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
286struct percpu_counter tcp_orphan_count;
287EXPORT_SYMBOL_GPL(tcp_orphan_count);
288
289long sysctl_tcp_mem[3] __read_mostly;
290EXPORT_SYMBOL(sysctl_tcp_mem);
291
292atomic_long_t tcp_memory_allocated; /* Current allocated memory. */
293EXPORT_SYMBOL(tcp_memory_allocated);
294
295#if IS_ENABLED(CONFIG_SMC)
296DEFINE_STATIC_KEY_FALSE(tcp_have_smc);
297EXPORT_SYMBOL(tcp_have_smc);
298#endif
299
300/*
301 * Current number of TCP sockets.
302 */
303struct percpu_counter tcp_sockets_allocated;
304EXPORT_SYMBOL(tcp_sockets_allocated);
305
306/*
307 * TCP splice context
308 */
309struct 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 */
321unsigned long tcp_memory_pressure __read_mostly;
322EXPORT_SYMBOL_GPL(tcp_memory_pressure);
323
324void 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}
337EXPORT_SYMBOL_GPL(tcp_enter_memory_pressure);
338
339void 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}
350EXPORT_SYMBOL_GPL(tcp_leave_memory_pressure);
351
352/* Convert seconds to retransmits based on initial and max timeout */
353static 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 */
373static 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
389static 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 */
407void 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}
458EXPORT_SYMBOL(tcp_init_sock);
459
460void 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
472static 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
488static 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}
603EXPORT_SYMBOL(tcp_poll);
604
605int 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}
647EXPORT_SYMBOL(tcp_ioctl);
648
649static 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
655static 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
660static 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
679static 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 */
695static 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
704static 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
738static 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
751static 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 **/
774ssize_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}
861EXPORT_SYMBOL(tcp_splice_read);
862
863struct 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
902static 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
927static 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
937ssize_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 if (IS_ENABLED(CONFIG_DEBUG_VM) &&
947 WARN_ONCE(PageSlab(page), "page must not be a Slab one"))
948 return -EINVAL;
949
950 /* Wait for a connection to finish. One exception is TCP Fast Open
951 * (passive side) where data is allowed to be sent before a connection
952 * is fully established.
953 */
954 if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
955 !tcp_passive_fastopen(sk)) {
956 err = sk_stream_wait_connect(sk, &timeo);
957 if (err != 0)
958 goto out_err;
959 }
960
961 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
962
963 mss_now = tcp_send_mss(sk, &size_goal, flags);
964 copied = 0;
965
966 err = -EPIPE;
967 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
968 goto out_err;
969
970 while (size > 0) {
971 struct sk_buff *skb = tcp_write_queue_tail(sk);
972 int copy, i;
973 bool can_coalesce;
974
975 if (!skb || (copy = size_goal - skb->len) <= 0 ||
976 !tcp_skb_can_collapse_to(skb)) {
977new_segment:
978 if (!sk_stream_memory_free(sk))
979 goto wait_for_sndbuf;
980
981 skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation,
982 tcp_rtx_and_write_queues_empty(sk));
983 if (!skb)
984 goto wait_for_memory;
985
986 skb_entail(sk, skb);
987 copy = size_goal;
988 }
989
990 if (copy > size)
991 copy = size;
992
993 i = skb_shinfo(skb)->nr_frags;
994 can_coalesce = skb_can_coalesce(skb, i, page, offset);
995 if (!can_coalesce && i >= sysctl_max_skb_frags) {
996 tcp_mark_push(tp, skb);
997 goto new_segment;
998 }
999 if (!sk_wmem_schedule(sk, copy))
1000 goto wait_for_memory;
1001
1002 if (can_coalesce) {
1003 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1004 } else {
1005 get_page(page);
1006 skb_fill_page_desc(skb, i, page, offset, copy);
1007 }
1008
1009 if (!(flags & MSG_NO_SHARED_FRAGS))
1010 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1011
1012 skb->len += copy;
1013 skb->data_len += copy;
1014 skb->truesize += copy;
1015 sk->sk_wmem_queued += copy;
1016 sk_mem_charge(sk, copy);
1017 skb->ip_summed = CHECKSUM_PARTIAL;
1018 tp->write_seq += copy;
1019 TCP_SKB_CB(skb)->end_seq += copy;
1020 tcp_skb_pcount_set(skb, 0);
1021
1022 if (!copied)
1023 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1024
1025 copied += copy;
1026 offset += copy;
1027 size -= copy;
1028 if (!size)
1029 goto out;
1030
1031 if (skb->len < size_goal || (flags & MSG_OOB))
1032 continue;
1033
1034 if (forced_push(tp)) {
1035 tcp_mark_push(tp, skb);
1036 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
1037 } else if (skb == tcp_send_head(sk))
1038 tcp_push_one(sk, mss_now);
1039 continue;
1040
1041wait_for_sndbuf:
1042 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1043wait_for_memory:
1044 tcp_push(sk, flags & ~MSG_MORE, mss_now,
1045 TCP_NAGLE_PUSH, size_goal);
1046
1047 err = sk_stream_wait_memory(sk, &timeo);
1048 if (err != 0)
1049 goto do_error;
1050
1051 mss_now = tcp_send_mss(sk, &size_goal, flags);
1052 }
1053
1054out:
1055 if (copied) {
1056 tcp_tx_timestamp(sk, sk->sk_tsflags);
1057 if (!(flags & MSG_SENDPAGE_NOTLAST))
1058 tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
1059 }
1060 return copied;
1061
1062do_error:
1063 if (copied)
1064 goto out;
1065out_err:
1066 /* make sure we wake any epoll edge trigger waiter */
1067 if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 &&
1068 err == -EAGAIN)) {
1069 sk->sk_write_space(sk);
1070 tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED);
1071 }
1072 return sk_stream_error(sk, flags, err);
1073}
1074EXPORT_SYMBOL_GPL(do_tcp_sendpages);
1075
1076int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
1077 size_t size, int flags)
1078{
1079 if (!(sk->sk_route_caps & NETIF_F_SG))
1080 return sock_no_sendpage_locked(sk, page, offset, size, flags);
1081
1082 tcp_rate_check_app_limited(sk); /* is sending application-limited? */
1083
1084 return do_tcp_sendpages(sk, page, offset, size, flags);
1085}
1086EXPORT_SYMBOL_GPL(tcp_sendpage_locked);
1087
1088int tcp_sendpage(struct sock *sk, struct page *page, int offset,
1089 size_t size, int flags)
1090{
1091 int ret;
1092
1093 lock_sock(sk);
1094 ret = tcp_sendpage_locked(sk, page, offset, size, flags);
1095 release_sock(sk);
1096
1097 return ret;
1098}
1099EXPORT_SYMBOL(tcp_sendpage);
1100
1101/* Do not bother using a page frag for very small frames.
1102 * But use this heuristic only for the first skb in write queue.
1103 *
1104 * Having no payload in skb->head allows better SACK shifting
1105 * in tcp_shift_skb_data(), reducing sack/rack overhead, because
1106 * write queue has less skbs.
1107 * Each skb can hold up to MAX_SKB_FRAGS * 32Kbytes, or ~0.5 MB.
1108 * This also speeds up tso_fragment(), since it wont fallback
1109 * to tcp_fragment().
1110 */
1111static int linear_payload_sz(bool first_skb)
1112{
1113 if (first_skb)
1114 return SKB_WITH_OVERHEAD(2048 - MAX_TCP_HEADER);
1115 return 0;
1116}
1117
1118static int select_size(bool first_skb, bool zc)
1119{
1120 if (zc)
1121 return 0;
1122 return linear_payload_sz(first_skb);
1123}
1124
1125void tcp_free_fastopen_req(struct tcp_sock *tp)
1126{
1127 if (tp->fastopen_req) {
1128 kfree(tp->fastopen_req);
1129 tp->fastopen_req = NULL;
1130 }
1131}
1132
1133static int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg,
1134 int *copied, size_t size,
1135 struct ubuf_info *uarg)
1136{
1137 struct tcp_sock *tp = tcp_sk(sk);
1138 struct inet_sock *inet = inet_sk(sk);
1139 struct sockaddr *uaddr = msg->msg_name;
1140 int err, flags;
1141
1142 if (!(sock_net(sk)->ipv4.sysctl_tcp_fastopen & TFO_CLIENT_ENABLE) ||
1143 (uaddr && msg->msg_namelen >= sizeof(uaddr->sa_family) &&
1144 uaddr->sa_family == AF_UNSPEC))
1145 return -EOPNOTSUPP;
1146 if (tp->fastopen_req)
1147 return -EALREADY; /* Another Fast Open is in progress */
1148
1149 tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request),
1150 sk->sk_allocation);
1151 if (unlikely(!tp->fastopen_req))
1152 return -ENOBUFS;
1153 tp->fastopen_req->data = msg;
1154 tp->fastopen_req->size = size;
1155 tp->fastopen_req->uarg = uarg;
1156
1157 if (inet->defer_connect) {
1158 err = tcp_connect(sk);
1159 /* Same failure procedure as in tcp_v4/6_connect */
1160 if (err) {
1161 tcp_set_state(sk, TCP_CLOSE);
1162 inet->inet_dport = 0;
1163 sk->sk_route_caps = 0;
1164 }
1165 }
1166 flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0;
1167 err = __inet_stream_connect(sk->sk_socket, uaddr,
1168 msg->msg_namelen, flags, 1);
1169 /* fastopen_req could already be freed in __inet_stream_connect
1170 * if the connection times out or gets rst
1171 */
1172 if (tp->fastopen_req) {
1173 *copied = tp->fastopen_req->copied;
1174 tcp_free_fastopen_req(tp);
1175 inet->defer_connect = 0;
1176 }
1177 return err;
1178}
1179
1180int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
1181{
1182 struct tcp_sock *tp = tcp_sk(sk);
1183 struct ubuf_info *uarg = NULL;
1184 struct sk_buff *skb;
1185 struct sockcm_cookie sockc;
1186 int flags, err, copied = 0;
1187 int mss_now = 0, size_goal, copied_syn = 0;
1188 bool process_backlog = false;
1189 bool zc = false;
1190 long timeo;
1191
1192 flags = msg->msg_flags;
1193
1194 if (flags & MSG_ZEROCOPY && size && sock_flag(sk, SOCK_ZEROCOPY)) {
1195 skb = tcp_write_queue_tail(sk);
1196 uarg = sock_zerocopy_realloc(sk, size, skb_zcopy(skb));
1197 if (!uarg) {
1198 err = -ENOBUFS;
1199 goto out_err;
1200 }
1201
1202 zc = sk->sk_route_caps & NETIF_F_SG;
1203 if (!zc)
1204 uarg->zerocopy = 0;
1205 }
1206
1207 if (unlikely(flags & MSG_FASTOPEN || inet_sk(sk)->defer_connect) &&
1208 !tp->repair) {
1209 err = tcp_sendmsg_fastopen(sk, msg, &copied_syn, size, uarg);
1210 if (err == -EINPROGRESS && copied_syn > 0)
1211 goto out;
1212 else if (err)
1213 goto out_err;
1214 }
1215
1216 timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1217
1218 tcp_rate_check_app_limited(sk); /* is sending application-limited? */
1219
1220 /* Wait for a connection to finish. One exception is TCP Fast Open
1221 * (passive side) where data is allowed to be sent before a connection
1222 * is fully established.
