xref: /openbmc/linux/net/ipv4/tcp.c (revision 7fc38225363dd8f19e667ad7c77b63bc4a5c065d)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the  BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		Implementation of the Transmission Control Protocol(TCP).
7  *
8  * Authors:	Ross Biro
9  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10  *		Mark Evans, <evansmp@uhura.aston.ac.uk>
11  *		Corey Minyard <wf-rch!minyard@relay.EU.net>
12  *		Florian La Roche, <flla@stud.uni-sb.de>
13  *		Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
14  *		Linus Torvalds, <torvalds@cs.helsinki.fi>
15  *		Alan Cox, <gw4pts@gw4pts.ampr.org>
16  *		Matthew Dillon, <dillon@apollo.west.oic.com>
17  *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
18  *		Jorge Cwik, <jorge@laser.satlink.net>
19  *
20  * Fixes:
21  *		Alan Cox	:	Numerous verify_area() calls
22  *		Alan Cox	:	Set the ACK bit on a reset
23  *		Alan Cox	:	Stopped it crashing if it closed while
24  *					sk->inuse=1 and was trying to connect
25  *					(tcp_err()).
26  *		Alan Cox	:	All icmp error handling was broken
27  *					pointers passed where wrong and the
28  *					socket was looked up backwards. Nobody
29  *					tested any icmp error code obviously.
30  *		Alan Cox	:	tcp_err() now handled properly. It
31  *					wakes people on errors. poll
32  *					behaves and the icmp error race
33  *					has gone by moving it into sock.c
34  *		Alan Cox	:	tcp_send_reset() fixed to work for
35  *					everything not just packets for
36  *					unknown sockets.
37  *		Alan Cox	:	tcp option processing.
38  *		Alan Cox	:	Reset tweaked (still not 100%) [Had
39  *					syn rule wrong]
40  *		Herp Rosmanith  :	More reset fixes
41  *		Alan Cox	:	No longer acks invalid rst frames.
42  *					Acking any kind of RST is right out.
43  *		Alan Cox	:	Sets an ignore me flag on an rst
44  *					receive otherwise odd bits of prattle
45  *					escape still
46  *		Alan Cox	:	Fixed another acking RST frame bug.
47  *					Should stop LAN workplace lockups.
48  *		Alan Cox	: 	Some tidyups using the new skb list
49  *					facilities
50  *		Alan Cox	:	sk->keepopen now seems to work
51  *		Alan Cox	:	Pulls options out correctly on accepts
52  *		Alan Cox	:	Fixed assorted sk->rqueue->next errors
53  *		Alan Cox	:	PSH doesn't end a TCP read. Switched a
54  *					bit to skb ops.
55  *		Alan Cox	:	Tidied tcp_data to avoid a potential
56  *					nasty.
57  *		Alan Cox	:	Added some better commenting, as the
58  *					tcp is hard to follow
59  *		Alan Cox	:	Removed incorrect check for 20 * psh
60  *	Michael O'Reilly	:	ack < copied bug fix.
61  *	Johannes Stille		:	Misc tcp fixes (not all in yet).
62  *		Alan Cox	:	FIN with no memory -> CRASH
63  *		Alan Cox	:	Added socket option proto entries.
64  *					Also added awareness of them to accept.
65  *		Alan Cox	:	Added TCP options (SOL_TCP)
66  *		Alan Cox	:	Switched wakeup calls to callbacks,
67  *					so the kernel can layer network
68  *					sockets.
69  *		Alan Cox	:	Use ip_tos/ip_ttl settings.
70  *		Alan Cox	:	Handle FIN (more) properly (we hope).
71  *		Alan Cox	:	RST frames sent on unsynchronised
72  *					state ack error.
73  *		Alan Cox	:	Put in missing check for SYN bit.
74  *		Alan Cox	:	Added tcp_select_window() aka NET2E
75  *					window non shrink trick.
76  *		Alan Cox	:	Added a couple of small NET2E timer
77  *					fixes
78  *		Charles Hedrick :	TCP fixes
79  *		Toomas Tamm	:	TCP window fixes
80  *		Alan Cox	:	Small URG fix to rlogin ^C ack fight
81  *		Charles Hedrick	:	Rewrote most of it to actually work
82  *		Linus		:	Rewrote tcp_read() and URG handling
83  *					completely
84  *		Gerhard Koerting:	Fixed some missing timer handling
85  *		Matthew Dillon  :	Reworked TCP machine states as per RFC
86  *		Gerhard Koerting:	PC/TCP workarounds
87  *		Adam Caldwell	:	Assorted timer/timing errors
88  *		Matthew Dillon	:	Fixed another RST bug
89  *		Alan Cox	:	Move to kernel side addressing changes.
90  *		Alan Cox	:	Beginning work on TCP fastpathing
91  *					(not yet usable)
92  *		Arnt Gulbrandsen:	Turbocharged tcp_check() routine.
93  *		Alan Cox	:	TCP fast path debugging
94  *		Alan Cox	:	Window clamping
95  *		Michael Riepe	:	Bug in tcp_check()
96  *		Matt Dillon	:	More TCP improvements and RST bug fixes
97  *		Matt Dillon	:	Yet more small nasties remove from the
98  *					TCP code (Be very nice to this man if
99  *					tcp finally works 100%) 8)
100  *		Alan Cox	:	BSD accept semantics.
101  *		Alan Cox	:	Reset on closedown bug.
102  *	Peter De Schrijver	:	ENOTCONN check missing in tcp_sendto().
103  *		Michael Pall	:	Handle poll() after URG properly in
104  *					all cases.
105  *		Michael Pall	:	Undo the last fix in tcp_read_urg()
106  *					(multi URG PUSH broke rlogin).
107  *		Michael Pall	:	Fix the multi URG PUSH problem in
108  *					tcp_readable(), poll() after URG
109  *					works now.
110  *		Michael Pall	:	recv(...,MSG_OOB) never blocks in the
111  *					BSD api.
112  *		Alan Cox	:	Changed the semantics of sk->socket to
113  *					fix a race and a signal problem with
114  *					accept() and async I/O.
115  *		Alan Cox	:	Relaxed the rules on tcp_sendto().
116  *		Yury Shevchuk	:	Really fixed accept() blocking problem.
117  *		Craig I. Hagan  :	Allow for BSD compatible TIME_WAIT for
118  *					clients/servers which listen in on
119  *					fixed ports.
120  *		Alan Cox	:	Cleaned the above up and shrank it to
121  *					a sensible code size.
122  *		Alan Cox	:	Self connect lockup fix.
123  *		Alan Cox	:	No connect to multicast.
124  *		Ross Biro	:	Close unaccepted children on master
125  *					socket close.
126  *		Alan Cox	:	Reset tracing code.
127  *		Alan Cox	:	Spurious resets on shutdown.
128  *		Alan Cox	:	Giant 15 minute/60 second timer error
129  *		Alan Cox	:	Small whoops in polling before an
130  *					accept.
131  *		Alan Cox	:	Kept the state trace facility since
132  *					it's handy for debugging.
133  *		Alan Cox	:	More reset handler fixes.
134  *		Alan Cox	:	Started rewriting the code based on
135  *					the RFC's for other useful protocol
136  *					references see: Comer, KA9Q NOS, and
137  *					for a reference on the difference
138  *					between specifications and how BSD
139  *					works see the 4.4lite source.
140  *		A.N.Kuznetsov	:	Don't time wait on completion of tidy
141  *					close.
142  *		Linus Torvalds	:	Fin/Shutdown & copied_seq changes.
143  *		Linus Torvalds	:	Fixed BSD port reuse to work first syn
144  *		Alan Cox	:	Reimplemented timers as per the RFC
145  *					and using multiple timers for sanity.
146  *		Alan Cox	:	Small bug fixes, and a lot of new
147  *					comments.
148  *		Alan Cox	:	Fixed dual reader crash by locking
149  *					the buffers (much like datagram.c)
150  *		Alan Cox	:	Fixed stuck sockets in probe. A probe
151  *					now gets fed up of retrying without
152  *					(even a no space) answer.
153  *		Alan Cox	:	Extracted closing code better
154  *		Alan Cox	:	Fixed the closing state machine to
155  *					resemble the RFC.
156  *		Alan Cox	:	More 'per spec' fixes.
157  *		Jorge Cwik	:	Even faster checksumming.
158  *		Alan Cox	:	tcp_data() doesn't ack illegal PSH
159  *					only frames. At least one pc tcp stack
160  *					generates them.
161  *		Alan Cox	:	Cache last socket.
162  *		Alan Cox	:	Per route irtt.
163  *		Matt Day	:	poll()->select() match BSD precisely on error
164  *		Alan Cox	:	New buffers
165  *		Marc Tamsky	:	Various sk->prot->retransmits and
166  *					sk->retransmits misupdating fixed.
167  *					Fixed tcp_write_timeout: stuck close,
168  *					and TCP syn retries gets used now.
169  *		Mark Yarvis	:	In tcp_read_wakeup(), don't send an
170  *					ack if state is TCP_CLOSED.
171  *		Alan Cox	:	Look up device on a retransmit - routes may
172  *					change. Doesn't yet cope with MSS shrink right
173  *					but it's a start!
174  *		Marc Tamsky	:	Closing in closing fixes.
175  *		Mike Shaver	:	RFC1122 verifications.
176  *		Alan Cox	:	rcv_saddr errors.
177  *		Alan Cox	:	Block double connect().
178  *		Alan Cox	:	Small hooks for enSKIP.
179  *		Alexey Kuznetsov:	Path MTU discovery.
180  *		Alan Cox	:	Support soft errors.
181  *		Alan Cox	:	Fix MTU discovery pathological case
182  *					when the remote claims no mtu!
183  *		Marc Tamsky	:	TCP_CLOSE fix.
184  *		Colin (G3TNE)	:	Send a reset on syn ack replies in
185  *					window but wrong (fixes NT lpd problems)
186  *		Pedro Roque	:	Better TCP window handling, delayed ack.
187  *		Joerg Reuter	:	No modification of locked buffers in
188  *					tcp_do_retransmit()
189  *		Eric Schenk	:	Changed receiver side silly window
190  *					avoidance algorithm to BSD style
191  *					algorithm. This doubles throughput
192  *					against machines running Solaris,
193  *					and seems to result in general
194  *					improvement.
195  *	Stefan Magdalinski	:	adjusted tcp_readable() to fix FIONREAD
196  *	Willy Konynenberg	:	Transparent proxying support.
197  *	Mike McLagan		:	Routing by source
198  *		Keith Owens	:	Do proper merging with partial SKB's in
199  *					tcp_do_sendmsg to avoid burstiness.
200  *		Eric Schenk	:	Fix fast close down bug with
201  *					shutdown() followed by close().
202  *		Andi Kleen 	:	Make poll agree with SIGIO
203  *	Salvatore Sanfilippo	:	Support SO_LINGER with linger == 1 and
204  *					lingertime == 0 (RFC 793 ABORT Call)
205  *	Hirokazu Takahashi	:	Use copy_from_user() instead of
206  *					csum_and_copy_from_user() if possible.
207  *
208  *		This program is free software; you can redistribute it and/or
209  *		modify it under the terms of the GNU General Public License
210  *		as published by the Free Software Foundation; either version
211  *		2 of the License, or(at your option) any later version.
212  *
213  * Description of States:
214  *
215  *	TCP_SYN_SENT		sent a connection request, waiting for ack
216  *
217  *	TCP_SYN_RECV		received a connection request, sent ack,
218  *				waiting for final ack in three-way handshake.
219  *
220  *	TCP_ESTABLISHED		connection established
221  *
222  *	TCP_FIN_WAIT1		our side has shutdown, waiting to complete
223  *				transmission of remaining buffered data
224  *
225  *	TCP_FIN_WAIT2		all buffered data sent, waiting for remote
226  *				to shutdown
227  *
228  *	TCP_CLOSING		both sides have shutdown but we still have
229  *				data we have to finish sending
230  *
231  *	TCP_TIME_WAIT		timeout to catch resent junk before entering
232  *				closed, can only be entered from FIN_WAIT2
233  *				or CLOSING.  Required because the other end
234  *				may not have gotten our last ACK causing it
235  *				to retransmit the data packet (which we ignore)
236  *
237  *	TCP_CLOSE_WAIT		remote side has shutdown and is waiting for
238  *				us to finish writing our data and to shutdown
239  *				(we have to close() to move on to LAST_ACK)
240  *
241  *	TCP_LAST_ACK		out side has shutdown after remote has
242  *				shutdown.  There may still be data in our
243  *				buffer that we have to finish sending
244  *
245  *	TCP_CLOSE		socket is finished
246  */
247 
248 #define pr_fmt(fmt) "TCP: " fmt
249 
250 #include <crypto/hash.h>
251 #include <linux/kernel.h>
252 #include <linux/module.h>
253 #include <linux/types.h>
254 #include <linux/fcntl.h>
255 #include <linux/poll.h>
256 #include <linux/inet_diag.h>
257 #include <linux/init.h>
258 #include <linux/fs.h>
259 #include <linux/skbuff.h>
260 #include <linux/scatterlist.h>
261 #include <linux/splice.h>
262 #include <linux/net.h>
263 #include <linux/socket.h>
264 #include <linux/random.h>
265 #include <linux/memblock.h>
266 #include <linux/highmem.h>
267 #include <linux/swap.h>
268 #include <linux/cache.h>
269 #include <linux/err.h>
270 #include <linux/time.h>
271 #include <linux/slab.h>
272 #include <linux/errqueue.h>
273 #include <linux/static_key.h>
274 
275 #include <net/icmp.h>
276 #include <net/inet_common.h>
277 #include <net/tcp.h>
278 #include <net/xfrm.h>
279 #include <net/ip.h>
280 #include <net/sock.h>
281 
282 #include <linux/uaccess.h>
283 #include <asm/ioctls.h>
284 #include <net/busy_poll.h>
285 
286 struct percpu_counter tcp_orphan_count;
287 EXPORT_SYMBOL_GPL(tcp_orphan_count);
288 
289 long sysctl_tcp_mem[3] __read_mostly;
290 EXPORT_SYMBOL(sysctl_tcp_mem);
291 
292 atomic_long_t tcp_memory_allocated;	/* Current allocated memory. */
293 EXPORT_SYMBOL(tcp_memory_allocated);
294 
295 #if IS_ENABLED(CONFIG_SMC)
296 DEFINE_STATIC_KEY_FALSE(tcp_have_smc);
297 EXPORT_SYMBOL(tcp_have_smc);
298 #endif
299 
300 /*
301  * Current number of TCP sockets.
302  */
303 struct percpu_counter tcp_sockets_allocated;
304 EXPORT_SYMBOL(tcp_sockets_allocated);
305 
306 /*
307  * TCP splice context
308  */
309 struct tcp_splice_state {
310 	struct pipe_inode_info *pipe;
311 	size_t len;
312 	unsigned int flags;
313 };
314 
315 /*
316  * Pressure flag: try to collapse.
317  * Technical note: it is used by multiple contexts non atomically.
318  * All the __sk_mem_schedule() is of this nature: accounting
319  * is strict, actions are advisory and have some latency.
320  */
321 unsigned long tcp_memory_pressure __read_mostly;
322 EXPORT_SYMBOL_GPL(tcp_memory_pressure);
323 
324 void tcp_enter_memory_pressure(struct sock *sk)
325 {
326 	unsigned long val;
327 
328 	if (tcp_memory_pressure)
329 		return;
330 	val = jiffies;
331 
332 	if (!val)
333 		val--;
334 	if (!cmpxchg(&tcp_memory_pressure, 0, val))
335 		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES);
336 }
337 EXPORT_SYMBOL_GPL(tcp_enter_memory_pressure);
338 
339 void tcp_leave_memory_pressure(struct sock *sk)
340 {
341 	unsigned long val;
342 
343 	if (!tcp_memory_pressure)
344 		return;
345 	val = xchg(&tcp_memory_pressure, 0);
346 	if (val)
347 		NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURESCHRONO,
348 			      jiffies_to_msecs(jiffies - val));
349 }
350 EXPORT_SYMBOL_GPL(tcp_leave_memory_pressure);
351 
352 /* Convert seconds to retransmits based on initial and max timeout */
353 static u8 secs_to_retrans(int seconds, int timeout, int rto_max)
354 {
355 	u8 res = 0;
356 
357 	if (seconds > 0) {
358 		int period = timeout;
359 
360 		res = 1;
361 		while (seconds > period && res < 255) {
362 			res++;
363 			timeout <<= 1;
364 			if (timeout > rto_max)
365 				timeout = rto_max;
366 			period += timeout;
367 		}
368 	}
369 	return res;
370 }
371 
372 /* Convert retransmits to seconds based on initial and max timeout */
373 static int retrans_to_secs(u8 retrans, int timeout, int rto_max)
374 {
375 	int period = 0;
376 
377 	if (retrans > 0) {
378 		period = timeout;
379 		while (--retrans) {
380 			timeout <<= 1;
381 			if (timeout > rto_max)
382 				timeout = rto_max;
383 			period += timeout;
384 		}
385 	}
386 	return period;
387 }
388 
389 static u64 tcp_compute_delivery_rate(const struct tcp_sock *tp)
390 {
391 	u32 rate = READ_ONCE(tp->rate_delivered);
392 	u32 intv = READ_ONCE(tp->rate_interval_us);
393 	u64 rate64 = 0;
394 
395 	if (rate && intv) {
396 		rate64 = (u64)rate * tp->mss_cache * USEC_PER_SEC;
397 		do_div(rate64, intv);
398 	}
399 	return rate64;
400 }
401 
402 /* Address-family independent initialization for a tcp_sock.
403  *
404  * NOTE: A lot of things set to zero explicitly by call to
405  *       sk_alloc() so need not be done here.
406  */
407 void tcp_init_sock(struct sock *sk)
408 {
409 	struct inet_connection_sock *icsk = inet_csk(sk);
410 	struct tcp_sock *tp = tcp_sk(sk);
411 
412 	tp->out_of_order_queue = RB_ROOT;
413 	sk->tcp_rtx_queue = RB_ROOT;
414 	tcp_init_xmit_timers(sk);
415 	INIT_LIST_HEAD(&tp->tsq_node);
416 	INIT_LIST_HEAD(&tp->tsorted_sent_queue);
417 
418 	icsk->icsk_rto = TCP_TIMEOUT_INIT;
419 	tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
420 	minmax_reset(&tp->rtt_min, tcp_jiffies32, ~0U);
421 
422 	/* So many TCP implementations out there (incorrectly) count the
423 	 * initial SYN frame in their delayed-ACK and congestion control
424 	 * algorithms that we must have the following bandaid to talk
425 	 * efficiently to them.  -DaveM
426 	 */
427 	tp->snd_cwnd = TCP_INIT_CWND;
428 
429 	/* There's a bubble in the pipe until at least the first ACK. */
430 	tp->app_limited = ~0U;
431 
432 	/* See draft-stevens-tcpca-spec-01 for discussion of the
433 	 * initialization of these values.
