002848
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849 if (optlen < sizeof(int))
2850 return -EINVAL;
2851
2852 if (get_user(val, (int __user *)optval))
2853 return -EFAULT;
2854
2855 lock_sock(sk);
2856
2857 switch (optname) {
2858 case TCP_MAXSEG:
2859 /* Values greater than interface MTU won't take effect. However
2860 * at the point when this call is done we typically don't yet
Rohit Chavana777f712017-05-22 11:59:15 +05302861 * know which interface is going to be used
2862 */
Gao Fengcfc62d82017-03-21 09:28:03 +08002863 if (val && (val < TCP_MIN_MSS || val > MAX_TCP_WINDOW)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864 err = -EINVAL;
2865 break;
2866 }
2867 tp->rx_opt.user_mss = val;
2868 break;
2869
2870 case TCP_NODELAY:
2871 if (val) {
2872 /* TCP_NODELAY is weaker than TCP_CORK, so that
2873 * this option on corked socket is remembered, but
2874 * it is not activated until cork is cleared.
2875 *
2876 * However, when TCP_NODELAY is set we make
2877 * an explicit push, which overrides even TCP_CORK
2878 * for currently queued segments.
2879 */
2880 tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH;
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07002881 tcp_push_pending_frames(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882 } else {
2883 tp->nonagle &= ~TCP_NAGLE_OFF;
2884 }
2885 break;
2886
Andreas Petlund36e31b0a2010-02-18 02:47:01 +00002887 case TCP_THIN_LINEAR_TIMEOUTS:
2888 if (val < 0 || val > 1)
2889 err = -EINVAL;
2890 else
2891 tp->thin_lto = val;
2892 break;
2893
Andreas Petlund7e380172010-02-18 04:48:19 +00002894 case TCP_THIN_DUPACK:
2895 if (val < 0 || val > 1)
2896 err = -EINVAL;
Andreas Petlund7e380172010-02-18 04:48:19 +00002897 break;
2898
Pavel Emelyanovee995282012-04-19 03:40:39 +00002899 case TCP_REPAIR:
2900 if (!tcp_can_repair_sock(sk))
2901 err = -EPERM;
Stefan Baranoff31048d72018-07-15 11:36:37 -04002902 else if (val == TCP_REPAIR_ON) {
Pavel Emelyanovee995282012-04-19 03:40:39 +00002903 tp->repair = 1;
2904 sk->sk_reuse = SK_FORCE_REUSE;
2905 tp->repair_queue = TCP_NO_QUEUE;
Stefan Baranoff31048d72018-07-15 11:36:37 -04002906 } else if (val == TCP_REPAIR_OFF) {
Pavel Emelyanovee995282012-04-19 03:40:39 +00002907 tp->repair = 0;
2908 sk->sk_reuse = SK_NO_REUSE;
2909 tcp_send_window_probe(sk);
Stefan Baranoff31048d72018-07-15 11:36:37 -04002910 } else if (val == TCP_REPAIR_OFF_NO_WP) {
2911 tp->repair = 0;
2912 sk->sk_reuse = SK_NO_REUSE;
Pavel Emelyanovee995282012-04-19 03:40:39 +00002913 } else
2914 err = -EINVAL;
2915
2916 break;
2917
2918 case TCP_REPAIR_QUEUE:
2919 if (!tp->repair)
2920 err = -EPERM;
Eric Dumazetbf2acc92018-04-29 18:55:20 -07002921 else if ((unsigned int)val < TCP_QUEUES_NR)
Pavel Emelyanovee995282012-04-19 03:40:39 +00002922 tp->repair_queue = val;
2923 else
2924 err = -EINVAL;
2925 break;
2926
2927 case TCP_QUEUE_SEQ:
2928 if (sk->sk_state != TCP_CLOSE)
2929 err = -EPERM;
2930 else if (tp->repair_queue == TCP_SEND_QUEUE)
2931 tp->write_seq = val;
2932 else if (tp->repair_queue == TCP_RECV_QUEUE)
2933 tp->rcv_nxt = val;
2934 else
2935 err = -EINVAL;
2936 break;
2937
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002938 case TCP_REPAIR_OPTIONS:
2939 if (!tp->repair)
2940 err = -EINVAL;
2941 else if (sk->sk_state == TCP_ESTABLISHED)
Douglas Caetano dos Santos15e56512017-05-26 14:28:00 -03002942 err = tcp_repair_options_est(sk,
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002943 (struct tcp_repair_opt __user *)optval,
2944 optlen);
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002945 else
2946 err = -EPERM;
2947 break;
2948
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 case TCP_CORK:
2950 /* When set indicates to always queue non-full frames.
2951 * Later the user clears this option and we transmit
2952 * any pending partial frames in the queue. This is
2953 * meant to be used alongside sendfile() to get properly
2954 * filled frames when the user (for example) must write
2955 * out headers with a write() call first and then use
2956 * sendfile to send out the data parts.
2957 *
2958 * TCP_CORK can be set together with TCP_NODELAY and it is
2959 * stronger than TCP_NODELAY.
2960 */
2961 if (val) {
2962 tp->nonagle |= TCP_NAGLE_CORK;
2963 } else {
2964 tp->nonagle &= ~TCP_NAGLE_CORK;
2965 if (tp->nonagle&TCP_NAGLE_OFF)
2966 tp->nonagle |= TCP_NAGLE_PUSH;
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07002967 tcp_push_pending_frames(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002968 }
2969 break;
2970
2971 case TCP_KEEPIDLE:
2972 if (val < 1 || val > MAX_TCP_KEEPIDLE)
2973 err = -EINVAL;
2974 else {
2975 tp->keepalive_time = val * HZ;
2976 if (sock_flag(sk, SOCK_KEEPOPEN) &&
2977 !((1 << sk->sk_state) &
2978 (TCPF_CLOSE | TCPF_LISTEN))) {
Flavio Leitner6c37e5d2010-04-26 18:33:27 +00002979 u32 elapsed = keepalive_time_elapsed(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980 if (tp->keepalive_time > elapsed)
2981 elapsed = tp->keepalive_time - elapsed;
2982 else
2983 elapsed = 0;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002984 inet_csk_reset_keepalive_timer(sk, elapsed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985 }
2986 }
2987 break;
2988 case TCP_KEEPINTVL:
2989 if (val < 1 || val > MAX_TCP_KEEPINTVL)
2990 err = -EINVAL;
2991 else
2992 tp->keepalive_intvl = val * HZ;
2993 break;
2994 case TCP_KEEPCNT:
2995 if (val < 1 || val > MAX_TCP_KEEPCNT)
2996 err = -EINVAL;
2997 else
2998 tp->keepalive_probes = val;
2999 break;
3000 case TCP_SYNCNT:
3001 if (val < 1 || val > MAX_TCP_SYNCNT)
3002 err = -EINVAL;
3003 else
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07003004 icsk->icsk_syn_retries = val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005 break;
3006
Eric Dumazetcd8ae852015-05-03 21:34:46 -07003007 case TCP_SAVE_SYN:
3008 if (val < 0 || val > 1)
3009 err = -EINVAL;
3010 else
3011 tp->save_syn = val;
3012 break;
3013
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 case TCP_LINGER2:
3015 if (val < 0)
3016 tp->linger2 = -1;
Nikolay Borisov1e579ca2016-02-03 09:46:56 +02003017 else if (val > net->ipv4.sysctl_tcp_fin_timeout / HZ)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018 tp->linger2 = 0;
3019 else
3020 tp->linger2 = val * HZ;
3021 break;
3022
3023 case TCP_DEFER_ACCEPT:
Julian Anastasovb103cf32009-10-19 10:10:40 +00003024 /* Translate value in seconds to number of retransmits */
3025 icsk->icsk_accept_queue.rskq_defer_accept =
3026 secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ,
3027 TCP_RTO_MAX / HZ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028 break;
3029
3030 case TCP_WINDOW_CLAMP:
3031 if (!val) {
3032 if (sk->sk_state != TCP_CLOSE) {
3033 err = -EINVAL;
3034 break;
3035 }
3036 tp->window_clamp = 0;
3037 } else
3038 tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ?