1223 */
1224 if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
1225 !tcp_passive_fastopen(sk)) {
1226 err = sk_stream_wait_connect(sk, &timeo);
1227 if (err != 0)
1228 goto do_error;
1229 }
1230
1231 if (unlikely(tp->repair)) {
1232 if (tp->repair_queue == TCP_RECV_QUEUE) {
1233 copied = tcp_send_rcvq(sk, msg, size);
1234 goto out_nopush;
1235 }
1236
1237 err = -EINVAL;
1238 if (tp->repair_queue == TCP_NO_QUEUE)
1239 goto out_err;
1240
1241 /* 'common' sending to sendq */
1242 }
1243
1244 sockcm_init(&sockc, sk);
1245 if (msg->msg_controllen) {
1246 err = sock_cmsg_send(sk, msg, &sockc);
1247 if (unlikely(err)) {
1248 err = -EINVAL;
1249 goto out_err;
1250 }
1251 }
1252
1253 /* This should be in poll */
1254 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1255
1256 /* Ok commence sending. */
1257 copied = 0;
1258
1259restart:
1260 mss_now = tcp_send_mss(sk, &size_goal, flags);
1261
1262 err = -EPIPE;
1263 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
1264 goto do_error;
1265
1266 while (msg_data_left(msg)) {
1267 int copy = 0;
1268
1269 skb = tcp_write_queue_tail(sk);
1270 if (skb)
1271 copy = size_goal - skb->len;
1272
1273 if (copy <= 0 || !tcp_skb_can_collapse_to(skb)) {
1274 bool first_skb;
1275 int linear;
1276
1277new_segment:
1278 if (!sk_stream_memory_free(sk))
1279 goto wait_for_sndbuf;
1280
1281 if (process_backlog && sk_flush_backlog(sk)) {
1282 process_backlog = false;
1283 goto restart;
1284 }
1285 first_skb = tcp_rtx_and_write_queues_empty(sk);
1286 linear = select_size(first_skb, zc);
1287 skb = sk_stream_alloc_skb(sk, linear, sk->sk_allocation,
1288 first_skb);
1289 if (!skb)
1290 goto wait_for_memory;
1291
1292 process_backlog = true;
1293 skb->ip_summed = CHECKSUM_PARTIAL;
1294
1295 skb_entail(sk, skb);
1296 copy = size_goal;
1297
1298 /* All packets are restored as if they have
1299 * already been sent. skb_mstamp_ns isn't set to
1300 * avoid wrong rtt estimation.
1301 */
1302 if (tp->repair)
1303 TCP_SKB_CB(skb)->sacked |= TCPCB_REPAIRED;
1304 }
1305
1306 /* Try to append data to the end of skb. */
1307 if (copy > msg_data_left(msg))
1308 copy = msg_data_left(msg);
1309
1310 /* Where to copy to? */
1311 if (skb_availroom(skb) > 0 && !zc) {
1312 /* We have some space in skb head. Superb! */
1313 copy = min_t(int, copy, skb_availroom(skb));
1314 err = skb_add_data_nocache(sk, skb, &msg->msg_iter, copy);
1315 if (err)
1316 goto do_fault;
1317 } else if (!zc) {
1318 bool merge = true;
1319 int i = skb_shinfo(skb)->nr_frags;
1320 struct page_frag *pfrag = sk_page_frag(sk);
1321
1322 if (!sk_page_frag_refill(sk, pfrag))
1323 goto wait_for_memory;
1324
1325 if (!skb_can_coalesce(skb, i, pfrag->page,
1326 pfrag->offset)) {
1327 if (i >= sysctl_max_skb_frags) {
1328 tcp_mark_push(tp, skb);
1329 goto new_segment;
1330 }
1331 merge = false;
1332 }
1333
1334 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1335
1336 if (!sk_wmem_schedule(sk, copy))
1337 goto wait_for_memory;
1338
1339 err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
1340 pfrag->page,
1341 pfrag->offset,
1342 copy);
1343 if (err)
1344 goto do_error;
1345
1346 /* Update the skb. */
1347 if (merge) {
1348 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1349 } else {
1350 skb_fill_page_desc(skb, i, pfrag->page,
1351 pfrag->offset, copy);
1352 page_ref_inc(pfrag->page);
1353 }
1354 pfrag->offset += copy;
1355 } else {
1356 err = skb_zerocopy_iter_stream(sk, skb, msg, copy, uarg);
1357 if (err == -EMSGSIZE || err == -EEXIST) {
1358 tcp_mark_push(tp, skb);
1359 goto new_segment;
1360 }
1361 if (err < 0)
1362 goto do_error;
1363 copy = err;
1364 }
1365
1366 if (!copied)
1367 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1368
1369 tp->write_seq += copy;
1370 TCP_SKB_CB(skb)->end_seq += copy;
1371 tcp_skb_pcount_set(skb, 0);
1372
1373 copied += copy;
1374 if (!msg_data_left(msg)) {
1375 if (unlikely(flags & MSG_EOR))
1376 TCP_SKB_CB(skb)->eor = 1;
1377 goto out;
1378 }
1379
1380 if (skb->len < size_goal || (flags & MSG_OOB) || unlikely(tp->repair))
1381 continue;
1382
1383 if (forced_push(tp)) {
1384 tcp_mark_push(tp, skb);
1385 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
1386 } else if (skb == tcp_send_head(sk))
1387 tcp_push_one(sk, mss_now);
1388 continue;
1389
1390wait_for_sndbuf:
1391 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1392wait_for_memory:
1393 if (copied)
1394 tcp_push(sk, flags & ~MSG_MORE, mss_now,
1395 TCP_NAGLE_PUSH, size_goal);
1396
1397 err = sk_stream_wait_memory(sk, &timeo);
1398 if (err != 0)
1399 goto do_error;
1400
1401 mss_now = tcp_send_mss(sk, &size_goal, flags);
1402 }
1403
1404out:
1405 if (copied) {
1406 tcp_tx_timestamp(sk, sockc.tsflags);
1407 tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
1408 }
1409out_nopush:
1410 sock_zerocopy_put(uarg);
1411 return copied + copied_syn;
1412
1413do_fault:
1414 if (!skb->len) {
1415 tcp_unlink_write_queue(skb, sk);
1416 /* It is the one place in all of TCP, except connection
1417 * reset, where we can be unlinking the send_head.
1418 */
1419 if (tcp_write_queue_empty(sk))
1420 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1421 sk_wmem_free_skb(sk, skb);
1422 }
1423
1424do_error:
1425 if (copied + copied_syn)
1426 goto out;
1427out_err:
1428 sock_zerocopy_put_abort(uarg, true);
1429 err = sk_stream_error(sk, flags, err);
1430 /* make sure we wake any epoll edge trigger waiter */
1431 if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 &&
1432 err == -EAGAIN)) {
1433 sk->sk_write_space(sk);
1434 tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED);
1435 }
1436 return err;
1437}
1438EXPORT_SYMBOL_GPL(tcp_sendmsg_locked);
1439
1440int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
1441{
1442 int ret;
1443
1444 lock_sock(sk);
1445 ret = tcp_sendmsg_locked(sk, msg, size);
1446 release_sock(sk);
1447
1448 return ret;
1449}
1450EXPORT_SYMBOL(tcp_sendmsg);
1451
1452/*
1453 * Handle reading urgent data. BSD has very simple semantics for
1454 * this, no blocking and very strange errors 8)
1455 */
1456
1457static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags)
1458{
1459 struct tcp_sock *tp = tcp_sk(sk);
1460
1461 /* No URG data to read. */
1462 if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data ||
1463 tp->urg_data == TCP_URG_READ)
1464 return -EINVAL; /* Yes this is right ! */
1465
1466 if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE))
1467 return -ENOTCONN;
1468
1469 if (tp->urg_data & TCP_URG_VALID) {
1470 int err = 0;
1471 char c = tp->urg_data;
1472
1473 if (!(flags & MSG_PEEK))
1474 tp->urg_data = TCP_URG_READ;
1475
1476 /* Read urgent data. */
1477 msg->msg_flags |= MSG_OOB;
1478
1479 if (len > 0) {
1480 if (!(flags & MSG_TRUNC))
1481 err = memcpy_to_msg(msg, &c, 1);
1482 len = 1;
1483 } else
1484 msg->msg_flags |= MSG_TRUNC;
1485
1486 return err ? -EFAULT : len;
1487 }
1488
1489 if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
1490 return 0;
1491
1492 /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and
1493 * the available implementations agree in this case:
1494 * this call should never block, independent of the
1495 * blocking state of the socket.
1496 * Mike <pall@rz.uni-karlsruhe.de>
1497 */
1498 return -EAGAIN;
1499}
1500
1501static int tcp_peek_sndq(struct sock *sk, struct msghdr *msg, int len)
1502{
1503 struct sk_buff *skb;
1504 int copied = 0, err = 0;
1505
1506 /* XXX -- need to support SO_PEEK_OFF */
1507
1508 skb_rbtree_walk(skb, &sk->tcp_rtx_queue) {
1509 err = skb_copy_datagram_msg(skb, 0, msg, skb->len);
1510 if (err)
1511 return err;
1512 copied += skb->len;
1513 }
1514
1515 skb_queue_walk(&sk->sk_write_queue, skb) {
1516 err = skb_copy_datagram_msg(skb, 0, msg, skb->len);
1517 if (err)
1518 break;
1519
1520 copied += skb->len;
1521 }
1522
1523 return err ?: copied;
1524}
1525
1526/* Clean up the receive buffer for full frames taken by the user,
1527 * then send an ACK if necessary. COPIED is the number of bytes
1528 * tcp_recvmsg has given to the user so far, it speeds up the
1529 * calculation of whether or not we must ACK for the sake of
1530 * a window update.