434 	 */
435 	tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
436 	tp->snd_cwnd_clamp = ~0;
437 	tp->mss_cache = TCP_MSS_DEFAULT;
438 
439 	tp->reordering = sock_net(sk)->ipv4.sysctl_tcp_reordering;
440 	tcp_assign_congestion_control(sk);
441 
442 	tp->tsoffset = 0;
443 	tp->rack.reo_wnd_steps = 1;
444 
445 	sk->sk_state = TCP_CLOSE;
446 
447 	sk->sk_write_space = sk_stream_write_space;
448 	sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
449 
450 	icsk->icsk_sync_mss = tcp_sync_mss;
451 
452 	sk->sk_sndbuf = sock_net(sk)->ipv4.sysctl_tcp_wmem[1];
453 	sk->sk_rcvbuf = sock_net(sk)->ipv4.sysctl_tcp_rmem[1];
454 
455 	sk_sockets_allocated_inc(sk);
456 	sk->sk_route_forced_caps = NETIF_F_GSO;
457 }
458 EXPORT_SYMBOL(tcp_init_sock);
459 
460 void tcp_init_transfer(struct sock *sk, int bpf_op)
461 {
462 	struct inet_connection_sock *icsk = inet_csk(sk);
463 
464 	tcp_mtup_init(sk);
465 	icsk->icsk_af_ops->rebuild_header(sk);
466 	tcp_init_metrics(sk);
467 	tcp_call_bpf(sk, bpf_op, 0, NULL);
468 	tcp_init_congestion_control(sk);
469 	tcp_init_buffer_space(sk);
470 }
471 
472 static void tcp_tx_timestamp(struct sock *sk, u16 tsflags)
473 {
474 	struct sk_buff *skb = tcp_write_queue_tail(sk);
475 
476 	if (tsflags && skb) {
477 		struct skb_shared_info *shinfo = skb_shinfo(skb);
478 		struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
479 
480 		sock_tx_timestamp(sk, tsflags, &shinfo->tx_flags);
481 		if (tsflags & SOF_TIMESTAMPING_TX_ACK)
482 			tcb->txstamp_ack = 1;
483 		if (tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
484 			shinfo->tskey = TCP_SKB_CB(skb)->seq + skb->len - 1;
485 	}
486 }
487 
488 static inline bool tcp_stream_is_readable(const struct tcp_sock *tp,
489 					  int target, struct sock *sk)
490 {
491 	return (tp->rcv_nxt - tp->copied_seq >= target) ||
492 		(sk->sk_prot->stream_memory_read ?
493 		sk->sk_prot->stream_memory_read(sk) : false);
494 }
495 
496 /*
497  *	Wait for a TCP event.
498  *
499  *	Note that we don't need to lock the socket, as the upper poll layers
500  *	take care of normal races (between the test and the event) and we don't
501  *	go look at any of the socket buffers directly.
502  */
503 __poll_t tcp_poll(struct file *file, struct socket *sock, poll_table *wait)
504 {
505 	__poll_t mask;
506 	struct sock *sk = sock->sk;
507 	const struct tcp_sock *tp = tcp_sk(sk);
508 	int state;
509 
510 	sock_poll_wait(file, sock, wait);
511 
512 	state = inet_sk_state_load(sk);
513 	if (state == TCP_LISTEN)
514 		return inet_csk_listen_poll(sk);
515 
516 	/* Socket is not locked. We are protected from async events
517 	 * by poll logic and correct handling of state changes
518 	 * made by other threads is impossible in any case.
519 	 */
520 
521 	mask = 0;
522 
523 	/*
524 	 * EPOLLHUP is certainly not done right. But poll() doesn't
525 	 * have a notion of HUP in just one direction, and for a
526 	 * socket the read side is more interesting.
527 	 *
528 	 * Some poll() documentation says that EPOLLHUP is incompatible
529 	 * with the EPOLLOUT/POLLWR flags, so somebody should check this
530 	 * all. But careful, it tends to be safer to return too many
531 	 * bits than too few, and you can easily break real applications
532 	 * if you don't tell them that something has hung up!
533 	 *
534 	 * Check-me.
535 	 *
536 	 * Check number 1. EPOLLHUP is _UNMASKABLE_ event (see UNIX98 and
537 	 * our fs/select.c). It means that after we received EOF,
538 	 * poll always returns immediately, making impossible poll() on write()
539 	 * in state CLOSE_WAIT. One solution is evident --- to set EPOLLHUP
540 	 * if and only if shutdown has been made in both directions.
541 	 * Actually, it is interesting to look how Solaris and DUX
542 	 * solve this dilemma. I would prefer, if EPOLLHUP were maskable,
543 	 * then we could set it on SND_SHUTDOWN. BTW examples given
544 	 * in Stevens' books assume exactly this behaviour, it explains
545 	 * why EPOLLHUP is incompatible with EPOLLOUT.	--ANK
546 	 *
547 	 * NOTE. Check for TCP_CLOSE is added. The goal is to prevent
548 	 * blocking on fresh not-connected or disconnected socket. --ANK
549 	 */
550 	if (sk->sk_shutdown == SHUTDOWN_MASK || state == TCP_CLOSE)
551 		mask |= EPOLLHUP;
552 	if (sk->sk_shutdown & RCV_SHUTDOWN)
553 		mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
554 
555 	/* Connected or passive Fast Open socket? */
556 	if (state != TCP_SYN_SENT &&
557 	    (state != TCP_SYN_RECV || tp->fastopen_rsk)) {
558 		int target = sock_rcvlowat(sk, 0, INT_MAX);
559 
560 		if (tp->urg_seq == tp->copied_seq &&
561 		    !sock_flag(sk, SOCK_URGINLINE) &&
562 		    tp->urg_data)
563 			target++;
564 
565 		if (tcp_stream_is_readable(tp, target, sk))
566 			mask |= EPOLLIN | EPOLLRDNORM;
567 
568 		if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
569 			if (sk_stream_is_writeable(sk)) {
570 				mask |= EPOLLOUT | EPOLLWRNORM;
571 			} else {  /* send SIGIO later */
572 				sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
573 				set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
574 
575 				/* Race breaker. If space is freed after
576 				 * wspace test but before the flags are set,
577 				 * IO signal will be lost. Memory barrier
578 				 * pairs with the input side.
579 				 */
580 				smp_mb__after_atomic();
581 				if (sk_stream_is_writeable(sk))
582 					mask |= EPOLLOUT | EPOLLWRNORM;
583 			}
584 		} else
585 			mask |= EPOLLOUT | EPOLLWRNORM;
586 
587 		if (tp->urg_data & TCP_URG_VALID)
588 			mask |= EPOLLPRI;
589 	} else if (state == TCP_SYN_SENT && inet_sk(sk)->defer_connect) {
590 		/* Active TCP fastopen socket with defer_connect
591 		 * Return EPOLLOUT so application can call write()
592 		 * in order for kernel to generate SYN+data
593 		 */
594 		mask |= EPOLLOUT | EPOLLWRNORM;
595 	}
596 	/* This barrier is coupled with smp_wmb() in tcp_reset() */
597 	smp_rmb();
598 	if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
599 		mask |= EPOLLERR;
600 
601 	return mask;
602 }
603 EXPORT_SYMBOL(tcp_poll);
604 
605 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
606 {
607 	struct tcp_sock *tp = tcp_sk(sk);
608 	int answ;
609 	bool slow;
610 
611 	switch (cmd) {
612 	case SIOCINQ:
613 		if (sk->sk_state == TCP_LISTEN)
614 			return -EINVAL;
615 
616 		slow = lock_sock_fast(sk);
617 		answ = tcp_inq(sk);
618 		unlock_sock_fast(sk, slow);
619 		break;
620 	case SIOCATMARK:
621 		answ = tp->urg_data && tp->urg_seq == tp->copied_seq;
622 		break;
623 	case SIOCOUTQ:
624 		if (sk->sk_state == TCP_LISTEN)
625 			return -EINVAL;
626 
627 		if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
628 			answ = 0;
629 		else
630 			answ = tp->write_seq - tp->snd_una;
631 		break;
632 	case SIOCOUTQNSD:
633 		if (sk->sk_state == TCP_LISTEN)
634 			return -EINVAL;
635 
636 		if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
637 			answ = 0;
638 		else
639 			answ = tp->write_seq - tp->snd_nxt;
640 		break;
641 	default:
642 		return -ENOIOCTLCMD;
643 	}
644 
645 	return put_user(answ, (int __user *)arg);
646 }
647 EXPORT_SYMBOL(tcp_ioctl);
648 
649 static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb)
650 {
651 	TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
652 	tp->pushed_seq = tp->write_seq;
653 }
654 
655 static inline bool forced_push(const struct tcp_sock *tp)
656 {
657 	return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1));
658 }
659 
660 static void skb_entail(struct sock *sk, struct sk_buff *skb)
661 {
662 	struct tcp_sock *tp = tcp_sk(sk);
663 	struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
664 
665 	skb->csum    = 0;
666 	tcb->seq     = tcb->end_seq = tp->write_seq;
667 	tcb->tcp_flags = TCPHDR_ACK;
668 	tcb->sacked  = 0;
669 	__skb_header_release(skb);
670 	tcp_add_write_queue_tail(sk, skb);
671 	sk->sk_wmem_queued += skb->truesize;
672 	sk_mem_charge(sk, skb->truesize);
673 	if (tp->nonagle & TCP_NAGLE_PUSH)
674 		tp->nonagle &= ~TCP_NAGLE_PUSH;
675 
676 	tcp_slow_start_after_idle_check(sk);
677 }
678 
679 static inline void tcp_mark_urg(struct tcp_sock *tp, int flags)
680 {
681 	if (flags & MSG_OOB)
682 		tp->snd_up = tp->write_seq;
683 }
684 
685 /* If a not yet filled skb is pushed, do not send it if
686  * we have data packets in Qdisc or NIC queues :
687  * Because TX completion will happen shortly, it gives a chance
688  * to coalesce future sendmsg() payload into this skb, without
689  * need for a timer, and with no latency trade off.
690  * As packets containing data payload have a bigger truesize
691  * than pure acks (dataless) packets, the last checks prevent
692  * autocorking if we only have an ACK in Qdisc/NIC queues,
693  * or if TX completion was delayed after we processed ACK packet.
694  */
695 static bool tcp_should_autocork(struct sock *sk, struct sk_buff *skb,
696 				int size_goal)
697 {
698 	return skb->len < size_goal &&
699 	       sock_net(sk)->ipv4.sysctl_tcp_autocorking &&
700 	       !tcp_rtx_queue_empty(sk) &&
701 	       refcount_read(&sk->sk_wmem_alloc) > skb->truesize;
702 }
703 
704 static void tcp_push(struct sock *sk, int flags, int mss_now,
705 		     int nonagle, int size_goal)
706 {
707 	struct tcp_sock *tp = tcp_sk(sk);
708 	struct sk_buff *skb;
709 
710 	skb = tcp_write_queue_tail(sk);
711 	if (!skb)
712 		return;
713 	if (!(flags & MSG_MORE) || forced_push(tp))
714 		tcp_mark_push(tp, skb);
715 
716 	tcp_mark_urg(tp, flags);
717 
718 	if (tcp_should_autocork(sk, skb, size_goal)) {
719 
720 		/* avoid atomic op if TSQ_THROTTLED bit is already set */
721 		if (!test_bit(TSQ_THROTTLED, &sk->sk_tsq_flags)) {
722 			NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAUTOCORKING);
723 			set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags);
724 		}
725 		/* It is possible TX completion already happened
726 		 * before we set TSQ_THROTTLED.
727 		 */
728 		if (refcount_read(&sk->sk_wmem_alloc) > skb->truesize)
729 			return;
730 	}
731 
732 	if (flags & MSG_MORE)
733 		nonagle = TCP_NAGLE_CORK;
734 
735 	__tcp_push_pending_frames(sk, mss_now, nonagle);
736 }
737 
738 static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
739 				unsigned int offset, size_t len)
740 {
741 	struct tcp_splice_state *tss = rd_desc->arg.data;
742 	int ret;
743 
744 	ret = skb_splice_bits(skb, skb->sk, offset, tss->pipe,
745 			      min(rd_desc->count, len), tss->flags);
746 	if (ret > 0)
747 		rd_desc->count -= ret;
748 	return ret;
749 }
750 
751 static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss)
752 {
753 	/* Store TCP splice context information in read_descriptor_t. */
754 	read_descriptor_t rd_desc = {
755 		.arg.data = tss,
756 		.count	  = tss->len,
757 	};
758 
759 	return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv);
760 }
761 
762 /**
763  *  tcp_splice_read - splice data from TCP socket to a pipe
764  * @sock:	socket to splice from
765  * @ppos:	position (not valid)
766  * @pipe:	pipe to splice to
767  * @len:	number of bytes to splice
768  * @flags:	splice modifier flags
769  *
770  * Description:
771  *    Will read pages from given socket and fill them into a pipe.
772  *
773  **/
774 ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos,
775 			struct pipe_inode_info *pipe, size_t len,
776 			unsigned int flags)
777 {
778 	struct sock *sk = sock->sk;
779 	struct tcp_splice_state tss = {
780 		.pipe = pipe,
781 		.len = len,
782 		.flags = flags,
783 	};
784 	long timeo;
785 	ssize_t spliced;
786 	int ret;
787 
788 	sock_rps_record_flow(sk);
789 	/*
790 	 * We can't seek on a socket input
791 	 */
792 	if (unlikely(*ppos))
793 		return -ESPIPE;
794 
795 	ret = spliced = 0;
796 
797 	lock_sock(sk);
798 
799 	timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK);
800 	while (tss.len) {
801 		ret = __tcp_splice_read(sk, &tss);
802 		if (ret < 0)
803 			break;
804 		else if (!ret) {
805 			if (spliced)
806 				break;
807 			if (sock_flag(sk, SOCK_DONE))
808 				break;
809 			if (sk->sk_err) {
810 				ret = sock_error(sk);
811 				break;
812 			}
813 			if (sk->sk_shutdown & RCV_SHUTDOWN)
814 				break;
815 			if (sk->sk_state == TCP_CLOSE) {
816 				/*
817 				 * This occurs when user tries to read
818 				 * from never connected socket.
819 				 */
820 				ret = -ENOTCONN;
821 				break;
822 			}
823 			if (!timeo) {
824 				ret = -EAGAIN;
825 				break;
826 			}
827 			/* if __tcp_splice_read() got nothing while we have
828 			 * an skb in receive queue, we do not want to loop.
829 			 * This might happen with URG data.
830 			 */
831 			if (!skb_queue_empty(&sk->sk_receive_queue))
832 				break;
833 			sk_wait_data(sk, &timeo, NULL);
834 			if (signal_pending(current)) {
835 				ret = sock_intr_errno(timeo);
836 				break;
837 			}
838 			continue;
839 		}
840 		tss.len -= ret;
841 		spliced += ret;
842 
843 		if (!timeo)
844 			break;
845 		release_sock(sk);
846 		lock_sock(sk);
847 
848 		if (sk->sk_err || sk->sk_state == TCP_CLOSE ||
849 		    (sk->sk_shutdown & RCV_SHUTDOWN) ||
850 		    signal_pending(current))
851 			break;
852 	}
853 
854 	release_sock(sk);
855 
856 	if (spliced)
857 		return spliced;
858 
859 	return ret;
860 }
861 EXPORT_SYMBOL(tcp_splice_read);
862 
863 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
864 				    bool force_schedule)
865 {
866 	struct sk_buff *skb;
867 
868 	/* The TCP header must be at least 32-bit aligned.  */
869 	size = ALIGN(size, 4);
870 
871 	if (unlikely(tcp_under_memory_pressure(sk)))
872 		sk_mem_reclaim_partial(sk);
873 
874 	skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp);
875 	if (likely(skb)) {
876 		bool mem_scheduled;
877 
878 		if (force_schedule) {
879 			mem_scheduled = true;
880 			sk_forced_mem_schedule(sk, skb->truesize);
881 		} else {
882 			mem_scheduled = sk_wmem_schedule(sk, skb->truesize);
883 		}
884 		if (likely(mem_scheduled)) {
885 			skb_reserve(skb, sk->sk_prot->max_header);
886 			/*
887 			 * Make sure that we have exactly size bytes
888 			 * available to the caller, no more, no less.
889 			 */
890 			skb->reserved_tailroom = skb->end - skb->tail - size;
891 			INIT_LIST_HEAD(&skb->tcp_tsorted_anchor);
892 			return skb;
893 		}
894 		__kfree_skb(skb);
895 	} else {
896 		sk->sk_prot->enter_memory_pressure(sk);
897 		sk_stream_moderate_sndbuf(sk);
898 	}
899 	return NULL;
900 }
901 
902 static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now,
903 				       int large_allowed)
904 {
905 	struct tcp_sock *tp = tcp_sk(sk);
906 	u32 new_size_goal, size_goal;
907 
908 	if (!large_allowed)
909 		return mss_now;
910 
911 	/* Note : tcp_tso_autosize() will eventually split this later */
912 	new_size_goal = sk->sk_gso_max_size - 1 - MAX_TCP_HEADER;
913 	new_size_goal = tcp_bound_to_half_wnd(tp, new_size_goal);
914 
915 	/* We try hard to avoid divides here */
916 	size_goal = tp->gso_segs * mss_now;
917 	if (unlikely(new_size_goal < size_goal ||
918 		     new_size_goal >= size_goal + mss_now)) {
919 		tp->gso_segs = min_t(u16, new_size_goal / mss_now,
920 				     sk->sk_gso_max_segs);
921 		size_goal = tp->gso_segs * mss_now;
922 	}
923 
924 	return max(size_goal, mss_now);
925 }
926 
927 static int tcp_send_mss(struct sock *sk, int *size_goal, int flags)
928 {
929 	int mss_now;
930 
931 	mss_now = tcp_current_mss(sk);
932 	*size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB));
933 
934 	return mss_now;
935 }
936 
937 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
938 			 size_t size, int flags)
939 {
940 	struct tcp_sock *tp = tcp_sk(sk);
941 	int mss_now, size_goal;
942 	int err;
943 	ssize_t copied;
944 	long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
945 
946 	/* Wait for a connection to finish. One exception is TCP Fast Open
947 	 * (passive side) where data is allowed to be sent before a connection
948 	 * is fully established.