3039 SOCK_MIN_RCVBUF / 2 : val;
3040 break;
3041
3042 case TCP_QUICKACK:
3043 if (!val) {
Wei Wang31954cd2019-01-25 10:53:19 -08003044 inet_csk_enter_pingpong_mode(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045 } else {
Wei Wang31954cd2019-01-25 10:53:19 -08003046 inet_csk_exit_pingpong_mode(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047 if ((1 << sk->sk_state) &
3048 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) &&
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07003049 inet_csk_ack_scheduled(sk)) {
3050 icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;
Chris Leech0e4b4992006-05-23 18:00:16 -07003051 tcp_cleanup_rbuf(sk, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003052 if (!(val & 1))
Wei Wang31954cd2019-01-25 10:53:19 -08003053 inet_csk_enter_pingpong_mode(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054 }
3055 }
3056 break;
3057
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003058#ifdef CONFIG_TCP_MD5SIG
3059 case TCP_MD5SIG:
Ivan Delalande8917a772017-06-15 18:07:07 -07003060 case TCP_MD5SIG_EXT:
Eric Dumazet72123032018-04-11 14:36:28 -07003061 if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
3062 err = tp->af_specific->md5_parse(sk, optname, optval, optlen);
3063 else
3064 err = -EINVAL;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003065 break;
3066#endif
Jerry Chudca43c72010-08-27 19:13:28 +00003067 case TCP_USER_TIMEOUT:
Yuchung Chengb2482302014-09-29 13:20:38 -07003068 /* Cap the max time in ms TCP will retry or probe the window
Jerry Chudca43c72010-08-27 19:13:28 +00003069 * before giving up and aborting (ETIMEDOUT) a connection.
3070 */
Hangbin Liu42493572012-07-26 22:52:21 +00003071 if (val < 0)
3072 err = -EINVAL;
3073 else
Jon Maxwell9bcc66e2018-07-19 11:14:42 +10003074 icsk->icsk_user_timeout = val;
Jerry Chudca43c72010-08-27 19:13:28 +00003075 break;
Jerry Chu83368862012-08-31 12:29:12 +00003076
3077 case TCP_FASTOPEN:
3078 if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE |
Christoph Paaschdfea2aa2015-06-18 09:15:34 -07003079 TCPF_LISTEN))) {
Haishuang Yan43713842017-09-27 11:35:42 +08003080 tcp_fastopen_init_key_once(net);
Christoph Paaschdfea2aa2015-06-18 09:15:34 -07003081
Eric Dumazet0536fcc2015-09-29 07:42:52 -07003082 fastopen_queue_tune(sk, val);
Christoph Paaschdfea2aa2015-06-18 09:15:34 -07003083 } else {
Jerry Chu83368862012-08-31 12:29:12 +00003084 err = -EINVAL;
Christoph Paaschdfea2aa2015-06-18 09:15:34 -07003085 }
Jerry Chu83368862012-08-31 12:29:12 +00003086 break;
Wei Wang19f6d3f2017-01-23 10:59:22 -08003087 case TCP_FASTOPEN_CONNECT:
3088 if (val > 1 || val < 0) {
3089 err = -EINVAL;
Haishuang Yane1cfcbe2017-09-27 11:35:40 +08003090 } else if (net->ipv4.sysctl_tcp_fastopen & TFO_CLIENT_ENABLE) {
Wei Wang19f6d3f2017-01-23 10:59:22 -08003091 if (sk->sk_state == TCP_CLOSE)
3092 tp->fastopen_connect = val;
3093 else
3094 err = -EINVAL;
3095 } else {
3096 err = -EOPNOTSUPP;
3097 }
3098 break;
Christoph Paasch71c02372017-10-23 13:22:23 -07003099 case TCP_FASTOPEN_NO_COOKIE:
3100 if (val > 1 || val < 0)
3101 err = -EINVAL;
3102 else if (!((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
3103 err = -EINVAL;
3104 else
3105 tp->fastopen_no_cookie = val;
3106 break;
Andrey Vagin93be6ce2013-02-11 05:50:18 +00003107 case TCP_TIMESTAMP:
3108 if (!tp->repair)
3109 err = -EPERM;
3110 else
Eric Dumazet9a568de2017-05-16 14:00:14 -07003111 tp->tsoffset = val - tcp_time_stamp_raw();
Andrey Vagin93be6ce2013-02-11 05:50:18 +00003112 break;
Andrey Vaginb1ed4c42016-06-27 15:33:56 -07003113 case TCP_REPAIR_WINDOW:
3114 err = tcp_repair_set_window(tp, optval, optlen);
3115 break;
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07003116 case TCP_NOTSENT_LOWAT:
3117 tp->notsent_lowat = val;
3118 sk->sk_write_space(sk);
3119 break;
Soheil Hassas Yeganehb75eba72018-05-01 15:39:15 -04003120 case TCP_INQ:
3121 if (val > 1 || val < 0)
3122 err = -EINVAL;
3123 else
3124 tp->recvmsg_inq = val;
3125 break;
Eric Dumazeta842fe12019-06-12 11:57:25 -07003126 case TCP_TX_DELAY:
3127 if (val)
3128 tcp_enable_tx_delay();
3129 tp->tcp_tx_delay = val;
3130 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131 default:
3132 err = -ENOPROTOOPT;
3133 break;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -07003134 }
3135
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136 release_sock(sk);
3137 return err;
3138}
3139
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08003140int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval,
David S. Millerb7058842009-09-30 16:12:20 -07003141 unsigned int optlen)
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08003142{
Eric Dumazetcf533ea2011-10-21 05:22:42 -04003143 const struct inet_connection_sock *icsk = inet_csk(sk);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08003144
3145 if (level != SOL_TCP)
3146 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
3147 optval, optlen);
3148 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
3149}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00003150EXPORT_SYMBOL(tcp_setsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08003151
3152#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08003153int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -07003154 char __user *optval, unsigned int optlen)
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08003155{
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -08003156 if (level != SOL_TCP)
3157 return inet_csk_compat_setsockopt(sk, level, optname,
3158 optval, optlen);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08003159 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
3160}
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08003161EXPORT_SYMBOL(compat_tcp_setsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08003162#endif
3163
Francis Yanefd90172016-11-27 23:07:17 -08003164static void tcp_get_info_chrono_stats(const struct tcp_sock *tp,
3165 struct tcp_info *info)
3166{
3167 u64 stats[__TCP_CHRONO_MAX], total = 0;
3168 enum tcp_chrono i;
3169
3170 for (i = TCP_CHRONO_BUSY; i < __TCP_CHRONO_MAX; ++i) {
3171 stats[i] = tp->chrono_stat[i - 1];
3172 if (i == tp->chrono_type)
Eric Dumazet628174c2017-05-16 14:00:09 -07003173 stats[i] += tcp_jiffies32 - tp->chrono_start;
Francis Yanefd90172016-11-27 23:07:17 -08003174 stats[i] *= USEC_PER_SEC / HZ;
3175 total += stats[i];
3176 }
3177
3178 info->tcpi_busy_time = total;
3179 info->tcpi_rwnd_limited = stats[TCP_CHRONO_RWND_LIMITED];
3180 info->tcpi_sndbuf_limited = stats[TCP_CHRONO_SNDBUF_LIMITED];
3181}
3182
Linus Torvalds1da177e2005-04-16 15:20:36 -07003183/* Return information about state of tcp endpoint in API format. */
Eric Dumazet0df48c22015-04-28 15:28:17 -07003184void tcp_get_info(struct sock *sk, struct tcp_info *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185{