1531 */
1532static void tcp_cleanup_rbuf(struct sock *sk, int copied)
1533{
1534 struct tcp_sock *tp = tcp_sk(sk);
1535 bool time_to_ack = false;
1536
1537 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
1538
1539 WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq),
1540 "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n",
1541 tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt);
1542
1543 if (inet_csk_ack_scheduled(sk)) {
1544 const struct inet_connection_sock *icsk = inet_csk(sk);
1545 /* Delayed ACKs frequently hit locked sockets during bulk
1546 * receive. */
1547 if (icsk->icsk_ack.blocked ||
1548 /* Once-per-two-segments ACK was not sent by tcp_input.c */
1549 tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss ||
1550 /*
1551 * If this read emptied read buffer, we send ACK, if
1552 * connection is not bidirectional, user drained
1553 * receive buffer and there was a small segment
1554 * in queue.
1555 */
1556 (copied > 0 &&
1557 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) ||
1558 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) &&
1559 !inet_csk_in_pingpong_mode(sk))) &&
1560 !atomic_read(&sk->sk_rmem_alloc)))
1561 time_to_ack = true;
1562 }
1563
1564 /* We send an ACK if we can now advertise a non-zero window
1565 * which has been raised "significantly".
1566 *
1567 * Even if window raised up to infinity, do not send window open ACK
1568 * in states, where we will not receive more. It is useless.
1569 */
1570 if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1571 __u32 rcv_window_now = tcp_receive_window(tp);
1572
1573 /* Optimize, __tcp_select_window() is not cheap. */
1574 if (2*rcv_window_now <= tp->window_clamp) {
1575 __u32 new_window = __tcp_select_window(sk);
1576
1577 /* Send ACK now, if this read freed lots of space
1578 * in our buffer. Certainly, new_window is new window.
1579 * We can advertise it now, if it is not less than current one.
1580 * "Lots" means "at least twice" here.
1581 */
1582 if (new_window && new_window >= 2 * rcv_window_now)
1583 time_to_ack = true;
1584 }
1585 }
1586 if (time_to_ack)
1587 tcp_send_ack(sk);
1588}
1589
1590static struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off)
1591{
1592 struct sk_buff *skb;
1593 u32 offset;
1594
1595 while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) {
1596 offset = seq - TCP_SKB_CB(skb)->seq;
1597 if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
1598 pr_err_once("%s: found a SYN, please report !\n", __func__);
1599 offset--;
1600 }
1601 if (offset < skb->len || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) {
1602 *off = offset;
1603 return skb;
1604 }
1605 /* This looks weird, but this can happen if TCP collapsing
1606 * splitted a fat GRO packet, while we released socket lock
1607 * in skb_splice_bits()
1608 */
1609 sk_eat_skb(sk, skb);
1610 }
1611 return NULL;
1612}
1613
1614/*
1615 * This routine provides an alternative to tcp_recvmsg() for routines
1616 * that would like to handle copying from skbuffs directly in 'sendfile'
1617 * fashion.
1618 * Note:
1619 * - It is assumed that the socket was locked by the caller.
1620 * - The routine does not block.
1621 * - At present, there is no support for reading OOB data
1622 * or for 'peeking' the socket using this routine
1623 * (although both would be easy to implement).
1624 */
1625int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
1626 sk_read_actor_t recv_actor)
1627{
1628 struct sk_buff *skb;
1629 struct tcp_sock *tp = tcp_sk(sk);
1630 u32 seq = tp->copied_seq;
1631 u32 offset;
1632 int copied = 0;
1633
1634 if (sk->sk_state == TCP_LISTEN)
1635 return -ENOTCONN;
1636 while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) {
1637 if (offset < skb->len) {
1638 int used;
1639 size_t len;
1640
1641 len = skb->len - offset;
1642 /* Stop reading if we hit a patch of urgent data */
1643 if (tp->urg_data) {
1644 u32 urg_offset = tp->urg_seq - seq;
1645 if (urg_offset < len)
1646 len = urg_offset;
1647 if (!len)
1648 break;
1649 }
1650 used = recv_actor(desc, skb, offset, len);
1651 if (used <= 0) {
1652 if (!copied)
1653 copied = used;
1654 break;
1655 } else if (used <= len) {
1656 seq += used;
1657 copied += used;
1658 offset += used;
1659 }
1660 /* If recv_actor drops the lock (e.g. TCP splice
1661 * receive) the skb pointer might be invalid when
1662 * getting here: tcp_collapse might have deleted it
1663 * while aggregating skbs from the socket queue.
1664 */
1665 skb = tcp_recv_skb(sk, seq - 1, &offset);
1666 if (!skb)
1667 break;
1668 /* TCP coalescing might have appended data to the skb.
1669 * Try to splice more frags
1670 */
1671 if (offset + 1 != skb->len)
1672 continue;
1673 }
1674 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) {
1675 sk_eat_skb(sk, skb);
1676 ++seq;
1677 break;
1678 }
1679 sk_eat_skb(sk, skb);
1680 if (!desc->count)
1681 break;
1682 tp->copied_seq = seq;
1683 }
1684 tp->copied_seq = seq;
1685
1686 tcp_rcv_space_adjust(sk);
1687
1688 /* Clean up data we have read: This will do ACK frames. */
1689 if (copied > 0) {
1690 tcp_recv_skb(sk, seq, &offset);
1691 tcp_cleanup_rbuf(sk, copied);
1692 }
1693 return copied;
1694}
1695EXPORT_SYMBOL(tcp_read_sock);
1696
1697int tcp_peek_len(struct socket *sock)
1698{
1699 return tcp_inq(sock->sk);
1700}
1701EXPORT_SYMBOL(tcp_peek_len);
1702
1703/* Make sure sk_rcvbuf is big enough to satisfy SO_RCVLOWAT hint */
1704int tcp_set_rcvlowat(struct sock *sk, int val)
1705{
1706 int cap;
1707
1708 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK)
1709 cap = sk->sk_rcvbuf >> 1;
1710 else
1711 cap = sock_net(sk)->ipv4.sysctl_tcp_rmem[2] >> 1;
1712 val = min(val, cap);
1713 sk->sk_rcvlowat = val ? : 1;
1714
1715 /* Check if we need to signal EPOLLIN right now */
1716 tcp_data_ready(sk);
1717
1718 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK)
1719 return 0;
1720
1721 val <<= 1;
1722 if (val > sk->sk_rcvbuf) {
1723 sk->sk_rcvbuf = val;
1724 tcp_sk(sk)->window_clamp = tcp_win_from_space(sk, val);
1725 }
1726 return 0;
1727}
1728EXPORT_SYMBOL(tcp_set_rcvlowat);
1729
1730#ifdef CONFIG_MMU
1731static const struct vm_operations_struct tcp_vm_ops = {
1732};
1733
1734int tcp_mmap(struct file *file, struct socket *sock,
1735 struct vm_area_struct *vma)
1736{
1737 if (vma->vm_flags & (VM_WRITE | VM_EXEC))
1738 return -EPERM;
1739 vma->vm_flags &= ~(VM_MAYWRITE | VM_MAYEXEC);
1740
1741 /* Instruct vm_insert_page() to not down_read(mmap_sem) */
1742 vma->vm_flags |= VM_MIXEDMAP;
1743
1744 vma->vm_ops = &tcp_vm_ops;
1745 return 0;
1746}
1747EXPORT_SYMBOL(tcp_mmap);
1748
1749static int tcp_zerocopy_receive(struct sock *sk,
1750 struct tcp_zerocopy_receive *zc)
1751{
1752 unsigned long address = (unsigned long)zc->address;
1753 const skb_frag_t *frags = NULL;
1754 u32 length = 0, seq, offset;
1755 struct vm_area_struct *vma;
1756 struct sk_buff *skb = NULL;
1757 struct tcp_sock *tp;
1758 int inq;
1759 int ret;
1760
1761 if (address & (PAGE_SIZE - 1) || address != zc->address)
1762 return -EINVAL;
1763
1764 if (sk->sk_state == TCP_LISTEN)
1765 return -ENOTCONN;
1766
1767 sock_rps_record_flow(sk);
1768
1769 down_read(&current->mm->mmap_sem);
1770
1771 ret = -EINVAL;
1772 vma = find_vma(current->mm, address);
1773 if (!vma || vma->vm_start > address || vma->vm_ops != &tcp_vm_ops)
1774 goto out;
1775 zc->length = min_t(unsigned long, zc->length, vma->vm_end - address);
1776
1777 tp = tcp_sk(sk);
1778 seq = tp->copied_seq;
1779 inq = tcp_inq(sk);
1780 zc->length = min_t(u32, zc->length, inq);
1781 zc->length &= ~(PAGE_SIZE - 1);
1782 if (zc->length) {
1783 zap_page_range(vma, address, zc->length);
1784 zc->recv_skip_hint = 0;
1785 } else {
1786 zc->recv_skip_hint = inq;
1787 }
1788 ret = 0;
1789 while (length + PAGE_SIZE <= zc->length) {
1790 if (zc->recv_skip_hint < PAGE_SIZE) {
1791 if (skb) {
1792 skb = skb->next;
1793 offset = seq - TCP_SKB_CB(skb)->seq;
1794 } else {
1795 skb = tcp_recv_skb(sk, seq, &offset);