949 	 */
950 	if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
951 	    !tcp_passive_fastopen(sk)) {
952 		err = sk_stream_wait_connect(sk, &timeo);
953 		if (err != 0)
954 			goto out_err;
955 	}
956 
957 	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
958 
959 	mss_now = tcp_send_mss(sk, &size_goal, flags);
960 	copied = 0;
961 
962 	err = -EPIPE;
963 	if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
964 		goto out_err;
965 
966 	while (size > 0) {
967 		struct sk_buff *skb = tcp_write_queue_tail(sk);
968 		int copy, i;
969 		bool can_coalesce;
970 
971 		if (!skb || (copy = size_goal - skb->len) <= 0 ||
972 		    !tcp_skb_can_collapse_to(skb)) {
973 new_segment:
974 			if (!sk_stream_memory_free(sk))
975 				goto wait_for_sndbuf;
976 
977 			skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation,
978 					tcp_rtx_and_write_queues_empty(sk));
979 			if (!skb)
980 				goto wait_for_memory;
981 
982 			skb_entail(sk, skb);
983 			copy = size_goal;
984 		}
985 
986 		if (copy > size)
987 			copy = size;
988 
989 		i = skb_shinfo(skb)->nr_frags;
990 		can_coalesce = skb_can_coalesce(skb, i, page, offset);
991 		if (!can_coalesce && i >= sysctl_max_skb_frags) {
992 			tcp_mark_push(tp, skb);
993 			goto new_segment;
994 		}
995 		if (!sk_wmem_schedule(sk, copy))
996 			goto wait_for_memory;
997 
998 		if (can_coalesce) {
999 			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1000 		} else {
1001 			get_page(page);
1002 			skb_fill_page_desc(skb, i, page, offset, copy);
1003 		}
1004 
1005 		if (!(flags & MSG_NO_SHARED_FRAGS))
1006 			skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1007 
1008 		skb->len += copy;
1009 		skb->data_len += copy;
1010 		skb->truesize += copy;
1011 		sk->sk_wmem_queued += copy;
1012 		sk_mem_charge(sk, copy);
1013 		skb->ip_summed = CHECKSUM_PARTIAL;
1014 		tp->write_seq += copy;
1015 		TCP_SKB_CB(skb)->end_seq += copy;
1016 		tcp_skb_pcount_set(skb, 0);
1017 
1018 		if (!copied)
1019 			TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1020 
1021 		copied += copy;
1022 		offset += copy;
1023 		size -= copy;
1024 		if (!size)
1025 			goto out;
1026 
1027 		if (skb->len < size_goal || (flags & MSG_OOB))
1028 			continue;
1029 
1030 		if (forced_push(tp)) {
1031 			tcp_mark_push(tp, skb);
1032 			__tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
1033 		} else if (skb == tcp_send_head(sk))
1034 			tcp_push_one(sk, mss_now);
1035 		continue;
1036 
1037 wait_for_sndbuf:
1038 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1039 wait_for_memory:
1040 		tcp_push(sk, flags & ~MSG_MORE, mss_now,
1041 			 TCP_NAGLE_PUSH, size_goal);
1042 
1043 		err = sk_stream_wait_memory(sk, &timeo);
1044 		if (err != 0)
1045 			goto do_error;
1046 
1047 		mss_now = tcp_send_mss(sk, &size_goal, flags);
1048 	}
1049 
1050 out:
1051 	if (copied) {
1052 		tcp_tx_timestamp(sk, sk->sk_tsflags);
1053 		if (!(flags & MSG_SENDPAGE_NOTLAST))
1054 			tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
1055 	}
1056 	return copied;
1057 
1058 do_error:
1059 	if (copied)
1060 		goto out;
1061 out_err:
1062 	/* make sure we wake any epoll edge trigger waiter */
1063 	if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 &&
1064 		     err == -EAGAIN)) {
1065 		sk->sk_write_space(sk);
1066 		tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED);
1067 	}
1068 	return sk_stream_error(sk, flags, err);
1069 }
1070 EXPORT_SYMBOL_GPL(do_tcp_sendpages);
1071 
1072 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
1073 			size_t size, int flags)
1074 {
1075 	if (!(sk->sk_route_caps & NETIF_F_SG))
1076 		return sock_no_sendpage_locked(sk, page, offset, size, flags);
1077 
1078 	tcp_rate_check_app_limited(sk);  /* is sending application-limited? */
1079 
1080 	return do_tcp_sendpages(sk, page, offset, size, flags);
1081 }
1082 EXPORT_SYMBOL_GPL(tcp_sendpage_locked);
1083 
1084 int tcp_sendpage(struct sock *sk, struct page *page, int offset,
1085 		 size_t size, int flags)
1086 {
1087 	int ret;
1088 
1089 	lock_sock(sk);
1090 	ret = tcp_sendpage_locked(sk, page, offset, size, flags);
1091 	release_sock(sk);
1092 
1093 	return ret;
1094 }
1095 EXPORT_SYMBOL(tcp_sendpage);
1096 
1097 /* Do not bother using a page frag for very small frames.
1098  * But use this heuristic only for the first skb in write queue.
1099  *
1100  * Having no payload in skb->head allows better SACK shifting
1101  * in tcp_shift_skb_data(), reducing sack/rack overhead, because
1102  * write queue has less skbs.
1103  * Each skb can hold up to MAX_SKB_FRAGS * 32Kbytes, or ~0.5 MB.
1104  * This also speeds up tso_fragment(), since it wont fallback
1105  * to tcp_fragment().
1106  */
1107 static int linear_payload_sz(bool first_skb)
1108 {
1109 	if (first_skb)
1110 		return SKB_WITH_OVERHEAD(2048 - MAX_TCP_HEADER);
1111 	return 0;
1112 }
1113 
1114 static int select_size(bool first_skb, bool zc)
1115 {
1116 	if (zc)
1117 		return 0;
1118 	return linear_payload_sz(first_skb);
1119 }
1120 
1121 void tcp_free_fastopen_req(struct tcp_sock *tp)
1122 {
1123 	if (tp->fastopen_req) {
1124 		kfree(tp->fastopen_req);
1125 		tp->fastopen_req = NULL;
1126 	}
1127 }
1128 
1129 static int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg,
1130 				int *copied, size_t size,
1131 				struct ubuf_info *uarg)
1132 {
1133 	struct tcp_sock *tp = tcp_sk(sk);
1134 	struct inet_sock *inet = inet_sk(sk);
1135 	struct sockaddr *uaddr = msg->msg_name;
1136 	int err, flags;
1137 
1138 	if (!(sock_net(sk)->ipv4.sysctl_tcp_fastopen & TFO_CLIENT_ENABLE) ||
1139 	    (uaddr && msg->msg_namelen >= sizeof(uaddr->sa_family) &&
1140 	     uaddr->sa_family == AF_UNSPEC))
1141 		return -EOPNOTSUPP;
1142 	if (tp->fastopen_req)
1143 		return -EALREADY; /* Another Fast Open is in progress */
1144 
1145 	tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request),
1146 				   sk->sk_allocation);
1147 	if (unlikely(!tp->fastopen_req))
1148 		return -ENOBUFS;
1149 	tp->fastopen_req->data = msg;
1150 	tp->fastopen_req->size = size;
1151 	tp->fastopen_req->uarg = uarg;
1152 
1153 	if (inet->defer_connect) {
1154 		err = tcp_connect(sk);
1155 		/* Same failure procedure as in tcp_v4/6_connect */
1156 		if (err) {
1157 			tcp_set_state(sk, TCP_CLOSE);
1158 			inet->inet_dport = 0;
1159 			sk->sk_route_caps = 0;
1160 		}
1161 	}
1162 	flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0;
1163 	err = __inet_stream_connect(sk->sk_socket, uaddr,
1164 				    msg->msg_namelen, flags, 1);
1165 	/* fastopen_req could already be freed in __inet_stream_connect
1166 	 * if the connection times out or gets rst
1167 	 */
1168 	if (tp->fastopen_req) {
1169 		*copied = tp->fastopen_req->copied;
1170 		tcp_free_fastopen_req(tp);
1171 		inet->defer_connect = 0;
1172 	}
1173 	return err;
1174 }
1175 
1176 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
1177 {
1178 	struct tcp_sock *tp = tcp_sk(sk);
1179 	struct ubuf_info *uarg = NULL;
1180 	struct sk_buff *skb;
1181 	struct sockcm_cookie sockc;
1182 	int flags, err, copied = 0;
1183 	int mss_now = 0, size_goal, copied_syn = 0;
1184 	bool process_backlog = false;
1185 	bool zc = false;
1186 	long timeo;
1187 
1188 	flags = msg->msg_flags;
1189 
1190 	if (flags & MSG_ZEROCOPY && size && sock_flag(sk, SOCK_ZEROCOPY)) {
1191 		skb = tcp_write_queue_tail(sk);
1192 		uarg = sock_zerocopy_realloc(sk, size, skb_zcopy(skb));
1193 		if (!uarg) {
1194 			err = -ENOBUFS;
1195 			goto out_err;
1196 		}
1197 
1198 		zc = sk->sk_route_caps & NETIF_F_SG;
1199 		if (!zc)
1200 			uarg->zerocopy = 0;
1201 	}
1202 
1203 	if (unlikely(flags & MSG_FASTOPEN || inet_sk(sk)->defer_connect) &&
1204 	    !tp->repair) {
1205 		err = tcp_sendmsg_fastopen(sk, msg, &copied_syn, size, uarg);
1206 		if (err == -EINPROGRESS && copied_syn > 0)
1207 			goto out;
1208 		else if (err)
1209 			goto out_err;
1210 	}
1211 
1212 	timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1213 
1214 	tcp_rate_check_app_limited(sk);  /* is sending application-limited? */
1215 
1216 	/* Wait for a connection to finish. One exception is TCP Fast Open
1217 	 * (passive side) where data is allowed to be sent before a connection
1218 	 * is fully established.
1219 	 */
1220 	if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
1221 	    !tcp_passive_fastopen(sk)) {
1222 		err = sk_stream_wait_connect(sk, &timeo);
1223 		if (err != 0)
1224 			goto do_error;
1225 	}
1226 
1227 	if (unlikely(tp->repair)) {
1228 		if (tp->repair_queue == TCP_RECV_QUEUE) {
1229 			copied = tcp_send_rcvq(sk, msg, size);
1230 			goto out_nopush;
1231 		}
1232 
1233 		err = -EINVAL;
1234 		if (tp->repair_queue == TCP_NO_QUEUE)
1235 			goto out_err;
1236 
1237 		/* 'common' sending to sendq */
1238 	}
1239 
1240 	sockcm_init(&sockc, sk);
1241 	if (msg->msg_controllen) {
1242 		err = sock_cmsg_send(sk, msg, &sockc);
1243 		if (unlikely(err)) {
1244 			err = -EINVAL;
1245 			goto out_err;
1246 		}
1247 	}
1248 
1249 	/* This should be in poll */
1250 	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1251 
1252 	/* Ok commence sending. */
1253 	copied = 0;
1254 
1255 restart:
1256 	mss_now = tcp_send_mss(sk, &size_goal, flags);
1257 
1258 	err = -EPIPE;
1259 	if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
1260 		goto do_error;
1261 
1262 	while (msg_data_left(msg)) {
1263 		int copy = 0;
1264 
1265 		skb = tcp_write_queue_tail(sk);
1266 		if (skb)
1267 			copy = size_goal - skb->len;
1268 
1269 		if (copy <= 0 || !tcp_skb_can_collapse_to(skb)) {
1270 			bool first_skb;
1271 			int linear;
1272 
1273 new_segment:
1274 			if (!sk_stream_memory_free(sk))
1275 				goto wait_for_sndbuf;
1276 
1277 			if (process_backlog && sk_flush_backlog(sk)) {
1278 				process_backlog = false;
1279 				goto restart;
1280 			}
1281 			first_skb = tcp_rtx_and_write_queues_empty(sk);
1282 			linear = select_size(first_skb, zc);
1283 			skb = sk_stream_alloc_skb(sk, linear, sk->sk_allocation,
1284 						  first_skb);
1285 			if (!skb)
1286 				goto wait_for_memory;
1287 
1288 			process_backlog = true;
1289 			skb->ip_summed = CHECKSUM_PARTIAL;
1290 
1291 			skb_entail(sk, skb);
1292 			copy = size_goal;
1293 
1294 			/* All packets are restored as if they have
1295 			 * already been sent. skb_mstamp_ns isn't set to
1296 			 * avoid wrong rtt estimation.
1297 			 */
1298 			if (tp->repair)
1299 				TCP_SKB_CB(skb)->sacked |= TCPCB_REPAIRED;
1300 		}
1301 
1302 		/* Try to append data to the end of skb. */
1303 		if (copy > msg_data_left(msg))
1304 			copy = msg_data_left(msg);
1305 
1306 		/* Where to copy to? */
1307 		if (skb_availroom(skb) > 0 && !zc) {
1308 			/* We have some space in skb head. Superb! */
1309 			copy = min_t(int, copy, skb_availroom(skb));
1310 			err = skb_add_data_nocache(sk, skb, &msg->msg_iter, copy);
1311 			if (err)
1312 				goto do_fault;
1313 		} else if (!zc) {
1314 			bool merge = true;
1315 			int i = skb_shinfo(skb)->nr_frags;
1316 			struct page_frag *pfrag = sk_page_frag(sk);
1317 
1318 			if (!sk_page_frag_refill(sk, pfrag))
1319 				goto wait_for_memory;
1320 
1321 			if (!skb_can_coalesce(skb, i, pfrag->page,
1322 					      pfrag->offset)) {
1323 				if (i >= sysctl_max_skb_frags) {
1324 					tcp_mark_push(tp, skb);
1325 					goto new_segment;
1326 				}
1327 				merge = false;
1328 			}
1329 
1330 			copy = min_t(int, copy, pfrag->size - pfrag->offset);
1331 
1332 			if (!sk_wmem_schedule(sk, copy))
1333 				goto wait_for_memory;
1334 
1335 			err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
1336 						       pfrag->page,
1337 						       pfrag->offset,
1338 						       copy);
1339 			if (err)
1340 				goto do_error;
1341 
1342 			/* Update the skb. */
1343 			if (merge) {
1344 				skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1345 			} else {
1346 				skb_fill_page_desc(skb, i, pfrag->page,
1347 						   pfrag->offset, copy);
1348 				page_ref_inc(pfrag->page);
1349 			}
1350 			pfrag->offset += copy;
1351 		} else {
1352 			err = skb_zerocopy_iter_stream(sk, skb, msg, copy, uarg);
1353 			if (err == -EMSGSIZE || err == -EEXIST) {
1354 				tcp_mark_push(tp, skb);
1355 				goto new_segment;
1356 			}
1357 			if (err < 0)
1358 				goto do_error;
1359 			copy = err;
1360 		}
1361 
1362 		if (!copied)
1363 			TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1364 
1365 		tp->write_seq += copy;
1366 		TCP_SKB_CB(skb)->end_seq += copy;
1367 		tcp_skb_pcount_set(skb, 0);
1368 
1369 		copied += copy;
1370 		if (!msg_data_left(msg)) {
1371 			if (unlikely(flags & MSG_EOR))
1372 				TCP_SKB_CB(skb)->eor = 1;
1373 			goto out;
1374 		}
1375 
1376 		if (skb->len < size_goal || (flags & MSG_OOB) || unlikely(tp->repair))
1377 			continue;
1378 
1379 		if (forced_push(tp)) {
1380 			tcp_mark_push(tp, skb);
1381 			__tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
1382 		} else if (skb == tcp_send_head(sk))
1383 			tcp_push_one(sk, mss_now);
1384 		continue;
1385 
1386 wait_for_sndbuf:
1387 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1388 wait_for_memory:
1389 		if (copied)
1390 			tcp_push(sk, flags & ~MSG_MORE, mss_now,
1391 				 TCP_NAGLE_PUSH, size_goal);
1392 
1393 		err = sk_stream_wait_memory(sk, &timeo);
1394 		if (err != 0)
1395 			goto do_error;
1396 
1397 		mss_now = tcp_send_mss(sk, &size_goal, flags);
1398 	}
1399 
1400 out:
1401 	if (copied) {
1402 		tcp_tx_timestamp(sk, sockc.tsflags);
1403 		tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
1404 	}
1405 out_nopush:
1406 	sock_zerocopy_put(uarg);
1407 	return copied + copied_syn;
1408 
1409 do_fault:
1410 	if (!skb->len) {
1411 		tcp_unlink_write_queue(skb, sk);
1412 		/* It is the one place in all of TCP, except connection
1413 		 * reset, where we can be unlinking the send_head.
1414 		 */
1415 		tcp_check_send_head(sk, skb);
1416 		sk_wmem_free_skb(sk, skb);
1417 	}
1418 
1419 do_error:
1420 	if (copied + copied_syn)
1421 		goto out;
1422 out_err:
1423 	sock_zerocopy_put_abort(uarg, true);
1424 	err = sk_stream_error(sk, flags, err);
1425 	/* make sure we wake any epoll edge trigger waiter */
1426 	if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 &&
1427 		     err == -EAGAIN)) {
1428 		sk->sk_write_space(sk);
1429 		tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED);
1430 	}
1431 	return err;
1432 }
1433 EXPORT_SYMBOL_GPL(tcp_sendmsg_locked);
1434 
1435 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
1436 {
1437 	int ret;
1438 
1439 	lock_sock(sk);
1440 	ret = tcp_sendmsg_locked(sk, msg, size);
1441 	release_sock(sk);
1442 
1443 	return ret;
1444 }
1445 EXPORT_SYMBOL(tcp_sendmsg);
1446 
1447 /*
1448  *	Handle reading urgent data. BSD has very simple semantics for
1449  *	this, no blocking and very strange errors 8)
1450  */
1451 
1452 static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags)
1453 {
1454 	struct tcp_sock *tp = tcp_sk(sk);
1455 
1456 	/* No URG data to read. */
1457 	if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data ||
1458 	    tp->urg_data == TCP_URG_READ)
1459 		return -EINVAL;	/* Yes this is right ! */
1460 
1461 	if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE))
1462 		return -ENOTCONN;
1463 
1464 	if (tp->urg_data & TCP_URG_VALID) {
1465 		int err = 0;
1466 		char c = tp->urg_data;
1467 
1468 		if (!(flags & MSG_PEEK))
1469 			tp->urg_data = TCP_URG_READ;
1470 
1471 		/* Read urgent data. */
1472 		msg->msg_flags |= MSG_OOB;
1473 
1474 		if (len > 0) {
1475 			if (!(flags & MSG_TRUNC))
1476 				err = memcpy_to_msg(msg, &c, 1);
1477 			len = 1;
1478 		} else
1479 			msg->msg_flags |= MSG_TRUNC;
1480 
1481 		return err ? -EFAULT : len;
1482 	}
1483 
1484 	if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
1485 		return 0;
1486 
1487 	/* Fixed the recv(..., MSG_OOB) behaviour.  BSD docs and
1488 	 * the available implementations agree in this case:
1489 	 * this call should never block, independent of the
1490 	 * blocking state of the socket.
1491 	 * Mike <pall@rz.uni-karlsruhe.de>
1492 	 */
1493 	return -EAGAIN;
1494 }
1495 
1496 static int tcp_peek_sndq(struct sock *sk, struct msghdr *msg, int len)
1497 {
1498 	struct sk_buff *skb;
1499 	int copied = 0, err = 0;
1500 
1501 	/* XXX -- need to support SO_PEEK_OFF */
1502 
1503 	skb_rbtree_walk(skb, &sk->tcp_rtx_queue) {
1504 		err = skb_copy_datagram_msg(skb, 0, msg, skb->len);
1505 		if (err)
1506 			return err;
1507 		copied += skb->len;
1508 	}
1509 
1510 	skb_queue_walk(&sk->sk_write_queue, skb) {
1511 		err = skb_copy_datagram_msg(skb, 0, msg, skb->len);
1512 		if (err)
1513 			break;
1514 
1515 		copied += skb->len;
1516 	}
1517 
1518 	return err ?: copied;
1519 }
1520 
1521 /* Clean up the receive buffer for full frames taken by the user,
1522  * then send an ACK if necessary.  COPIED is the number of bytes
1523  * tcp_recvmsg has given to the user so far, it speeds up the
1524  * calculation of whether or not we must ACK for the sake of
1525  * a window update.