Craig Gallek35ac8382015-06-15 11:26:20 -04003186 const struct tcp_sock *tp = tcp_sk(sk); /* iff sk_type == SOCK_STREAM */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07003187 const struct inet_connection_sock *icsk = inet_csk(sk);
Eric Dumazet76a9ebe2018-10-15 09:37:53 -07003188 unsigned long rate;
Wei Wang02635982017-07-28 10:28:20 -07003189 u32 now;
Eric Dumazetff5d7492016-01-27 10:52:43 -08003190 u64 rate64;
Eric Dumazet67db3e42016-11-04 11:54:32 -07003191 bool slow;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192
3193 memset(info, 0, sizeof(*info));
Craig Gallek35ac8382015-06-15 11:26:20 -04003194 if (sk->sk_type != SOCK_STREAM)
3195 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003196
Yafang Shao986ffdf2017-12-20 11:12:52 +08003197 info->tcpi_state = inet_sk_state_load(sk);
Eric Dumazet00fd38d2015-11-12 08:43:18 -08003198
Eric Dumazetccbf3bf2016-11-04 11:54:31 -07003199 /* Report meaningful fields for all TCP states, including listeners */
3200 rate = READ_ONCE(sk->sk_pacing_rate);
Eric Dumazet76a9ebe2018-10-15 09:37:53 -07003201 rate64 = (rate != ~0UL) ? rate : ~0ULL;
Eric Dumazetf522a5f2016-11-09 11:24:22 -08003202 info->tcpi_pacing_rate = rate64;
Eric Dumazetccbf3bf2016-11-04 11:54:31 -07003203
3204 rate = READ_ONCE(sk->sk_max_pacing_rate);
Eric Dumazet76a9ebe2018-10-15 09:37:53 -07003205 rate64 = (rate != ~0UL) ? rate : ~0ULL;
Eric Dumazetf522a5f2016-11-09 11:24:22 -08003206 info->tcpi_max_pacing_rate = rate64;
Eric Dumazetccbf3bf2016-11-04 11:54:31 -07003207
3208 info->tcpi_reordering = tp->reordering;
3209 info->tcpi_snd_cwnd = tp->snd_cwnd;
3210
3211 if (info->tcpi_state == TCP_LISTEN) {
3212 /* listeners aliased fields :
3213 * tcpi_unacked -> Number of children ready for accept()
3214 * tcpi_sacked -> max backlog
3215 */
3216 info->tcpi_unacked = sk->sk_ack_backlog;
3217 info->tcpi_sacked = sk->sk_max_ack_backlog;
3218 return;
3219 }
Eric Dumazetb369e7f2017-01-09 10:29:27 -08003220
3221 slow = lock_sock_fast(sk);
3222
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03003223 info->tcpi_ca_state = icsk->icsk_ca_state;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07003224 info->tcpi_retransmits = icsk->icsk_retransmits;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03003225 info->tcpi_probes = icsk->icsk_probes_out;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07003226 info->tcpi_backoff = icsk->icsk_backoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003227
3228 if (tp->rx_opt.tstamp_ok)
3229 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
Ilpo Järvinene60402d2007-08-09 15:14:46 +03003230 if (tcp_is_sack(tp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003231 info->tcpi_options |= TCPI_OPT_SACK;
3232 if (tp->rx_opt.wscale_ok) {
3233 info->tcpi_options |= TCPI_OPT_WSCALE;
3234 info->tcpi_snd_wscale = tp->rx_opt.snd_wscale;
3235 info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09003236 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003237
Eric Dumazetb5c56932011-10-03 14:01:21 -04003238 if (tp->ecn_flags & TCP_ECN_OK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003239 info->tcpi_options |= TCPI_OPT_ECN;
Eric Dumazetb5c56932011-10-03 14:01:21 -04003240 if (tp->ecn_flags & TCP_ECN_SEEN)
3241 info->tcpi_options |= TCPI_OPT_ECN_SEEN;
Yuchung Cheng6f736012012-10-19 15:14:44 +00003242 if (tp->syn_data_acked)
3243 info->tcpi_options |= TCPI_OPT_SYN_DATA;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003244
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07003245 info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto);
3246 info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato);
David S. Millerc1b4a7e2005-07-05 15:24:38 -07003247 info->tcpi_snd_mss = tp->mss_cache;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07003248 info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003249
Eric Dumazetccbf3bf2016-11-04 11:54:31 -07003250 info->tcpi_unacked = tp->packets_out;
3251 info->tcpi_sacked = tp->sacked_out;
3252
Linus Torvalds1da177e2005-04-16 15:20:36 -07003253 info->tcpi_lost = tp->lost_out;
3254 info->tcpi_retrans = tp->retrans_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255
Eric Dumazetd635fbe2017-05-16 14:00:03 -07003256 now = tcp_jiffies32;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257 info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07003258 info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003259 info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp);
3260
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -08003261 info->tcpi_pmtu = icsk->icsk_pmtu_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003262 info->tcpi_rcv_ssthresh = tp->rcv_ssthresh;
Eric Dumazet740b0f12014-02-26 14:02:48 -08003263 info->tcpi_rtt = tp->srtt_us >> 3;
3264 info->tcpi_rttvar = tp->mdev_us >> 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003265 info->tcpi_snd_ssthresh = tp->snd_ssthresh;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003266 info->tcpi_advmss = tp->advmss;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003267
Eric Dumazet645f4c62017-04-25 10:15:41 -07003268 info->tcpi_rcv_rtt = tp->rcv_rtt_est.rtt_us >> 3;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269 info->tcpi_rcv_space = tp->rcvq_space.space;
3270
3271 info->tcpi_total_retrans = tp->total_retrans;
Eric Dumazet977cb0e2014-02-13 14:27:40 -08003272
Eric Dumazetf522a5f2016-11-09 11:24:22 -08003273 info->tcpi_bytes_acked = tp->bytes_acked;
3274 info->tcpi_bytes_received = tp->bytes_received;
Eric Dumazet67db3e42016-11-04 11:54:32 -07003275 info->tcpi_notsent_bytes = max_t(int, 0, tp->write_seq - tp->snd_nxt);
Francis Yanefd90172016-11-27 23:07:17 -08003276 tcp_get_info_chrono_stats(tp, info);
Eric Dumazet67db3e42016-11-04 11:54:32 -07003277
Marcelo Ricardo Leitner2efd0552015-05-20 16:35:41 -07003278 info->tcpi_segs_out = tp->segs_out;
3279 info->tcpi_segs_in = tp->segs_in;
Eric Dumazetcd9b2662016-02-11 22:02:53 -08003280
Eric Dumazetcd9b2662016-02-11 22:02:53 -08003281 info->tcpi_min_rtt = tcp_min_rtt(tp);
Martin KaFai Laua44d6ea2016-03-14 10:52:15 -07003282 info->tcpi_data_segs_in = tp->data_segs_in;
3283 info->tcpi_data_segs_out = tp->data_segs_out;
Yuchung Chengeb8329e2016-09-19 23:39:16 -04003284
3285 info->tcpi_delivery_rate_app_limited = tp->rate_app_limited ? 1 : 0;
Wei Wang02635982017-07-28 10:28:20 -07003286 rate64 = tcp_compute_delivery_rate(tp);
3287 if (rate64)
Eric Dumazetf522a5f2016-11-09 11:24:22 -08003288 info->tcpi_delivery_rate = rate64;
Yuchung Chengfeb5f2e2018-04-17 23:18:49 -07003289 info->tcpi_delivered = tp->delivered;
3290 info->tcpi_delivered_ce = tp->delivered_ce;
Wei Wangba113c32018-07-31 17:46:21 -07003291 info->tcpi_bytes_sent = tp->bytes_sent;
Wei Wangfb31c9b2018-07-31 17:46:22 -07003292 info->tcpi_bytes_retrans = tp->bytes_retrans;
Wei Wang7e10b652018-07-31 17:46:23 -07003293 info->tcpi_dsack_dups = tp->dsack_dups;