1796 }
1797
1798 zc->recv_skip_hint = skb->len - offset;
1799 offset -= skb_headlen(skb);
1800 if ((int)offset < 0 || skb_has_frag_list(skb))
1801 break;
1802 frags = skb_shinfo(skb)->frags;
1803 while (offset) {
1804 if (frags->size > offset)
1805 goto out;
1806 offset -= frags->size;
1807 frags++;
1808 }
1809 }
1810 if (frags->size != PAGE_SIZE || frags->page_offset) {
1811 int remaining = zc->recv_skip_hint;
1812
1813 while (remaining && (frags->size != PAGE_SIZE ||
1814 frags->page_offset)) {
1815 remaining -= frags->size;
1816 frags++;
1817 }
1818 zc->recv_skip_hint -= remaining;
1819 break;
1820 }
1821 ret = vm_insert_page(vma, address + length,
1822 skb_frag_page(frags));
1823 if (ret)
1824 break;
1825 length += PAGE_SIZE;
1826 seq += PAGE_SIZE;
1827 zc->recv_skip_hint -= PAGE_SIZE;
1828 frags++;
1829 }
1830out:
1831 up_read(&current->mm->mmap_sem);
1832 if (length) {
1833 tp->copied_seq = seq;
1834 tcp_rcv_space_adjust(sk);
1835
1836 /* Clean up data we have read: This will do ACK frames. */
1837 tcp_recv_skb(sk, seq, &offset);
1838 tcp_cleanup_rbuf(sk, length);
1839 ret = 0;
1840 if (length == zc->length)
1841 zc->recv_skip_hint = 0;
1842 } else {
1843 if (!zc->recv_skip_hint && sock_flag(sk, SOCK_DONE))
1844 ret = -EIO;
1845 }
1846 zc->length = length;
1847 return ret;
1848}
1849#endif
1850
1851static void tcp_update_recv_tstamps(struct sk_buff *skb,
1852 struct scm_timestamping_internal *tss)
1853{
1854 if (skb->tstamp)
1855 tss->ts[0] = ktime_to_timespec64(skb->tstamp);
1856 else
1857 tss->ts[0] = (struct timespec64) {0};
1858
1859 if (skb_hwtstamps(skb)->hwtstamp)
1860 tss->ts[2] = ktime_to_timespec64(skb_hwtstamps(skb)->hwtstamp);
1861 else
1862 tss->ts[2] = (struct timespec64) {0};
1863}
1864
1865/* Similar to __sock_recv_timestamp, but does not require an skb */
1866static void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk,
1867 struct scm_timestamping_internal *tss)
1868{
1869 int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW);
1870 bool has_timestamping = false;
1871
1872 if (tss->ts[0].tv_sec || tss->ts[0].tv_nsec) {
1873 if (sock_flag(sk, SOCK_RCVTSTAMP)) {
1874 if (sock_flag(sk, SOCK_RCVTSTAMPNS)) {
1875 if (new_tstamp) {
1876 struct __kernel_timespec kts = {tss->ts[0].tv_sec, tss->ts[0].tv_nsec};
1877
1878 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW,
1879 sizeof(kts), &kts);
1880 } else {
1881 struct timespec ts_old = timespec64_to_timespec(tss->ts[0]);
1882
1883 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD,
1884 sizeof(ts_old), &ts_old);
1885 }
1886 } else {
1887 if (new_tstamp) {
1888 struct __kernel_sock_timeval stv;
1889
1890 stv.tv_sec = tss->ts[0].tv_sec;
1891 stv.tv_usec = tss->ts[0].tv_nsec / 1000;
1892 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW,
1893 sizeof(stv), &stv);
1894 } else {
1895 struct __kernel_old_timeval tv;
1896
1897 tv.tv_sec = tss->ts[0].tv_sec;
1898 tv.tv_usec = tss->ts[0].tv_nsec / 1000;
1899 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD,
1900 sizeof(tv), &tv);
1901 }
1902 }
1903 }
1904
1905 if (sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE)
1906 has_timestamping = true;
1907 else
1908 tss->ts[0] = (struct timespec64) {0};
1909 }
1910
1911 if (tss->ts[2].tv_sec || tss->ts[2].tv_nsec) {
1912 if (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)
1913 has_timestamping = true;
1914 else
1915 tss->ts[2] = (struct timespec64) {0};
1916 }
1917
1918 if (has_timestamping) {
1919 tss->ts[1] = (struct timespec64) {0};
1920 if (sock_flag(sk, SOCK_TSTAMP_NEW))
1921 put_cmsg_scm_timestamping64(msg, tss);
1922 else
1923 put_cmsg_scm_timestamping(msg, tss);
1924 }
1925}
1926
1927static int tcp_inq_hint(struct sock *sk)
1928{
1929 const struct tcp_sock *tp = tcp_sk(sk);
1930 u32 copied_seq = READ_ONCE(tp->copied_seq);
1931 u32 rcv_nxt = READ_ONCE(tp->rcv_nxt);
1932 int inq;
1933
1934 inq = rcv_nxt - copied_seq;
1935 if (unlikely(inq < 0 || copied_seq != READ_ONCE(tp->copied_seq))) {
1936 lock_sock(sk);
1937 inq = tp->rcv_nxt - tp->copied_seq;
1938 release_sock(sk);
1939 }
1940 /* After receiving a FIN, tell the user-space to continue reading
1941 * by returning a non-zero inq.
1942 */
1943 if (inq == 0 && sock_flag(sk, SOCK_DONE))
1944 inq = 1;
1945 return inq;
1946}
1947
1948/*
1949 * This routine copies from a sock struct into the user buffer.
1950 *
1951 * Technical note: in 2.3 we work on _locked_ socket, so that
1952 * tricks with *seq access order and skb->users are not required.
1953 * Probably, code can be easily improved even more.
1954 */
1955
1956int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
1957 int flags, int *addr_len)
1958{
1959 struct tcp_sock *tp = tcp_sk(sk);
1960 int copied = 0;
1961 u32 peek_seq;
1962 u32 *seq;
1963 unsigned long used;
1964 int err, inq;
1965 int target; /* Read at least this many bytes */
1966 long timeo;
1967 struct sk_buff *skb, *last;
1968 u32 urg_hole = 0;
1969 struct scm_timestamping_internal tss;
1970 bool has_tss = false;
1971 bool has_cmsg;
1972
1973 if (unlikely(flags & MSG_ERRQUEUE))
1974 return inet_recv_error(sk, msg, len, addr_len);
1975
1976 if (sk_can_busy_loop(sk) && skb_queue_empty(&sk->sk_receive_queue) &&
1977 (sk->sk_state == TCP_ESTABLISHED))
1978 sk_busy_loop(sk, nonblock);
1979
1980 lock_sock(sk);
1981
1982 err = -ENOTCONN;
1983 if (sk->sk_state == TCP_LISTEN)
1984 goto out;
1985
1986 has_cmsg = tp->recvmsg_inq;
1987 timeo = sock_rcvtimeo(sk, nonblock);
1988
1989 /* Urgent data needs to be handled specially. */
1990 if (flags & MSG_OOB)
1991 goto recv_urg;
1992
1993 if (unlikely(tp->repair)) {
1994 err = -EPERM;
1995 if (!(flags & MSG_PEEK))
1996 goto out;
1997
1998 if (tp->repair_queue == TCP_SEND_QUEUE)
1999 goto recv_sndq;
2000
2001 err = -EINVAL;
2002 if (tp->repair_queue == TCP_NO_QUEUE)
2003 goto out;
2004
2005 /* 'common' recv queue MSG_PEEK-ing */
2006 }
2007
2008 seq = &tp->copied_seq;
2009 if (flags & MSG_PEEK) {
2010 peek_seq = tp->copied_seq;
2011 seq = &peek_seq;
2012 }
2013
2014 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
2015
2016 do {
2017 u32 offset;
2018
2019 /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */
2020 if (tp->urg_data && tp->urg_seq == *seq) {
2021 if (copied)
2022 break;
2023 if (signal_pending(current)) {
2024 copied = timeo ? sock_intr_errno(timeo) : -EAGAIN;
2025 break;
2026 }
2027 }
2028
2029 /* Next get a buffer. */
2030
2031 last = skb_peek_tail(&sk->sk_receive_queue);
2032 skb_queue_walk(&sk->sk_receive_queue, skb) {
2033 last = skb;
2034 /* Now that we have two receive queues this
2035 * shouldn't happen.
2036 */
2037 if (WARN(before(*seq, TCP_SKB_CB(skb)->seq),
2038 "TCP recvmsg seq # bug: copied %X, seq %X, rcvnxt %X, fl %X\n",
2039 *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt,
2040 flags))
2041 break;
2042
2043 offset = *seq - TCP_SKB_CB(skb)->seq;
2044 if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
2045 pr_err_once("%s: found a SYN, please report !\n", __func__);
2046 offset--;
2047 }
2048 if (offset < skb->len)
2049 goto found_ok_skb;
2050 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
2051 goto found_fin_ok;
2052 WARN(!(flags & MSG_PEEK),
2053 "TCP recvmsg seq # bug 2: copied %X, seq %X, rcvnxt %X, fl %X\n",
2054 *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags);
2055 }
2056
2057 /* Well, if we have backlog, try to process it now yet. */
2058
2059 if (copied >= target && !sk->sk_backlog.tail)
2060 break;
2061
2062 if (copied) {
2063 if (sk->sk_err ||
2064 sk->sk_state == TCP_CLOSE ||
2065 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2066 !timeo ||
2067 signal_pending(current))
2068 break;
2069 } else {
2070 if (sock_flag(sk, SOCK_DONE))
2071 break;
2072
2073 if (sk->sk_err) {
2074 copied = sock_error(sk);
2075 break;
2076 }
2077
2078 if (sk->sk_shutdown & RCV_SHUTDOWN)
2079 break;
2080
2081 if (sk->sk_state == TCP_CLOSE) {
2082 /* This occurs when user tries to read
2083 * from never connected socket.