1526  */
1527 static void tcp_cleanup_rbuf(struct sock *sk, int copied)
1528 {
1529 	struct tcp_sock *tp = tcp_sk(sk);
1530 	bool time_to_ack = false;
1531 
1532 	struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
1533 
1534 	WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq),
1535 	     "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n",
1536 	     tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt);
1537 
1538 	if (inet_csk_ack_scheduled(sk)) {
1539 		const struct inet_connection_sock *icsk = inet_csk(sk);
1540 		   /* Delayed ACKs frequently hit locked sockets during bulk
1541 		    * receive. */
1542 		if (icsk->icsk_ack.blocked ||
1543 		    /* Once-per-two-segments ACK was not sent by tcp_input.c */
1544 		    tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss ||
1545 		    /*
1546 		     * If this read emptied read buffer, we send ACK, if
1547 		     * connection is not bidirectional, user drained
1548 		     * receive buffer and there was a small segment
1549 		     * in queue.
1550 		     */
1551 		    (copied > 0 &&
1552 		     ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) ||
1553 		      ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) &&
1554 		       !inet_csk_in_pingpong_mode(sk))) &&
1555 		      !atomic_read(&sk->sk_rmem_alloc)))
1556 			time_to_ack = true;
1557 	}
1558 
1559 	/* We send an ACK if we can now advertise a non-zero window
1560 	 * which has been raised "significantly".
1561 	 *
1562 	 * Even if window raised up to infinity, do not send window open ACK
1563 	 * in states, where we will not receive more. It is useless.
1564 	 */
1565 	if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1566 		__u32 rcv_window_now = tcp_receive_window(tp);
1567 
1568 		/* Optimize, __tcp_select_window() is not cheap. */
1569 		if (2*rcv_window_now <= tp->window_clamp) {
1570 			__u32 new_window = __tcp_select_window(sk);
1571 
1572 			/* Send ACK now, if this read freed lots of space
1573 			 * in our buffer. Certainly, new_window is new window.
1574 			 * We can advertise it now, if it is not less than current one.
1575 			 * "Lots" means "at least twice" here.
1576 			 */
1577 			if (new_window && new_window >= 2 * rcv_window_now)
1578 				time_to_ack = true;
1579 		}
1580 	}
1581 	if (time_to_ack)
1582 		tcp_send_ack(sk);
1583 }
1584 
1585 static struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off)
1586 {
1587 	struct sk_buff *skb;
1588 	u32 offset;
1589 
1590 	while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) {
1591 		offset = seq - TCP_SKB_CB(skb)->seq;
1592 		if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
1593 			pr_err_once("%s: found a SYN, please report !\n", __func__);
1594 			offset--;
1595 		}
1596 		if (offset < skb->len || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) {
1597 			*off = offset;
1598 			return skb;
1599 		}
1600 		/* This looks weird, but this can happen if TCP collapsing
1601 		 * splitted a fat GRO packet, while we released socket lock
1602 		 * in skb_splice_bits()
1603 		 */
1604 		sk_eat_skb(sk, skb);
1605 	}
1606 	return NULL;
1607 }
1608 
1609 /*
1610  * This routine provides an alternative to tcp_recvmsg() for routines
1611  * that would like to handle copying from skbuffs directly in 'sendfile'
1612  * fashion.
1613  * Note:
1614  *	- It is assumed that the socket was locked by the caller.
1615  *	- The routine does not block.
1616  *	- At present, there is no support for reading OOB data
1617  *	  or for 'peeking' the socket using this routine
1618  *	  (although both would be easy to implement).
1619  */
1620 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
1621 		  sk_read_actor_t recv_actor)
1622 {
1623 	struct sk_buff *skb;
1624 	struct tcp_sock *tp = tcp_sk(sk);
1625 	u32 seq = tp->copied_seq;
1626 	u32 offset;
1627 	int copied = 0;
1628 
1629 	if (sk->sk_state == TCP_LISTEN)
1630 		return -ENOTCONN;
1631 	while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) {
1632 		if (offset < skb->len) {
1633 			int used;
1634 			size_t len;
1635 
1636 			len = skb->len - offset;
1637 			/* Stop reading if we hit a patch of urgent data */
1638 			if (tp->urg_data) {
1639 				u32 urg_offset = tp->urg_seq - seq;
1640 				if (urg_offset < len)
1641 					len = urg_offset;
1642 				if (!len)
1643 					break;
1644 			}
1645 			used = recv_actor(desc, skb, offset, len);
1646 			if (used <= 0) {
1647 				if (!copied)
1648 					copied = used;
1649 				break;
1650 			} else if (used <= len) {
1651 				seq += used;
1652 				copied += used;
1653 				offset += used;
1654 			}
1655 			/* If recv_actor drops the lock (e.g. TCP splice
1656 			 * receive) the skb pointer might be invalid when
1657 			 * getting here: tcp_collapse might have deleted it
1658 			 * while aggregating skbs from the socket queue.
1659 			 */
1660 			skb = tcp_recv_skb(sk, seq - 1, &offset);
1661 			if (!skb)
1662 				break;
1663 			/* TCP coalescing might have appended data to the skb.
1664 			 * Try to splice more frags
1665 			 */
1666 			if (offset + 1 != skb->len)
1667 				continue;
1668 		}
1669 		if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) {
1670 			sk_eat_skb(sk, skb);
1671 			++seq;
1672 			break;
1673 		}
1674 		sk_eat_skb(sk, skb);
1675 		if (!desc->count)
1676 			break;
1677 		tp->copied_seq = seq;
1678 	}
1679 	tp->copied_seq = seq;
1680 
1681 	tcp_rcv_space_adjust(sk);
1682 
1683 	/* Clean up data we have read: This will do ACK frames. */
1684 	if (copied > 0) {
1685 		tcp_recv_skb(sk, seq, &offset);
1686 		tcp_cleanup_rbuf(sk, copied);
1687 	}
1688 	return copied;
1689 }
1690 EXPORT_SYMBOL(tcp_read_sock);
1691 
1692 int tcp_peek_len(struct socket *sock)
1693 {
1694 	return tcp_inq(sock->sk);
1695 }
1696 EXPORT_SYMBOL(tcp_peek_len);
1697 
1698 /* Make sure sk_rcvbuf is big enough to satisfy SO_RCVLOWAT hint */
1699 int tcp_set_rcvlowat(struct sock *sk, int val)
1700 {
1701 	int cap;
1702 
1703 	if (sk->sk_userlocks & SOCK_RCVBUF_LOCK)
1704 		cap = sk->sk_rcvbuf >> 1;
1705 	else
1706 		cap = sock_net(sk)->ipv4.sysctl_tcp_rmem[2] >> 1;
1707 	val = min(val, cap);
1708 	sk->sk_rcvlowat = val ? : 1;
1709 
1710 	/* Check if we need to signal EPOLLIN right now */
1711 	tcp_data_ready(sk);
1712 
1713 	if (sk->sk_userlocks & SOCK_RCVBUF_LOCK)
1714 		return 0;
1715 
1716 	val <<= 1;
1717 	if (val > sk->sk_rcvbuf) {
1718 		sk->sk_rcvbuf = val;
1719 		tcp_sk(sk)->window_clamp = tcp_win_from_space(sk, val);
1720 	}
1721 	return 0;
1722 }
1723 EXPORT_SYMBOL(tcp_set_rcvlowat);
1724 
1725 #ifdef CONFIG_MMU
1726 static const struct vm_operations_struct tcp_vm_ops = {
1727 };
1728 
1729 int tcp_mmap(struct file *file, struct socket *sock,
1730 	     struct vm_area_struct *vma)
1731 {
1732 	if (vma->vm_flags & (VM_WRITE | VM_EXEC))
1733 		return -EPERM;
1734 	vma->vm_flags &= ~(VM_MAYWRITE | VM_MAYEXEC);
1735 
1736 	/* Instruct vm_insert_page() to not down_read(mmap_sem) */
1737 	vma->vm_flags |= VM_MIXEDMAP;
1738 
1739 	vma->vm_ops = &tcp_vm_ops;
1740 	return 0;
1741 }
1742 EXPORT_SYMBOL(tcp_mmap);
1743 
1744 static int tcp_zerocopy_receive(struct sock *sk,
1745 				struct tcp_zerocopy_receive *zc)
1746 {
1747 	unsigned long address = (unsigned long)zc->address;
1748 	const skb_frag_t *frags = NULL;
1749 	u32 length = 0, seq, offset;
1750 	struct vm_area_struct *vma;
1751 	struct sk_buff *skb = NULL;
1752 	struct tcp_sock *tp;
1753 	int inq;
1754 	int ret;
1755 
1756 	if (address & (PAGE_SIZE - 1) || address != zc->address)
1757 		return -EINVAL;
1758 
1759 	if (sk->sk_state == TCP_LISTEN)
1760 		return -ENOTCONN;
1761 
1762 	sock_rps_record_flow(sk);
1763 
1764 	down_read(&current->mm->mmap_sem);
1765 
1766 	ret = -EINVAL;
1767 	vma = find_vma(current->mm, address);
1768 	if (!vma || vma->vm_start > address || vma->vm_ops != &tcp_vm_ops)
1769 		goto out;
1770 	zc->length = min_t(unsigned long, zc->length, vma->vm_end - address);
1771 
1772 	tp = tcp_sk(sk);
1773 	seq = tp->copied_seq;
1774 	inq = tcp_inq(sk);
1775 	zc->length = min_t(u32, zc->length, inq);
1776 	zc->length &= ~(PAGE_SIZE - 1);
1777 	if (zc->length) {
1778 		zap_page_range(vma, address, zc->length);
1779 		zc->recv_skip_hint = 0;
1780 	} else {
1781 		zc->recv_skip_hint = inq;
1782 	}
1783 	ret = 0;
1784 	while (length + PAGE_SIZE <= zc->length) {
1785 		if (zc->recv_skip_hint < PAGE_SIZE) {
1786 			if (skb) {
1787 				skb = skb->next;
1788 				offset = seq - TCP_SKB_CB(skb)->seq;
1789 			} else {
1790 				skb = tcp_recv_skb(sk, seq, &offset);
1791 			}
1792 
1793 			zc->recv_skip_hint = skb->len - offset;
1794 			offset -= skb_headlen(skb);
1795 			if ((int)offset < 0 || skb_has_frag_list(skb))
1796 				break;
1797 			frags = skb_shinfo(skb)->frags;
1798 			while (offset) {
1799 				if (frags->size > offset)
1800 					goto out;
1801 				offset -= frags->size;
1802 				frags++;
1803 			}
1804 		}
1805 		if (frags->size != PAGE_SIZE || frags->page_offset) {
1806 			int remaining = zc->recv_skip_hint;
1807 
1808 			while (remaining && (frags->size != PAGE_SIZE ||
1809 					     frags->page_offset)) {
1810 				remaining -= frags->size;
1811 				frags++;
1812 			}
1813 			zc->recv_skip_hint -= remaining;
1814 			break;
1815 		}
1816 		ret = vm_insert_page(vma, address + length,
1817 				     skb_frag_page(frags));
1818 		if (ret)
1819 			break;
1820 		length += PAGE_SIZE;
1821 		seq += PAGE_SIZE;
1822 		zc->recv_skip_hint -= PAGE_SIZE;
1823 		frags++;
1824 	}
1825 out:
1826 	up_read(&current->mm->mmap_sem);
1827 	if (length) {
1828 		tp->copied_seq = seq;
1829 		tcp_rcv_space_adjust(sk);
1830 
1831 		/* Clean up data we have read: This will do ACK frames. */
1832 		tcp_recv_skb(sk, seq, &offset);
1833 		tcp_cleanup_rbuf(sk, length);
1834 		ret = 0;
1835 		if (length == zc->length)
1836 			zc->recv_skip_hint = 0;
1837 	} else {
1838 		if (!zc->recv_skip_hint && sock_flag(sk, SOCK_DONE))
1839 			ret = -EIO;
1840 	}
1841 	zc->length = length;
1842 	return ret;
1843 }
1844 #endif
1845 
1846 static void tcp_update_recv_tstamps(struct sk_buff *skb,
1847 				    struct scm_timestamping *tss)
1848 {
1849 	if (skb->tstamp)
1850 		tss->ts[0] = ktime_to_timespec(skb->tstamp);
1851 	else
1852 		tss->ts[0] = (struct timespec) {0};
1853 
1854 	if (skb_hwtstamps(skb)->hwtstamp)
1855 		tss->ts[2] = ktime_to_timespec(skb_hwtstamps(skb)->hwtstamp);
1856 	else
1857 		tss->ts[2] = (struct timespec) {0};
1858 }
1859 
1860 /* Similar to __sock_recv_timestamp, but does not require an skb */
1861 static void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk,
1862 			       struct scm_timestamping *tss)
1863 {
1864 	struct timeval tv;
1865 	bool has_timestamping = false;
1866 
1867 	if (tss->ts[0].tv_sec || tss->ts[0].tv_nsec) {
1868 		if (sock_flag(sk, SOCK_RCVTSTAMP)) {
1869 			if (sock_flag(sk, SOCK_RCVTSTAMPNS)) {
1870 				put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPNS,
1871 					 sizeof(tss->ts[0]), &tss->ts[0]);
1872 			} else {
1873 				tv.tv_sec = tss->ts[0].tv_sec;
1874 				tv.tv_usec = tss->ts[0].tv_nsec / 1000;
1875 
1876 				put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP,
1877 					 sizeof(tv), &tv);
1878 			}
1879 		}
1880 
1881 		if (sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE)
1882 			has_timestamping = true;
1883 		else
1884 			tss->ts[0] = (struct timespec) {0};
1885 	}
1886 
1887 	if (tss->ts[2].tv_sec || tss->ts[2].tv_nsec) {
1888 		if (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)
1889 			has_timestamping = true;
1890 		else
1891 			tss->ts[2] = (struct timespec) {0};
1892 	}
1893 
1894 	if (has_timestamping) {
1895 		tss->ts[1] = (struct timespec) {0};
1896 		put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING,
1897 			 sizeof(*tss), tss);
1898 	}
1899 }
1900 
1901 static int tcp_inq_hint(struct sock *sk)
1902 {
1903 	const struct tcp_sock *tp = tcp_sk(sk);
1904 	u32 copied_seq = READ_ONCE(tp->copied_seq);
1905 	u32 rcv_nxt = READ_ONCE(tp->rcv_nxt);
1906 	int inq;
1907 
1908 	inq = rcv_nxt - copied_seq;
1909 	if (unlikely(inq < 0 || copied_seq != READ_ONCE(tp->copied_seq))) {
1910 		lock_sock(sk);
1911 		inq = tp->rcv_nxt - tp->copied_seq;
1912 		release_sock(sk);
1913 	}
1914 	return inq;
1915 }
1916 
1917 /*
1918  *	This routine copies from a sock struct into the user buffer.
1919  *
1920  *	Technical note: in 2.3 we work on _locked_ socket, so that
1921  *	tricks with *seq access order and skb->users are not required.
1922  *	Probably, code can be easily improved even more.
1923  */
1924 
1925 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
1926 		int flags, int *addr_len)
1927 {
1928 	struct tcp_sock *tp = tcp_sk(sk);
1929 	int copied = 0;
1930 	u32 peek_seq;
1931 	u32 *seq;
1932 	unsigned long used;
1933 	int err, inq;
1934 	int target;		/* Read at least this many bytes */
1935 	long timeo;
1936 	struct sk_buff *skb, *last;
1937 	u32 urg_hole = 0;
1938 	struct scm_timestamping tss;
1939 	bool has_tss = false;
1940 	bool has_cmsg;
1941 
1942 	if (unlikely(flags & MSG_ERRQUEUE))
1943 		return inet_recv_error(sk, msg, len, addr_len);
1944 
1945 	if (sk_can_busy_loop(sk) && skb_queue_empty(&sk->sk_receive_queue) &&
1946 	    (sk->sk_state == TCP_ESTABLISHED))
1947 		sk_busy_loop(sk, nonblock);
1948 
1949 	lock_sock(sk);
1950 
1951 	err = -ENOTCONN;
1952 	if (sk->sk_state == TCP_LISTEN)
1953 		goto out;
1954 
1955 	has_cmsg = tp->recvmsg_inq;
1956 	timeo = sock_rcvtimeo(sk, nonblock);
1957 
1958 	/* Urgent data needs to be handled specially. */
1959 	if (flags & MSG_OOB)
1960 		goto recv_urg;
1961 
1962 	if (unlikely(tp->repair)) {
1963 		err = -EPERM;
1964 		if (!(flags & MSG_PEEK))
1965 			goto out;
1966 
1967 		if (tp->repair_queue == TCP_SEND_QUEUE)
1968 			goto recv_sndq;
1969 
1970 		err = -EINVAL;
1971 		if (tp->repair_queue == TCP_NO_QUEUE)
1972 			goto out;
1973 
1974 		/* 'common' recv queue MSG_PEEK-ing */
1975 	}
1976 
1977 	seq = &tp->copied_seq;
1978 	if (flags & MSG_PEEK) {
1979 		peek_seq = tp->copied_seq;
1980 		seq = &peek_seq;
1981 	}
1982 
1983 	target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1984 
1985 	do {
1986 		u32 offset;
1987 
1988 		/* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */
1989 		if (tp->urg_data && tp->urg_seq == *seq) {
1990 			if (copied)
1991 				break;
1992 			if (signal_pending(current)) {
1993 				copied = timeo ? sock_intr_errno(timeo) : -EAGAIN;
1994 				break;
1995 			}
1996 		}
1997 
1998 		/* Next get a buffer. */
1999 
2000 		last = skb_peek_tail(&sk->sk_receive_queue);
2001 		skb_queue_walk(&sk->sk_receive_queue, skb) {
2002 			last = skb;
2003 			/* Now that we have two receive queues this
2004 			 * shouldn't happen.
2005 			 */
2006 			if (WARN(before(*seq, TCP_SKB_CB(skb)->seq),
2007 				 "TCP recvmsg seq # bug: copied %X, seq %X, rcvnxt %X, fl %X\n",
2008 				 *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt,
2009 				 flags))
2010 				break;
2011 
2012 			offset = *seq - TCP_SKB_CB(skb)->seq;
2013 			if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
2014 				pr_err_once("%s: found a SYN, please report !\n", __func__);
2015 				offset--;
2016 			}
2017 			if (offset < skb->len)
2018 				goto found_ok_skb;
2019 			if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
2020 				goto found_fin_ok;
2021 			WARN(!(flags & MSG_PEEK),
2022 			     "TCP recvmsg seq # bug 2: copied %X, seq %X, rcvnxt %X, fl %X\n",
2023 			     *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags);
2024 		}
2025 
2026 		/* Well, if we have backlog, try to process it now yet. */
2027 
2028 		if (copied >= target && !sk->sk_backlog.tail)
2029 			break;
2030 
2031 		if (copied) {
2032 			if (sk->sk_err ||
2033 			    sk->sk_state == TCP_CLOSE ||
2034 			    (sk->sk_shutdown & RCV_SHUTDOWN) ||
2035 			    !timeo ||
2036 			    signal_pending(current))
2037 				break;
2038 		} else {
2039 			if (sock_flag(sk, SOCK_DONE))
2040 				break;
2041 
2042 			if (sk->sk_err) {
2043 				copied = sock_error(sk);
2044 				break;
2045 			}
2046 
2047 			if (sk->sk_shutdown & RCV_SHUTDOWN)
2048 				break;
2049 
2050 			if (sk->sk_state == TCP_CLOSE) {
2051 				/* This occurs when user tries to read
2052 				 * from never connected socket.