Wei Wang7ec65372018-07-31 17:46:24 -07003294 info->tcpi_reord_seen = tp->reord_seen;
Eric Dumazetb369e7f2017-01-09 10:29:27 -08003295 unlock_sock_fast(sk, slow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003296}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003297EXPORT_SYMBOL_GPL(tcp_get_info);
3298
Wei Wang984988a2018-07-31 17:46:20 -07003299static size_t tcp_opt_stats_get_size(void)
3300{
3301 return
3302 nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_BUSY */
3303 nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_RWND_LIMITED */
3304 nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_SNDBUF_LIMITED */
3305 nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_DATA_SEGS_OUT */
3306 nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_TOTAL_RETRANS */
3307 nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_PACING_RATE */
3308 nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_DELIVERY_RATE */
3309 nla_total_size(sizeof(u32)) + /* TCP_NLA_SND_CWND */
3310 nla_total_size(sizeof(u32)) + /* TCP_NLA_REORDERING */
3311 nla_total_size(sizeof(u32)) + /* TCP_NLA_MIN_RTT */
3312 nla_total_size(sizeof(u8)) + /* TCP_NLA_RECUR_RETRANS */
3313 nla_total_size(sizeof(u8)) + /* TCP_NLA_DELIVERY_RATE_APP_LMT */
3314 nla_total_size(sizeof(u32)) + /* TCP_NLA_SNDQ_SIZE */
3315 nla_total_size(sizeof(u8)) + /* TCP_NLA_CA_STATE */
3316 nla_total_size(sizeof(u32)) + /* TCP_NLA_SND_SSTHRESH */
3317 nla_total_size(sizeof(u32)) + /* TCP_NLA_DELIVERED */
3318 nla_total_size(sizeof(u32)) + /* TCP_NLA_DELIVERED_CE */
Wei Wangba113c32018-07-31 17:46:21 -07003319 nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_BYTES_SENT */
Wei Wangfb31c9b2018-07-31 17:46:22 -07003320 nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_BYTES_RETRANS */
Wei Wang7e10b652018-07-31 17:46:23 -07003321 nla_total_size(sizeof(u32)) + /* TCP_NLA_DSACK_DUPS */
Wei Wang7ec65372018-07-31 17:46:24 -07003322 nla_total_size(sizeof(u32)) + /* TCP_NLA_REORD_SEEN */
Yousuk Seunge8bd8fc2018-11-15 16:44:12 -08003323 nla_total_size(sizeof(u32)) + /* TCP_NLA_SRTT */
Wei Wang984988a2018-07-31 17:46:20 -07003324 0;
3325}
3326
Francis Yan1c885802016-11-27 23:07:18 -08003327struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk)
3328{
3329 const struct tcp_sock *tp = tcp_sk(sk);
3330 struct sk_buff *stats;
3331 struct tcp_info info;
Eric Dumazet76a9ebe2018-10-15 09:37:53 -07003332 unsigned long rate;
Wei Wangbb7c19f2017-07-28 10:28:21 -07003333 u64 rate64;
Francis Yan1c885802016-11-27 23:07:18 -08003334
Wei Wang984988a2018-07-31 17:46:20 -07003335 stats = alloc_skb(tcp_opt_stats_get_size(), GFP_ATOMIC);
Francis Yan1c885802016-11-27 23:07:18 -08003336 if (!stats)
3337 return NULL;
3338
3339 tcp_get_info_chrono_stats(tp, &info);
3340 nla_put_u64_64bit(stats, TCP_NLA_BUSY,
3341 info.tcpi_busy_time, TCP_NLA_PAD);
3342 nla_put_u64_64bit(stats, TCP_NLA_RWND_LIMITED,
3343 info.tcpi_rwnd_limited, TCP_NLA_PAD);
3344 nla_put_u64_64bit(stats, TCP_NLA_SNDBUF_LIMITED,
3345 info.tcpi_sndbuf_limited, TCP_NLA_PAD);
Yuchung Cheng7e981022017-01-27 16:24:38 -08003346 nla_put_u64_64bit(stats, TCP_NLA_DATA_SEGS_OUT,
3347 tp->data_segs_out, TCP_NLA_PAD);
3348 nla_put_u64_64bit(stats, TCP_NLA_TOTAL_RETRANS,
3349 tp->total_retrans, TCP_NLA_PAD);
Wei Wangbb7c19f2017-07-28 10:28:21 -07003350
3351 rate = READ_ONCE(sk->sk_pacing_rate);
Eric Dumazet76a9ebe2018-10-15 09:37:53 -07003352 rate64 = (rate != ~0UL) ? rate : ~0ULL;
Wei Wangbb7c19f2017-07-28 10:28:21 -07003353 nla_put_u64_64bit(stats, TCP_NLA_PACING_RATE, rate64, TCP_NLA_PAD);
3354
3355 rate64 = tcp_compute_delivery_rate(tp);
3356 nla_put_u64_64bit(stats, TCP_NLA_DELIVERY_RATE, rate64, TCP_NLA_PAD);
3357
3358 nla_put_u32(stats, TCP_NLA_SND_CWND, tp->snd_cwnd);
3359 nla_put_u32(stats, TCP_NLA_REORDERING, tp->reordering);
3360 nla_put_u32(stats, TCP_NLA_MIN_RTT, tcp_min_rtt(tp));
3361
3362 nla_put_u8(stats, TCP_NLA_RECUR_RETRANS, inet_csk(sk)->icsk_retransmits);
3363 nla_put_u8(stats, TCP_NLA_DELIVERY_RATE_APP_LMT, !!tp->rate_app_limited);
Yousuk Seung7156d192018-03-16 10:51:07 -07003364 nla_put_u32(stats, TCP_NLA_SND_SSTHRESH, tp->snd_ssthresh);
Yuchung Chengfeb5f2e2018-04-17 23:18:49 -07003365 nla_put_u32(stats, TCP_NLA_DELIVERED, tp->delivered);
3366 nla_put_u32(stats, TCP_NLA_DELIVERED_CE, tp->delivered_ce);
Priyaranjan Jha87ecc952018-03-04 10:38:35 -08003367
3368 nla_put_u32(stats, TCP_NLA_SNDQ_SIZE, tp->write_seq - tp->snd_una);
Priyaranjan Jhabe631892018-03-04 10:38:36 -08003369 nla_put_u8(stats, TCP_NLA_CA_STATE, inet_csk(sk)->icsk_ca_state);
Yuchung Chengfeb5f2e2018-04-17 23:18:49 -07003370
Wei Wangba113c32018-07-31 17:46:21 -07003371 nla_put_u64_64bit(stats, TCP_NLA_BYTES_SENT, tp->bytes_sent,
3372 TCP_NLA_PAD);
Wei Wangfb31c9b2018-07-31 17:46:22 -07003373 nla_put_u64_64bit(stats, TCP_NLA_BYTES_RETRANS, tp->bytes_retrans,
3374 TCP_NLA_PAD);
Wei Wang7e10b652018-07-31 17:46:23 -07003375 nla_put_u32(stats, TCP_NLA_DSACK_DUPS, tp->dsack_dups);
Wei Wang7ec65372018-07-31 17:46:24 -07003376 nla_put_u32(stats, TCP_NLA_REORD_SEEN, tp->reord_seen);
Yousuk Seunge8bd8fc2018-11-15 16:44:12 -08003377 nla_put_u32(stats, TCP_NLA_SRTT, tp->srtt_us >> 3);
Wei Wangba113c32018-07-31 17:46:21 -07003378
Francis Yan1c885802016-11-27 23:07:18 -08003379 return stats;
3380}
3381
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08003382static int do_tcp_getsockopt(struct sock *sk, int level,
3383 int optname, char __user *optval, int __user *optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003384{
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -07003385 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003386 struct tcp_sock *tp = tcp_sk(sk);
Nikolay Borisov6fa25162016-02-03 09:46:49 +02003387 struct net *net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003388 int val, len;
3389
Linus Torvalds1da177e2005-04-16 15:20:36 -07003390 if (get_user(len, optlen))
3391 return -EFAULT;
3392
3393 len = min_t(unsigned int, len, sizeof(int));
3394
3395 if (len < 0)
3396 return -EINVAL;
3397
3398 switch (optname) {
3399 case TCP_MAXSEG:
David S. Millerc1b4a7e2005-07-05 15:24:38 -07003400 val = tp->mss_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003401 if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
3402 val = tp->rx_opt.user_mss;
Pavel Emelyanov5e6a3ce2012-04-19 03:41:32 +00003403 if (tp->repair)
3404 val = tp->rx_opt.mss_clamp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405 break;
3406 case TCP_NODELAY:
3407 val = !!(tp->nonagle&TCP_NAGLE_OFF);
3408 break;
3409 case TCP_CORK:
3410 val = !!(tp->nonagle&TCP_NAGLE_CORK);
3411 break;
3412 case TCP_KEEPIDLE:
Eric Dumazetdf19a622009-08-28 23:48:54 -07003413 val = keepalive_time_when(tp) / HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003414 break;
3415 case TCP_KEEPINTVL:
Eric Dumazetdf19a622009-08-28 23:48:54 -07003416 val = keepalive_intvl_when(tp) / HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003417 break;
3418 case TCP_KEEPCNT:
Eric Dumazetdf19a622009-08-28 23:48:54 -07003419 val = keepalive_probes(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003420 break;
3421 case TCP_SYNCNT:
Nikolay Borisov6fa25162016-02-03 09:46:49 +02003422 val = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003423 break;
3424 case TCP_LINGER2:
3425 val = tp->linger2;
3426 if (val >= 0)
Nikolay Borisov1e579ca2016-02-03 09:46:56 +02003427 val = (val ? : net->ipv4.sysctl_tcp_fin_timeout) / HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003428 break;
3429 case TCP_DEFER_ACCEPT:
Julian Anastasovb103cf32009-10-19 10:10:40 +00003430 val = retrans_to_secs(icsk->icsk_accept_queue.rskq_defer_accept,
3431 TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003432 break;
3433 case TCP_WINDOW_CLAMP:
3434 val = tp->window_clamp;
3435 break;
3436 case TCP_INFO: {
3437 struct tcp_info info;
3438
3439 if (get_user(len, optlen))
3440 return -EFAULT;
3441
3442 tcp_get_info(sk, &info);
3443
3444 len = min_t(unsigned int, len, sizeof(info));
3445 if (put_user(len, optlen))
3446 return -EFAULT;
3447 if (copy_to_user(optval, &info, len))
3448 return -EFAULT;
3449 return 0;
3450 }
Eric Dumazet6e9250f2015-04-28 16:23:49 -07003451 case TCP_CC_INFO: {
3452 const struct tcp_congestion_ops *ca_ops;
3453 union tcp_cc_info info;
3454 size_t sz = 0;
3455 int attr;
3456
3457 if (get_user(len, optlen))
3458 return -EFAULT;
3459
3460 ca_ops = icsk->icsk_ca_ops;
3461 if (ca_ops && ca_ops->get_info)
3462 sz = ca_ops->get_info(sk, ~0U, &attr, &info);
3463
3464 len = min_t(unsigned int, len, sz);
3465 if (put_user(len, optlen))
3466 return -EFAULT;
3467 if (copy_to_user(optval, &info, len))
3468 return -EFAULT;
3469 return 0;
3470 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003471 case TCP_QUICKACK:
Wei Wang31954cd2019-01-25 10:53:19 -08003472 val = !inet_csk_in_pingpong_mode(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003473 break;
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07003474
3475 case TCP_CONGESTION:
3476 if (get_user(len, optlen))
3477 return -EFAULT;
3478 len = min_t(unsigned int, len, TCP_CA_NAME_MAX);
3479 if (put_user(len, optlen))
3480 return -EFAULT;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03003481 if (copy_to_user(optval, icsk->icsk_ca_ops->name, len))
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07003482 return -EFAULT;
3483 return 0;
William Allen Simpsone56fb502009-12-02 18:19:30 +00003484
Dave Watson734942c2017-06-14 11:37:14 -07003485 case TCP_ULP:
3486 if (get_user(len, optlen))
3487 return -EFAULT;
3488 len = min_t(unsigned int, len, TCP_ULP_NAME_MAX);
Dave Watsond97af302017-06-26 08:36:47 -07003489 if (!icsk->icsk_ulp_ops) {
3490 if (put_user(0, optlen))
3491 return -EFAULT;
3492 return 0;
3493 }
Dave Watson734942c2017-06-14 11:37:14 -07003494 if (put_user(len, optlen))
3495 return -EFAULT;
3496 if (copy_to_user(optval, icsk->icsk_ulp_ops->name, len))
3497 return -EFAULT;
3498 return 0;
3499
Yuchung Cheng1fba70e2017-10-18 11:22:51 -07003500 case TCP_FASTOPEN_KEY: {
Jason Baron0f1ce022019-05-29 12:33:58 -04003501 __u8 key[TCP_FASTOPEN_KEY_BUF_LENGTH];
Yuchung Cheng1fba70e2017-10-18 11:22:51 -07003502 struct tcp_fastopen_context *ctx;
Jason Baron0f1ce022019-05-29 12:33:58 -04003503 unsigned int key_len = 0;
Yuchung Cheng1fba70e2017-10-18 11:22:51 -07003504
3505 if (get_user(len, optlen))
3506 return -EFAULT;
3507
3508 rcu_read_lock();
3509 ctx = rcu_dereference(icsk->icsk_accept_queue.fastopenq.ctx);
Jason Baron0f1ce022019-05-29 12:33:58 -04003510 if (ctx) {
3511 key_len = tcp_fastopen_context_len(ctx) *
3512 TCP_FASTOPEN_KEY_LENGTH;
3513 memcpy(&key[0], &ctx->key[0], key_len);
3514 }
Yuchung Cheng1fba70e2017-10-18 11:22:51 -07003515 rcu_read_unlock();
3516
Jason Baron0f1ce022019-05-29 12:33:58 -04003517 len = min_t(unsigned int, len, key_len);
Yuchung Cheng1fba70e2017-10-18 11:22:51 -07003518 if (put_user(len, optlen))
3519 return -EFAULT;
3520 if (copy_to_user(optval, key, len))
3521 return -EFAULT;
3522 return 0;
3523 }
Josh Hunt3c0fef02010-07-30 13:49:35 +00003524 case TCP_THIN_LINEAR_TIMEOUTS:
3525 val = tp->thin_lto;
3526 break;
Yuchung Cheng4a7f6002017-01-12 22:11:41 -08003527
Josh Hunt3c0fef02010-07-30 13:49:35 +00003528 case TCP_THIN_DUPACK:
Yuchung Cheng4a7f6002017-01-12 22:11:41 -08003529 val = 0;
Josh Hunt3c0fef02010-07-30 13:49:35 +00003530 break;
Jerry Chudca43c72010-08-27 19:13:28 +00003531
Pavel Emelyanovee995282012-04-19 03:40:39 +00003532 case TCP_REPAIR:
3533 val = tp->repair;
3534 break;
3535
3536 case TCP_REPAIR_QUEUE:
3537 if (tp->repair)
3538 val = tp->repair_queue;
3539 else
3540 return -EINVAL;
3541 break;
3542
Andrey Vaginb1ed4c42016-06-27 15:33:56 -07003543 case TCP_REPAIR_WINDOW: {
3544 struct tcp_repair_window opt;
3545
3546 if (get_user(len, optlen))
3547 return -EFAULT;
3548
3549 if (len != sizeof(opt))
3550 return -EINVAL;
3551
3552 if (!tp->repair)
3553 return -EPERM;
3554
3555 opt.snd_wl1 = tp->snd_wl1;
3556 opt.snd_wnd = tp->snd_wnd;
3557 opt.max_window = tp->max_window;
3558 opt.rcv_wnd = tp->rcv_wnd;
3559 opt.rcv_wup = tp->rcv_wup;
3560
3561 if (copy_to_user(optval, &opt, len))
3562 return -EFAULT;
3563 return 0;
3564 }
Pavel Emelyanovee995282012-04-19 03:40:39 +00003565 case TCP_QUEUE_SEQ:
3566 if (tp->repair_queue == TCP_SEND_QUEUE)
3567 val = tp->write_seq;
3568 else if (tp->repair_queue == TCP_RECV_QUEUE)
3569 val = tp->rcv_nxt;
3570 else
3571 return -EINVAL;
3572 break;
3573
Jerry Chudca43c72010-08-27 19:13:28 +00003574 case TCP_USER_TIMEOUT:
Jon Maxwell9bcc66e2018-07-19 11:14:42 +10003575 val = icsk->icsk_user_timeout;
Jerry Chudca43c72010-08-27 19:13:28 +00003576 break;
Kenjiro Nakayama1536e282014-04-17 01:25:01 +09003577
3578 case TCP_FASTOPEN:
Eric Dumazet0536fcc2015-09-29 07:42:52 -07003579 val = icsk->icsk_accept_queue.fastopenq.max_qlen;
Kenjiro Nakayama1536e282014-04-17 01:25:01 +09003580 break;
3581
Wei Wang19f6d3f2017-01-23 10:59:22 -08003582 case TCP_FASTOPEN_CONNECT:
3583 val = tp->fastopen_connect;
3584 break;
3585
Christoph Paasch71c02372017-10-23 13:22:23 -07003586 case TCP_FASTOPEN_NO_COOKIE:
3587 val = tp->fastopen_no_cookie;
3588 break;
3589
Eric Dumazeta842fe12019-06-12 11:57:25 -07003590 case TCP_TX_DELAY:
3591 val = tp->tcp_tx_delay;
3592 break;
3593
Andrey Vagin93be6ce2013-02-11 05:50:18 +00003594 case TCP_TIMESTAMP:
Eric Dumazet9a568de2017-05-16 14:00:14 -07003595 val = tcp_time_stamp_raw() + tp->tsoffset;
Andrey Vagin93be6ce2013-02-11 05:50:18 +00003596 break;
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07003597 case TCP_NOTSENT_LOWAT:
3598 val = tp->notsent_lowat;
3599 break;
Soheil Hassas Yeganehb75eba72018-05-01 15:39:15 -04003600 case TCP_INQ:
3601 val = tp->recvmsg_inq;
3602 break;
Eric Dumazetcd8ae852015-05-03 21:34:46 -07003603 case TCP_SAVE_SYN:
3604 val = tp->save_syn;
3605 break;
3606 case TCP_SAVED_SYN: {
3607 if (get_user(len, optlen))
3608 return -EFAULT;
3609
3610 lock_sock(sk);
3611 if (tp->saved_syn) {
Eric B Munsonaea09292015-05-18 14:35:58 -04003612 if (len < tp->saved_syn[0]) {
3613 if (put_user(tp->saved_syn[0], optlen)) {
3614 release_sock(sk);
3615 return -EFAULT;
3616 }
3617 release_sock(sk);
3618 return -EINVAL;
3619 }
3620 len = tp->saved_syn[0];
Eric Dumazetcd8ae852015-05-03 21:34:46 -07003621 if (put_user(len, optlen)) {
3622 release_sock(sk);
3623 return -EFAULT;
3624 }
3625 if (copy_to_user(optval, tp->saved_syn + 1, len)) {
3626 release_sock(sk);
3627 return -EFAULT;
3628 }
3629 tcp_saved_syn_free(tp);
3630 release_sock(sk);
3631 } else {
3632 release_sock(sk);
3633 len = 0;
3634 if (put_user(len, optlen))
3635 return -EFAULT;
3636 }
3637 return 0;
3638 }
Eric Dumazet05255b82018-04-27 08:58:08 -07003639#ifdef CONFIG_MMU
3640 case TCP_ZEROCOPY_RECEIVE: {
3641 struct tcp_zerocopy_receive zc;
3642 int err;
3643
3644 if (get_user(len, optlen))
3645 return -EFAULT;
3646 if (len != sizeof(zc))
3647 return -EINVAL;
3648 if (copy_from_user(&zc, optval, len))
3649 return -EFAULT;
3650 lock_sock(sk);
3651 err = tcp_zerocopy_receive(sk, &zc);
3652 release_sock(sk);
3653 if (!err && copy_to_user(optval, &zc, len))
3654 err = -EFAULT;
3655 return err;
3656 }
3657#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003658 default:
3659 return -ENOPROTOOPT;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -07003660 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003661
3662 if (put_user(len, optlen))
3663 return -EFAULT;
3664 if (copy_to_user(optval, &val, len))
3665 return -EFAULT;
3666 return 0;
3667}
3668
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08003669int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval,
3670 int __user *optlen)
3671{
3672 struct inet_connection_sock *icsk = inet_csk(sk);
3673
3674 if (level != SOL_TCP)
3675 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
3676 optval, optlen);
3677 return do_tcp_getsockopt(sk, level, optname, optval, optlen);
3678}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00003679EXPORT_SYMBOL(tcp_getsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08003680
3681#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08003682int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
3683 char __user *optval, int __user *optlen)
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08003684{
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -08003685 if (level != SOL_TCP)
3686 return inet_csk_compat_getsockopt(sk, level, optname,
3687 optval, optlen);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08003688 return do_tcp_getsockopt(sk, level, optname, optval, optlen);
3689}
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08003690EXPORT_SYMBOL(compat_tcp_getsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08003691#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003692
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003693#ifdef CONFIG_TCP_MD5SIG
Eric Dumazet349ce992014-10-23 12:58:58 -07003694static DEFINE_PER_CPU(struct tcp_md5sig_pool, tcp_md5sig_pool);
Eric Dumazet71cea172013-05-20 06:52:26 +00003695static DEFINE_MUTEX(tcp_md5sig_mutex);
Eric Dumazet349ce992014-10-23 12:58:58 -07003696static bool tcp_md5sig_pool_populated = false;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003697
Eric Dumazet71cea172013-05-20 06:52:26 +00003698static void __tcp_alloc_md5sig_pool(void)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003699{
Herbert Xucf80e0e2016-01-24 21:20:23 +08003700 struct crypto_ahash *hash;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003701 int cpu;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003702
Herbert Xucf80e0e2016-01-24 21:20:23 +08003703 hash = crypto_alloc_ahash("md5", 0, CRYPTO_ALG_ASYNC);
Linus Torvalds70477372016-03-17 11:22:54 -07003704 if (IS_ERR(hash))
Herbert Xucf80e0e2016-01-24 21:20:23 +08003705 return;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003706
Herbert Xucf80e0e2016-01-24 21:20:23 +08003707 for_each_possible_cpu(cpu) {
Eric Dumazet19689e32016-06-27 18:51:53 +02003708 void *scratch = per_cpu(tcp_md5sig_pool, cpu).scratch;
Herbert Xucf80e0e2016-01-24 21:20:23 +08003709 struct ahash_request *req;
3710
Eric Dumazet19689e32016-06-27 18:51:53 +02003711 if (!scratch) {
3712 scratch = kmalloc_node(sizeof(union tcp_md5sum_block) +
3713 sizeof(struct tcphdr),
3714 GFP_KERNEL,
3715 cpu_to_node(cpu));
3716 if (!scratch)
3717 return;
3718 per_cpu(tcp_md5sig_pool, cpu).scratch = scratch;
3719 }
Herbert Xucf80e0e2016-01-24 21:20:23 +08003720 if (per_cpu(tcp_md5sig_pool, cpu).md5_req)
3721 continue;
3722
3723 req = ahash_request_alloc(hash, GFP_KERNEL);
3724 if (!req)
3725 return;
3726
3727 ahash_request_set_callback(req, 0, NULL, NULL);
3728
3729 per_cpu(tcp_md5sig_pool, cpu).md5_req = req;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003730 }
Eric Dumazet349ce992014-10-23 12:58:58 -07003731 /* before setting tcp_md5sig_pool_populated, we must commit all writes
3732 * to memory. See smp_rmb() in tcp_get_md5sig_pool()
Eric Dumazet71cea172013-05-20 06:52:26 +00003733 */
3734 smp_wmb();
Eric Dumazet349ce992014-10-23 12:58:58 -07003735 tcp_md5sig_pool_populated = true;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003736}
3737
Eric Dumazet71cea172013-05-20 06:52:26 +00003738bool tcp_alloc_md5sig_pool(void)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003739{
Eric Dumazet349ce992014-10-23 12:58:58 -07003740 if (unlikely(!tcp_md5sig_pool_populated)) {
Eric Dumazet71cea172013-05-20 06:52:26 +00003741 mutex_lock(&tcp_md5sig_mutex);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003742
Eric Dumazet6015c712018-11-27 15:03:21 -08003743 if (!tcp_md5sig_pool_populated) {
Eric Dumazet71cea172013-05-20 06:52:26 +00003744 __tcp_alloc_md5sig_pool();
Eric Dumazet6015c712018-11-27 15:03:21 -08003745 if (tcp_md5sig_pool_populated)
Eric Dumazet921f9a02019-02-26 09:49:11 -08003746 static_branch_inc(&tcp_md5_needed);
Eric Dumazet6015c712018-11-27 15:03:21 -08003747 }
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003748
Eric Dumazet71cea172013-05-20 06:52:26 +00003749 mutex_unlock(&tcp_md5sig_mutex);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003750 }
Eric Dumazet349ce992014-10-23 12:58:58 -07003751 return tcp_md5sig_pool_populated;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003752}
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003753EXPORT_SYMBOL(tcp_alloc_md5sig_pool);
3754
Eric Dumazet35790c02010-05-16 00:34:04 -07003755
3756/**
3757 * tcp_get_md5sig_pool - get md5sig_pool for this user
3758 *
3759 * We use percpu structure, so if we succeed, we exit with preemption
3760 * and BH disabled, to make sure another thread or softirq handling
3761 * wont try to get same context.
3762 */
3763struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003764{
Eric Dumazet35790c02010-05-16 00:34:04 -07003765 local_bh_disable();
Eric Dumazet35790c02010-05-16 00:34:04 -07003766
Eric Dumazet349ce992014-10-23 12:58:58 -07003767 if (tcp_md5sig_pool_populated) {
3768 /* coupled with smp_wmb() in __tcp_alloc_md5sig_pool() */
3769 smp_rmb();
3770 return this_cpu_ptr(&tcp_md5sig_pool);
3771 }
Eric Dumazet35790c02010-05-16 00:34:04 -07003772 local_bh_enable();
3773 return NULL;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003774}
Eric Dumazet35790c02010-05-16 00:34:04 -07003775EXPORT_SYMBOL(tcp_get_md5sig_pool);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003776
Adam Langley49a72df2008-07-19 00:01:42 -07003777int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp,
Eric Dumazetcf533ea2011-10-21 05:22:42 -04003778 const struct sk_buff *skb, unsigned int header_len)
Adam Langley49a72df2008-07-19 00:01:42 -07003779{
3780 struct scatterlist sg;
3781 const struct tcphdr *tp = tcp_hdr(skb);
Herbert Xucf80e0e2016-01-24 21:20:23 +08003782 struct ahash_request *req = hp->md5_req;
Eric Dumazet95c96172012-04-15 05:58:06 +00003783 unsigned int i;
3784 const unsigned int head_data_len = skb_headlen(skb) > header_len ?
3785 skb_headlen(skb) - header_len : 0;
Adam Langley49a72df2008-07-19 00:01:42 -07003786 const struct skb_shared_info *shi = skb_shinfo(skb);
Eric Dumazetd7fd1b572010-05-17 20:40:51 +00003787 struct sk_buff *frag_iter;
Adam Langley49a72df2008-07-19 00:01:42 -07003788
3789 sg_init_table(&sg, 1);
3790
3791 sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len);
Herbert Xucf80e0e2016-01-24 21:20:23 +08003792 ahash_request_set_crypt(req, &sg, NULL, head_data_len);
3793 if (crypto_ahash_update(req))
Adam Langley49a72df2008-07-19 00:01:42 -07003794 return 1;
3795
3796 for (i = 0; i < shi->nr_frags; ++i) {
3797 const struct skb_frag_struct *f = &shi->frags[i];
Eric Dumazet54d27fc2013-05-13 21:25:52 +00003798 unsigned int offset = f->page_offset;
3799 struct page *page = skb_frag_page(f) + (offset >> PAGE_SHIFT);
3800
3801 sg_set_page(&sg, page, skb_frag_size(f),
3802 offset_in_page(offset));
Herbert Xucf80e0e2016-01-24 21:20:23 +08003803 ahash_request_set_crypt(req, &sg, NULL, skb_frag_size(f));
3804 if (crypto_ahash_update(req))
Adam Langley49a72df2008-07-19 00:01:42 -07003805 return 1;
3806 }
3807
Eric Dumazetd7fd1b572010-05-17 20:40:51 +00003808 skb_walk_frags(skb, frag_iter)
3809 if (tcp_md5_hash_skb_data(hp, frag_iter, 0))
3810 return 1;
3811
Adam Langley49a72df2008-07-19 00:01:42 -07003812 return 0;
3813}
Adam Langley49a72df2008-07-19 00:01:42 -07003814EXPORT_SYMBOL(tcp_md5_hash_skb_data);
3815
Eric Dumazetcf533ea2011-10-21 05:22:42 -04003816int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, const struct tcp_md5sig_key *key)
Adam Langley49a72df2008-07-19 00:01:42 -07003817{
3818 struct scatterlist sg;
3819
3820 sg_init_one(&sg, key->key, key->keylen);
Herbert Xucf80e0e2016-01-24 21:20:23 +08003821 ahash_request_set_crypt(hp->md5_req, &sg, NULL, key->keylen);
3822 return crypto_ahash_update(hp->md5_req);
Adam Langley49a72df2008-07-19 00:01:42 -07003823}
Adam Langley49a72df2008-07-19 00:01:42 -07003824EXPORT_SYMBOL(tcp_md5_hash_key);
3825
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003826#endif
3827
Andi Kleen4ac02ba2007-04-20 17:11:46 -07003828void tcp_done(struct sock *sk)
3829{
Jerry Chu83368862012-08-31 12:29:12 +00003830 struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
3831
Jianjun Kong5a5f3a82008-11-03 00:24:34 -08003832 if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
Eric Dumazetc10d9312016-04-29 14:16:47 -07003833 TCP_INC_STATS(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
Andi Kleen4ac02ba2007-04-20 17:11:46 -07003834
3835 tcp_set_state(sk, TCP_CLOSE);
3836 tcp_clear_xmit_timers(sk);
Ian Morris00db4122015-04-03 09:17:27 +01003837 if (req)
Jerry Chu83368862012-08-31 12:29:12 +00003838 reqsk_fastopen_remove(sk, req, false);
Andi Kleen4ac02ba2007-04-20 17:11:46 -07003839
3840 sk->sk_shutdown = SHUTDOWN_MASK;
3841
3842 if (!sock_flag(sk, SOCK_DEAD))
3843 sk->sk_state_change(sk);
3844 else
3845 inet_csk_destroy_sock(sk);
3846}
3847EXPORT_SYMBOL_GPL(tcp_done);
3848
Lorenzo Colittic1e64e22015-12-16 12:30:05 +09003849int tcp_abort(struct sock *sk, int err)
3850{
3851 if (!sk_fullsock(sk)) {
Eric Dumazet07f6f4a2015-12-17 16:14:11 -08003852 if (sk->sk_state == TCP_NEW_SYN_RECV) {
3853 struct request_sock *req = inet_reqsk(sk);
3854
3855 local_bh_disable();
Lorenzo Colittiacc2cf42018-07-07 16:31:40 +09003856 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
Eric Dumazet07f6f4a2015-12-17 16:14:11 -08003857 local_bh_enable();
3858 return 0;
3859 }
Lorenzo Colittic1e64e22015-12-16 12:30:05 +09003860 return -EOPNOTSUPP;
3861 }
3862
3863 /* Don't race with userspace socket closes such as tcp_close. */
3864 lock_sock(sk);
3865
Lorenzo Colitti2010b932015-12-22 00:03:44 +09003866 if (sk->sk_state == TCP_LISTEN) {
3867 tcp_set_state(sk, TCP_CLOSE);
3868 inet_csk_listen_stop(sk);
3869 }
3870
Lorenzo Colittic1e64e22015-12-16 12:30:05 +09003871 /* Don't race with BH socket closes such as inet_csk_listen_stop. */
3872 local_bh_disable();
3873 bh_lock_sock(sk);
3874
3875 if (!sock_flag(sk, SOCK_DEAD)) {
3876 sk->sk_err = err;
3877 /* This barrier is coupled with smp_rmb() in tcp_poll() */
3878 smp_wmb();
3879 sk->sk_error_report(sk);
3880 if (tcp_need_reset(sk->sk_state))
3881 tcp_send_active_reset(sk, GFP_ATOMIC);
3882 tcp_done(sk);
3883 }
3884
3885 bh_unlock_sock(sk);
3886 local_bh_enable();
Soheil Hassas Yeganehe05836a2018-03-06 17:15:12 -05003887 tcp_write_queue_purge(sk);
Lorenzo Colittic1e64e22015-12-16 12:30:05 +09003888 release_sock(sk);
Lorenzo Colittic1e64e22015-12-16 12:30:05 +09003889 return 0;
3890}
3891EXPORT_SYMBOL_GPL(tcp_abort);
3892
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07003893extern struct tcp_congestion_ops tcp_reno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003894
3895static __initdata unsigned long thash_entries;
3896static int __init set_thash_entries(char *str)
3897{
Eldad Zack413c27d2012-05-19 14:13:18 +00003898 ssize_t ret;
3899
Linus Torvalds1da177e2005-04-16 15:20:36 -07003900 if (!str)
3901 return 0;
Eldad Zack413c27d2012-05-19 14:13:18 +00003902
3903 ret = kstrtoul(str, 0, &thash_entries);
3904 if (ret)
3905 return 0;
3906
Linus Torvalds1da177e2005-04-16 15:20:36 -07003907 return 1;
3908}
3909__setup("thash_entries=", set_thash_entries);
3910
Fabian Frederick47d7a88c2014-10-01 18:27:50 +02003911static void __init tcp_init_mem(void)
Glauber Costa4acb4192012-01-30 01:20:17 +00003912{
Eric Dumazetb66e91c2015-05-15 12:39:30 -07003913 unsigned long limit = nr_free_buffer_pages() / 16;
3914
Glauber Costa4acb4192012-01-30 01:20:17 +00003915 limit = max(limit, 128UL);
Eric Dumazetb66e91c2015-05-15 12:39:30 -07003916 sysctl_tcp_mem[0] = limit / 4 * 3; /* 4.68 % */
3917 sysctl_tcp_mem[1] = limit; /* 6.25 % */
3918 sysctl_tcp_mem[2] = sysctl_tcp_mem[0] * 2; /* 9.37 % */
Glauber Costa4acb4192012-01-30 01:20:17 +00003919}
3920
Linus Torvalds1da177e2005-04-16 15:20:36 -07003921void __init tcp_init(void)
3922{
Eric Dumazetb49960a2012-05-02 02:28:41 +00003923 int max_rshare, max_wshare, cnt;
Eric Dumazetb2d3ea42016-09-19 23:39:12 -04003924 unsigned long limit;
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003925 unsigned int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003926
Eric Dumazet3b4929f2019-05-17 17:17:22 -07003927 BUILD_BUG_ON(TCP_MIN_SND_MSS <= MAX_TCP_OPTION_SPACE);
Eric Dumazetb2d3ea42016-09-19 23:39:12 -04003928 BUILD_BUG_ON(sizeof(struct tcp_skb_cb) >
3929 FIELD_SIZEOF(struct sk_buff, cb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003930
Tejun Heo908c7f12014-09-08 09:51:29 +09003931 percpu_counter_init(&tcp_sockets_allocated, 0, GFP_KERNEL);
3932 percpu_counter_init(&tcp_orphan_count, 0, GFP_KERNEL);
Haishuang Yan1946e672016-12-28 17:52:32 +08003933 inet_hashinfo_init(&tcp_hashinfo);
Martin KaFai Lau27da6d32017-12-01 12:52:32 -08003934 inet_hashinfo2_init(&tcp_hashinfo, "tcp_listen_portaddr_hash",
3935 thash_entries, 21, /* one slot per 2 MB*/
3936 0, 64 * 1024);
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003937 tcp_hashinfo.bind_bucket_cachep =
3938 kmem_cache_create("tcp_bind_bucket",
3939 sizeof(struct inet_bind_bucket), 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003940 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003941
Linus Torvalds1da177e2005-04-16 15:20:36 -07003942 /* Size and allocate the main established and bind bucket
3943 * hash tables.
3944 *
3945 * The methodology is similar to that of the buffer cache.
3946 */
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003947 tcp_hashinfo.ehash =
Linus Torvalds1da177e2005-04-16 15:20:36 -07003948 alloc_large_system_hash("TCP established",
Arnaldo Carvalho de Melo0f7ff922005-08-09 19:59:44 -07003949 sizeof(struct inet_ehash_bucket),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003950 thash_entries,
Eric Dumazetfd90b292012-11-30 10:08:52 +00003951 17, /* one slot per 128 KB of memory */
John Heffner9e950ef2006-11-06 23:10:51 -08003952 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003953 NULL,
Eric Dumazetf373b532009-10-09 00:16:19 +00003954 &tcp_hashinfo.ehash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003955 0,
Jean Delvare0ccfe612007-10-30 00:59:25 -07003956 thash_entries ? 0 : 512 * 1024);
Eric Dumazet05dbc7b2013-10-03 00:22:02 -07003957 for (i = 0; i <= tcp_hashinfo.ehash_mask; i++)
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08003958 INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i);
Eric Dumazet05dbc7b2013-10-03 00:22:02 -07003959
Eric Dumazet230140c2007-11-07 02:40:20 -08003960 if (inet_ehash_locks_alloc(&tcp_hashinfo))
3961 panic("TCP: failed to alloc ehash_locks");
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003962 tcp_hashinfo.bhash =
Linus Torvalds1da177e2005-04-16 15:20:36 -07003963 alloc_large_system_hash("TCP bind",
Arnaldo Carvalho de Melo0f7ff922005-08-09 19:59:44 -07003964 sizeof(struct inet_bind_hashbucket),
Eric Dumazetf373b532009-10-09 00:16:19 +00003965 tcp_hashinfo.ehash_mask + 1,
Eric Dumazetfd90b292012-11-30 10:08:52 +00003966 17, /* one slot per 128 KB of memory */
John Heffner9e950ef2006-11-06 23:10:51 -08003967 0,
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003968 &tcp_hashinfo.bhash_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003969 NULL,
Tim Bird31fe62b2012-05-23 13:33:35 +00003970 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003971 64 * 1024);
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003972 tcp_hashinfo.bhash_size = 1U << tcp_hashinfo.bhash_size;
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003973 for (i = 0; i < tcp_hashinfo.bhash_size; i++) {
3974 spin_lock_init(&tcp_hashinfo.bhash[i].lock);
3975 INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003976 }
3977
Eric Dumazetc5ed63d2010-08-25 23:02:17 -07003978
3979 cnt = tcp_hashinfo.ehash_mask + 1;
Eric Dumazetc5ed63d2010-08-25 23:02:17 -07003980 sysctl_tcp_max_orphans = cnt / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003981
Eric W. Biedermana4fe34b2013-10-19 16:25:36 -07003982 tcp_init_mem();
Jason Wangc43b8742012-02-02 00:07:00 +00003983 /* Set per-socket limits to no more than 1/128 the pressure threshold */
Eric Dumazet5fb84b12012-04-10 00:56:42 +00003984 limit = nr_free_buffer_pages() << (PAGE_SHIFT - 7);
Eric Dumazetb49960a2012-05-02 02:28:41 +00003985 max_wshare = min(4UL*1024*1024, limit);
3986 max_rshare = min(6UL*1024*1024, limit);
John Heffner7b4f4b52006-03-25 01:34:07 -08003987
Eric Dumazet356d1832017-11-07 00:29:28 -08003988 init_net.ipv4.sysctl_tcp_wmem[0] = SK_MEM_QUANTUM;
3989 init_net.ipv4.sysctl_tcp_wmem[1] = 16*1024;
3990 init_net.ipv4.sysctl_tcp_wmem[2] = max(64*1024, max_wshare);
John Heffner7b4f4b52006-03-25 01:34:07 -08003991
Eric Dumazet356d1832017-11-07 00:29:28 -08003992 init_net.ipv4.sysctl_tcp_rmem[0] = SK_MEM_QUANTUM;
Yuchung Chenga3375312018-09-27 11:21:19 -07003993 init_net.ipv4.sysctl_tcp_rmem[1] = 131072;
3994 init_net.ipv4.sysctl_tcp_rmem[2] = max(131072, max_rshare);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003995
Joe Perchesafd465032012-03-12 07:03:32 +00003996 pr_info("Hash tables configured (established %u bind %u)\n",
Joe Perches058bd4d2012-03-11 18:36:11 +00003997 tcp_hashinfo.ehash_mask + 1, tcp_hashinfo.bhash_size);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07003998
Haishuang Yan1946e672016-12-28 17:52:32 +08003999 tcp_v4_init();
David S. Miller51c5d0c2012-07-10 00:49:14 -07004000 tcp_metrics_init();
Florian Westphal55d86942014-09-26 22:37:32 +02004001 BUG_ON(tcp_register_congestion_control(&tcp_reno) != 0);
Eric Dumazet46d3cea2012-07-11 05:50:31 +00004002 tcp_tasklet_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004003}