2084 */
2085 copied = -ENOTCONN;
2086 break;
2087 }
2088
2089 if (!timeo) {
2090 copied = -EAGAIN;
2091 break;
2092 }
2093
2094 if (signal_pending(current)) {
2095 copied = sock_intr_errno(timeo);
2096 break;
2097 }
2098 }
2099
2100 tcp_cleanup_rbuf(sk, copied);
2101
2102 if (copied >= target) {
2103 /* Do not sleep, just process backlog. */
2104 release_sock(sk);
2105 lock_sock(sk);
2106 } else {
2107 sk_wait_data(sk, &timeo, last);
2108 }
2109
2110 if ((flags & MSG_PEEK) &&
2111 (peek_seq - copied - urg_hole != tp->copied_seq)) {
2112 net_dbg_ratelimited("TCP(%s:%d): Application bug, race in MSG_PEEK\n",
2113 current->comm,
2114 task_pid_nr(current));
2115 peek_seq = tp->copied_seq;
2116 }
2117 continue;
2118
2119found_ok_skb:
2120 /* Ok so how much can we use? */
2121 used = skb->len - offset;
2122 if (len < used)
2123 used = len;
2124
2125 /* Do we have urgent data here? */
2126 if (tp->urg_data) {
2127 u32 urg_offset = tp->urg_seq - *seq;
2128 if (urg_offset < used) {
2129 if (!urg_offset) {
2130 if (!sock_flag(sk, SOCK_URGINLINE)) {
2131 ++*seq;
2132 urg_hole++;
2133 offset++;
2134 used--;
2135 if (!used)
2136 goto skip_copy;
2137 }
2138 } else
2139 used = urg_offset;
2140 }
2141 }
2142
2143 if (!(flags & MSG_TRUNC)) {
2144 err = skb_copy_datagram_msg(skb, offset, msg, used);
2145 if (err) {
2146 /* Exception. Bailout! */
2147 if (!copied)
2148 copied = -EFAULT;
2149 break;
2150 }
2151 }
2152
2153 *seq += used;
2154 copied += used;
2155 len -= used;
2156
2157 tcp_rcv_space_adjust(sk);
2158
2159skip_copy:
2160 if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) {
2161 tp->urg_data = 0;
2162 tcp_fast_path_check(sk);
2163 }
2164 if (used + offset < skb->len)
2165 continue;
2166
2167 if (TCP_SKB_CB(skb)->has_rxtstamp) {
2168 tcp_update_recv_tstamps(skb, &tss);
2169 has_tss = true;
2170 has_cmsg = true;
2171 }
2172 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
2173 goto found_fin_ok;
2174 if (!(flags & MSG_PEEK))
2175 sk_eat_skb(sk, skb);
2176 continue;
2177
2178found_fin_ok:
2179 /* Process the FIN. */
2180 ++*seq;
2181 if (!(flags & MSG_PEEK))
2182 sk_eat_skb(sk, skb);
2183 break;
2184 } while (len > 0);
2185
2186 /* According to UNIX98, msg_name/msg_namelen are ignored
2187 * on connected socket. I was just happy when found this 8) --ANK
2188 */
2189
2190 /* Clean up data we have read: This will do ACK frames. */
2191 tcp_cleanup_rbuf(sk, copied);
2192
2193 release_sock(sk);
2194
2195 if (has_cmsg) {
2196 if (has_tss)
2197 tcp_recv_timestamp(msg, sk, &tss);
2198 if (tp->recvmsg_inq) {
2199 inq = tcp_inq_hint(sk);
2200 put_cmsg(msg, SOL_TCP, TCP_CM_INQ, sizeof(inq), &inq);
2201 }
2202 }
2203
2204 return copied;
2205
2206out:
2207 release_sock(sk);
2208 return err;
2209
2210recv_urg:
2211 err = tcp_recv_urg(sk, msg, len, flags);
2212 goto out;
2213
2214recv_sndq:
2215 err = tcp_peek_sndq(sk, msg, len);
2216 goto out;
2217}
2218EXPORT_SYMBOL(tcp_recvmsg);
2219
2220void tcp_set_state(struct sock *sk, int state)
2221{
2222 int oldstate = sk->sk_state;
2223
2224 /* We defined a new enum for TCP states that are exported in BPF
2225 * so as not force the internal TCP states to be frozen. The
2226 * following checks will detect if an internal state value ever
2227 * differs from the BPF value. If this ever happens, then we will
2228 * need to remap the internal value to the BPF value before calling
2229 * tcp_call_bpf_2arg.
2230 */
2231 BUILD_BUG_ON((int)BPF_TCP_ESTABLISHED != (int)TCP_ESTABLISHED);
2232 BUILD_BUG_ON((int)BPF_TCP_SYN_SENT != (int)TCP_SYN_SENT);
2233 BUILD_BUG_ON((int)BPF_TCP_SYN_RECV != (int)TCP_SYN_RECV);
2234 BUILD_BUG_ON((int)BPF_TCP_FIN_WAIT1 != (int)TCP_FIN_WAIT1);
2235 BUILD_BUG_ON((int)BPF_TCP_FIN_WAIT2 != (int)TCP_FIN_WAIT2);
2236 BUILD_BUG_ON((int)BPF_TCP_TIME_WAIT != (int)TCP_TIME_WAIT);
2237 BUILD_BUG_ON((int)BPF_TCP_CLOSE != (int)TCP_CLOSE);
2238 BUILD_BUG_ON((int)BPF_TCP_CLOSE_WAIT != (int)TCP_CLOSE_WAIT);
2239 BUILD_BUG_ON((int)BPF_TCP_LAST_ACK != (int)TCP_LAST_ACK);
2240 BUILD_BUG_ON((int)BPF_TCP_LISTEN != (int)TCP_LISTEN);
2241 BUILD_BUG_ON((int)BPF_TCP_CLOSING != (int)TCP_CLOSING);
2242 BUILD_BUG_ON((int)BPF_TCP_NEW_SYN_RECV != (int)TCP_NEW_SYN_RECV);
2243 BUILD_BUG_ON((int)BPF_TCP_MAX_STATES != (int)TCP_MAX_STATES);
2244
2245 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_STATE_CB_FLAG))
2246 tcp_call_bpf_2arg(sk, BPF_SOCK_OPS_STATE_CB, oldstate, state);
2247
2248 switch (state) {
2249 case TCP_ESTABLISHED:
2250 if (oldstate != TCP_ESTABLISHED)
2251 TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
2252 break;
2253
2254 case TCP_CLOSE:
2255 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
2256 TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS);
2257
2258 sk->sk_prot->unhash(sk);
2259 if (inet_csk(sk)->icsk_bind_hash &&
2260 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
2261 inet_put_port(sk);
2262 /* fall through */
2263 default:
2264 if (oldstate == TCP_ESTABLISHED)
2265 TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
2266 }
2267
2268 /* Change state AFTER socket is unhashed to avoid closed
2269 * socket sitting in hash tables.
2270 */
2271 inet_sk_state_store(sk, state);
2272}
2273EXPORT_SYMBOL_GPL(tcp_set_state);
2274
2275/*
2276 * State processing on a close. This implements the state shift for
2277 * sending our FIN frame. Note that we only send a FIN for some
2278 * states. A shutdown() may have already sent the FIN, or we may be
2279 * closed.
2280 */
2281
2282static const unsigned char new_state[16] = {
2283 /* current state: new state: action: */
2284 [0 /* (Invalid) */] = TCP_CLOSE,
2285 [TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2286 [TCP_SYN_SENT] = TCP_CLOSE,
2287 [TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2288 [TCP_FIN_WAIT1] = TCP_FIN_WAIT1,
2289 [TCP_FIN_WAIT2] = TCP_FIN_WAIT2,
2290 [TCP_TIME_WAIT] = TCP_CLOSE,
2291 [TCP_CLOSE] = TCP_CLOSE,
2292 [TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN,
2293 [TCP_LAST_ACK] = TCP_LAST_ACK,
2294 [TCP_LISTEN] = TCP_CLOSE,
2295 [TCP_CLOSING] = TCP_CLOSING,
2296 [TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */
2297};
2298
2299static int tcp_close_state(struct sock *sk)
2300{
2301 int next = (int)new_state[sk->sk_state];
2302 int ns = next & TCP_STATE_MASK;
2303
2304 tcp_set_state(sk, ns);
2305
2306 return next & TCP_ACTION_FIN;
2307}
2308
2309/*
2310 * Shutdown the sending side of a connection. Much like close except
2311 * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD).
2312 */
2313
2314void tcp_shutdown(struct sock *sk, int how)
2315{
2316 /* We need to grab some memory, and put together a FIN,
2317 * and then put it into the queue to be sent.
2318 * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92.
2319 */
2320 if (!(how & SEND_SHUTDOWN))
2321 return;
2322
2323 /* If we've already sent a FIN, or it's a closed state, skip this. */
2324 if ((1 << sk->sk_state) &
2325 (TCPF_ESTABLISHED | TCPF_SYN_SENT |
2326 TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) {
2327 /* Clear out any half completed packets. FIN if needed. */
2328 if (tcp_close_state(sk))
2329 tcp_send_fin(sk);
2330 }
2331}
2332EXPORT_SYMBOL(tcp_shutdown);
2333
2334bool tcp_check_oom(struct sock *sk, int shift)
2335{
2336 bool too_many_orphans, out_of_socket_memory;
2337
2338 too_many_orphans = tcp_too_many_orphans(sk, shift);
2339 out_of_socket_memory = tcp_out_of_memory(sk);
2340
2341 if (too_many_orphans)
2342 net_info_ratelimited("too many orphaned sockets\n");
2343 if (out_of_socket_memory)
2344 net_info_ratelimited("out of memory -- consider tuning tcp_mem\n");
2345 return too_many_orphans || out_of_socket_memory;
2346}
2347
2348void tcp_close(struct sock *sk, long timeout)
2349{
2350 struct sk_buff *skb;
2351 int data_was_unread = 0;
2352 int state;
2353
2354 lock_sock(sk);
2355 sk->sk_shutdown = SHUTDOWN_MASK;
2356
2357 if (sk->sk_state == TCP_LISTEN) {
2358 tcp_set_state(sk, TCP_CLOSE);
2359
2360 /* Special case. */
2361 inet_csk_listen_stop(sk);
2362
2363 goto adjudge_to_death;
2364 }
2365
2366 /* We need to flush the recv. buffs. We do this only on the
2367 * descriptor close, not protocol-sourced closes, because the
2368 * reader process may not have drained the data yet!
2369 */
2370 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
2371 u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq;
2372
2373 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
2374 len--;
2375 data_was_unread += len;
2376 __kfree_skb(skb);
2377 }
2378
2379 sk_mem_reclaim(sk);
2380
2381 /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */
2382 if (sk->sk_state == TCP_CLOSE)
2383 goto adjudge_to_death;
2384
2385 /* As outlined in RFC 2525, section 2.17, we send a RST here because
2386 * data was lost. To witness the awful effects of the old behavior of
2387 * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk
2388 * GET in an FTP client, suspend the process, wait for the client to
2389 * advertise a zero window, then kill -9 the FTP client, wheee...
2390 * Note: timeout is always zero in such a case.