2053 				 */
2054 				copied = -ENOTCONN;
2055 				break;
2056 			}
2057 
2058 			if (!timeo) {
2059 				copied = -EAGAIN;
2060 				break;
2061 			}
2062 
2063 			if (signal_pending(current)) {
2064 				copied = sock_intr_errno(timeo);
2065 				break;
2066 			}
2067 		}
2068 
2069 		tcp_cleanup_rbuf(sk, copied);
2070 
2071 		if (copied >= target) {
2072 			/* Do not sleep, just process backlog. */
2073 			release_sock(sk);
2074 			lock_sock(sk);
2075 		} else {
2076 			sk_wait_data(sk, &timeo, last);
2077 		}
2078 
2079 		if ((flags & MSG_PEEK) &&
2080 		    (peek_seq - copied - urg_hole != tp->copied_seq)) {
2081 			net_dbg_ratelimited("TCP(%s:%d): Application bug, race in MSG_PEEK\n",
2082 					    current->comm,
2083 					    task_pid_nr(current));
2084 			peek_seq = tp->copied_seq;
2085 		}
2086 		continue;
2087 
2088 found_ok_skb:
2089 		/* Ok so how much can we use? */
2090 		used = skb->len - offset;
2091 		if (len < used)
2092 			used = len;
2093 
2094 		/* Do we have urgent data here? */
2095 		if (tp->urg_data) {
2096 			u32 urg_offset = tp->urg_seq - *seq;
2097 			if (urg_offset < used) {
2098 				if (!urg_offset) {
2099 					if (!sock_flag(sk, SOCK_URGINLINE)) {
2100 						++*seq;
2101 						urg_hole++;
2102 						offset++;
2103 						used--;
2104 						if (!used)
2105 							goto skip_copy;
2106 					}
2107 				} else
2108 					used = urg_offset;
2109 			}
2110 		}
2111 
2112 		if (!(flags & MSG_TRUNC)) {
2113 			err = skb_copy_datagram_msg(skb, offset, msg, used);
2114 			if (err) {
2115 				/* Exception. Bailout! */
2116 				if (!copied)
2117 					copied = -EFAULT;
2118 				break;
2119 			}
2120 		}
2121 
2122 		*seq += used;
2123 		copied += used;
2124 		len -= used;
2125 
2126 		tcp_rcv_space_adjust(sk);
2127 
2128 skip_copy:
2129 		if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) {
2130 			tp->urg_data = 0;
2131 			tcp_fast_path_check(sk);
2132 		}
2133 		if (used + offset < skb->len)
2134 			continue;
2135 
2136 		if (TCP_SKB_CB(skb)->has_rxtstamp) {
2137 			tcp_update_recv_tstamps(skb, &tss);
2138 			has_tss = true;
2139 			has_cmsg = true;
2140 		}
2141 		if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
2142 			goto found_fin_ok;
2143 		if (!(flags & MSG_PEEK))
2144 			sk_eat_skb(sk, skb);
2145 		continue;
2146 
2147 found_fin_ok:
2148 		/* Process the FIN. */
2149 		++*seq;
2150 		if (!(flags & MSG_PEEK))
2151 			sk_eat_skb(sk, skb);
2152 		break;
2153 	} while (len > 0);
2154 
2155 	/* According to UNIX98, msg_name/msg_namelen are ignored
2156 	 * on connected socket. I was just happy when found this 8) --ANK
2157 	 */
2158 
2159 	/* Clean up data we have read: This will do ACK frames. */
2160 	tcp_cleanup_rbuf(sk, copied);
2161 
2162 	release_sock(sk);
2163 
2164 	if (has_cmsg) {
2165 		if (has_tss)
2166 			tcp_recv_timestamp(msg, sk, &tss);
2167 		if (tp->recvmsg_inq) {
2168 			inq = tcp_inq_hint(sk);
2169 			put_cmsg(msg, SOL_TCP, TCP_CM_INQ, sizeof(inq), &inq);
2170 		}
2171 	}
2172 
2173 	return copied;
2174 
2175 out:
2176 	release_sock(sk);
2177 	return err;
2178 
2179 recv_urg:
2180 	err = tcp_recv_urg(sk, msg, len, flags);
2181 	goto out;
2182 
2183 recv_sndq:
2184 	err = tcp_peek_sndq(sk, msg, len);
2185 	goto out;
2186 }
2187 EXPORT_SYMBOL(tcp_recvmsg);
2188 
2189 void tcp_set_state(struct sock *sk, int state)
2190 {
2191 	int oldstate = sk->sk_state;
2192 
2193 	/* We defined a new enum for TCP states that are exported in BPF
2194 	 * so as not force the internal TCP states to be frozen. The
2195 	 * following checks will detect if an internal state value ever
2196 	 * differs from the BPF value. If this ever happens, then we will
2197 	 * need to remap the internal value to the BPF value before calling
2198 	 * tcp_call_bpf_2arg.
2199 	 */
2200 	BUILD_BUG_ON((int)BPF_TCP_ESTABLISHED != (int)TCP_ESTABLISHED);
2201 	BUILD_BUG_ON((int)BPF_TCP_SYN_SENT != (int)TCP_SYN_SENT);
2202 	BUILD_BUG_ON((int)BPF_TCP_SYN_RECV != (int)TCP_SYN_RECV);
2203 	BUILD_BUG_ON((int)BPF_TCP_FIN_WAIT1 != (int)TCP_FIN_WAIT1);
2204 	BUILD_BUG_ON((int)BPF_TCP_FIN_WAIT2 != (int)TCP_FIN_WAIT2);
2205 	BUILD_BUG_ON((int)BPF_TCP_TIME_WAIT != (int)TCP_TIME_WAIT);
2206 	BUILD_BUG_ON((int)BPF_TCP_CLOSE != (int)TCP_CLOSE);
2207 	BUILD_BUG_ON((int)BPF_TCP_CLOSE_WAIT != (int)TCP_CLOSE_WAIT);
2208 	BUILD_BUG_ON((int)BPF_TCP_LAST_ACK != (int)TCP_LAST_ACK);
2209 	BUILD_BUG_ON((int)BPF_TCP_LISTEN != (int)TCP_LISTEN);
2210 	BUILD_BUG_ON((int)BPF_TCP_CLOSING != (int)TCP_CLOSING);
2211 	BUILD_BUG_ON((int)BPF_TCP_NEW_SYN_RECV != (int)TCP_NEW_SYN_RECV);
2212 	BUILD_BUG_ON((int)BPF_TCP_MAX_STATES != (int)TCP_MAX_STATES);
2213 
2214 	if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_STATE_CB_FLAG))
2215 		tcp_call_bpf_2arg(sk, BPF_SOCK_OPS_STATE_CB, oldstate, state);
2216 
2217 	switch (state) {
2218 	case TCP_ESTABLISHED:
2219 		if (oldstate != TCP_ESTABLISHED)
2220 			TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
2221 		break;
2222 
2223 	case TCP_CLOSE:
2224 		if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
2225 			TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS);
2226 
2227 		sk->sk_prot->unhash(sk);
2228 		if (inet_csk(sk)->icsk_bind_hash &&
2229 		    !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
2230 			inet_put_port(sk);
2231 		/* fall through */
2232 	default:
2233 		if (oldstate == TCP_ESTABLISHED)
2234 			TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
2235 	}
2236 
2237 	/* Change state AFTER socket is unhashed to avoid closed
2238 	 * socket sitting in hash tables.
2239 	 */
2240 	inet_sk_state_store(sk, state);
2241 }
2242 EXPORT_SYMBOL_GPL(tcp_set_state);
2243 
2244 /*
2245  *	State processing on a close. This implements the state shift for
2246  *	sending our FIN frame. Note that we only send a FIN for some
2247  *	states. A shutdown() may have already sent the FIN, or we may be
2248  *	closed.
2249  */
2250 
2251 static const unsigned char new_state[16] = {
2252   /* current state:        new state:      action:	*/
2253   [0 /* (Invalid) */]	= TCP_CLOSE,
2254   [TCP_ESTABLISHED]	= TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2255   [TCP_SYN_SENT]	= TCP_CLOSE,
2256   [TCP_SYN_RECV]	= TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2257   [TCP_FIN_WAIT1]	= TCP_FIN_WAIT1,
2258   [TCP_FIN_WAIT2]	= TCP_FIN_WAIT2,
2259   [TCP_TIME_WAIT]	= TCP_CLOSE,
2260   [TCP_CLOSE]		= TCP_CLOSE,
2261   [TCP_CLOSE_WAIT]	= TCP_LAST_ACK  | TCP_ACTION_FIN,
2262   [TCP_LAST_ACK]	= TCP_LAST_ACK,
2263   [TCP_LISTEN]		= TCP_CLOSE,
2264   [TCP_CLOSING]		= TCP_CLOSING,
2265   [TCP_NEW_SYN_RECV]	= TCP_CLOSE,	/* should not happen ! */
2266 };
2267 
2268 static int tcp_close_state(struct sock *sk)
2269 {
2270 	int next = (int)new_state[sk->sk_state];
2271 	int ns = next & TCP_STATE_MASK;
2272 
2273 	tcp_set_state(sk, ns);
2274 
2275 	return next & TCP_ACTION_FIN;
2276 }
2277 
2278 /*
2279  *	Shutdown the sending side of a connection. Much like close except
2280  *	that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD).
2281  */
2282 
2283 void tcp_shutdown(struct sock *sk, int how)
2284 {
2285 	/*	We need to grab some memory, and put together a FIN,
2286 	 *	and then put it into the queue to be sent.
2287 	 *		Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92.
2288 	 */
2289 	if (!(how & SEND_SHUTDOWN))
2290 		return;
2291 
2292 	/* If we've already sent a FIN, or it's a closed state, skip this. */
2293 	if ((1 << sk->sk_state) &
2294 	    (TCPF_ESTABLISHED | TCPF_SYN_SENT |
2295 	     TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) {
2296 		/* Clear out any half completed packets.  FIN if needed. */
2297 		if (tcp_close_state(sk))
2298 			tcp_send_fin(sk);
2299 	}
2300 }
2301 EXPORT_SYMBOL(tcp_shutdown);
2302 
2303 bool tcp_check_oom(struct sock *sk, int shift)
2304 {
2305 	bool too_many_orphans, out_of_socket_memory;
2306 
2307 	too_many_orphans = tcp_too_many_orphans(sk, shift);
2308 	out_of_socket_memory = tcp_out_of_memory(sk);
2309 
2310 	if (too_many_orphans)
2311 		net_info_ratelimited("too many orphaned sockets\n");
2312 	if (out_of_socket_memory)
2313 		net_info_ratelimited("out of memory -- consider tuning tcp_mem\n");
2314 	return too_many_orphans || out_of_socket_memory;
2315 }
2316 
2317 void tcp_close(struct sock *sk, long timeout)
2318 {
2319 	struct sk_buff *skb;
2320 	int data_was_unread = 0;
2321 	int state;
2322 
2323 	lock_sock(sk);
2324 	sk->sk_shutdown = SHUTDOWN_MASK;
2325 
2326 	if (sk->sk_state == TCP_LISTEN) {
2327 		tcp_set_state(sk, TCP_CLOSE);
2328 
2329 		/* Special case. */
2330 		inet_csk_listen_stop(sk);
2331 
2332 		goto adjudge_to_death;
2333 	}
2334 
2335 	/*  We need to flush the recv. buffs.  We do this only on the
2336 	 *  descriptor close, not protocol-sourced closes, because the
2337 	 *  reader process may not have drained the data yet!
2338 	 */
2339 	while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
2340 		u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq;
2341 
2342 		if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
2343 			len--;
2344 		data_was_unread += len;
2345 		__kfree_skb(skb);
2346 	}
2347 
2348 	sk_mem_reclaim(sk);
2349 
2350 	/* If socket has been already reset (e.g. in tcp_reset()) - kill it. */
2351 	if (sk->sk_state == TCP_CLOSE)
2352 		goto adjudge_to_death;
2353 
2354 	/* As outlined in RFC 2525, section 2.17, we send a RST here because
2355 	 * data was lost. To witness the awful effects of the old behavior of
2356 	 * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk
2357 	 * GET in an FTP client, suspend the process, wait for the client to
2358 	 * advertise a zero window, then kill -9 the FTP client, wheee...
2359 	 * Note: timeout is always zero in such a case.
2360 	 */
2361 	if (unlikely(tcp_sk(sk)->repair)) {
2362 		sk->sk_prot->disconnect(sk, 0);
2363 	} else if (data_was_unread) {
2364 		/* Unread data was tossed, zap the connection. */
2365 		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE);
2366 		tcp_set_state(sk, TCP_CLOSE);
2367 		tcp_send_active_reset(sk, sk->sk_allocation);
2368 	} else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
2369 		/* Check zero linger _after_ checking for unread data. */
2370 		sk->sk_prot->disconnect(sk, 0);
2371 		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONDATA);
2372 	} else if (tcp_close_state(sk)) {
2373 		/* We FIN if the application ate all the data before
2374 		 * zapping the connection.
2375 		 */
2376 
2377 		/* RED-PEN. Formally speaking, we have broken TCP state
2378 		 * machine. State transitions:
2379 		 *
2380 		 * TCP_ESTABLISHED -> TCP_FIN_WAIT1
2381 		 * TCP_SYN_RECV	-> TCP_FIN_WAIT1 (forget it, it's impossible)
2382 		 * TCP_CLOSE_WAIT -> TCP_LAST_ACK
2383 		 *
2384 		 * are legal only when FIN has been sent (i.e. in window),
2385 		 * rather than queued out of window. Purists blame.
2386 		 *
2387 		 * F.e. "RFC state" is ESTABLISHED,
2388 		 * if Linux state is FIN-WAIT-1, but FIN is still not sent.
2389 		 *
2390 		 * The visible declinations are that sometimes
2391 		 * we enter time-wait state, when it is not required really
2392 		 * (harmless), do not send active resets, when they are
2393 		 * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when
2394 		 * they look as CLOSING or LAST_ACK for Linux)
2395 		 * Probably, I missed some more holelets.
2396 		 * 						--ANK
2397 		 * XXX (TFO) - To start off we don't support SYN+ACK+FIN
2398 		 * in a single packet! (May consider it later but will
2399 		 * probably need API support or TCP_CORK SYN-ACK until
2400 		 * data is written and socket is closed.)
2401 		 */
2402 		tcp_send_fin(sk);
2403 	}
2404 
2405 	sk_stream_wait_close(sk, timeout);
2406 
2407 adjudge_to_death:
2408 	state = sk->sk_state;
2409 	sock_hold(sk);
2410 	sock_orphan(sk);
2411 
2412 	local_bh_disable();
2413 	bh_lock_sock(sk);
2414 	/* remove backlog if any, without releasing ownership. */
2415 	__release_sock(sk);
2416 
2417 	percpu_counter_inc(sk->sk_prot->orphan_count);
2418 
2419 	/* Have we already been destroyed by a softirq or backlog? */
2420 	if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE)
2421 		goto out;
2422 
2423 	/*	This is a (useful) BSD violating of the RFC. There is a
2424 	 *	problem with TCP as specified in that the other end could
2425 	 *	keep a socket open forever with no application left this end.
2426 	 *	We use a 1 minute timeout (about the same as BSD) then kill
2427 	 *	our end. If they send after that then tough - BUT: long enough
2428 	 *	that we won't make the old 4*rto = almost no time - whoops
2429 	 *	reset mistake.
2430 	 *
2431 	 *	Nope, it was not mistake. It is really desired behaviour
2432 	 *	f.e. on http servers, when such sockets are useless, but
2433 	 *	consume significant resources. Let's do it with special
2434 	 *	linger2	option.					--ANK
2435 	 */
2436 
2437 	if (sk->sk_state == TCP_FIN_WAIT2) {
2438 		struct tcp_sock *tp = tcp_sk(sk);
2439 		if (tp->linger2 < 0) {
2440 			tcp_set_state(sk, TCP_CLOSE);
2441 			tcp_send_active_reset(sk, GFP_ATOMIC);
2442 			__NET_INC_STATS(sock_net(sk),
2443 					LINUX_MIB_TCPABORTONLINGER);
2444 		} else {
2445 			const int tmo = tcp_fin_time(sk);
2446 
2447 			if (tmo > TCP_TIMEWAIT_LEN) {
2448 				inet_csk_reset_keepalive_timer(sk,
2449 						tmo - TCP_TIMEWAIT_LEN);
2450 			} else {
2451 				tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
2452 				goto out;
2453 			}
2454 		}
2455 	}
2456 	if (sk->sk_state != TCP_CLOSE) {
2457 		sk_mem_reclaim(sk);
2458 		if (tcp_check_oom(sk, 0)) {
2459 			tcp_set_state(sk, TCP_CLOSE);
2460 			tcp_send_active_reset(sk, GFP_ATOMIC);
2461 			__NET_INC_STATS(sock_net(sk),
2462 					LINUX_MIB_TCPABORTONMEMORY);
2463 		} else if (!check_net(sock_net(sk))) {
2464 			/* Not possible to send reset; just close */
2465 			tcp_set_state(sk, TCP_CLOSE);
2466 		}
2467 	}
2468 
2469 	if (sk->sk_state == TCP_CLOSE) {
2470 		struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
2471 		/* We could get here with a non-NULL req if the socket is
2472 		 * aborted (e.g., closed with unread data) before 3WHS
2473 		 * finishes.
2474 		 */
2475 		if (req)
2476 			reqsk_fastopen_remove(sk, req, false);
2477 		inet_csk_destroy_sock(sk);
2478 	}
2479 	/* Otherwise, socket is reprieved until protocol close. */
2480 
2481 out:
2482 	bh_unlock_sock(sk);
2483 	local_bh_enable();
2484 	release_sock(sk);
2485 	sock_put(sk);
2486 }
2487 EXPORT_SYMBOL(tcp_close);
2488 
2489 /* These states need RST on ABORT according to RFC793 */
2490 
2491 static inline bool tcp_need_reset(int state)
2492 {
2493 	return (1 << state) &
2494 	       (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 |
2495 		TCPF_FIN_WAIT2 | TCPF_SYN_RECV);
2496 }
2497 
2498 static void tcp_rtx_queue_purge(struct sock *sk)
2499 {
2500 	struct rb_node *p = rb_first(&sk->tcp_rtx_queue);
2501 
2502 	while (p) {
2503 		struct sk_buff *skb = rb_to_skb(p);
2504 
2505 		p = rb_next(p);
2506 		/* Since we are deleting whole queue, no need to
2507 		 * list_del(&skb->tcp_tsorted_anchor)
2508 		 */
2509 		tcp_rtx_queue_unlink(skb, sk);
2510 		sk_wmem_free_skb(sk, skb);
2511 	}
2512 }
2513 
2514 void tcp_write_queue_purge(struct sock *sk)
2515 {
2516 	struct sk_buff *skb;
2517 
2518 	tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
2519 	while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
2520 		tcp_skb_tsorted_anchor_cleanup(skb);
2521 		sk_wmem_free_skb(sk, skb);
2522 	}
2523 	tcp_rtx_queue_purge(sk);
2524 	INIT_LIST_HEAD(&tcp_sk(sk)->tsorted_sent_queue);
2525 	sk_mem_reclaim(sk);
2526 	tcp_clear_all_retrans_hints(tcp_sk(sk));
2527 	tcp_sk(sk)->packets_out = 0;
2528 }
2529 
2530 int tcp_disconnect(struct sock *sk, int flags)
2531 {
2532 	struct inet_sock *inet = inet_sk(sk);
2533 	struct inet_connection_sock *icsk = inet_csk(sk);
2534 	struct tcp_sock *tp = tcp_sk(sk);
2535 	int old_state = sk->sk_state;
2536 
2537 	if (old_state != TCP_CLOSE)
2538 		tcp_set_state(sk, TCP_CLOSE);
2539 
2540 	/* ABORT function of RFC793 */
2541 	if (old_state == TCP_LISTEN) {
2542 		inet_csk_listen_stop(sk);
2543 	} else if (unlikely(tp->repair)) {
2544 		sk->sk_err = ECONNABORTED;
2545 	} else if (tcp_need_reset(old_state) ||
2546 		   (tp->snd_nxt != tp->write_seq &&
2547 		    (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) {
2548 		/* The last check adjusts for discrepancy of Linux wrt. RFC
2549 		 * states
2550 		 */
2551 		tcp_send_active_reset(sk, gfp_any());
2552 		sk->sk_err = ECONNRESET;
2553 	} else if (old_state == TCP_SYN_SENT)
2554 		sk->sk_err = ECONNRESET;
2555 
2556 	tcp_clear_xmit_timers(sk);
2557 	__skb_queue_purge(&sk->sk_receive_queue);
2558 	tp->copied_seq = tp->rcv_nxt;
2559 	tp->urg_data = 0;
2560 	tcp_write_queue_purge(sk);
2561 	tcp_fastopen_active_disable_ofo_check(sk);
2562 	skb_rbtree_purge(&tp->out_of_order_queue);
2563 
2564 	inet->inet_dport = 0;
2565 
2566 	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
2567 		inet_reset_saddr(sk);
2568 
2569 	sk->sk_shutdown = 0;
2570 	sock_reset_flag(sk, SOCK_DONE);
2571 	tp->srtt_us = 0;
2572 	tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
2573 	tp->rcv_rtt_last_tsecr = 0;
2574 	tp->write_seq += tp->max_window + 2;
2575 	if (tp->write_seq == 0)
2576 		tp->write_seq = 1;
2577 	icsk->icsk_backoff = 0;
2578 	icsk->icsk_probes_out = 0;
2579 	icsk->icsk_rto = TCP_TIMEOUT_INIT;
2580 	tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
2581 	tp->snd_cwnd = TCP_INIT_CWND;
2582 	tp->snd_cwnd_cnt = 0;
2583 	tp->window_clamp = 0;
2584 	tp->delivered_ce = 0;
2585 	tcp_set_ca_state(sk, TCP_CA_Open);
2586 	tp->is_sack_reneg = 0;
2587 	tcp_clear_retrans(tp);
2588 	inet_csk_delack_init(sk);
2589 	/* Initialize rcv_mss to TCP_MIN_MSS to avoid division by 0
2590 	 * issue in __tcp_select_window()
2591 	 */
2592 	icsk->icsk_ack.rcv_mss = TCP_MIN_MSS;
2593 	memset(&tp->rx_opt, 0, sizeof(tp->rx_opt));
2594 	__sk_dst_reset(sk);
2595 	dst_release(sk->sk_rx_dst);
2596 	sk->sk_rx_dst = NULL;
2597 	tcp_saved_syn_free(tp);
2598 	tp->compressed_ack = 0;
2599 	tp->bytes_sent = 0;
2600 	tp->bytes_retrans = 0;
2601 	tp->duplicate_sack[0].start_seq = 0;
2602 	tp->duplicate_sack[0].end_seq = 0;
2603 	tp->dsack_dups = 0;
2604 	tp->reord_seen = 0;
2605 	tp->retrans_out = 0;
2606 	tp->sacked_out = 0;
2607 	tp->tlp_high_seq = 0;
2608 	tp->last_oow_ack_time = 0;
2609 	/* There's a bubble in the pipe until at least the first ACK. */
2610 	tp->app_limited = ~0U;
2611 	tp->rack.mstamp = 0;
2612 	tp->rack.advanced = 0;
2613 	tp->rack.reo_wnd_steps = 1;
2614 	tp->rack.last_delivered = 0;
2615 	tp->rack.reo_wnd_persist = 0;
2616 	tp->rack.dsack_seen = 0;
2617 	tp->syn_data_acked = 0;
2618 	tp->rx_opt.saw_tstamp = 0;
2619 	tp->rx_opt.dsack = 0;
2620 	tp->rx_opt.num_sacks = 0;
2621 
2622 
2623 	/* Clean up fastopen related fields */
2624 	tcp_free_fastopen_req(tp);
2625 	inet->defer_connect = 0;
2626 
2627 	WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
2628 
2629 	if (sk->sk_frag.page) {
2630 		put_page(sk->sk_frag.page);
2631 		sk->sk_frag.page = NULL;
2632 		sk->sk_frag.offset = 0;
2633 	}
2634 
2635 	sk->sk_error_report(sk);
2636 	return 0;
2637 }
2638 EXPORT_SYMBOL(tcp_disconnect);
2639 
2640 static inline bool tcp_can_repair_sock(const struct sock *sk)
2641 {
2642 	return ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN) &&
2643 		(sk->sk_state != TCP_LISTEN);
2644 }
2645 
2646 static int tcp_repair_set_window(struct tcp_sock *tp, char __user *optbuf, int len)
2647 {
2648 	struct tcp_repair_window opt;
2649 
2650 	if (!tp->repair)
2651 		return -EPERM;
2652 
2653 	if (len != sizeof(opt))
2654 		return -EINVAL;
2655 
2656 	if (copy_from_user(&opt, optbuf, sizeof(opt)))
2657 		return -EFAULT;
2658 
2659 	if (opt.max_window < opt.snd_wnd)
2660 		return -EINVAL;
2661 
2662 	if (after(opt.snd_wl1, tp->rcv_nxt + opt.rcv_wnd))
2663 		return -EINVAL;
2664 
2665 	if (after(opt.rcv_wup, tp->rcv_nxt))
2666 		return -EINVAL;
2667 
2668 	tp->snd_wl1	= opt.snd_wl1;
2669 	tp->snd_wnd	= opt.snd_wnd;
2670 	tp->max_window	= opt.max_window;
2671 
2672 	tp->rcv_wnd	= opt.rcv_wnd;
2673 	tp->rcv_wup	= opt.rcv_wup;
2674 
2675 	return 0;
2676 }
2677 
2678 static int tcp_repair_options_est(struct sock *sk,
2679 		struct tcp_repair_opt __user *optbuf, unsigned int len)
2680 {
2681 	struct tcp_sock *tp = tcp_sk(sk);
2682 	struct tcp_repair_opt opt;
2683 
2684 	while (len >= sizeof(opt)) {
2685 		if (copy_from_user(&opt, optbuf, sizeof(opt)))
2686 			return -EFAULT;
2687 
2688 		optbuf++;
2689 		len -= sizeof(opt);
2690 
2691 		switch (opt.opt_code) {
2692 		case TCPOPT_MSS:
2693 			tp->rx_opt.mss_clamp = opt.opt_val;
2694 			tcp_mtup_init(sk);
2695 			break;
2696 		case TCPOPT_WINDOW:
2697 			{
2698 				u16 snd_wscale = opt.opt_val & 0xFFFF;
2699 				u16 rcv_wscale = opt.opt_val >> 16;
2700 
2701 				if (snd_wscale > TCP_MAX_WSCALE || rcv_wscale > TCP_MAX_WSCALE)
2702 					return -EFBIG;
2703 
2704 				tp->rx_opt.snd_wscale = snd_wscale;
2705 				tp->rx_opt.rcv_wscale = rcv_wscale;
2706 				tp->rx_opt.wscale_ok = 1;
2707 			}
2708 			break;
2709 		case TCPOPT_SACK_PERM:
2710 			if (opt.opt_val != 0)
2711 				return -EINVAL;
2712 
2713 			tp->rx_opt.sack_ok |= TCP_SACK_SEEN;
2714 			break;
2715 		case TCPOPT_TIMESTAMP:
2716 			if (opt.opt_val != 0)
2717 				return -EINVAL;
2718 
2719 			tp->rx_opt.tstamp_ok = 1;
2720 			break;
2721 		}
2722 	}
2723 
2724 	return 0;
2725 }
2726 
2727 /*
2728  *	Socket option code for TCP.
2729  */
2730 static int do_tcp_setsockopt(struct sock *sk, int level,
2731 		int optname, char __user *optval, unsigned int optlen)
2732 {
2733 	struct tcp_sock *tp = tcp_sk(sk);
2734 	struct inet_connection_sock *icsk = inet_csk(sk);
2735 	struct net *net = sock_net(sk);
2736 	int val;
2737 	int err = 0;
2738 
2739 	/* These are data/string values, all the others are ints */
2740 	switch (optname) {
2741 	case TCP_CONGESTION: {
2742 		char name[TCP_CA_NAME_MAX];
2743 
2744 		if (optlen < 1)
2745 			return -EINVAL;
2746 
2747 		val = strncpy_from_user(name, optval,
2748 					min_t(long, TCP_CA_NAME_MAX-1, optlen));
2749 		if (val < 0)
2750 			return -EFAULT;
2751 		name[val] = 0;
2752 
2753 		lock_sock(sk);
2754 		err = tcp_set_congestion_control(sk, name, true, true);
2755 		release_sock(sk);
2756 		return err;
2757 	}
2758 	case TCP_ULP: {
2759 		char name[TCP_ULP_NAME_MAX];
2760 
2761 		if (optlen < 1)
2762 			return -EINVAL;
2763 
2764 		val = strncpy_from_user(name, optval,
2765 					min_t(long, TCP_ULP_NAME_MAX - 1,
2766 					      optlen));
2767 		if (val < 0)
2768 			return -EFAULT;
2769 		name[val] = 0;
2770 
2771 		lock_sock(sk);
2772 		err = tcp_set_ulp(sk, name);
2773 		release_sock(sk);
2774 		return err;
2775 	}
2776 	case TCP_FASTOPEN_KEY: {
2777 		__u8 key[TCP_FASTOPEN_KEY_LENGTH];
2778 
2779 		if (optlen != sizeof(key))
2780 			return -EINVAL;
2781 
2782 		if (copy_from_user(key, optval, optlen))
2783 			return -EFAULT;
2784 
2785 		return tcp_fastopen_reset_cipher(net, sk, key, sizeof(key));
2786 	}
2787 	default:
2788 		/* fallthru */
2789 		break;
2790 	}
2791 
2792 	if (optlen < sizeof(int))
2793 		return -EINVAL;
2794 
2795 	if (get_user(val, (int __user *)optval))
2796 		return -EFAULT;
2797 
2798 	lock_sock(sk);
2799 
2800 	switch (optname) {
2801 	case TCP_MAXSEG:
2802 		/* Values greater than interface MTU won't take effect. However
2803 		 * at the point when this call is done we typically don't yet
2804 		 * know which interface is going to be used
2805 		 */
2806 		if (val && (val < TCP_MIN_MSS || val > MAX_TCP_WINDOW)) {
2807 			err = -EINVAL;
2808 			break;
2809 		}
2810 		tp->rx_opt.user_mss = val;
2811 		break;
2812 
2813 	case TCP_NODELAY:
2814 		if (val) {
2815 			/* TCP_NODELAY is weaker than TCP_CORK, so that
2816 			 * this option on corked socket is remembered, but
2817 			 * it is not activated until cork is cleared.
2818 			 *
2819 			 * However, when TCP_NODELAY is set we make
2820 			 * an explicit push, which overrides even TCP_CORK
2821 			 * for currently queued segments.
2822 			 */
2823 			tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH;
2824 			tcp_push_pending_frames(sk);
2825 		} else {
2826 			tp->nonagle &= ~TCP_NAGLE_OFF;
2827 		}
2828 		break;
2829 
2830 	case TCP_THIN_LINEAR_TIMEOUTS:
2831 		if (val < 0 || val > 1)
2832 			err = -EINVAL;
2833 		else
2834 			tp->thin_lto = val;
2835 		break;
2836 
2837 	case TCP_THIN_DUPACK:
2838 		if (val < 0 || val > 1)
2839 			err = -EINVAL;
2840 		break;
2841 
2842 	case TCP_REPAIR:
2843 		if (!tcp_can_repair_sock(sk))
2844 			err = -EPERM;
2845 		else if (val == TCP_REPAIR_ON) {
2846 			tp->repair = 1;
2847 			sk->sk_reuse = SK_FORCE_REUSE;
2848 			tp->repair_queue = TCP_NO_QUEUE;
2849 		} else if (val == TCP_REPAIR_OFF) {
2850 			tp->repair = 0;
2851 			sk->sk_reuse = SK_NO_REUSE;
2852 			tcp_send_window_probe(sk);
2853 		} else if (val == TCP_REPAIR_OFF_NO_WP) {
2854 			tp->repair = 0;
2855 			sk->sk_reuse = SK_NO_REUSE;
2856 		} else
2857 			err = -EINVAL;
2858 
2859 		break;
2860 
2861 	case TCP_REPAIR_QUEUE:
2862 		if (!tp->repair)
2863 			err = -EPERM;
2864 		else if ((unsigned int)val < TCP_QUEUES_NR)
2865 			tp->repair_queue = val;
2866 		else
2867 			err = -EINVAL;
2868 		break;
2869 
2870 	case TCP_QUEUE_SEQ:
2871 		if (sk->sk_state != TCP_CLOSE)
2872 			err = -EPERM;
2873 		else if (tp->repair_queue == TCP_SEND_QUEUE)
2874 			tp->write_seq = val;
2875 		else if (tp->repair_queue == TCP_RECV_QUEUE)
2876 			tp->rcv_nxt = val;
2877 		else
2878 			err = -EINVAL;
2879 		break;
2880 
2881 	case TCP_REPAIR_OPTIONS:
2882 		if (!tp->repair)
2883 			err = -EINVAL;
2884 		else if (sk->sk_state == TCP_ESTABLISHED)
2885 			err = tcp_repair_options_est(sk,
2886 					(struct tcp_repair_opt __user *)optval,
2887 					optlen);
2888 		else
2889 			err = -EPERM;
2890 		break;
2891 
2892 	case TCP_CORK:
2893 		/* When set indicates to always queue non-full frames.
2894 		 * Later the user clears this option and we transmit
2895 		 * any pending partial frames in the queue.  This is
2896 		 * meant to be used alongside sendfile() to get properly
2897 		 * filled frames when the user (for example) must write
2898 		 * out headers with a write() call first and then use
2899 		 * sendfile to send out the data parts.
2900 		 *
2901 		 * TCP_CORK can be set together with TCP_NODELAY and it is
2902 		 * stronger than TCP_NODELAY.
2903 		 */
2904 		if (val) {
2905 			tp->nonagle |= TCP_NAGLE_CORK;
2906 		} else {
2907 			tp->nonagle &= ~TCP_NAGLE_CORK;
2908 			if (tp->nonagle&TCP_NAGLE_OFF)
2909 				tp->nonagle |= TCP_NAGLE_PUSH;
2910 			tcp_push_pending_frames(sk);
2911 		}
2912 		break;
2913 
2914 	case TCP_KEEPIDLE:
2915 		if (val < 1 || val > MAX_TCP_KEEPIDLE)
2916 			err = -EINVAL;
2917 		else {
2918 			tp->keepalive_time = val * HZ;
2919 			if (sock_flag(sk, SOCK_KEEPOPEN) &&
2920 			    !((1 << sk->sk_state) &
2921 			      (TCPF_CLOSE | TCPF_LISTEN))) {
2922 				u32 elapsed = keepalive_time_elapsed(tp);
2923 				if (tp->keepalive_time > elapsed)
2924 					elapsed = tp->keepalive_time - elapsed;
2925 				else
2926 					elapsed = 0;
2927 				inet_csk_reset_keepalive_timer(sk, elapsed);
2928 			}
2929 		}
2930 		break;
2931 	case TCP_KEEPINTVL:
2932 		if (val < 1 || val > MAX_TCP_KEEPINTVL)
2933 			err = -EINVAL;
2934 		else
2935 			tp->keepalive_intvl = val * HZ;
2936 		break;
2937 	case TCP_KEEPCNT:
2938 		if (val < 1 || val > MAX_TCP_KEEPCNT)
2939 			err = -EINVAL;
2940 		else
2941 			tp->keepalive_probes = val;
2942 		break;
2943 	case TCP_SYNCNT:
2944 		if (val < 1 || val > MAX_TCP_SYNCNT)
2945 			err = -EINVAL;
2946 		else
2947 			icsk->icsk_syn_retries = val;
2948 		break;
2949 
2950 	case TCP_SAVE_SYN:
2951 		if (val < 0 || val > 1)
2952 			err = -EINVAL;
2953 		else
2954 			tp->save_syn = val;
2955 		break;
2956 
2957 	case TCP_LINGER2:
2958 		if (val < 0)
2959 			tp->linger2 = -1;
2960 		else if (val > net->ipv4.sysctl_tcp_fin_timeout / HZ)
2961 			tp->linger2 = 0;
2962 		else
2963 			tp->linger2 = val * HZ;
2964 		break;
2965 
2966 	case TCP_DEFER_ACCEPT:
2967 		/* Translate value in seconds to number of retransmits */
2968 		icsk->icsk_accept_queue.rskq_defer_accept =
2969 			secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ,
2970 					TCP_RTO_MAX / HZ);
2971 		break;
2972 
2973 	case TCP_WINDOW_CLAMP:
2974 		if (!val) {
2975 			if (sk->sk_state != TCP_CLOSE) {
2976 				err = -EINVAL;
2977 				break;
2978 			}
2979 			tp->window_clamp = 0;
2980 		} else
2981 			tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ?
2982 						SOCK_MIN_RCVBUF / 2 : val;
2983 		break;
2984 
2985 	case TCP_QUICKACK:
2986 		if (!val) {
2987 			inet_csk_enter_pingpong_mode(sk);
2988 		} else {
2989 			inet_csk_exit_pingpong_mode(sk);
2990 			if ((1 << sk->sk_state) &
2991 			    (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) &&
2992 			    inet_csk_ack_scheduled(sk)) {
2993 				icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;
2994 				tcp_cleanup_rbuf(sk, 1);
2995 				if (!(val & 1))
2996 					inet_csk_enter_pingpong_mode(sk);
2997 			}
2998 		}
2999 		break;
3000 
3001 #ifdef CONFIG_TCP_MD5SIG
3002 	case TCP_MD5SIG:
3003 	case TCP_MD5SIG_EXT:
3004 		if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
3005 			err = tp->af_specific->md5_parse(sk, optname, optval, optlen);
3006 		else
3007 			err = -EINVAL;
3008 		break;
3009 #endif
3010 	case TCP_USER_TIMEOUT:
3011 		/* Cap the max time in ms TCP will retry or probe the window
3012 		 * before giving up and aborting (ETIMEDOUT) a connection.