2391 */
2392 if (unlikely(tcp_sk(sk)->repair)) {
2393 sk->sk_prot->disconnect(sk, 0);
2394 } else if (data_was_unread) {
2395 /* Unread data was tossed, zap the connection. */
2396 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE);
2397 tcp_set_state(sk, TCP_CLOSE);
2398 tcp_send_active_reset(sk, sk->sk_allocation);
2399 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
2400 /* Check zero linger _after_ checking for unread data. */
2401 sk->sk_prot->disconnect(sk, 0);
2402 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONDATA);
2403 } else if (tcp_close_state(sk)) {
2404 /* We FIN if the application ate all the data before
2405 * zapping the connection.
2406 */
2407
2408 /* RED-PEN. Formally speaking, we have broken TCP state
2409 * machine. State transitions:
2410 *
2411 * TCP_ESTABLISHED -> TCP_FIN_WAIT1
2412 * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible)
2413 * TCP_CLOSE_WAIT -> TCP_LAST_ACK
2414 *
2415 * are legal only when FIN has been sent (i.e. in window),
2416 * rather than queued out of window. Purists blame.
2417 *
2418 * F.e. "RFC state" is ESTABLISHED,
2419 * if Linux state is FIN-WAIT-1, but FIN is still not sent.
2420 *
2421 * The visible declinations are that sometimes
2422 * we enter time-wait state, when it is not required really
2423 * (harmless), do not send active resets, when they are
2424 * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when
2425 * they look as CLOSING or LAST_ACK for Linux)
2426 * Probably, I missed some more holelets.
2427 * --ANK
2428 * XXX (TFO) - To start off we don't support SYN+ACK+FIN
2429 * in a single packet! (May consider it later but will
2430 * probably need API support or TCP_CORK SYN-ACK until
2431 * data is written and socket is closed.)
2432 */
2433 tcp_send_fin(sk);
2434 }
2435
2436 sk_stream_wait_close(sk, timeout);
2437
2438adjudge_to_death:
2439 state = sk->sk_state;
2440 sock_hold(sk);
2441 sock_orphan(sk);
2442
2443 local_bh_disable();
2444 bh_lock_sock(sk);
2445 /* remove backlog if any, without releasing ownership. */
2446 __release_sock(sk);
2447
2448 percpu_counter_inc(sk->sk_prot->orphan_count);
2449
2450 /* Have we already been destroyed by a softirq or backlog? */
2451 if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE)
2452 goto out;
2453
2454 /* This is a (useful) BSD violating of the RFC. There is a
2455 * problem with TCP as specified in that the other end could
2456 * keep a socket open forever with no application left this end.
2457 * We use a 1 minute timeout (about the same as BSD) then kill
2458 * our end. If they send after that then tough - BUT: long enough
2459 * that we won't make the old 4*rto = almost no time - whoops
2460 * reset mistake.
2461 *
2462 * Nope, it was not mistake. It is really desired behaviour
2463 * f.e. on http servers, when such sockets are useless, but
2464 * consume significant resources. Let's do it with special
2465 * linger2 option. --ANK
2466 */
2467
2468 if (sk->sk_state == TCP_FIN_WAIT2) {
2469 struct tcp_sock *tp = tcp_sk(sk);
2470 if (tp->linger2 < 0) {
2471 tcp_set_state(sk, TCP_CLOSE);
2472 tcp_send_active_reset(sk, GFP_ATOMIC);
2473 __NET_INC_STATS(sock_net(sk),
2474 LINUX_MIB_TCPABORTONLINGER);
2475 } else {
2476 const int tmo = tcp_fin_time(sk);
2477
2478 if (tmo > TCP_TIMEWAIT_LEN) {
2479 inet_csk_reset_keepalive_timer(sk,
2480 tmo - TCP_TIMEWAIT_LEN);
2481 } else {
2482 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
2483 goto out;
2484 }
2485 }
2486 }
2487 if (sk->sk_state != TCP_CLOSE) {
2488 sk_mem_reclaim(sk);
2489 if (tcp_check_oom(sk, 0)) {
2490 tcp_set_state(sk, TCP_CLOSE);
2491 tcp_send_active_reset(sk, GFP_ATOMIC);
2492 __NET_INC_STATS(sock_net(sk),
2493 LINUX_MIB_TCPABORTONMEMORY);
2494 } else if (!check_net(sock_net(sk))) {
2495 /* Not possible to send reset; just close */
2496 tcp_set_state(sk, TCP_CLOSE);
2497 }
2498 }
2499
2500 if (sk->sk_state == TCP_CLOSE) {
2501 struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
2502 /* We could get here with a non-NULL req if the socket is
2503 * aborted (e.g., closed with unread data) before 3WHS
2504 * finishes.
2505 */
2506 if (req)
2507 reqsk_fastopen_remove(sk, req, false);
2508 inet_csk_destroy_sock(sk);
2509 }
2510 /* Otherwise, socket is reprieved until protocol close. */
2511
2512out:
2513 bh_unlock_sock(sk);
2514 local_bh_enable();
2515 release_sock(sk);
2516 sock_put(sk);
2517}
2518EXPORT_SYMBOL(tcp_close);
2519
2520/* These states need RST on ABORT according to RFC793 */
2521
2522static inline bool tcp_need_reset(int state)
2523{
2524 return (1 << state) &
2525 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 |
2526 TCPF_FIN_WAIT2 | TCPF_SYN_RECV);
2527}
2528
2529static void tcp_rtx_queue_purge(struct sock *sk)
2530{
2531 struct rb_node *p = rb_first(&sk->tcp_rtx_queue);
2532
2533 while (p) {
2534 struct sk_buff *skb = rb_to_skb(p);
2535
2536 p = rb_next(p);
2537 /* Since we are deleting whole queue, no need to
2538 * list_del(&skb->tcp_tsorted_anchor)
2539 */
2540 tcp_rtx_queue_unlink(skb, sk);
2541 sk_wmem_free_skb(sk, skb);
2542 }
2543}
2544
2545void tcp_write_queue_purge(struct sock *sk)
2546{
2547 struct sk_buff *skb;
2548
2549 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
2550 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
2551 tcp_skb_tsorted_anchor_cleanup(skb);
2552 sk_wmem_free_skb(sk, skb);
2553 }
2554 tcp_rtx_queue_purge(sk);
2555 INIT_LIST_HEAD(&tcp_sk(sk)->tsorted_sent_queue);
2556 sk_mem_reclaim(sk);
2557 tcp_clear_all_retrans_hints(tcp_sk(sk));
2558 tcp_sk(sk)->packets_out = 0;
2559 inet_csk(sk)->icsk_backoff = 0;
2560}
2561
2562int tcp_disconnect(struct sock *sk, int flags)
2563{
2564 struct inet_sock *inet = inet_sk(sk);
2565 struct inet_connection_sock *icsk = inet_csk(sk);
2566 struct tcp_sock *tp = tcp_sk(sk);
2567 int old_state = sk->sk_state;
2568
2569 if (old_state != TCP_CLOSE)
2570 tcp_set_state(sk, TCP_CLOSE);
2571
2572 /* ABORT function of RFC793 */
2573 if (old_state == TCP_LISTEN) {
2574 inet_csk_listen_stop(sk);
2575 } else if (unlikely(tp->repair)) {
2576 sk->sk_err = ECONNABORTED;
2577 } else if (tcp_need_reset(old_state) ||
2578 (tp->snd_nxt != tp->write_seq &&
2579 (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) {
2580 /* The last check adjusts for discrepancy of Linux wrt. RFC
2581 * states
2582 */
2583 tcp_send_active_reset(sk, gfp_any());
2584 sk->sk_err = ECONNRESET;
2585 } else if (old_state == TCP_SYN_SENT)
2586 sk->sk_err = ECONNRESET;
2587
2588 tcp_clear_xmit_timers(sk);
2589 __skb_queue_purge(&sk->sk_receive_queue);
2590 tp->copied_seq = tp->rcv_nxt;
2591 tp->urg_data = 0;
2592 tcp_write_queue_purge(sk);
2593 tcp_fastopen_active_disable_ofo_check(sk);
2594 skb_rbtree_purge(&tp->out_of_order_queue);
2595
2596 inet->inet_dport = 0;
2597
2598 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
2599 inet_reset_saddr(sk);
2600
2601 sk->sk_shutdown = 0;
2602 sock_reset_flag(sk, SOCK_DONE);
2603 tp->srtt_us = 0;
2604 tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
2605 tp->rcv_rtt_last_tsecr = 0;
2606 tp->write_seq += tp->max_window + 2;
2607 if (tp->write_seq == 0)
2608 tp->write_seq = 1;
2609 icsk->icsk_backoff = 0;
2610 tp->snd_cwnd = 2;
2611 icsk->icsk_probes_out = 0;
2612 icsk->icsk_rto = TCP_TIMEOUT_INIT;
2613 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
2614 tp->snd_cwnd = TCP_INIT_CWND;
2615 tp->snd_cwnd_cnt = 0;
2616 tp->window_clamp = 0;
2617 tp->delivered_ce = 0;
2618 tcp_set_ca_state(sk, TCP_CA_Open);
2619 tp->is_sack_reneg = 0;
2620 tcp_clear_retrans(tp);
2621 inet_csk_delack_init(sk);
2622 /* Initialize rcv_mss to TCP_MIN_MSS to avoid division by 0
2623 * issue in __tcp_select_window()
2624 */
2625 icsk->icsk_ack.rcv_mss = TCP_MIN_MSS;
2626 memset(&tp->rx_opt, 0, sizeof(tp->rx_opt));
2627 __sk_dst_reset(sk);
2628 dst_release(sk->sk_rx_dst);