3013 		 */
3014 		if (val < 0)
3015 			err = -EINVAL;
3016 		else
3017 			icsk->icsk_user_timeout = val;
3018 		break;
3019 
3020 	case TCP_FASTOPEN:
3021 		if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE |
3022 		    TCPF_LISTEN))) {
3023 			tcp_fastopen_init_key_once(net);
3024 
3025 			fastopen_queue_tune(sk, val);
3026 		} else {
3027 			err = -EINVAL;
3028 		}
3029 		break;
3030 	case TCP_FASTOPEN_CONNECT:
3031 		if (val > 1 || val < 0) {
3032 			err = -EINVAL;
3033 		} else if (net->ipv4.sysctl_tcp_fastopen & TFO_CLIENT_ENABLE) {
3034 			if (sk->sk_state == TCP_CLOSE)
3035 				tp->fastopen_connect = val;
3036 			else
3037 				err = -EINVAL;
3038 		} else {
3039 			err = -EOPNOTSUPP;
3040 		}
3041 		break;
3042 	case TCP_FASTOPEN_NO_COOKIE:
3043 		if (val > 1 || val < 0)
3044 			err = -EINVAL;
3045 		else if (!((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
3046 			err = -EINVAL;
3047 		else
3048 			tp->fastopen_no_cookie = val;
3049 		break;
3050 	case TCP_TIMESTAMP:
3051 		if (!tp->repair)
3052 			err = -EPERM;
3053 		else
3054 			tp->tsoffset = val - tcp_time_stamp_raw();
3055 		break;
3056 	case TCP_REPAIR_WINDOW:
3057 		err = tcp_repair_set_window(tp, optval, optlen);
3058 		break;
3059 	case TCP_NOTSENT_LOWAT:
3060 		tp->notsent_lowat = val;
3061 		sk->sk_write_space(sk);
3062 		break;
3063 	case TCP_INQ:
3064 		if (val > 1 || val < 0)
3065 			err = -EINVAL;
3066 		else
3067 			tp->recvmsg_inq = val;
3068 		break;
3069 	default:
3070 		err = -ENOPROTOOPT;
3071 		break;
3072 	}
3073 
3074 	release_sock(sk);
3075 	return err;
3076 }
3077 
3078 int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval,
3079 		   unsigned int optlen)
3080 {
3081 	const struct inet_connection_sock *icsk = inet_csk(sk);
3082 
3083 	if (level != SOL_TCP)
3084 		return icsk->icsk_af_ops->setsockopt(sk, level, optname,
3085 						     optval, optlen);
3086 	return do_tcp_setsockopt(sk, level, optname, optval, optlen);
3087 }
3088 EXPORT_SYMBOL(tcp_setsockopt);
3089 
3090 #ifdef CONFIG_COMPAT
3091 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
3092 			  char __user *optval, unsigned int optlen)
3093 {
3094 	if (level != SOL_TCP)
3095 		return inet_csk_compat_setsockopt(sk, level, optname,
3096 						  optval, optlen);
3097 	return do_tcp_setsockopt(sk, level, optname, optval, optlen);
3098 }
3099 EXPORT_SYMBOL(compat_tcp_setsockopt);
3100 #endif
3101 
3102 static void tcp_get_info_chrono_stats(const struct tcp_sock *tp,
3103 				      struct tcp_info *info)
3104 {
3105 	u64 stats[__TCP_CHRONO_MAX], total = 0;
3106 	enum tcp_chrono i;
3107 
3108 	for (i = TCP_CHRONO_BUSY; i < __TCP_CHRONO_MAX; ++i) {
3109 		stats[i] = tp->chrono_stat[i - 1];
3110 		if (i == tp->chrono_type)
3111 			stats[i] += tcp_jiffies32 - tp->chrono_start;
3112 		stats[i] *= USEC_PER_SEC / HZ;
3113 		total += stats[i];
3114 	}
3115 
3116 	info->tcpi_busy_time = total;
3117 	info->tcpi_rwnd_limited = stats[TCP_CHRONO_RWND_LIMITED];
3118 	info->tcpi_sndbuf_limited = stats[TCP_CHRONO_SNDBUF_LIMITED];
3119 }
3120 
3121 /* Return information about state of tcp endpoint in API format. */
3122 void tcp_get_info(struct sock *sk, struct tcp_info *info)
3123 {
3124 	const struct tcp_sock *tp = tcp_sk(sk); /* iff sk_type == SOCK_STREAM */
3125 	const struct inet_connection_sock *icsk = inet_csk(sk);
3126 	unsigned long rate;
3127 	u32 now;
3128 	u64 rate64;
3129 	bool slow;
3130 
3131 	memset(info, 0, sizeof(*info));
3132 	if (sk->sk_type != SOCK_STREAM)
3133 		return;
3134 
3135 	info->tcpi_state = inet_sk_state_load(sk);
3136 
3137 	/* Report meaningful fields for all TCP states, including listeners */
3138 	rate = READ_ONCE(sk->sk_pacing_rate);
3139 	rate64 = (rate != ~0UL) ? rate : ~0ULL;
3140 	info->tcpi_pacing_rate = rate64;
3141 
3142 	rate = READ_ONCE(sk->sk_max_pacing_rate);
3143 	rate64 = (rate != ~0UL) ? rate : ~0ULL;
3144 	info->tcpi_max_pacing_rate = rate64;
3145 
3146 	info->tcpi_reordering = tp->reordering;
3147 	info->tcpi_snd_cwnd = tp->snd_cwnd;
3148 
3149 	if (info->tcpi_state == TCP_LISTEN) {
3150 		/* listeners aliased fields :
3151 		 * tcpi_unacked -> Number of children ready for accept()
3152 		 * tcpi_sacked  -> max backlog
3153 		 */
3154 		info->tcpi_unacked = sk->sk_ack_backlog;
3155 		info->tcpi_sacked = sk->sk_max_ack_backlog;
3156 		return;
3157 	}
3158 
3159 	slow = lock_sock_fast(sk);
3160 
3161 	info->tcpi_ca_state = icsk->icsk_ca_state;
3162 	info->tcpi_retransmits = icsk->icsk_retransmits;
3163 	info->tcpi_probes = icsk->icsk_probes_out;
3164 	info->tcpi_backoff = icsk->icsk_backoff;
3165 
3166 	if (tp->rx_opt.tstamp_ok)
3167 		info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
3168 	if (tcp_is_sack(tp))
3169 		info->tcpi_options |= TCPI_OPT_SACK;
3170 	if (tp->rx_opt.wscale_ok) {
3171 		info->tcpi_options |= TCPI_OPT_WSCALE;
3172 		info->tcpi_snd_wscale = tp->rx_opt.snd_wscale;
3173 		info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale;
3174 	}
3175 
3176 	if (tp->ecn_flags & TCP_ECN_OK)
3177 		info->tcpi_options |= TCPI_OPT_ECN;
3178 	if (tp->ecn_flags & TCP_ECN_SEEN)
3179 		info->tcpi_options |= TCPI_OPT_ECN_SEEN;
3180 	if (tp->syn_data_acked)
3181 		info->tcpi_options |= TCPI_OPT_SYN_DATA;
3182 
3183 	info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto);
3184 	info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato);
3185 	info->tcpi_snd_mss = tp->mss_cache;
3186 	info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss;
3187 
3188 	info->tcpi_unacked = tp->packets_out;
3189 	info->tcpi_sacked = tp->sacked_out;
3190 
3191 	info->tcpi_lost = tp->lost_out;
3192 	info->tcpi_retrans = tp->retrans_out;
3193 
3194 	now = tcp_jiffies32;
3195 	info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime);
3196 	info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime);
3197 	info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp);
3198 
3199 	info->tcpi_pmtu = icsk->icsk_pmtu_cookie;
3200 	info->tcpi_rcv_ssthresh = tp->rcv_ssthresh;
3201 	info->tcpi_rtt = tp->srtt_us >> 3;
3202 	info->tcpi_rttvar = tp->mdev_us >> 2;
3203 	info->tcpi_snd_ssthresh = tp->snd_ssthresh;
3204 	info->tcpi_advmss = tp->advmss;
3205 
3206 	info->tcpi_rcv_rtt = tp->rcv_rtt_est.rtt_us >> 3;
3207 	info->tcpi_rcv_space = tp->rcvq_space.space;
3208 
3209 	info->tcpi_total_retrans = tp->total_retrans;
3210 
3211 	info->tcpi_bytes_acked = tp->bytes_acked;
3212 	info->tcpi_bytes_received = tp->bytes_received;
3213 	info->tcpi_notsent_bytes = max_t(int, 0, tp->write_seq - tp->snd_nxt);
3214 	tcp_get_info_chrono_stats(tp, info);
3215 
3216 	info->tcpi_segs_out = tp->segs_out;
3217 	info->tcpi_segs_in = tp->segs_in;
3218 
3219 	info->tcpi_min_rtt = tcp_min_rtt(tp);
3220 	info->tcpi_data_segs_in = tp->data_segs_in;
3221 	info->tcpi_data_segs_out = tp->data_segs_out;
3222 
3223 	info->tcpi_delivery_rate_app_limited = tp->rate_app_limited ? 1 : 0;
3224 	rate64 = tcp_compute_delivery_rate(tp);
3225 	if (rate64)
3226 		info->tcpi_delivery_rate = rate64;
3227 	info->tcpi_delivered = tp->delivered;
3228 	info->tcpi_delivered_ce = tp->delivered_ce;
3229 	info->tcpi_bytes_sent = tp->bytes_sent;
3230 	info->tcpi_bytes_retrans = tp->bytes_retrans;
3231 	info->tcpi_dsack_dups = tp->dsack_dups;
3232 	info->tcpi_reord_seen = tp->reord_seen;
3233 	unlock_sock_fast(sk, slow);
3234 }
3235 EXPORT_SYMBOL_GPL(tcp_get_info);
3236 
3237 static size_t tcp_opt_stats_get_size(void)
3238 {
3239 	return
3240 		nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_BUSY */
3241 		nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_RWND_LIMITED */
3242 		nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_SNDBUF_LIMITED */
3243 		nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_DATA_SEGS_OUT */
3244 		nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_TOTAL_RETRANS */
3245 		nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_PACING_RATE */
3246 		nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_DELIVERY_RATE */
3247 		nla_total_size(sizeof(u32)) + /* TCP_NLA_SND_CWND */
3248 		nla_total_size(sizeof(u32)) + /* TCP_NLA_REORDERING */
3249 		nla_total_size(sizeof(u32)) + /* TCP_NLA_MIN_RTT */
3250 		nla_total_size(sizeof(u8)) + /* TCP_NLA_RECUR_RETRANS */
3251 		nla_total_size(sizeof(u8)) + /* TCP_NLA_DELIVERY_RATE_APP_LMT */
3252 		nla_total_size(sizeof(u32)) + /* TCP_NLA_SNDQ_SIZE */
3253 		nla_total_size(sizeof(u8)) + /* TCP_NLA_CA_STATE */
3254 		nla_total_size(sizeof(u32)) + /* TCP_NLA_SND_SSTHRESH */
3255 		nla_total_size(sizeof(u32)) + /* TCP_NLA_DELIVERED */
3256 		nla_total_size(sizeof(u32)) + /* TCP_NLA_DELIVERED_CE */
3257 		nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_BYTES_SENT */
3258 		nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_BYTES_RETRANS */
3259 		nla_total_size(sizeof(u32)) + /* TCP_NLA_DSACK_DUPS */
3260 		nla_total_size(sizeof(u32)) + /* TCP_NLA_REORD_SEEN */
3261 		nla_total_size(sizeof(u32)) + /* TCP_NLA_SRTT */
3262 		0;
3263 }
3264 
3265 struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk)
3266 {
3267 	const struct tcp_sock *tp = tcp_sk(sk);
3268 	struct sk_buff *stats;
3269 	struct tcp_info info;
3270 	unsigned long rate;
3271 	u64 rate64;
3272 
3273 	stats = alloc_skb(tcp_opt_stats_get_size(), GFP_ATOMIC);
3274 	if (!stats)
3275 		return NULL;
3276 
3277 	tcp_get_info_chrono_stats(tp, &info);
3278 	nla_put_u64_64bit(stats, TCP_NLA_BUSY,
3279 			  info.tcpi_busy_time, TCP_NLA_PAD);
3280 	nla_put_u64_64bit(stats, TCP_NLA_RWND_LIMITED,
3281 			  info.tcpi_rwnd_limited, TCP_NLA_PAD);
3282 	nla_put_u64_64bit(stats, TCP_NLA_SNDBUF_LIMITED,
3283 			  info.tcpi_sndbuf_limited, TCP_NLA_PAD);
3284 	nla_put_u64_64bit(stats, TCP_NLA_DATA_SEGS_OUT,
3285 			  tp->data_segs_out, TCP_NLA_PAD);
3286 	nla_put_u64_64bit(stats, TCP_NLA_TOTAL_RETRANS,
3287 			  tp->total_retrans, TCP_NLA_PAD);
3288 
3289 	rate = READ_ONCE(sk->sk_pacing_rate);
3290 	rate64 = (rate != ~0UL) ? rate : ~0ULL;
3291 	nla_put_u64_64bit(stats, TCP_NLA_PACING_RATE, rate64, TCP_NLA_PAD);
3292 
3293 	rate64 = tcp_compute_delivery_rate(tp);
3294 	nla_put_u64_64bit(stats, TCP_NLA_DELIVERY_RATE, rate64, TCP_NLA_PAD);
3295 
3296 	nla_put_u32(stats, TCP_NLA_SND_CWND, tp->snd_cwnd);
3297 	nla_put_u32(stats, TCP_NLA_REORDERING, tp->reordering);
3298 	nla_put_u32(stats, TCP_NLA_MIN_RTT, tcp_min_rtt(tp));
3299 
3300 	nla_put_u8(stats, TCP_NLA_RECUR_RETRANS, inet_csk(sk)->icsk_retransmits);
3301 	nla_put_u8(stats, TCP_NLA_DELIVERY_RATE_APP_LMT, !!tp->rate_app_limited);
3302 	nla_put_u32(stats, TCP_NLA_SND_SSTHRESH, tp->snd_ssthresh);
3303 	nla_put_u32(stats, TCP_NLA_DELIVERED, tp->delivered);
3304 	nla_put_u32(stats, TCP_NLA_DELIVERED_CE, tp->delivered_ce);
3305 
3306 	nla_put_u32(stats, TCP_NLA_SNDQ_SIZE, tp->write_seq - tp->snd_una);
3307 	nla_put_u8(stats, TCP_NLA_CA_STATE, inet_csk(sk)->icsk_ca_state);
3308 
3309 	nla_put_u64_64bit(stats, TCP_NLA_BYTES_SENT, tp->bytes_sent,
3310 			  TCP_NLA_PAD);
3311 	nla_put_u64_64bit(stats, TCP_NLA_BYTES_RETRANS, tp->bytes_retrans,
3312 			  TCP_NLA_PAD);
3313 	nla_put_u32(stats, TCP_NLA_DSACK_DUPS, tp->dsack_dups);
3314 	nla_put_u32(stats, TCP_NLA_REORD_SEEN, tp->reord_seen);
3315 	nla_put_u32(stats, TCP_NLA_SRTT, tp->srtt_us >> 3);
3316 
3317 	return stats;
3318 }
3319 
3320 static int do_tcp_getsockopt(struct sock *sk, int level,
3321 		int optname, char __user *optval, int __user *optlen)
3322 {
3323 	struct inet_connection_sock *icsk = inet_csk(sk);
3324 	struct tcp_sock *tp = tcp_sk(sk);
3325 	struct net *net = sock_net(sk);
3326 	int val, len;
3327 
3328 	if (get_user(len, optlen))
3329 		return -EFAULT;
3330 
3331 	len = min_t(unsigned int, len, sizeof(int));
3332 
3333 	if (len < 0)
3334 		return -EINVAL;
3335 
3336 	switch (optname) {
3337 	case TCP_MAXSEG:
3338 		val = tp->mss_cache;
3339 		if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
3340 			val = tp->rx_opt.user_mss;
3341 		if (tp->repair)
3342 			val = tp->rx_opt.mss_clamp;
3343 		break;
3344 	case TCP_NODELAY:
3345 		val = !!(tp->nonagle&TCP_NAGLE_OFF);
3346 		break;
3347 	case TCP_CORK:
3348 		val = !!(tp->nonagle&TCP_NAGLE_CORK);
3349 		break;
3350 	case TCP_KEEPIDLE:
3351 		val = keepalive_time_when(tp) / HZ;
3352 		break;
3353 	case TCP_KEEPINTVL:
3354 		val = keepalive_intvl_when(tp) / HZ;
3355 		break;
3356 	case TCP_KEEPCNT:
3357 		val = keepalive_probes(tp);
3358 		break;
3359 	case TCP_SYNCNT:
3360 		val = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries;
3361 		break;
3362 	case TCP_LINGER2:
3363 		val = tp->linger2;
3364 		if (val >= 0)
3365 			val = (val ? : net->ipv4.sysctl_tcp_fin_timeout) / HZ;
3366 		break;
3367 	case TCP_DEFER_ACCEPT:
3368 		val = retrans_to_secs(icsk->icsk_accept_queue.rskq_defer_accept,
3369 				      TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ);
3370 		break;
3371 	case TCP_WINDOW_CLAMP:
3372 		val = tp->window_clamp;
3373 		break;
3374 	case TCP_INFO: {
3375 		struct tcp_info info;
3376 
3377 		if (get_user(len, optlen))
3378 			return -EFAULT;
3379 
3380 		tcp_get_info(sk, &info);
3381 
3382 		len = min_t(unsigned int, len, sizeof(info));
3383 		if (put_user(len, optlen))
3384 			return -EFAULT;
3385 		if (copy_to_user(optval, &info, len))
3386 			return -EFAULT;
3387 		return 0;
3388 	}
3389 	case TCP_CC_INFO: {
3390 		const struct tcp_congestion_ops *ca_ops;
3391 		union tcp_cc_info info;
3392 		size_t sz = 0;
3393 		int attr;
3394 
3395 		if (get_user(len, optlen))
3396 			return -EFAULT;
3397 
3398 		ca_ops = icsk->icsk_ca_ops;
3399 		if (ca_ops && ca_ops->get_info)
3400 			sz = ca_ops->get_info(sk, ~0U, &attr, &info);
3401 
3402 		len = min_t(unsigned int, len, sz);
3403 		if (put_user(len, optlen))
3404 			return -EFAULT;
3405 		if (copy_to_user(optval, &info, len))
3406 			return -EFAULT;
3407 		return 0;
3408 	}
3409 	case TCP_QUICKACK:
3410 		val = !inet_csk_in_pingpong_mode(sk);
3411 		break;
3412 
3413 	case TCP_CONGESTION:
3414 		if (get_user(len, optlen))
3415 			return -EFAULT;
3416 		len = min_t(unsigned int, len, TCP_CA_NAME_MAX);
3417 		if (put_user(len, optlen))
3418 			return -EFAULT;
3419 		if (copy_to_user(optval, icsk->icsk_ca_ops->name, len))
3420 			return -EFAULT;
3421 		return 0;
3422 
3423 	case TCP_ULP:
3424 		if (get_user(len, optlen))
3425 			return -EFAULT;
3426 		len = min_t(unsigned int, len, TCP_ULP_NAME_MAX);
3427 		if (!icsk->icsk_ulp_ops) {
3428 			if (put_user(0, optlen))
3429 				return -EFAULT;
3430 			return 0;
3431 		}
3432 		if (put_user(len, optlen))
3433 			return -EFAULT;
3434 		if (copy_to_user(optval, icsk->icsk_ulp_ops->name, len))
3435 			return -EFAULT;
3436 		return 0;
3437 
3438 	case TCP_FASTOPEN_KEY: {
3439 		__u8 key[TCP_FASTOPEN_KEY_LENGTH];
3440 		struct tcp_fastopen_context *ctx;
3441 
3442 		if (get_user(len, optlen))
3443 			return -EFAULT;
3444 
3445 		rcu_read_lock();
3446 		ctx = rcu_dereference(icsk->icsk_accept_queue.fastopenq.ctx);
3447 		if (ctx)
3448 			memcpy(key, ctx->key, sizeof(key));
3449 		else
3450 			len = 0;
3451 		rcu_read_unlock();
3452 
3453 		len = min_t(unsigned int, len, sizeof(key));
3454 		if (put_user(len, optlen))
3455 			return -EFAULT;
3456 		if (copy_to_user(optval, key, len))
3457 			return -EFAULT;
3458 		return 0;
3459 	}
3460 	case TCP_THIN_LINEAR_TIMEOUTS:
3461 		val = tp->thin_lto;
3462 		break;
3463 
3464 	case TCP_THIN_DUPACK:
3465 		val = 0;
3466 		break;
3467 
3468 	case TCP_REPAIR:
3469 		val = tp->repair;
3470 		break;
3471 
3472 	case TCP_REPAIR_QUEUE:
3473 		if (tp->repair)
3474 			val = tp->repair_queue;
3475 		else
3476 			return -EINVAL;
3477 		break;
3478 
3479 	case TCP_REPAIR_WINDOW: {
3480 		struct tcp_repair_window opt;
3481 
3482 		if (get_user(len, optlen))
3483 			return -EFAULT;
3484 
3485 		if (len != sizeof(opt))
3486 			return -EINVAL;
3487 
3488 		if (!tp->repair)
3489 			return -EPERM;
3490 
3491 		opt.snd_wl1	= tp->snd_wl1;
3492 		opt.snd_wnd	= tp->snd_wnd;
3493 		opt.max_window	= tp->max_window;
3494 		opt.rcv_wnd	= tp->rcv_wnd;
3495 		opt.rcv_wup	= tp->rcv_wup;
3496 
3497 		if (copy_to_user(optval, &opt, len))
3498 			return -EFAULT;
3499 		return 0;
3500 	}
3501 	case TCP_QUEUE_SEQ:
3502 		if (tp->repair_queue == TCP_SEND_QUEUE)
3503 			val = tp->write_seq;
3504 		else if (tp->repair_queue == TCP_RECV_QUEUE)
3505 			val = tp->rcv_nxt;
3506 		else
3507 			return -EINVAL;
3508 		break;
3509 
3510 	case TCP_USER_TIMEOUT:
3511 		val = icsk->icsk_user_timeout;
3512 		break;
3513 
3514 	case TCP_FASTOPEN:
3515 		val = icsk->icsk_accept_queue.fastopenq.max_qlen;
3516 		break;
3517 
3518 	case TCP_FASTOPEN_CONNECT:
3519 		val = tp->fastopen_connect;
3520 		break;
3521 
3522 	case TCP_FASTOPEN_NO_COOKIE:
3523 		val = tp->fastopen_no_cookie;
3524 		break;
3525 
3526 	case TCP_TIMESTAMP:
3527 		val = tcp_time_stamp_raw() + tp->tsoffset;
3528 		break;
3529 	case TCP_NOTSENT_LOWAT:
3530 		val = tp->notsent_lowat;
3531 		break;
3532 	case TCP_INQ:
3533 		val = tp->recvmsg_inq;
3534 		break;
3535 	case TCP_SAVE_SYN:
3536 		val = tp->save_syn;
3537 		break;
3538 	case TCP_SAVED_SYN: {
3539 		if (get_user(len, optlen))
3540 			return -EFAULT;
3541 
3542 		lock_sock(sk);
3543 		if (tp->saved_syn) {
3544 			if (len < tp->saved_syn[0]) {
3545 				if (put_user(tp->saved_syn[0], optlen)) {
3546 					release_sock(sk);
3547 					return -EFAULT;
3548 				}
3549 				release_sock(sk);
3550 				return -EINVAL;
3551 			}
3552 			len = tp->saved_syn[0];
3553 			if (put_user(len, optlen)) {
3554 				release_sock(sk);
3555 				return -EFAULT;
3556 			}
3557 			if (copy_to_user(optval, tp->saved_syn + 1, len)) {
3558 				release_sock(sk);
3559 				return -EFAULT;
3560 			}
3561 			tcp_saved_syn_free(tp);
3562 			release_sock(sk);
3563 		} else {
3564 			release_sock(sk);
3565 			len = 0;
3566 			if (put_user(len, optlen))
3567 				return -EFAULT;
3568 		}
3569 		return 0;
3570 	}
3571 #ifdef CONFIG_MMU
3572 	case TCP_ZEROCOPY_RECEIVE: {
3573 		struct tcp_zerocopy_receive zc;
3574 		int err;
3575 
3576 		if (get_user(len, optlen))
3577 			return -EFAULT;
3578 		if (len != sizeof(zc))
3579 			return -EINVAL;
3580 		if (copy_from_user(&zc, optval, len))
3581 			return -EFAULT;
3582 		lock_sock(sk);
3583 		err = tcp_zerocopy_receive(sk, &zc);
3584 		release_sock(sk);
3585 		if (!err && copy_to_user(optval, &zc, len))
3586 			err = -EFAULT;
3587 		return err;
3588 	}
3589 #endif
3590 	default:
3591 		return -ENOPROTOOPT;
3592 	}
3593 
3594 	if (put_user(len, optlen))
3595 		return -EFAULT;
3596 	if (copy_to_user(optval, &val, len))
3597 		return -EFAULT;
3598 	return 0;
3599 }
3600 
3601 int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval,
3602 		   int __user *optlen)
3603 {
3604 	struct inet_connection_sock *icsk = inet_csk(sk);
3605 
3606 	if (level != SOL_TCP)
3607 		return icsk->icsk_af_ops->getsockopt(sk, level, optname,
3608 						     optval, optlen);
3609 	return do_tcp_getsockopt(sk, level, optname, optval, optlen);
3610 }
3611 EXPORT_SYMBOL(tcp_getsockopt);
3612 
3613 #ifdef CONFIG_COMPAT
3614 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
3615 			  char __user *optval, int __user *optlen)
3616 {
3617 	if (level != SOL_TCP)
3618 		return inet_csk_compat_getsockopt(sk, level, optname,
3619 						  optval, optlen);
3620 	return do_tcp_getsockopt(sk, level, optname, optval, optlen);
3621 }
3622 EXPORT_SYMBOL(compat_tcp_getsockopt);
3623 #endif
3624 
3625 #ifdef CONFIG_TCP_MD5SIG
3626 static DEFINE_PER_CPU(struct tcp_md5sig_pool, tcp_md5sig_pool);
3627 static DEFINE_MUTEX(tcp_md5sig_mutex);
3628 static bool tcp_md5sig_pool_populated = false;
3629 
3630 static void __tcp_alloc_md5sig_pool(void)
3631 {
3632 	struct crypto_ahash *hash;
3633 	int cpu;
3634 
3635 	hash = crypto_alloc_ahash("md5", 0, CRYPTO_ALG_ASYNC);
3636 	if (IS_ERR(hash))
3637 		return;
3638 
3639 	for_each_possible_cpu(cpu) {
3640 		void *scratch = per_cpu(tcp_md5sig_pool, cpu).scratch;
3641 		struct ahash_request *req;
3642 
3643 		if (!scratch) {
3644 			scratch = kmalloc_node(sizeof(union tcp_md5sum_block) +
3645 					       sizeof(struct tcphdr),
3646 					       GFP_KERNEL,
3647 					       cpu_to_node(cpu));
3648 			if (!scratch)
3649 				return;
3650 			per_cpu(tcp_md5sig_pool, cpu).scratch = scratch;
3651 		}
3652 		if (per_cpu(tcp_md5sig_pool, cpu).md5_req)
3653 			continue;
3654 
3655 		req = ahash_request_alloc(hash, GFP_KERNEL);
3656 		if (!req)
3657 			return;
3658 
3659 		ahash_request_set_callback(req, 0, NULL, NULL);
3660 
3661 		per_cpu(tcp_md5sig_pool, cpu).md5_req = req;
3662 	}
3663 	/* before setting tcp_md5sig_pool_populated, we must commit all writes
3664 	 * to memory. See smp_rmb() in tcp_get_md5sig_pool()
3665 	 */
3666 	smp_wmb();
3667 	tcp_md5sig_pool_populated = true;
3668 }
3669 
3670 bool tcp_alloc_md5sig_pool(void)
3671 {
3672 	if (unlikely(!tcp_md5sig_pool_populated)) {
3673 		mutex_lock(&tcp_md5sig_mutex);
3674 
3675 		if (!tcp_md5sig_pool_populated) {
3676 			__tcp_alloc_md5sig_pool();
3677 			if (tcp_md5sig_pool_populated)
3678 				static_key_slow_inc(&tcp_md5_needed);
3679 		}
3680 
3681 		mutex_unlock(&tcp_md5sig_mutex);
3682 	}
3683 	return tcp_md5sig_pool_populated;
3684 }
3685 EXPORT_SYMBOL(tcp_alloc_md5sig_pool);
3686 
3687 
3688 /**
3689  *	tcp_get_md5sig_pool - get md5sig_pool for this user
3690  *
3691  *	We use percpu structure, so if we succeed, we exit with preemption
3692  *	and BH disabled, to make sure another thread or softirq handling
3693  *	wont try to get same context.
3694  */
3695 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
3696 {
3697 	local_bh_disable();
3698 
3699 	if (tcp_md5sig_pool_populated) {
3700 		/* coupled with smp_wmb() in __tcp_alloc_md5sig_pool() */
3701 		smp_rmb();
3702 		return this_cpu_ptr(&tcp_md5sig_pool);
3703 	}
3704 	local_bh_enable();
3705 	return NULL;
3706 }
3707 EXPORT_SYMBOL(tcp_get_md5sig_pool);
3708 
3709 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp,
3710 			  const struct sk_buff *skb, unsigned int header_len)
3711 {
3712 	struct scatterlist sg;
3713 	const struct tcphdr *tp = tcp_hdr(skb);
3714 	struct ahash_request *req = hp->md5_req;
3715 	unsigned int i;
3716 	const unsigned int head_data_len = skb_headlen(skb) > header_len ?
3717 					   skb_headlen(skb) - header_len : 0;
3718 	const struct skb_shared_info *shi = skb_shinfo(skb);
3719 	struct sk_buff *frag_iter;
3720 
3721 	sg_init_table(&sg, 1);
3722 
3723 	sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len);
3724 	ahash_request_set_crypt(req, &sg, NULL, head_data_len);
3725 	if (crypto_ahash_update(req))
3726 		return 1;
3727 
3728 	for (i = 0; i < shi->nr_frags; ++i) {
3729 		const struct skb_frag_struct *f = &shi->frags[i];
3730 		unsigned int offset = f->page_offset;
3731 		struct page *page = skb_frag_page(f) + (offset >> PAGE_SHIFT);
3732 
3733 		sg_set_page(&sg, page, skb_frag_size(f),
3734 			    offset_in_page(offset));
3735 		ahash_request_set_crypt(req, &sg, NULL, skb_frag_size(f));
3736 		if (crypto_ahash_update(req))
3737 			return 1;
3738 	}
3739 
3740 	skb_walk_frags(skb, frag_iter)
3741 		if (tcp_md5_hash_skb_data(hp, frag_iter, 0))
3742 			return 1;
3743 
3744 	return 0;
3745 }
3746 EXPORT_SYMBOL(tcp_md5_hash_skb_data);
3747 
3748 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, const struct tcp_md5sig_key *key)
3749 {
3750 	struct scatterlist sg;
3751 
3752 	sg_init_one(&sg, key->key, key->keylen);
3753 	ahash_request_set_crypt(hp->md5_req, &sg, NULL, key->keylen);
3754 	return crypto_ahash_update(hp->md5_req);
3755 }
3756 EXPORT_SYMBOL(tcp_md5_hash_key);
3757 
3758 #endif
3759 
3760 void tcp_done(struct sock *sk)
3761 {
3762 	struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
3763 
3764 	if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
3765 		TCP_INC_STATS(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
3766 
3767 	tcp_set_state(sk, TCP_CLOSE);
3768 	tcp_clear_xmit_timers(sk);
3769 	if (req)
3770 		reqsk_fastopen_remove(sk, req, false);
3771 
3772 	sk->sk_shutdown = SHUTDOWN_MASK;
3773 
3774 	if (!sock_flag(sk, SOCK_DEAD))
3775 		sk->sk_state_change(sk);
3776 	else
3777 		inet_csk_destroy_sock(sk);
3778 }
3779 EXPORT_SYMBOL_GPL(tcp_done);
3780 
3781 int tcp_abort(struct sock *sk, int err)
3782 {
3783 	if (!sk_fullsock(sk)) {
3784 		if (sk->sk_state == TCP_NEW_SYN_RECV) {
3785 			struct request_sock *req = inet_reqsk(sk);
3786 
3787 			local_bh_disable();
3788 			inet_csk_reqsk_queue_drop(req->rsk_listener, req);
3789 			local_bh_enable();
3790 			return 0;
3791 		}
3792 		return -EOPNOTSUPP;
3793 	}
3794 
3795 	/* Don't race with userspace socket closes such as tcp_close. */
3796 	lock_sock(sk);
3797 
3798 	if (sk->sk_state == TCP_LISTEN) {
3799 		tcp_set_state(sk, TCP_CLOSE);
3800 		inet_csk_listen_stop(sk);
3801 	}
3802 
3803 	/* Don't race with BH socket closes such as inet_csk_listen_stop. */
3804 	local_bh_disable();
3805 	bh_lock_sock(sk);
3806 
3807 	if (!sock_flag(sk, SOCK_DEAD)) {
3808 		sk->sk_err = err;
3809 		/* This barrier is coupled with smp_rmb() in tcp_poll() */
3810 		smp_wmb();
3811 		sk->sk_error_report(sk);
3812 		if (tcp_need_reset(sk->sk_state))
3813 			tcp_send_active_reset(sk, GFP_ATOMIC);
3814 		tcp_done(sk);
3815 	}
3816 
3817 	bh_unlock_sock(sk);
3818 	local_bh_enable();
3819 	tcp_write_queue_purge(sk);
3820 	release_sock(sk);
3821 	return 0;
3822 }
3823 EXPORT_SYMBOL_GPL(tcp_abort);
3824 
3825 extern struct tcp_congestion_ops tcp_reno;
3826 
3827 static __initdata unsigned long thash_entries;
3828 static int __init set_thash_entries(char *str)
3829 {
3830 	ssize_t ret;
3831 
3832 	if (!str)
3833 		return 0;
3834 
3835 	ret = kstrtoul(str, 0, &thash_entries);
3836 	if (ret)
3837 		return 0;
3838 
3839 	return 1;
3840 }
3841 __setup("thash_entries=", set_thash_entries);
3842 
3843 static void __init tcp_init_mem(void)
3844 {
3845 	unsigned long limit = nr_free_buffer_pages() / 16;
3846 
3847 	limit = max(limit, 128UL);
3848 	sysctl_tcp_mem[0] = limit / 4 * 3;		/* 4.68 % */
3849 	sysctl_tcp_mem[1] = limit;			/* 6.25 % */
3850 	sysctl_tcp_mem[2] = sysctl_tcp_mem[0] * 2;	/* 9.37 % */
3851 }
3852 
3853 void __init tcp_init(void)
3854 {
3855 	int max_rshare, max_wshare, cnt;
3856 	unsigned long limit;
3857 	unsigned int i;
3858 
3859 	BUILD_BUG_ON(sizeof(struct tcp_skb_cb) >
3860 		     FIELD_SIZEOF(struct sk_buff, cb));
3861 
3862 	percpu_counter_init(&tcp_sockets_allocated, 0, GFP_KERNEL);
3863 	percpu_counter_init(&tcp_orphan_count, 0, GFP_KERNEL);
3864 	inet_hashinfo_init(&tcp_hashinfo);
3865 	inet_hashinfo2_init(&tcp_hashinfo, "tcp_listen_portaddr_hash",
3866 			    thash_entries, 21,  /* one slot per 2 MB*/
3867 			    0, 64 * 1024);
3868 	tcp_hashinfo.bind_bucket_cachep =
3869 		kmem_cache_create("tcp_bind_bucket",
3870 				  sizeof(struct inet_bind_bucket), 0,
3871 				  SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3872 
3873 	/* Size and allocate the main established and bind bucket
3874 	 * hash tables.
3875 	 *
3876 	 * The methodology is similar to that of the buffer cache.
3877 	 */
3878 	tcp_hashinfo.ehash =
3879 		alloc_large_system_hash("TCP established",
3880 					sizeof(struct inet_ehash_bucket),
3881 					thash_entries,
3882 					17, /* one slot per 128 KB of memory */
3883 					0,
3884 					NULL,
3885 					&tcp_hashinfo.ehash_mask,
3886 					0,
3887 					thash_entries ? 0 : 512 * 1024);
3888 	for (i = 0; i <= tcp_hashinfo.ehash_mask; i++)
3889 		INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i);
3890 
3891 	if (inet_ehash_locks_alloc(&tcp_hashinfo))
3892 		panic("TCP: failed to alloc ehash_locks");
3893 	tcp_hashinfo.bhash =
3894 		alloc_large_system_hash("TCP bind",
3895 					sizeof(struct inet_bind_hashbucket),
3896 					tcp_hashinfo.ehash_mask + 1,
3897 					17, /* one slot per 128 KB of memory */
3898 					0,
3899 					&tcp_hashinfo.bhash_size,
3900 					NULL,
3901 					0,
3902 					64 * 1024);
3903 	tcp_hashinfo.bhash_size = 1U << tcp_hashinfo.bhash_size;
3904 	for (i = 0; i < tcp_hashinfo.bhash_size; i++) {
3905 		spin_lock_init(&tcp_hashinfo.bhash[i].lock);
3906 		INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain);
3907 	}
3908 
3909 
3910 	cnt = tcp_hashinfo.ehash_mask + 1;
3911 	sysctl_tcp_max_orphans = cnt / 2;
3912 
3913 	tcp_init_mem();
3914 	/* Set per-socket limits to no more than 1/128 the pressure threshold */
3915 	limit = nr_free_buffer_pages() << (PAGE_SHIFT - 7);
3916 	max_wshare = min(4UL*1024*1024, limit);
3917 	max_rshare = min(6UL*1024*1024, limit);
3918 
3919 	init_net.ipv4.sysctl_tcp_wmem[0] = SK_MEM_QUANTUM;
3920 	init_net.ipv4.sysctl_tcp_wmem[1] = 16*1024;
3921 	init_net.ipv4.sysctl_tcp_wmem[2] = max(64*1024, max_wshare);
3922 
3923 	init_net.ipv4.sysctl_tcp_rmem[0] = SK_MEM_QUANTUM;
3924 	init_net.ipv4.sysctl_tcp_rmem[1] = 131072;
3925 	init_net.ipv4.sysctl_tcp_rmem[2] = max(131072, max_rshare);
3926 
3927 	pr_info("Hash tables configured (established %u bind %u)\n",
3928 		tcp_hashinfo.ehash_mask + 1, tcp_hashinfo.bhash_size);
3929 
3930 	tcp_v4_init();
3931 	tcp_metrics_init();
3932 	BUG_ON(tcp_register_congestion_control(&tcp_reno) != 0);
3933 	tcp_tasklet_init();
3934 }
3935