2629 sk->sk_rx_dst = NULL;
2630 tcp_saved_syn_free(tp);
2631 tp->compressed_ack = 0;
2632 tp->bytes_sent = 0;
2633 tp->bytes_retrans = 0;
2634 tp->duplicate_sack[0].start_seq = 0;
2635 tp->duplicate_sack[0].end_seq = 0;
2636 tp->dsack_dups = 0;
2637 tp->reord_seen = 0;
2638 tp->retrans_out = 0;
2639 tp->sacked_out = 0;
2640 tp->tlp_high_seq = 0;
2641 tp->last_oow_ack_time = 0;
2642 /* There's a bubble in the pipe until at least the first ACK. */
2643 tp->app_limited = ~0U;
2644 tp->rack.mstamp = 0;
2645 tp->rack.advanced = 0;
2646 tp->rack.reo_wnd_steps = 1;
2647 tp->rack.last_delivered = 0;
2648 tp->rack.reo_wnd_persist = 0;
2649 tp->rack.dsack_seen = 0;
2650 tp->syn_data_acked = 0;
2651 tp->rx_opt.saw_tstamp = 0;
2652 tp->rx_opt.dsack = 0;
2653 tp->rx_opt.num_sacks = 0;
2654
2655
2656 /* Clean up fastopen related fields */
2657 tcp_free_fastopen_req(tp);
2658 inet->defer_connect = 0;
2659
2660 WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
2661
2662 if (sk->sk_frag.page) {
2663 put_page(sk->sk_frag.page);
2664 sk->sk_frag.page = NULL;
2665 sk->sk_frag.offset = 0;
2666 }
2667
2668 sk->sk_error_report(sk);
2669 return 0;
2670}
2671EXPORT_SYMBOL(tcp_disconnect);
2672
2673static inline bool tcp_can_repair_sock(const struct sock *sk)
2674{
2675 return ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN) &&
2676 (sk->sk_state != TCP_LISTEN);
2677}
2678
2679static int tcp_repair_set_window(struct tcp_sock *tp, char __user *optbuf, int len)
2680{
2681 struct tcp_repair_window opt;
2682
2683 if (!tp->repair)
2684 return -EPERM;
2685
2686 if (len != sizeof(opt))
2687 return -EINVAL;
2688
2689 if (copy_from_user(&opt, optbuf, sizeof(opt)))
2690 return -EFAULT;
2691
2692 if (opt.max_window < opt.snd_wnd)
2693 return -EINVAL;
2694
2695 if (after(opt.snd_wl1, tp->rcv_nxt + opt.rcv_wnd))
2696 return -EINVAL;
2697
2698 if (after(opt.rcv_wup, tp->rcv_nxt))
2699 return -EINVAL;
2700
2701 tp->snd_wl1 = opt.snd_wl1;
2702 tp->snd_wnd = opt.snd_wnd;
2703 tp->max_window = opt.max_window;
2704
2705 tp->rcv_wnd = opt.rcv_wnd;
2706 tp->rcv_wup = opt.rcv_wup;
2707
2708 return 0;
2709}
2710
2711static int tcp_repair_options_est(struct sock *sk,
2712 struct tcp_repair_opt __user *optbuf, unsigned int len)
2713{
2714 struct tcp_sock *tp = tcp_sk(sk);
2715 struct tcp_repair_opt opt;
2716
2717 while (len >= sizeof(opt)) {
2718 if (copy_from_user(&opt, optbuf, sizeof(opt)))
2719 return -EFAULT;
2720
2721 optbuf++;
2722 len -= sizeof(opt);
2723
2724 switch (opt.opt_code) {
2725 case TCPOPT_MSS:
2726 tp->rx_opt.mss_clamp = opt.opt_val;
2727 tcp_mtup_init(sk);
2728 break;
2729 case TCPOPT_WINDOW:
2730 {
2731 u16 snd_wscale = opt.opt_val & 0xFFFF;
2732 u16 rcv_wscale = opt.opt_val >> 16;
2733
2734 if (snd_wscale > TCP_MAX_WSCALE || rcv_wscale > TCP_MAX_WSCALE)
2735 return -EFBIG;
2736
2737 tp->rx_opt.snd_wscale = snd_wscale;
2738 tp->rx_opt.rcv_wscale = rcv_wscale;
2739 tp->rx_opt.wscale_ok = 1;
2740 }
2741 break;
2742 case TCPOPT_SACK_PERM:
2743 if (opt.opt_val != 0)
2744 return -EINVAL;
2745
2746 tp->rx_opt.sack_ok |= TCP_SACK_SEEN;
2747 break;
2748 case TCPOPT_TIMESTAMP:
2749 if (opt.opt_val != 0)
2750 return -EINVAL;
2751
2752 tp->rx_opt.tstamp_ok = 1;
2753 break;
2754 }
2755 }
2756
2757 return 0;
2758}
2759
2760/*
2761 * Socket option code for TCP.
2762 */
2763static int do_tcp_setsockopt(struct sock *sk, int level,
2764 int optname, char __user *optval, unsigned int optlen)
2765{
2766 struct tcp_sock *tp = tcp_sk(sk);
2767 struct inet_connection_sock *icsk = inet_csk(sk);
2768 struct net *net = sock_net(sk);
2769 int val;
2770 int err = 0;
2771
2772 /* These are data/string values, all the others are ints */
2773 switch (optname) {
2774 case TCP_CONGESTION: {
2775 char name[TCP_CA_NAME_MAX];
2776
2777 if (optlen < 1)
2778 return -EINVAL;
2779
2780 val = strncpy_from_user(name, optval,
2781 min_t(long, TCP_CA_NAME_MAX-1, optlen));
2782 if (val < 0)
2783 return -EFAULT;
2784 name[val] = 0;
2785
2786 lock_sock(sk);
2787 err = tcp_set_congestion_control(sk, name, true, true);
2788 release_sock(sk);
2789 return err;
2790 }
2791 case TCP_ULP: {
2792 char name[TCP_ULP_NAME_MAX];
2793
2794 if (optlen < 1)
2795 return -EINVAL;
2796
2797 val = strncpy_from_user(name, optval,
2798 min_t(long, TCP_ULP_NAME_MAX - 1,
2799 optlen));
2800 if (val < 0)
2801 return -EFAULT;
2802 name[val] = 0;
2803
2804 lock_sock(sk);
2805 err = tcp_set_ulp(sk, name);
2806 release_sock(sk);
2807 return err;
2808 }
2809 case TCP_FASTOPEN_KEY: {
2810 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
2811
2812 if (optlen != sizeof(key))
2813 return -EINVAL;
2814
2815 if (copy_from_user(key, optval, optlen))
2816 return -EFAULT;
2817
2818 return tcp_fastopen_reset_cipher(net, sk, key, sizeof(key));
2819 }
2820 default:
2821 /* fallthru */
2822 break;
2823 }
2824
2825 if (optlen < sizeof(int))
2826 return -EINVAL;
2827
2828 if (get_user(val, (int __user *)optval))
2829 return -EFAULT;
2830
2831 lock_sock(sk);
2832
2833 switch (optname) {
2834 case TCP_MAXSEG:
2835 /* Values greater than interface MTU won't take effect. However
2836 * at the point when this call is done we typically don't yet
2837 * know which interface is going to be used
2838 */
2839 if (val && (val < TCP_MIN_MSS || val > MAX_TCP_WINDOW)) {
2840 err = -EINVAL;
2841 break;
2842 }
2843 tp->rx_opt.user_mss = val;
2844 break;
2845
2846 case TCP_NODELAY:
2847 if (val) {
2848 /* TCP_NODELAY is weaker than TCP_CORK, so that
2849 * this option on corked socket is remembered, but
2850 * it is not activated until cork is cleared.
2851 *
2852 * However, when TCP_NODELAY is set we make
2853 * an explicit push, which overrides even TCP_CORK
2854 * for currently queued segments.
2855 */
2856 tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH;
2857 tcp_push_pending_frames(sk);
2858 } else {
2859 tp->nonagle &= ~TCP_NAGLE_OFF;
2860 }
2861 break;
2862
2863 case TCP_THIN_LINEAR_TIMEOUTS:
2864 if (val < 0 || val > 1)
2865 err = -EINVAL;
2866 else
2867 tp->thin_lto = val;
2868 break;
2869
2870 case TCP_THIN_DUPACK:
2871 if (val < 0 || val > 1)
2872 err = -EINVAL;
2873 break;
2874
2875 case TCP_REPAIR:
2876 if (!tcp_can_repair_sock(sk))
2877 err = -EPERM;
2878 else if (val == TCP_REPAIR_ON) {
2879 tp->repair = 1;
2880 sk->sk_reuse = SK_FORCE_REUSE;
2881 tp->repair_queue = TCP_NO_QUEUE;
2882 } else if (val == TCP_REPAIR_OFF) {
2883 tp->repair = 0;
2884 sk->sk_reuse = SK_NO_REUSE;
2885 tcp_send_window_probe(sk);
2886 } else if (val == TCP_REPAIR_OFF_NO_WP) {
2887 tp->repair = 0;
2888 sk->sk_reuse = SK_NO_REUSE;
2889 } else
2890 err = -EINVAL;
2891
2892 break;
2893
2894 case TCP_REPAIR_QUEUE:
2895 if (!tp->repair)
2896 err = -EPERM;
2897 else if ((unsigned int)val < TCP_QUEUES_NR)
2898 tp->repair_queue = val;
2899 else
2900 err = -EINVAL;
2901 break;
2902
2903 case TCP_QUEUE_SEQ:
2904 if (sk->sk_state != TCP_CLOSE)
2905 err = -EPERM;
2906 else if (tp->repair_queue == TCP_SEND_QUEUE)
2907 tp->write_seq = val;
2908 else if (tp->repair_queue == TCP_RECV_QUEUE)
2909 tp->rcv_nxt = val;
2910 else
2911 err = -EINVAL;
2912 break;
2913
2914 case TCP_REPAIR_OPTIONS:
2915 if (!tp->repair)
2916 err = -EINVAL;
2917 else if (sk->sk_state == TCP_ESTABLISHED)
2918 err = tcp_repair_options_est(sk,
2919 (struct tcp_repair_opt __user *)optval,
2920 optlen);
2921 else
2922 err = -EPERM;
2923 break;
2924
2925 case TCP_CORK:
2926 /* When set indicates to always queue non-full frames.
2927 * Later the user clears this option and we transmit
2928 * any pending partial frames in the queue. This is
2929 * meant to be used alongside sendfile() to get properly
2930 * filled frames when the user (for example) must write
2931 * out headers with a write() call first and then use
2932 * sendfile to send out the data parts.
2933 *
2934 * TCP_CORK can be set together with TCP_NODELAY and it is
2935 * stronger than TCP_NODELAY.
2936 */
2937 if (val) {
2938 tp->nonagle |= TCP_NAGLE_CORK;
2939 } else {
2940 tp->nonagle &= ~TCP_NAGLE_CORK;
2941 if (tp->nonagle&TCP_NAGLE_OFF)
2942 tp->nonagle |= TCP_NAGLE_PUSH;
2943 tcp_push_pending_frames(sk);
2944 }
2945 break;
2946
2947 case TCP_KEEPIDLE:
2948 if (val < 1 || val > MAX_TCP_KEEPIDLE)
2949 err = -EINVAL;
2950 else {
2951 tp->keepalive_time = val * HZ;
2952 if (sock_flag(sk, SOCK_KEEPOPEN) &&
2953 !((1 << sk->sk_state) &
2954 (TCPF_CLOSE | TCPF_LISTEN))) {
2955 u32 elapsed = keepalive_time_elapsed(tp);
2956 if (tp->keepalive_time > elapsed)
2957 elapsed = tp->keepalive_time - elapsed;
2958 else
2959 elapsed = 0;
2960 inet_csk_reset_keepalive_timer(sk, elapsed);
2961 }
2962 }
2963 break;
2964 case TCP_KEEPINTVL:
2965 if (val < 1 || val > MAX_TCP_KEEPINTVL)
2966 err = -EINVAL;
2967 else
2968 tp->keepalive_intvl = val * HZ;
2969 break;
2970 case TCP_KEEPCNT:
2971 if (val < 1 || val > MAX_TCP_KEEPCNT)
2972 err = -EINVAL;
2973 else
2974 tp->keepalive_probes = val;
2975 break;
2976 case TCP_SYNCNT:
2977 if (val < 1 || val > MAX_TCP_SYNCNT)
2978 err = -EINVAL;
2979 else
2980 icsk->icsk_syn_retries = val;
2981 break;
2982
2983 case TCP_SAVE_SYN:
2984 if (val < 0 || val > 1)
2985 err = -EINVAL;
2986 else
2987 tp->save_syn = val;
2988 break;
2989
2990 case TCP_LINGER2:
2991 if (val < 0)
2992 tp->linger2 = -1;
2993 else if (val > net->ipv4.sysctl_tcp_fin_timeout / HZ)
2994 tp->linger2 = 0;
2995 else
2996 tp->linger2 = val * HZ;
2997 break;
2998
2999 case TCP_DEFER_ACCEPT:
3000 /* Translate value in seconds to number of retransmits */
3001 icsk->icsk_accept_queue.rskq_defer_accept =
3002 secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ,
3003 TCP_RTO_MAX / HZ);
3004 break;
3005
3006 case TCP_WINDOW_CLAMP:
3007 if (!val) {
3008 if (sk->sk_state != TCP_CLOSE) {
3009 err = -EINVAL;
3010 break;
3011 }
3012 tp->window_clamp = 0;
3013 } else
3014 tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ?
3015 SOCK_MIN_RCVBUF / 2 : val;
3016 break;
3017
3018 case TCP_QUICKACK:
3019 if (!val) {
3020 inet_csk_enter_pingpong_mode(sk);
3021 } else {
3022 inet_csk_exit_pingpong_mode(sk);
3023 if ((1 << sk->sk_state) &
3024 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) &&
3025 inet_csk_ack_scheduled(sk)) {
3026 icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;
3027 tcp_cleanup_rbuf(sk, 1);
3028 if (!(val & 1))
3029 inet_csk_enter_pingpong_mode(sk);
3030 }
3031 }
3032 break;
3033
3034#ifdef CONFIG_TCP_MD5SIG
3035 case TCP_MD5SIG:
3036 case TCP_MD5SIG_EXT:
3037 if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
3038 err = tp->af_specific->md5_parse(sk, optname, optval, optlen);
3039 else
3040 err = -EINVAL;
3041 break;
3042#endif
3043 case TCP_USER_TIMEOUT:
3044 /* Cap the max time in ms TCP will retry or probe the window
3045 * before giving up and aborting (ETIMEDOUT) a connection.
3046 */
3047 if (val < 0)
3048 err = -EINVAL;
3049 else
3050 icsk->icsk_user_timeout = val;
3051 break;
3052
3053 case TCP_FASTOPEN:
3054 if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE |
3055 TCPF_LISTEN))) {
3056 tcp_fastopen_init_key_once(net);
3057
3058 fastopen_queue_tune(sk, val);
3059 } else {
3060 err = -EINVAL;
3061 }
3062 break;
3063 case TCP_FASTOPEN_CONNECT:
3064 if (val > 1 || val < 0) {
3065 err = -EINVAL;
3066 } else if (net->ipv4.sysctl_tcp_fastopen & TFO_CLIENT_ENABLE) {
3067 if (sk->sk_state == TCP_CLOSE)
3068 tp->fastopen_connect = val;
3069 else
3070 err = -EINVAL;
3071 } else {
3072 err = -EOPNOTSUPP;
3073 }
3074 break;
3075 case TCP_FASTOPEN_NO_COOKIE:
3076 if (val > 1 || val < 0)
3077 err = -EINVAL;
3078 else if (!((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
3079 err = -EINVAL;
3080 else
3081 tp->fastopen_no_cookie = val;
3082 break;
3083 case TCP_TIMESTAMP:
3084 if (!tp->repair)
3085 err = -EPERM;
3086 else
3087 tp->tsoffset = val - tcp_time_stamp_raw();
3088 break;
3089 case TCP_REPAIR_WINDOW:
3090 err = tcp_repair_set_window(tp, optval, optlen);
3091 break;
3092 case TCP_NOTSENT_LOWAT:
3093 tp->notsent_lowat = val;
3094 sk->sk_write_space(sk);
3095 break;
3096 case TCP_INQ:
3097 if (val > 1 || val < 0)
3098 err = -EINVAL;
3099 else
3100 tp->recvmsg_inq = val;
3101 break;
3102 default:
3103 err = -ENOPROTOOPT;
3104 break;
3105 }
3106
3107 release_sock(sk);
3108 return err;
3109}
3110
3111int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval,
3112 unsigned int optlen)
3113{
3114 const struct inet_connection_sock *icsk = inet_csk(sk);
3115
3116 if (level != SOL_TCP)
3117 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
3118 optval, optlen);
3119 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
3120}
3121EXPORT_SYMBOL(tcp_setsockopt);
3122
3123#ifdef CONFIG_COMPAT
3124int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
3125 char __user *optval, unsigned int optlen)
3126{
3127 if (level != SOL_TCP)
3128 return inet_csk_compat_setsockopt(sk, level, optname,
3129 optval, optlen);
3130 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
3131}
3132EXPORT_SYMBOL(compat_tcp_setsockopt);
3133#endif
3134
3135static void tcp_get_info_chrono_stats(const struct tcp_sock *tp,
3136 struct tcp_info *info)
3137{
3138 u64 stats[__TCP_CHRONO_MAX], total = 0;
3139 enum tcp_chrono i;
3140
3141 for (i = TCP_CHRONO_BUSY; i < __TCP_CHRONO_MAX; ++i) {
3142 stats[i] = tp->chrono_stat[i - 1];
3143 if (i == tp->chrono_type)
3144 stats[i] += tcp_jiffies32 - tp->chrono_start;
3145 stats[i] *= USEC_PER_SEC / HZ;
3146 total += stats[i];
3147 }
3148
3149 info->tcpi_busy_time = total;
3150 info->tcpi_rwnd_limited = stats[TCP_CHRONO_RWND_LIMITED];
3151 info->tcpi_sndbuf_limited = stats[TCP_CHRONO_SNDBUF_LIMITED];
3152}
3153
3154/* Return information about state of tcp endpoint in API format. */
3155void tcp_get_info(struct sock *sk, struct tcp_info *info)
3156{
3157 const struct tcp_sock *tp = tcp_sk(sk); /* iff sk_type == SOCK_STREAM */
3158 const struct inet_connection_sock *icsk = inet_csk(sk);
3159 unsigned long rate;
3160 u32 now;
3161 u64 rate64;
3162 bool slow;
3163
3164 memset(info, 0, sizeof(*info));
3165 if (sk->sk_type != SOCK_STREAM)
3166 return;
3167
3168 info->tcpi_state = inet_sk_state_load(sk);
3169
3170 /* Report meaningful fields for all TCP states, including listeners */
3171 rate = READ_ONCE(sk->sk_pacing_rate);
3172 rate64 = (rate != ~0UL) ? rate : ~0ULL;
3173 info->tcpi_pacing_rate = rate64;
3174
3175 rate = READ_ONCE(sk->sk_max_pacing_rate);
3176 rate64 = (rate != ~0UL) ? rate : ~0ULL;
3177 info->tcpi_max_pacing_rate = rate64;
3178
3179 info->tcpi_reordering = tp->reordering;
3180 info->tcpi_snd_cwnd = tp->snd_cwnd;
3181
3182 if (info->tcpi_state == TCP_LISTEN) {
3183 /* listeners aliased fields :
3184 * tcpi_unacked -> Number of children ready for accept()
3185 * tcpi_sacked -> max backlog
3186 */
3187 info->tcpi_unacked = sk->sk_ack_backlog;
3188 info->tcpi_sacked = sk->sk_max_ack_backlog;
3189 return;
3190 }
3191
3192 slow = lock_sock_fast(sk);
3193
3194 info->tcpi_ca_state = icsk->icsk_ca_state;
3195 info->tcpi_retransmits = icsk->icsk_retransmits;
3196 info->tcpi_probes = icsk->icsk_probes_out;
3197 info->tcpi_backoff = icsk->icsk_backoff;
3198
3199 if (tp->rx_opt.tstamp_ok)
3200 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
3201 if (tcp_is_sack(tp))
3202 info->tcpi_options |= TCPI_OPT_SACK;
3203 if (tp->rx_opt.wscale_ok) {
3204 info->tcpi_options |= TCPI_OPT_WSCALE;
3205 info->tcpi_snd_wscale = tp->rx_opt.snd_wscale;
3206 info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale;
3207 }
3208
3209 if (tp->ecn_flags & TCP_ECN_OK)
3210 info->tcpi_options |= TCPI_OPT_ECN;
3211 if (tp->ecn_flags & TCP_ECN_SEEN)
3212 info->tcpi_options |= TCPI_OPT_ECN_SEEN;
3213 if (tp->syn_data_acked)
3214 info->tcpi_options |= TCPI_OPT_SYN_DATA;
3215
3216 info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto);
3217 info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato);
3218 info->tcpi_snd_mss = tp->mss_cache;
3219 info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss;
3220
3221 info->tcpi_unacked = tp->packets_out;
3222 info->tcpi_sacked = tp->sacked_out;
3223
3224 info->tcpi_lost = tp->lost_out;
3225 info->tcpi_retrans = tp->retrans_out;
3226
3227 now = tcp_jiffies32;
3228 info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime);
3229 info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime);