]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
3 | * operating system. INET is implemented using the BSD Socket | |
4 | * interface as the means of communication with the user level. | |
5 | * | |
6 | * Implementation of the Transmission Control Protocol(TCP). | |
7 | * | |
02c30a84 | 8 | * Authors: Ross Biro |
1da177e4 LT |
9 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
10 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
11 | * Corey Minyard <wf-rch!minyard@relay.EU.net> | |
12 | * Florian La Roche, <flla@stud.uni-sb.de> | |
13 | * Charles Hedrick, <hedrick@klinzhai.rutgers.edu> | |
14 | * Linus Torvalds, <torvalds@cs.helsinki.fi> | |
15 | * Alan Cox, <gw4pts@gw4pts.ampr.org> | |
16 | * Matthew Dillon, <dillon@apollo.west.oic.com> | |
17 | * Arnt Gulbrandsen, <agulbra@nvg.unit.no> | |
18 | * Jorge Cwik, <jorge@laser.satlink.net> | |
19 | * | |
20 | * Fixes: | |
21 | * Alan Cox : Numerous verify_area() calls | |
22 | * Alan Cox : Set the ACK bit on a reset | |
23 | * Alan Cox : Stopped it crashing if it closed while | |
24 | * sk->inuse=1 and was trying to connect | |
25 | * (tcp_err()). | |
26 | * Alan Cox : All icmp error handling was broken | |
27 | * pointers passed where wrong and the | |
28 | * socket was looked up backwards. Nobody | |
29 | * tested any icmp error code obviously. | |
30 | * Alan Cox : tcp_err() now handled properly. It | |
31 | * wakes people on errors. poll | |
32 | * behaves and the icmp error race | |
33 | * has gone by moving it into sock.c | |
34 | * Alan Cox : tcp_send_reset() fixed to work for | |
35 | * everything not just packets for | |
36 | * unknown sockets. | |
37 | * Alan Cox : tcp option processing. | |
38 | * Alan Cox : Reset tweaked (still not 100%) [Had | |
39 | * syn rule wrong] | |
40 | * Herp Rosmanith : More reset fixes | |
41 | * Alan Cox : No longer acks invalid rst frames. | |
42 | * Acking any kind of RST is right out. | |
43 | * Alan Cox : Sets an ignore me flag on an rst | |
44 | * receive otherwise odd bits of prattle | |
45 | * escape still | |
46 | * Alan Cox : Fixed another acking RST frame bug. | |
47 | * Should stop LAN workplace lockups. | |
48 | * Alan Cox : Some tidyups using the new skb list | |
49 | * facilities | |
50 | * Alan Cox : sk->keepopen now seems to work | |
51 | * Alan Cox : Pulls options out correctly on accepts | |
52 | * Alan Cox : Fixed assorted sk->rqueue->next errors | |
53 | * Alan Cox : PSH doesn't end a TCP read. Switched a | |
54 | * bit to skb ops. | |
55 | * Alan Cox : Tidied tcp_data to avoid a potential | |
56 | * nasty. | |
57 | * Alan Cox : Added some better commenting, as the | |
58 | * tcp is hard to follow | |
59 | * Alan Cox : Removed incorrect check for 20 * psh | |
60 | * Michael O'Reilly : ack < copied bug fix. | |
61 | * Johannes Stille : Misc tcp fixes (not all in yet). | |
62 | * Alan Cox : FIN with no memory -> CRASH | |
63 | * Alan Cox : Added socket option proto entries. | |
64 | * Also added awareness of them to accept. | |
65 | * Alan Cox : Added TCP options (SOL_TCP) | |
66 | * Alan Cox : Switched wakeup calls to callbacks, | |
67 | * so the kernel can layer network | |
68 | * sockets. | |
69 | * Alan Cox : Use ip_tos/ip_ttl settings. | |
70 | * Alan Cox : Handle FIN (more) properly (we hope). | |
71 | * Alan Cox : RST frames sent on unsynchronised | |
72 | * state ack error. | |
73 | * Alan Cox : Put in missing check for SYN bit. | |
74 | * Alan Cox : Added tcp_select_window() aka NET2E | |
75 | * window non shrink trick. | |
76 | * Alan Cox : Added a couple of small NET2E timer | |
77 | * fixes | |
78 | * Charles Hedrick : TCP fixes | |
79 | * Toomas Tamm : TCP window fixes | |
80 | * Alan Cox : Small URG fix to rlogin ^C ack fight | |
81 | * Charles Hedrick : Rewrote most of it to actually work | |
82 | * Linus : Rewrote tcp_read() and URG handling | |
83 | * completely | |
84 | * Gerhard Koerting: Fixed some missing timer handling | |
85 | * Matthew Dillon : Reworked TCP machine states as per RFC | |
86 | * Gerhard Koerting: PC/TCP workarounds | |
87 | * Adam Caldwell : Assorted timer/timing errors | |
88 | * Matthew Dillon : Fixed another RST bug | |
89 | * Alan Cox : Move to kernel side addressing changes. | |
90 | * Alan Cox : Beginning work on TCP fastpathing | |
91 | * (not yet usable) | |
92 | * Arnt Gulbrandsen: Turbocharged tcp_check() routine. | |
93 | * Alan Cox : TCP fast path debugging | |
94 | * Alan Cox : Window clamping | |
95 | * Michael Riepe : Bug in tcp_check() | |
96 | * Matt Dillon : More TCP improvements and RST bug fixes | |
97 | * Matt Dillon : Yet more small nasties remove from the | |
98 | * TCP code (Be very nice to this man if | |
99 | * tcp finally works 100%) 8) | |
100 | * Alan Cox : BSD accept semantics. | |
101 | * Alan Cox : Reset on closedown bug. | |
102 | * Peter De Schrijver : ENOTCONN check missing in tcp_sendto(). | |
103 | * Michael Pall : Handle poll() after URG properly in | |
104 | * all cases. | |
105 | * Michael Pall : Undo the last fix in tcp_read_urg() | |
106 | * (multi URG PUSH broke rlogin). | |
107 | * Michael Pall : Fix the multi URG PUSH problem in | |
108 | * tcp_readable(), poll() after URG | |
109 | * works now. | |
110 | * Michael Pall : recv(...,MSG_OOB) never blocks in the | |
111 | * BSD api. | |
112 | * Alan Cox : Changed the semantics of sk->socket to | |
113 | * fix a race and a signal problem with | |
114 | * accept() and async I/O. | |
115 | * Alan Cox : Relaxed the rules on tcp_sendto(). | |
116 | * Yury Shevchuk : Really fixed accept() blocking problem. | |
117 | * Craig I. Hagan : Allow for BSD compatible TIME_WAIT for | |
118 | * clients/servers which listen in on | |
119 | * fixed ports. | |
120 | * Alan Cox : Cleaned the above up and shrank it to | |
121 | * a sensible code size. | |
122 | * Alan Cox : Self connect lockup fix. | |
123 | * Alan Cox : No connect to multicast. | |
124 | * Ross Biro : Close unaccepted children on master | |
125 | * socket close. | |
126 | * Alan Cox : Reset tracing code. | |
127 | * Alan Cox : Spurious resets on shutdown. | |
128 | * Alan Cox : Giant 15 minute/60 second timer error | |
129 | * Alan Cox : Small whoops in polling before an | |
130 | * accept. | |
131 | * Alan Cox : Kept the state trace facility since | |
132 | * it's handy for debugging. | |
133 | * Alan Cox : More reset handler fixes. | |
134 | * Alan Cox : Started rewriting the code based on | |
135 | * the RFC's for other useful protocol | |
136 | * references see: Comer, KA9Q NOS, and | |
137 | * for a reference on the difference | |
138 | * between specifications and how BSD | |
139 | * works see the 4.4lite source. | |
140 | * A.N.Kuznetsov : Don't time wait on completion of tidy | |
141 | * close. | |
142 | * Linus Torvalds : Fin/Shutdown & copied_seq changes. | |
143 | * Linus Torvalds : Fixed BSD port reuse to work first syn | |
144 | * Alan Cox : Reimplemented timers as per the RFC | |
145 | * and using multiple timers for sanity. | |
146 | * Alan Cox : Small bug fixes, and a lot of new | |
147 | * comments. | |
148 | * Alan Cox : Fixed dual reader crash by locking | |
149 | * the buffers (much like datagram.c) | |
150 | * Alan Cox : Fixed stuck sockets in probe. A probe | |
151 | * now gets fed up of retrying without | |
152 | * (even a no space) answer. | |
153 | * Alan Cox : Extracted closing code better | |
154 | * Alan Cox : Fixed the closing state machine to | |
155 | * resemble the RFC. | |
156 | * Alan Cox : More 'per spec' fixes. | |
157 | * Jorge Cwik : Even faster checksumming. | |
158 | * Alan Cox : tcp_data() doesn't ack illegal PSH | |
159 | * only frames. At least one pc tcp stack | |
160 | * generates them. | |
161 | * Alan Cox : Cache last socket. | |
162 | * Alan Cox : Per route irtt. | |
163 | * Matt Day : poll()->select() match BSD precisely on error | |
164 | * Alan Cox : New buffers | |
165 | * Marc Tamsky : Various sk->prot->retransmits and | |
166 | * sk->retransmits misupdating fixed. | |
167 | * Fixed tcp_write_timeout: stuck close, | |
168 | * and TCP syn retries gets used now. | |
169 | * Mark Yarvis : In tcp_read_wakeup(), don't send an | |
170 | * ack if state is TCP_CLOSED. | |
171 | * Alan Cox : Look up device on a retransmit - routes may | |
172 | * change. Doesn't yet cope with MSS shrink right | |
173 | * but it's a start! | |
174 | * Marc Tamsky : Closing in closing fixes. | |
175 | * Mike Shaver : RFC1122 verifications. | |
176 | * Alan Cox : rcv_saddr errors. | |
177 | * Alan Cox : Block double connect(). | |
178 | * Alan Cox : Small hooks for enSKIP. | |
179 | * Alexey Kuznetsov: Path MTU discovery. | |
180 | * Alan Cox : Support soft errors. | |
181 | * Alan Cox : Fix MTU discovery pathological case | |
182 | * when the remote claims no mtu! | |
183 | * Marc Tamsky : TCP_CLOSE fix. | |
184 | * Colin (G3TNE) : Send a reset on syn ack replies in | |
185 | * window but wrong (fixes NT lpd problems) | |
186 | * Pedro Roque : Better TCP window handling, delayed ack. | |
187 | * Joerg Reuter : No modification of locked buffers in | |
188 | * tcp_do_retransmit() | |
189 | * Eric Schenk : Changed receiver side silly window | |
190 | * avoidance algorithm to BSD style | |
191 | * algorithm. This doubles throughput | |
192 | * against machines running Solaris, | |
193 | * and seems to result in general | |
194 | * improvement. | |
195 | * Stefan Magdalinski : adjusted tcp_readable() to fix FIONREAD | |
196 | * Willy Konynenberg : Transparent proxying support. | |
197 | * Mike McLagan : Routing by source | |
198 | * Keith Owens : Do proper merging with partial SKB's in | |
199 | * tcp_do_sendmsg to avoid burstiness. | |
200 | * Eric Schenk : Fix fast close down bug with | |
201 | * shutdown() followed by close(). | |
202 | * Andi Kleen : Make poll agree with SIGIO | |
203 | * Salvatore Sanfilippo : Support SO_LINGER with linger == 1 and | |
204 | * lingertime == 0 (RFC 793 ABORT Call) | |
205 | * Hirokazu Takahashi : Use copy_from_user() instead of | |
206 | * csum_and_copy_from_user() if possible. | |
207 | * | |
208 | * This program is free software; you can redistribute it and/or | |
209 | * modify it under the terms of the GNU General Public License | |
210 | * as published by the Free Software Foundation; either version | |
211 | * 2 of the License, or(at your option) any later version. | |
212 | * | |
213 | * Description of States: | |
214 | * | |
215 | * TCP_SYN_SENT sent a connection request, waiting for ack | |
216 | * | |
217 | * TCP_SYN_RECV received a connection request, sent ack, | |
218 | * waiting for final ack in three-way handshake. | |
219 | * | |
220 | * TCP_ESTABLISHED connection established | |
221 | * | |
222 | * TCP_FIN_WAIT1 our side has shutdown, waiting to complete | |
223 | * transmission of remaining buffered data | |
224 | * | |
225 | * TCP_FIN_WAIT2 all buffered data sent, waiting for remote | |
226 | * to shutdown | |
227 | * | |
228 | * TCP_CLOSING both sides have shutdown but we still have | |
229 | * data we have to finish sending | |
230 | * | |
231 | * TCP_TIME_WAIT timeout to catch resent junk before entering | |
232 | * closed, can only be entered from FIN_WAIT2 | |
233 | * or CLOSING. Required because the other end | |
234 | * may not have gotten our last ACK causing it | |
235 | * to retransmit the data packet (which we ignore) | |
236 | * | |
237 | * TCP_CLOSE_WAIT remote side has shutdown and is waiting for | |
238 | * us to finish writing our data and to shutdown | |
239 | * (we have to close() to move on to LAST_ACK) | |
240 | * | |
241 | * TCP_LAST_ACK out side has shutdown after remote has | |
242 | * shutdown. There may still be data in our | |
243 | * buffer that we have to finish sending | |
244 | * | |
245 | * TCP_CLOSE socket is finished | |
246 | */ | |
247 | ||
afd46503 JP |
248 | #define pr_fmt(fmt) "TCP: " fmt |
249 | ||
172589cc | 250 | #include <linux/kernel.h> |
1da177e4 LT |
251 | #include <linux/module.h> |
252 | #include <linux/types.h> | |
253 | #include <linux/fcntl.h> | |
254 | #include <linux/poll.h> | |
255 | #include <linux/init.h> | |
1da177e4 | 256 | #include <linux/fs.h> |
9c55e01c | 257 | #include <linux/skbuff.h> |
81b23b4a | 258 | #include <linux/scatterlist.h> |
9c55e01c JA |
259 | #include <linux/splice.h> |
260 | #include <linux/net.h> | |
261 | #include <linux/socket.h> | |
1da177e4 LT |
262 | #include <linux/random.h> |
263 | #include <linux/bootmem.h> | |
57413ebc MS |
264 | #include <linux/highmem.h> |
265 | #include <linux/swap.h> | |
b8059ead | 266 | #include <linux/cache.h> |
f4c50d99 | 267 | #include <linux/err.h> |
cfb6eeb4 | 268 | #include <linux/crypto.h> |
da5c78c8 | 269 | #include <linux/time.h> |
5a0e3ad6 | 270 | #include <linux/slab.h> |
1da177e4 LT |
271 | |
272 | #include <net/icmp.h> | |
cf60af03 | 273 | #include <net/inet_common.h> |
1da177e4 LT |
274 | #include <net/tcp.h> |
275 | #include <net/xfrm.h> | |
276 | #include <net/ip.h> | |
1a2449a8 | 277 | #include <net/netdma.h> |
9c55e01c | 278 | #include <net/sock.h> |
1da177e4 LT |
279 | |
280 | #include <asm/uaccess.h> | |
281 | #include <asm/ioctls.h> | |
076bb0c8 | 282 | #include <net/busy_poll.h> |
1da177e4 | 283 | |
ab32ea5d | 284 | int sysctl_tcp_fin_timeout __read_mostly = TCP_FIN_TIMEOUT; |
1da177e4 | 285 | |
95bd09eb ED |
286 | int sysctl_tcp_min_tso_segs __read_mostly = 2; |
287 | ||
dd24c001 | 288 | struct percpu_counter tcp_orphan_count; |
0a5578cf ACM |
289 | EXPORT_SYMBOL_GPL(tcp_orphan_count); |
290 | ||
a4fe34bf | 291 | long sysctl_tcp_mem[3] __read_mostly; |
b8059ead DM |
292 | int sysctl_tcp_wmem[3] __read_mostly; |
293 | int sysctl_tcp_rmem[3] __read_mostly; | |
1da177e4 | 294 | |
a4fe34bf | 295 | EXPORT_SYMBOL(sysctl_tcp_mem); |
1da177e4 LT |
296 | EXPORT_SYMBOL(sysctl_tcp_rmem); |
297 | EXPORT_SYMBOL(sysctl_tcp_wmem); | |
298 | ||
8d987e5c | 299 | atomic_long_t tcp_memory_allocated; /* Current allocated memory. */ |
1da177e4 | 300 | EXPORT_SYMBOL(tcp_memory_allocated); |
1748376b ED |
301 | |
302 | /* | |
303 | * Current number of TCP sockets. | |
304 | */ | |
305 | struct percpu_counter tcp_sockets_allocated; | |
1da177e4 LT |
306 | EXPORT_SYMBOL(tcp_sockets_allocated); |
307 | ||
9c55e01c JA |
308 | /* |
309 | * TCP splice context | |
310 | */ | |
311 | struct tcp_splice_state { | |
312 | struct pipe_inode_info *pipe; | |
313 | size_t len; | |
314 | unsigned int flags; | |
315 | }; | |
316 | ||
1da177e4 LT |
317 | /* |
318 | * Pressure flag: try to collapse. | |
319 | * Technical note: it is used by multiple contexts non atomically. | |
3ab224be | 320 | * All the __sk_mem_schedule() is of this nature: accounting |
1da177e4 LT |
321 | * is strict, actions are advisory and have some latency. |
322 | */ | |
4103f8cd | 323 | int tcp_memory_pressure __read_mostly; |
1da177e4 LT |
324 | EXPORT_SYMBOL(tcp_memory_pressure); |
325 | ||
5c52ba17 | 326 | void tcp_enter_memory_pressure(struct sock *sk) |
1da177e4 LT |
327 | { |
328 | if (!tcp_memory_pressure) { | |
4e673444 | 329 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES); |
1da177e4 LT |
330 | tcp_memory_pressure = 1; |
331 | } | |
332 | } | |
1da177e4 LT |
333 | EXPORT_SYMBOL(tcp_enter_memory_pressure); |
334 | ||
b103cf34 JA |
335 | /* Convert seconds to retransmits based on initial and max timeout */ |
336 | static u8 secs_to_retrans(int seconds, int timeout, int rto_max) | |
337 | { | |
338 | u8 res = 0; | |
339 | ||
340 | if (seconds > 0) { | |
341 | int period = timeout; | |
342 | ||
343 | res = 1; | |
344 | while (seconds > period && res < 255) { | |
345 | res++; | |
346 | timeout <<= 1; | |
347 | if (timeout > rto_max) | |
348 | timeout = rto_max; | |
349 | period += timeout; | |
350 | } | |
351 | } | |
352 | return res; | |
353 | } | |
354 | ||
355 | /* Convert retransmits to seconds based on initial and max timeout */ | |
356 | static int retrans_to_secs(u8 retrans, int timeout, int rto_max) | |
357 | { | |
358 | int period = 0; | |
359 | ||
360 | if (retrans > 0) { | |
361 | period = timeout; | |
362 | while (--retrans) { | |
363 | timeout <<= 1; | |
364 | if (timeout > rto_max) | |
365 | timeout = rto_max; | |
366 | period += timeout; | |
367 | } | |
368 | } | |
369 | return period; | |
370 | } | |
371 | ||
900f65d3 NC |
372 | /* Address-family independent initialization for a tcp_sock. |
373 | * | |
374 | * NOTE: A lot of things set to zero explicitly by call to | |
375 | * sk_alloc() so need not be done here. | |
376 | */ | |
377 | void tcp_init_sock(struct sock *sk) | |
378 | { | |
379 | struct inet_connection_sock *icsk = inet_csk(sk); | |
380 | struct tcp_sock *tp = tcp_sk(sk); | |
381 | ||
382 | skb_queue_head_init(&tp->out_of_order_queue); | |
383 | tcp_init_xmit_timers(sk); | |
384 | tcp_prequeue_init(tp); | |
46d3ceab | 385 | INIT_LIST_HEAD(&tp->tsq_node); |
900f65d3 NC |
386 | |
387 | icsk->icsk_rto = TCP_TIMEOUT_INIT; | |
388 | tp->mdev = TCP_TIMEOUT_INIT; | |
389 | ||
390 | /* So many TCP implementations out there (incorrectly) count the | |
391 | * initial SYN frame in their delayed-ACK and congestion control | |
392 | * algorithms that we must have the following bandaid to talk | |
393 | * efficiently to them. -DaveM | |
394 | */ | |
395 | tp->snd_cwnd = TCP_INIT_CWND; | |
396 | ||
397 | /* See draft-stevens-tcpca-spec-01 for discussion of the | |
398 | * initialization of these values. | |
399 | */ | |
400 | tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; | |
401 | tp->snd_cwnd_clamp = ~0; | |
402 | tp->mss_cache = TCP_MSS_DEFAULT; | |
403 | ||
404 | tp->reordering = sysctl_tcp_reordering; | |
eed530b6 | 405 | tcp_enable_early_retrans(tp); |
900f65d3 NC |
406 | icsk->icsk_ca_ops = &tcp_init_congestion_ops; |
407 | ||
ceaa1fef AV |
408 | tp->tsoffset = 0; |
409 | ||
900f65d3 NC |
410 | sk->sk_state = TCP_CLOSE; |
411 | ||
412 | sk->sk_write_space = sk_stream_write_space; | |
413 | sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); | |
414 | ||
415 | icsk->icsk_sync_mss = tcp_sync_mss; | |
416 | ||
900f65d3 NC |
417 | sk->sk_sndbuf = sysctl_tcp_wmem[1]; |
418 | sk->sk_rcvbuf = sysctl_tcp_rmem[1]; | |
419 | ||
420 | local_bh_disable(); | |
421 | sock_update_memcg(sk); | |
422 | sk_sockets_allocated_inc(sk); | |
423 | local_bh_enable(); | |
424 | } | |
425 | EXPORT_SYMBOL(tcp_init_sock); | |
426 | ||
1da177e4 LT |
427 | /* |
428 | * Wait for a TCP event. | |
429 | * | |
430 | * Note that we don't need to lock the socket, as the upper poll layers | |
431 | * take care of normal races (between the test and the event) and we don't | |
432 | * go look at any of the socket buffers directly. | |
433 | */ | |
434 | unsigned int tcp_poll(struct file *file, struct socket *sock, poll_table *wait) | |
435 | { | |
436 | unsigned int mask; | |
437 | struct sock *sk = sock->sk; | |
cf533ea5 | 438 | const struct tcp_sock *tp = tcp_sk(sk); |
1da177e4 | 439 | |
c3f1dbaf DM |
440 | sock_rps_record_flow(sk); |
441 | ||
aa395145 | 442 | sock_poll_wait(file, sk_sleep(sk), wait); |
1da177e4 | 443 | if (sk->sk_state == TCP_LISTEN) |
dc40c7bc | 444 | return inet_csk_listen_poll(sk); |
1da177e4 LT |
445 | |
446 | /* Socket is not locked. We are protected from async events | |
70efce27 WN |
447 | * by poll logic and correct handling of state changes |
448 | * made by other threads is impossible in any case. | |
1da177e4 LT |
449 | */ |
450 | ||
451 | mask = 0; | |
1da177e4 LT |
452 | |
453 | /* | |
454 | * POLLHUP is certainly not done right. But poll() doesn't | |
455 | * have a notion of HUP in just one direction, and for a | |
456 | * socket the read side is more interesting. | |
457 | * | |
458 | * Some poll() documentation says that POLLHUP is incompatible | |
459 | * with the POLLOUT/POLLWR flags, so somebody should check this | |
460 | * all. But careful, it tends to be safer to return too many | |
461 | * bits than too few, and you can easily break real applications | |
462 | * if you don't tell them that something has hung up! | |
463 | * | |
464 | * Check-me. | |
465 | * | |
466 | * Check number 1. POLLHUP is _UNMASKABLE_ event (see UNIX98 and | |
467 | * our fs/select.c). It means that after we received EOF, | |
468 | * poll always returns immediately, making impossible poll() on write() | |
469 | * in state CLOSE_WAIT. One solution is evident --- to set POLLHUP | |
470 | * if and only if shutdown has been made in both directions. | |
471 | * Actually, it is interesting to look how Solaris and DUX | |
70efce27 | 472 | * solve this dilemma. I would prefer, if POLLHUP were maskable, |
1da177e4 LT |
473 | * then we could set it on SND_SHUTDOWN. BTW examples given |
474 | * in Stevens' books assume exactly this behaviour, it explains | |
70efce27 | 475 | * why POLLHUP is incompatible with POLLOUT. --ANK |
1da177e4 LT |
476 | * |
477 | * NOTE. Check for TCP_CLOSE is added. The goal is to prevent | |
478 | * blocking on fresh not-connected or disconnected socket. --ANK | |
479 | */ | |
480 | if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == TCP_CLOSE) | |
481 | mask |= POLLHUP; | |
482 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
f348d70a | 483 | mask |= POLLIN | POLLRDNORM | POLLRDHUP; |
1da177e4 | 484 | |
8336886f JC |
485 | /* Connected or passive Fast Open socket? */ |
486 | if (sk->sk_state != TCP_SYN_SENT && | |
487 | (sk->sk_state != TCP_SYN_RECV || tp->fastopen_rsk != NULL)) { | |
c7004482 DM |
488 | int target = sock_rcvlowat(sk, 0, INT_MAX); |
489 | ||
490 | if (tp->urg_seq == tp->copied_seq && | |
491 | !sock_flag(sk, SOCK_URGINLINE) && | |
492 | tp->urg_data) | |
b634f875 | 493 | target++; |
c7004482 | 494 | |
1da177e4 LT |
495 | /* Potential race condition. If read of tp below will |
496 | * escape above sk->sk_state, we can be illegally awaken | |
497 | * in SYN_* states. */ | |
c7004482 | 498 | if (tp->rcv_nxt - tp->copied_seq >= target) |
1da177e4 LT |
499 | mask |= POLLIN | POLLRDNORM; |
500 | ||
501 | if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { | |
64dc6130 | 502 | if (sk_stream_is_writeable(sk)) { |
1da177e4 LT |
503 | mask |= POLLOUT | POLLWRNORM; |
504 | } else { /* send SIGIO later */ | |
505 | set_bit(SOCK_ASYNC_NOSPACE, | |
506 | &sk->sk_socket->flags); | |
507 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
508 | ||
509 | /* Race breaker. If space is freed after | |
510 | * wspace test but before the flags are set, | |
511 | * IO signal will be lost. | |
512 | */ | |
64dc6130 | 513 | if (sk_stream_is_writeable(sk)) |
1da177e4 LT |
514 | mask |= POLLOUT | POLLWRNORM; |
515 | } | |
d84ba638 KM |
516 | } else |
517 | mask |= POLLOUT | POLLWRNORM; | |
1da177e4 LT |
518 | |
519 | if (tp->urg_data & TCP_URG_VALID) | |
520 | mask |= POLLPRI; | |
521 | } | |
a4d25803 TM |
522 | /* This barrier is coupled with smp_wmb() in tcp_reset() */ |
523 | smp_rmb(); | |
524 | if (sk->sk_err) | |
525 | mask |= POLLERR; | |
526 | ||
1da177e4 LT |
527 | return mask; |
528 | } | |
4bc2f18b | 529 | EXPORT_SYMBOL(tcp_poll); |
1da177e4 LT |
530 | |
531 | int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg) | |
532 | { | |
533 | struct tcp_sock *tp = tcp_sk(sk); | |
534 | int answ; | |
0e71c55c | 535 | bool slow; |
1da177e4 LT |
536 | |
537 | switch (cmd) { | |
538 | case SIOCINQ: | |
539 | if (sk->sk_state == TCP_LISTEN) | |
540 | return -EINVAL; | |
541 | ||
0e71c55c | 542 | slow = lock_sock_fast(sk); |
1da177e4 LT |
543 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) |
544 | answ = 0; | |
545 | else if (sock_flag(sk, SOCK_URGINLINE) || | |
546 | !tp->urg_data || | |
547 | before(tp->urg_seq, tp->copied_seq) || | |
548 | !before(tp->urg_seq, tp->rcv_nxt)) { | |
91521944 | 549 | |
1da177e4 LT |
550 | answ = tp->rcv_nxt - tp->copied_seq; |
551 | ||
a3374c42 ED |
552 | /* Subtract 1, if FIN was received */ |
553 | if (answ && sock_flag(sk, SOCK_DONE)) | |
554 | answ--; | |
1da177e4 LT |
555 | } else |
556 | answ = tp->urg_seq - tp->copied_seq; | |
0e71c55c | 557 | unlock_sock_fast(sk, slow); |
1da177e4 LT |
558 | break; |
559 | case SIOCATMARK: | |
560 | answ = tp->urg_data && tp->urg_seq == tp->copied_seq; | |
561 | break; | |
562 | case SIOCOUTQ: | |
563 | if (sk->sk_state == TCP_LISTEN) | |
564 | return -EINVAL; | |
565 | ||
566 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) | |
567 | answ = 0; | |
568 | else | |
569 | answ = tp->write_seq - tp->snd_una; | |
570 | break; | |
2f4e1b39 MS |
571 | case SIOCOUTQNSD: |
572 | if (sk->sk_state == TCP_LISTEN) | |
573 | return -EINVAL; | |
574 | ||
575 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) | |
576 | answ = 0; | |
577 | else | |
578 | answ = tp->write_seq - tp->snd_nxt; | |
579 | break; | |
1da177e4 LT |
580 | default: |
581 | return -ENOIOCTLCMD; | |
3ff50b79 | 582 | } |
1da177e4 LT |
583 | |
584 | return put_user(answ, (int __user *)arg); | |
585 | } | |
4bc2f18b | 586 | EXPORT_SYMBOL(tcp_ioctl); |
1da177e4 | 587 | |
1da177e4 LT |
588 | static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb) |
589 | { | |
4de075e0 | 590 | TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH; |
1da177e4 LT |
591 | tp->pushed_seq = tp->write_seq; |
592 | } | |
593 | ||
a2a385d6 | 594 | static inline bool forced_push(const struct tcp_sock *tp) |
1da177e4 LT |
595 | { |
596 | return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1)); | |
597 | } | |
598 | ||
9e412ba7 | 599 | static inline void skb_entail(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 600 | { |
9e412ba7 | 601 | struct tcp_sock *tp = tcp_sk(sk); |
352d4800 ACM |
602 | struct tcp_skb_cb *tcb = TCP_SKB_CB(skb); |
603 | ||
604 | skb->csum = 0; | |
605 | tcb->seq = tcb->end_seq = tp->write_seq; | |
4de075e0 | 606 | tcb->tcp_flags = TCPHDR_ACK; |
352d4800 | 607 | tcb->sacked = 0; |
1da177e4 | 608 | skb_header_release(skb); |
fe067e8a | 609 | tcp_add_write_queue_tail(sk, skb); |
3ab224be HA |
610 | sk->sk_wmem_queued += skb->truesize; |
611 | sk_mem_charge(sk, skb->truesize); | |
89ebd197 | 612 | if (tp->nonagle & TCP_NAGLE_PUSH) |
e905a9ed | 613 | tp->nonagle &= ~TCP_NAGLE_PUSH; |
1da177e4 LT |
614 | } |
615 | ||
afeca340 | 616 | static inline void tcp_mark_urg(struct tcp_sock *tp, int flags) |
1da177e4 | 617 | { |
33f5f57e | 618 | if (flags & MSG_OOB) |
1da177e4 | 619 | tp->snd_up = tp->write_seq; |
1da177e4 LT |
620 | } |
621 | ||
9e412ba7 IJ |
622 | static inline void tcp_push(struct sock *sk, int flags, int mss_now, |
623 | int nonagle) | |
1da177e4 | 624 | { |
fe067e8a | 625 | if (tcp_send_head(sk)) { |
afeca340 KK |
626 | struct tcp_sock *tp = tcp_sk(sk); |
627 | ||
1da177e4 | 628 | if (!(flags & MSG_MORE) || forced_push(tp)) |
afeca340 KK |
629 | tcp_mark_push(tp, tcp_write_queue_tail(sk)); |
630 | ||
631 | tcp_mark_urg(tp, flags); | |
9e412ba7 | 632 | __tcp_push_pending_frames(sk, mss_now, |
1da177e4 LT |
633 | (flags & MSG_MORE) ? TCP_NAGLE_CORK : nonagle); |
634 | } | |
635 | } | |
636 | ||
6ff7751d AB |
637 | static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, |
638 | unsigned int offset, size_t len) | |
9c55e01c JA |
639 | { |
640 | struct tcp_splice_state *tss = rd_desc->arg.data; | |
33966dd0 | 641 | int ret; |
9c55e01c | 642 | |
9fa5fdf2 DM |
643 | ret = skb_splice_bits(skb, offset, tss->pipe, min(rd_desc->count, len), |
644 | tss->flags); | |
33966dd0 WT |
645 | if (ret > 0) |
646 | rd_desc->count -= ret; | |
647 | return ret; | |
9c55e01c JA |
648 | } |
649 | ||
650 | static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss) | |
651 | { | |
652 | /* Store TCP splice context information in read_descriptor_t. */ | |
653 | read_descriptor_t rd_desc = { | |
654 | .arg.data = tss, | |
33966dd0 | 655 | .count = tss->len, |
9c55e01c JA |
656 | }; |
657 | ||
658 | return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv); | |
659 | } | |
660 | ||
661 | /** | |
662 | * tcp_splice_read - splice data from TCP socket to a pipe | |
663 | * @sock: socket to splice from | |
664 | * @ppos: position (not valid) | |
665 | * @pipe: pipe to splice to | |
666 | * @len: number of bytes to splice | |
667 | * @flags: splice modifier flags | |
668 | * | |
669 | * Description: | |
670 | * Will read pages from given socket and fill them into a pipe. | |
671 | * | |
672 | **/ | |
673 | ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos, | |
674 | struct pipe_inode_info *pipe, size_t len, | |
675 | unsigned int flags) | |
676 | { | |
677 | struct sock *sk = sock->sk; | |
678 | struct tcp_splice_state tss = { | |
679 | .pipe = pipe, | |
680 | .len = len, | |
681 | .flags = flags, | |
682 | }; | |
683 | long timeo; | |
684 | ssize_t spliced; | |
685 | int ret; | |
686 | ||
3a047bf8 | 687 | sock_rps_record_flow(sk); |
9c55e01c JA |
688 | /* |
689 | * We can't seek on a socket input | |
690 | */ | |
691 | if (unlikely(*ppos)) | |
692 | return -ESPIPE; | |
693 | ||
694 | ret = spliced = 0; | |
695 | ||
696 | lock_sock(sk); | |
697 | ||
42324c62 | 698 | timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK); |
9c55e01c JA |
699 | while (tss.len) { |
700 | ret = __tcp_splice_read(sk, &tss); | |
701 | if (ret < 0) | |
702 | break; | |
703 | else if (!ret) { | |
704 | if (spliced) | |
705 | break; | |
9c55e01c JA |
706 | if (sock_flag(sk, SOCK_DONE)) |
707 | break; | |
708 | if (sk->sk_err) { | |
709 | ret = sock_error(sk); | |
710 | break; | |
711 | } | |
712 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
713 | break; | |
714 | if (sk->sk_state == TCP_CLOSE) { | |
715 | /* | |
716 | * This occurs when user tries to read | |
717 | * from never connected socket. | |
718 | */ | |
719 | if (!sock_flag(sk, SOCK_DONE)) | |
720 | ret = -ENOTCONN; | |
721 | break; | |
722 | } | |
723 | if (!timeo) { | |
724 | ret = -EAGAIN; | |
725 | break; | |
726 | } | |
727 | sk_wait_data(sk, &timeo); | |
728 | if (signal_pending(current)) { | |
729 | ret = sock_intr_errno(timeo); | |
730 | break; | |
731 | } | |
732 | continue; | |
733 | } | |
734 | tss.len -= ret; | |
735 | spliced += ret; | |
736 | ||
33966dd0 WT |
737 | if (!timeo) |
738 | break; | |
9c55e01c JA |
739 | release_sock(sk); |
740 | lock_sock(sk); | |
741 | ||
742 | if (sk->sk_err || sk->sk_state == TCP_CLOSE || | |
33966dd0 | 743 | (sk->sk_shutdown & RCV_SHUTDOWN) || |
9c55e01c JA |
744 | signal_pending(current)) |
745 | break; | |
746 | } | |
747 | ||
748 | release_sock(sk); | |
749 | ||
750 | if (spliced) | |
751 | return spliced; | |
752 | ||
753 | return ret; | |
754 | } | |
4bc2f18b | 755 | EXPORT_SYMBOL(tcp_splice_read); |
9c55e01c | 756 | |
df97c708 | 757 | struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp) |
f561d0f2 PE |
758 | { |
759 | struct sk_buff *skb; | |
760 | ||
761 | /* The TCP header must be at least 32-bit aligned. */ | |
762 | size = ALIGN(size, 4); | |
763 | ||
764 | skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp); | |
765 | if (skb) { | |
3ab224be | 766 | if (sk_wmem_schedule(sk, skb->truesize)) { |
a21d4572 | 767 | skb_reserve(skb, sk->sk_prot->max_header); |
f561d0f2 PE |
768 | /* |
769 | * Make sure that we have exactly size bytes | |
770 | * available to the caller, no more, no less. | |
771 | */ | |
16fad69c | 772 | skb->reserved_tailroom = skb->end - skb->tail - size; |
f561d0f2 PE |
773 | return skb; |
774 | } | |
775 | __kfree_skb(skb); | |
776 | } else { | |
5c52ba17 | 777 | sk->sk_prot->enter_memory_pressure(sk); |
f561d0f2 PE |
778 | sk_stream_moderate_sndbuf(sk); |
779 | } | |
780 | return NULL; | |
781 | } | |
782 | ||
0c54b85f IJ |
783 | static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now, |
784 | int large_allowed) | |
785 | { | |
786 | struct tcp_sock *tp = tcp_sk(sk); | |
2a3a041c | 787 | u32 xmit_size_goal, old_size_goal; |
0c54b85f IJ |
788 | |
789 | xmit_size_goal = mss_now; | |
790 | ||
791 | if (large_allowed && sk_can_gso(sk)) { | |
95bd09eb ED |
792 | u32 gso_size, hlen; |
793 | ||
794 | /* Maybe we should/could use sk->sk_prot->max_header here ? */ | |
795 | hlen = inet_csk(sk)->icsk_af_ops->net_header_len + | |
796 | inet_csk(sk)->icsk_ext_hdr_len + | |
797 | tp->tcp_header_len; | |
798 | ||
799 | /* Goal is to send at least one packet per ms, | |
800 | * not one big TSO packet every 100 ms. | |
801 | * This preserves ACK clocking and is consistent | |
802 | * with tcp_tso_should_defer() heuristic. | |
803 | */ | |
804 | gso_size = sk->sk_pacing_rate / (2 * MSEC_PER_SEC); | |
805 | gso_size = max_t(u32, gso_size, | |
806 | sysctl_tcp_min_tso_segs * mss_now); | |
807 | ||
808 | xmit_size_goal = min_t(u32, gso_size, | |
809 | sk->sk_gso_max_size - 1 - hlen); | |
810 | ||
0c54b85f | 811 | xmit_size_goal = tcp_bound_to_half_wnd(tp, xmit_size_goal); |
2a3a041c IJ |
812 | |
813 | /* We try hard to avoid divides here */ | |
814 | old_size_goal = tp->xmit_size_goal_segs * mss_now; | |
815 | ||
816 | if (likely(old_size_goal <= xmit_size_goal && | |
817 | old_size_goal + mss_now > xmit_size_goal)) { | |
818 | xmit_size_goal = old_size_goal; | |
819 | } else { | |
1485348d BH |
820 | tp->xmit_size_goal_segs = |
821 | min_t(u16, xmit_size_goal / mss_now, | |
822 | sk->sk_gso_max_segs); | |
2a3a041c IJ |
823 | xmit_size_goal = tp->xmit_size_goal_segs * mss_now; |
824 | } | |
0c54b85f IJ |
825 | } |
826 | ||
afece1c6 | 827 | return max(xmit_size_goal, mss_now); |
0c54b85f IJ |
828 | } |
829 | ||
830 | static int tcp_send_mss(struct sock *sk, int *size_goal, int flags) | |
831 | { | |
832 | int mss_now; | |
833 | ||
834 | mss_now = tcp_current_mss(sk); | |
835 | *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB)); | |
836 | ||
837 | return mss_now; | |
838 | } | |
839 | ||
64022d0b ED |
840 | static ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, |
841 | size_t size, int flags) | |
1da177e4 LT |
842 | { |
843 | struct tcp_sock *tp = tcp_sk(sk); | |
c1b4a7e6 | 844 | int mss_now, size_goal; |
1da177e4 LT |
845 | int err; |
846 | ssize_t copied; | |
847 | long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); | |
848 | ||
8336886f JC |
849 | /* Wait for a connection to finish. One exception is TCP Fast Open |
850 | * (passive side) where data is allowed to be sent before a connection | |
851 | * is fully established. | |
852 | */ | |
853 | if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) && | |
854 | !tcp_passive_fastopen(sk)) { | |
1da177e4 LT |
855 | if ((err = sk_stream_wait_connect(sk, &timeo)) != 0) |
856 | goto out_err; | |
8336886f | 857 | } |
1da177e4 LT |
858 | |
859 | clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); | |
860 | ||
0c54b85f | 861 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
862 | copied = 0; |
863 | ||
864 | err = -EPIPE; | |
865 | if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) | |
0d6a775e | 866 | goto out_err; |
1da177e4 | 867 | |
64022d0b | 868 | while (size > 0) { |
fe067e8a | 869 | struct sk_buff *skb = tcp_write_queue_tail(sk); |
38ba0a65 | 870 | int copy, i; |
38ba0a65 | 871 | bool can_coalesce; |
1da177e4 | 872 | |
fe067e8a | 873 | if (!tcp_send_head(sk) || (copy = size_goal - skb->len) <= 0) { |
1da177e4 LT |
874 | new_segment: |
875 | if (!sk_stream_memory_free(sk)) | |
876 | goto wait_for_sndbuf; | |
877 | ||
df97c708 | 878 | skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation); |
1da177e4 LT |
879 | if (!skb) |
880 | goto wait_for_memory; | |
881 | ||
9e412ba7 | 882 | skb_entail(sk, skb); |
c1b4a7e6 | 883 | copy = size_goal; |
1da177e4 LT |
884 | } |
885 | ||
886 | if (copy > size) | |
887 | copy = size; | |
888 | ||
889 | i = skb_shinfo(skb)->nr_frags; | |
890 | can_coalesce = skb_can_coalesce(skb, i, page, offset); | |
891 | if (!can_coalesce && i >= MAX_SKB_FRAGS) { | |
892 | tcp_mark_push(tp, skb); | |
893 | goto new_segment; | |
894 | } | |
3ab224be | 895 | if (!sk_wmem_schedule(sk, copy)) |
1da177e4 | 896 | goto wait_for_memory; |
e905a9ed | 897 | |
1da177e4 | 898 | if (can_coalesce) { |
9e903e08 | 899 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); |
1da177e4 LT |
900 | } else { |
901 | get_page(page); | |
902 | skb_fill_page_desc(skb, i, page, offset, copy); | |
903 | } | |
c9af6db4 | 904 | skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; |
cef401de | 905 | |
1da177e4 LT |
906 | skb->len += copy; |
907 | skb->data_len += copy; | |
908 | skb->truesize += copy; | |
909 | sk->sk_wmem_queued += copy; | |
3ab224be | 910 | sk_mem_charge(sk, copy); |
84fa7933 | 911 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 LT |
912 | tp->write_seq += copy; |
913 | TCP_SKB_CB(skb)->end_seq += copy; | |
7967168c | 914 | skb_shinfo(skb)->gso_segs = 0; |
1da177e4 LT |
915 | |
916 | if (!copied) | |
4de075e0 | 917 | TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH; |
1da177e4 LT |
918 | |
919 | copied += copy; | |
64022d0b ED |
920 | offset += copy; |
921 | if (!(size -= copy)) | |
1da177e4 LT |
922 | goto out; |
923 | ||
69d15067 | 924 | if (skb->len < size_goal || (flags & MSG_OOB)) |
1da177e4 LT |
925 | continue; |
926 | ||
927 | if (forced_push(tp)) { | |
928 | tcp_mark_push(tp, skb); | |
9e412ba7 | 929 | __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH); |
fe067e8a | 930 | } else if (skb == tcp_send_head(sk)) |
1da177e4 LT |
931 | tcp_push_one(sk, mss_now); |
932 | continue; | |
933 | ||
934 | wait_for_sndbuf: | |
935 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
936 | wait_for_memory: | |
bad115cf | 937 | tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH); |
1da177e4 LT |
938 | |
939 | if ((err = sk_stream_wait_memory(sk, &timeo)) != 0) | |
940 | goto do_error; | |
941 | ||
0c54b85f | 942 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
943 | } |
944 | ||
945 | out: | |
35f9c09f | 946 | if (copied && !(flags & MSG_SENDPAGE_NOTLAST)) |
9e412ba7 | 947 | tcp_push(sk, flags, mss_now, tp->nonagle); |
1da177e4 LT |
948 | return copied; |
949 | ||
950 | do_error: | |
951 | if (copied) | |
952 | goto out; | |
953 | out_err: | |
954 | return sk_stream_error(sk, flags, err); | |
955 | } | |
956 | ||
7ba42910 CG |
957 | int tcp_sendpage(struct sock *sk, struct page *page, int offset, |
958 | size_t size, int flags) | |
1da177e4 LT |
959 | { |
960 | ssize_t res; | |
1da177e4 | 961 | |
1da177e4 | 962 | if (!(sk->sk_route_caps & NETIF_F_SG) || |
8648b305 | 963 | !(sk->sk_route_caps & NETIF_F_ALL_CSUM)) |
7ba42910 CG |
964 | return sock_no_sendpage(sk->sk_socket, page, offset, size, |
965 | flags); | |
1da177e4 | 966 | |
1da177e4 | 967 | lock_sock(sk); |
64022d0b | 968 | res = do_tcp_sendpages(sk, page, offset, size, flags); |
1da177e4 LT |
969 | release_sock(sk); |
970 | return res; | |
971 | } | |
4bc2f18b | 972 | EXPORT_SYMBOL(tcp_sendpage); |
1da177e4 | 973 | |
690e99c4 | 974 | static inline int select_size(const struct sock *sk, bool sg) |
1da177e4 | 975 | { |
cf533ea5 | 976 | const struct tcp_sock *tp = tcp_sk(sk); |
c1b4a7e6 | 977 | int tmp = tp->mss_cache; |
1da177e4 | 978 | |
def87cf4 | 979 | if (sg) { |
f07d960d ED |
980 | if (sk_can_gso(sk)) { |
981 | /* Small frames wont use a full page: | |
982 | * Payload will immediately follow tcp header. | |
983 | */ | |
984 | tmp = SKB_WITH_OVERHEAD(2048 - MAX_TCP_HEADER); | |
985 | } else { | |
b4e26f5e DM |
986 | int pgbreak = SKB_MAX_HEAD(MAX_TCP_HEADER); |
987 | ||
988 | if (tmp >= pgbreak && | |
989 | tmp <= pgbreak + (MAX_SKB_FRAGS - 1) * PAGE_SIZE) | |
990 | tmp = pgbreak; | |
991 | } | |
992 | } | |
1da177e4 | 993 | |
1da177e4 LT |
994 | return tmp; |
995 | } | |
996 | ||
cf60af03 YC |
997 | void tcp_free_fastopen_req(struct tcp_sock *tp) |
998 | { | |
999 | if (tp->fastopen_req != NULL) { | |
1000 | kfree(tp->fastopen_req); | |
1001 | tp->fastopen_req = NULL; | |
1002 | } | |
1003 | } | |
1004 | ||
1005 | static int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *size) | |
1006 | { | |
1007 | struct tcp_sock *tp = tcp_sk(sk); | |
1008 | int err, flags; | |
1009 | ||
1010 | if (!(sysctl_tcp_fastopen & TFO_CLIENT_ENABLE)) | |
1011 | return -EOPNOTSUPP; | |
1012 | if (tp->fastopen_req != NULL) | |
1013 | return -EALREADY; /* Another Fast Open is in progress */ | |
1014 | ||
1015 | tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request), | |
1016 | sk->sk_allocation); | |
1017 | if (unlikely(tp->fastopen_req == NULL)) | |
1018 | return -ENOBUFS; | |
1019 | tp->fastopen_req->data = msg; | |
1020 | ||
1021 | flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0; | |
1022 | err = __inet_stream_connect(sk->sk_socket, msg->msg_name, | |
1023 | msg->msg_namelen, flags); | |
1024 | *size = tp->fastopen_req->copied; | |
1025 | tcp_free_fastopen_req(tp); | |
1026 | return err; | |
1027 | } | |
1028 | ||
7ba42910 | 1029 | int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, |
1da177e4 LT |
1030 | size_t size) |
1031 | { | |
1032 | struct iovec *iov; | |
1033 | struct tcp_sock *tp = tcp_sk(sk); | |
1034 | struct sk_buff *skb; | |
cf60af03 YC |
1035 | int iovlen, flags, err, copied = 0; |
1036 | int mss_now = 0, size_goal, copied_syn = 0, offset = 0; | |
690e99c4 | 1037 | bool sg; |
1da177e4 LT |
1038 | long timeo; |
1039 | ||
1040 | lock_sock(sk); | |
1da177e4 LT |
1041 | |
1042 | flags = msg->msg_flags; | |
cf60af03 YC |
1043 | if (flags & MSG_FASTOPEN) { |
1044 | err = tcp_sendmsg_fastopen(sk, msg, &copied_syn); | |
1045 | if (err == -EINPROGRESS && copied_syn > 0) | |
1046 | goto out; | |
1047 | else if (err) | |
1048 | goto out_err; | |
1049 | offset = copied_syn; | |
1050 | } | |
1051 | ||
1da177e4 LT |
1052 | timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); |
1053 | ||
8336886f JC |
1054 | /* Wait for a connection to finish. One exception is TCP Fast Open |
1055 | * (passive side) where data is allowed to be sent before a connection | |
1056 | * is fully established. | |
1057 | */ | |
1058 | if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) && | |
1059 | !tcp_passive_fastopen(sk)) { | |
1da177e4 | 1060 | if ((err = sk_stream_wait_connect(sk, &timeo)) != 0) |
cf60af03 | 1061 | goto do_error; |
8336886f | 1062 | } |
1da177e4 | 1063 | |
c0e88ff0 PE |
1064 | if (unlikely(tp->repair)) { |
1065 | if (tp->repair_queue == TCP_RECV_QUEUE) { | |
1066 | copied = tcp_send_rcvq(sk, msg, size); | |
1067 | goto out; | |
1068 | } | |
1069 | ||
1070 | err = -EINVAL; | |
1071 | if (tp->repair_queue == TCP_NO_QUEUE) | |
1072 | goto out_err; | |
1073 | ||
1074 | /* 'common' sending to sendq */ | |
1075 | } | |
1076 | ||
1da177e4 LT |
1077 | /* This should be in poll */ |
1078 | clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); | |
1079 | ||
0c54b85f | 1080 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
1081 | |
1082 | /* Ok commence sending. */ | |
1083 | iovlen = msg->msg_iovlen; | |
1084 | iov = msg->msg_iov; | |
1085 | copied = 0; | |
1086 | ||
1087 | err = -EPIPE; | |
1088 | if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) | |
0d6a775e | 1089 | goto out_err; |
1da177e4 | 1090 | |
690e99c4 | 1091 | sg = !!(sk->sk_route_caps & NETIF_F_SG); |
def87cf4 | 1092 | |
1da177e4 | 1093 | while (--iovlen >= 0) { |
01db403c | 1094 | size_t seglen = iov->iov_len; |
1da177e4 LT |
1095 | unsigned char __user *from = iov->iov_base; |
1096 | ||
1097 | iov++; | |
cf60af03 YC |
1098 | if (unlikely(offset > 0)) { /* Skip bytes copied in SYN */ |
1099 | if (offset >= seglen) { | |
1100 | offset -= seglen; | |
1101 | continue; | |
1102 | } | |
1103 | seglen -= offset; | |
1104 | from += offset; | |
1105 | offset = 0; | |
1106 | } | |
1da177e4 LT |
1107 | |
1108 | while (seglen > 0) { | |
6828b92b HX |
1109 | int copy = 0; |
1110 | int max = size_goal; | |
1da177e4 | 1111 | |
fe067e8a | 1112 | skb = tcp_write_queue_tail(sk); |
6828b92b HX |
1113 | if (tcp_send_head(sk)) { |
1114 | if (skb->ip_summed == CHECKSUM_NONE) | |
1115 | max = mss_now; | |
1116 | copy = max - skb->len; | |
1117 | } | |
1da177e4 | 1118 | |
6828b92b | 1119 | if (copy <= 0) { |
1da177e4 LT |
1120 | new_segment: |
1121 | /* Allocate new segment. If the interface is SG, | |
1122 | * allocate skb fitting to single page. | |
1123 | */ | |
1124 | if (!sk_stream_memory_free(sk)) | |
1125 | goto wait_for_sndbuf; | |
1126 | ||
def87cf4 KK |
1127 | skb = sk_stream_alloc_skb(sk, |
1128 | select_size(sk, sg), | |
1129 | sk->sk_allocation); | |
1da177e4 LT |
1130 | if (!skb) |
1131 | goto wait_for_memory; | |
1132 | ||
7ed5c5ae AV |
1133 | /* |
1134 | * All packets are restored as if they have | |
1135 | * already been sent. | |
1136 | */ | |
1137 | if (tp->repair) | |
1138 | TCP_SKB_CB(skb)->when = tcp_time_stamp; | |
1139 | ||
1da177e4 LT |
1140 | /* |
1141 | * Check whether we can use HW checksum. | |
1142 | */ | |
8648b305 | 1143 | if (sk->sk_route_caps & NETIF_F_ALL_CSUM) |
84fa7933 | 1144 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 | 1145 | |
9e412ba7 | 1146 | skb_entail(sk, skb); |
c1b4a7e6 | 1147 | copy = size_goal; |
6828b92b | 1148 | max = size_goal; |
1da177e4 LT |
1149 | } |
1150 | ||
1151 | /* Try to append data to the end of skb. */ | |
1152 | if (copy > seglen) | |
1153 | copy = seglen; | |
1154 | ||
1155 | /* Where to copy to? */ | |
a21d4572 | 1156 | if (skb_availroom(skb) > 0) { |
1da177e4 | 1157 | /* We have some space in skb head. Superb! */ |
a21d4572 | 1158 | copy = min_t(int, copy, skb_availroom(skb)); |
c6e1a0d1 TH |
1159 | err = skb_add_data_nocache(sk, skb, from, copy); |
1160 | if (err) | |
1da177e4 LT |
1161 | goto do_fault; |
1162 | } else { | |
5640f768 | 1163 | bool merge = true; |
1da177e4 | 1164 | int i = skb_shinfo(skb)->nr_frags; |
5640f768 ED |
1165 | struct page_frag *pfrag = sk_page_frag(sk); |
1166 | ||
1167 | if (!sk_page_frag_refill(sk, pfrag)) | |
1168 | goto wait_for_memory; | |
1169 | ||
1170 | if (!skb_can_coalesce(skb, i, pfrag->page, | |
1171 | pfrag->offset)) { | |
1172 | if (i == MAX_SKB_FRAGS || !sg) { | |
1173 | tcp_mark_push(tp, skb); | |
1174 | goto new_segment; | |
1da177e4 | 1175 | } |
5640f768 ED |
1176 | merge = false; |
1177 | } | |
ef015786 | 1178 | |
5640f768 | 1179 | copy = min_t(int, copy, pfrag->size - pfrag->offset); |
ef015786 | 1180 | |
3ab224be | 1181 | if (!sk_wmem_schedule(sk, copy)) |
ef015786 | 1182 | goto wait_for_memory; |
1da177e4 | 1183 | |
c6e1a0d1 | 1184 | err = skb_copy_to_page_nocache(sk, from, skb, |
5640f768 ED |
1185 | pfrag->page, |
1186 | pfrag->offset, | |
1187 | copy); | |
1188 | if (err) | |
1da177e4 | 1189 | goto do_error; |
1da177e4 LT |
1190 | |
1191 | /* Update the skb. */ | |
1192 | if (merge) { | |
9e903e08 | 1193 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); |
1da177e4 | 1194 | } else { |
5640f768 ED |
1195 | skb_fill_page_desc(skb, i, pfrag->page, |
1196 | pfrag->offset, copy); | |
1197 | get_page(pfrag->page); | |
1da177e4 | 1198 | } |
5640f768 | 1199 | pfrag->offset += copy; |
1da177e4 LT |
1200 | } |
1201 | ||
1202 | if (!copied) | |
4de075e0 | 1203 | TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH; |
1da177e4 LT |
1204 | |
1205 | tp->write_seq += copy; | |
1206 | TCP_SKB_CB(skb)->end_seq += copy; | |
7967168c | 1207 | skb_shinfo(skb)->gso_segs = 0; |
1da177e4 LT |
1208 | |
1209 | from += copy; | |
1210 | copied += copy; | |
1211 | if ((seglen -= copy) == 0 && iovlen == 0) | |
1212 | goto out; | |
1213 | ||
c0e88ff0 | 1214 | if (skb->len < max || (flags & MSG_OOB) || unlikely(tp->repair)) |
1da177e4 LT |
1215 | continue; |
1216 | ||
1217 | if (forced_push(tp)) { | |
1218 | tcp_mark_push(tp, skb); | |
9e412ba7 | 1219 | __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH); |
fe067e8a | 1220 | } else if (skb == tcp_send_head(sk)) |
1da177e4 LT |
1221 | tcp_push_one(sk, mss_now); |
1222 | continue; | |
1223 | ||
1224 | wait_for_sndbuf: | |
1225 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
1226 | wait_for_memory: | |
ec342325 | 1227 | if (copied) |
9e412ba7 | 1228 | tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH); |
1da177e4 LT |
1229 | |
1230 | if ((err = sk_stream_wait_memory(sk, &timeo)) != 0) | |
1231 | goto do_error; | |
1232 | ||
0c54b85f | 1233 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
1234 | } |
1235 | } | |
1236 | ||
1237 | out: | |
ec342325 | 1238 | if (copied) |
9e412ba7 | 1239 | tcp_push(sk, flags, mss_now, tp->nonagle); |
1da177e4 | 1240 | release_sock(sk); |
cf60af03 | 1241 | return copied + copied_syn; |
1da177e4 LT |
1242 | |
1243 | do_fault: | |
1244 | if (!skb->len) { | |
fe067e8a DM |
1245 | tcp_unlink_write_queue(skb, sk); |
1246 | /* It is the one place in all of TCP, except connection | |
1247 | * reset, where we can be unlinking the send_head. | |
1248 | */ | |
1249 | tcp_check_send_head(sk, skb); | |
3ab224be | 1250 | sk_wmem_free_skb(sk, skb); |
1da177e4 LT |
1251 | } |
1252 | ||
1253 | do_error: | |
cf60af03 | 1254 | if (copied + copied_syn) |
1da177e4 LT |
1255 | goto out; |
1256 | out_err: | |
1257 | err = sk_stream_error(sk, flags, err); | |
1da177e4 LT |
1258 | release_sock(sk); |
1259 | return err; | |
1260 | } | |
4bc2f18b | 1261 | EXPORT_SYMBOL(tcp_sendmsg); |
1da177e4 LT |
1262 | |
1263 | /* | |
1264 | * Handle reading urgent data. BSD has very simple semantics for | |
1265 | * this, no blocking and very strange errors 8) | |
1266 | */ | |
1267 | ||
377f0a08 | 1268 | static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags) |
1da177e4 LT |
1269 | { |
1270 | struct tcp_sock *tp = tcp_sk(sk); | |
1271 | ||
1272 | /* No URG data to read. */ | |
1273 | if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data || | |
1274 | tp->urg_data == TCP_URG_READ) | |
1275 | return -EINVAL; /* Yes this is right ! */ | |
1276 | ||
1277 | if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE)) | |
1278 | return -ENOTCONN; | |
1279 | ||
1280 | if (tp->urg_data & TCP_URG_VALID) { | |
1281 | int err = 0; | |
1282 | char c = tp->urg_data; | |
1283 | ||
1284 | if (!(flags & MSG_PEEK)) | |
1285 | tp->urg_data = TCP_URG_READ; | |
1286 | ||
1287 | /* Read urgent data. */ | |
1288 | msg->msg_flags |= MSG_OOB; | |
1289 | ||
1290 | if (len > 0) { | |
1291 | if (!(flags & MSG_TRUNC)) | |
1292 | err = memcpy_toiovec(msg->msg_iov, &c, 1); | |
1293 | len = 1; | |
1294 | } else | |
1295 | msg->msg_flags |= MSG_TRUNC; | |
1296 | ||
1297 | return err ? -EFAULT : len; | |
1298 | } | |
1299 | ||
1300 | if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN)) | |
1301 | return 0; | |
1302 | ||
1303 | /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and | |
1304 | * the available implementations agree in this case: | |
1305 | * this call should never block, independent of the | |
1306 | * blocking state of the socket. | |
1307 | * Mike <pall@rz.uni-karlsruhe.de> | |
1308 | */ | |
1309 | return -EAGAIN; | |
1310 | } | |
1311 | ||
c0e88ff0 PE |
1312 | static int tcp_peek_sndq(struct sock *sk, struct msghdr *msg, int len) |
1313 | { | |
1314 | struct sk_buff *skb; | |
1315 | int copied = 0, err = 0; | |
1316 | ||
1317 | /* XXX -- need to support SO_PEEK_OFF */ | |
1318 | ||
1319 | skb_queue_walk(&sk->sk_write_queue, skb) { | |
1320 | err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, skb->len); | |
1321 | if (err) | |
1322 | break; | |
1323 | ||
1324 | copied += skb->len; | |
1325 | } | |
1326 | ||
1327 | return err ?: copied; | |
1328 | } | |
1329 | ||
1da177e4 LT |
1330 | /* Clean up the receive buffer for full frames taken by the user, |
1331 | * then send an ACK if necessary. COPIED is the number of bytes | |
1332 | * tcp_recvmsg has given to the user so far, it speeds up the | |
1333 | * calculation of whether or not we must ACK for the sake of | |
1334 | * a window update. | |
1335 | */ | |
0e4b4992 | 1336 | void tcp_cleanup_rbuf(struct sock *sk, int copied) |
1da177e4 LT |
1337 | { |
1338 | struct tcp_sock *tp = tcp_sk(sk); | |
a2a385d6 | 1339 | bool time_to_ack = false; |
1da177e4 | 1340 | |
1da177e4 LT |
1341 | struct sk_buff *skb = skb_peek(&sk->sk_receive_queue); |
1342 | ||
d792c100 | 1343 | WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq), |
2af6fd8b | 1344 | "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n", |
d792c100 | 1345 | tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt); |
1da177e4 | 1346 | |
463c84b9 ACM |
1347 | if (inet_csk_ack_scheduled(sk)) { |
1348 | const struct inet_connection_sock *icsk = inet_csk(sk); | |
1da177e4 LT |
1349 | /* Delayed ACKs frequently hit locked sockets during bulk |
1350 | * receive. */ | |
463c84b9 | 1351 | if (icsk->icsk_ack.blocked || |
1da177e4 | 1352 | /* Once-per-two-segments ACK was not sent by tcp_input.c */ |
463c84b9 | 1353 | tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss || |
1da177e4 LT |
1354 | /* |
1355 | * If this read emptied read buffer, we send ACK, if | |
1356 | * connection is not bidirectional, user drained | |
1357 | * receive buffer and there was a small segment | |
1358 | * in queue. | |
1359 | */ | |
1ef9696c AK |
1360 | (copied > 0 && |
1361 | ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) || | |
1362 | ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) && | |
1363 | !icsk->icsk_ack.pingpong)) && | |
1364 | !atomic_read(&sk->sk_rmem_alloc))) | |
a2a385d6 | 1365 | time_to_ack = true; |
1da177e4 LT |
1366 | } |
1367 | ||
1368 | /* We send an ACK if we can now advertise a non-zero window | |
1369 | * which has been raised "significantly". | |
1370 | * | |
1371 | * Even if window raised up to infinity, do not send window open ACK | |
1372 | * in states, where we will not receive more. It is useless. | |
1373 | */ | |
1374 | if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) { | |
1375 | __u32 rcv_window_now = tcp_receive_window(tp); | |
1376 | ||
1377 | /* Optimize, __tcp_select_window() is not cheap. */ | |
1378 | if (2*rcv_window_now <= tp->window_clamp) { | |
1379 | __u32 new_window = __tcp_select_window(sk); | |
1380 | ||
1381 | /* Send ACK now, if this read freed lots of space | |
1382 | * in our buffer. Certainly, new_window is new window. | |
1383 | * We can advertise it now, if it is not less than current one. | |
1384 | * "Lots" means "at least twice" here. | |
1385 | */ | |
1386 | if (new_window && new_window >= 2 * rcv_window_now) | |
a2a385d6 | 1387 | time_to_ack = true; |
1da177e4 LT |
1388 | } |
1389 | } | |
1390 | if (time_to_ack) | |
1391 | tcp_send_ack(sk); | |
1392 | } | |
1393 | ||
1394 | static void tcp_prequeue_process(struct sock *sk) | |
1395 | { | |
1396 | struct sk_buff *skb; | |
1397 | struct tcp_sock *tp = tcp_sk(sk); | |
1398 | ||
6f67c817 | 1399 | NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPPREQUEUED); |
1da177e4 LT |
1400 | |
1401 | /* RX process wants to run with disabled BHs, though it is not | |
1402 | * necessary */ | |
1403 | local_bh_disable(); | |
1404 | while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) | |
c57943a1 | 1405 | sk_backlog_rcv(sk, skb); |
1da177e4 LT |
1406 | local_bh_enable(); |
1407 | ||
1408 | /* Clear memory counter. */ | |
1409 | tp->ucopy.memory = 0; | |
1410 | } | |
1411 | ||
73852e81 SM |
1412 | #ifdef CONFIG_NET_DMA |
1413 | static void tcp_service_net_dma(struct sock *sk, bool wait) | |
1414 | { | |
1415 | dma_cookie_t done, used; | |
1416 | dma_cookie_t last_issued; | |
1417 | struct tcp_sock *tp = tcp_sk(sk); | |
1418 | ||
1419 | if (!tp->ucopy.dma_chan) | |
1420 | return; | |
1421 | ||
1422 | last_issued = tp->ucopy.dma_cookie; | |
b9ee8683 | 1423 | dma_async_issue_pending(tp->ucopy.dma_chan); |
73852e81 SM |
1424 | |
1425 | do { | |
e239345f | 1426 | if (dma_async_is_tx_complete(tp->ucopy.dma_chan, |
73852e81 | 1427 | last_issued, &done, |
27bf6970 | 1428 | &used) == DMA_COMPLETE) { |
73852e81 SM |
1429 | /* Safe to free early-copied skbs now */ |
1430 | __skb_queue_purge(&sk->sk_async_wait_queue); | |
1431 | break; | |
1432 | } else { | |
1433 | struct sk_buff *skb; | |
1434 | while ((skb = skb_peek(&sk->sk_async_wait_queue)) && | |
1435 | (dma_async_is_complete(skb->dma_cookie, done, | |
27bf6970 | 1436 | used) == DMA_COMPLETE)) { |
73852e81 SM |
1437 | __skb_dequeue(&sk->sk_async_wait_queue); |
1438 | kfree_skb(skb); | |
1439 | } | |
1440 | } | |
1441 | } while (wait); | |
1442 | } | |
1443 | #endif | |
1444 | ||
f26845b4 | 1445 | static struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off) |
1da177e4 LT |
1446 | { |
1447 | struct sk_buff *skb; | |
1448 | u32 offset; | |
1449 | ||
f26845b4 | 1450 | while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) { |
1da177e4 | 1451 | offset = seq - TCP_SKB_CB(skb)->seq; |
aa8223c7 | 1452 | if (tcp_hdr(skb)->syn) |
1da177e4 | 1453 | offset--; |
aa8223c7 | 1454 | if (offset < skb->len || tcp_hdr(skb)->fin) { |
1da177e4 LT |
1455 | *off = offset; |
1456 | return skb; | |
1457 | } | |
f26845b4 ED |
1458 | /* This looks weird, but this can happen if TCP collapsing |
1459 | * splitted a fat GRO packet, while we released socket lock | |
1460 | * in skb_splice_bits() | |
1461 | */ | |
1462 | sk_eat_skb(sk, skb, false); | |
1da177e4 LT |
1463 | } |
1464 | return NULL; | |
1465 | } | |
1466 | ||
1467 | /* | |
1468 | * This routine provides an alternative to tcp_recvmsg() for routines | |
1469 | * that would like to handle copying from skbuffs directly in 'sendfile' | |
1470 | * fashion. | |
1471 | * Note: | |
1472 | * - It is assumed that the socket was locked by the caller. | |
1473 | * - The routine does not block. | |
1474 | * - At present, there is no support for reading OOB data | |
1475 | * or for 'peeking' the socket using this routine | |
1476 | * (although both would be easy to implement). | |
1477 | */ | |
1478 | int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, | |
1479 | sk_read_actor_t recv_actor) | |
1480 | { | |
1481 | struct sk_buff *skb; | |
1482 | struct tcp_sock *tp = tcp_sk(sk); | |
1483 | u32 seq = tp->copied_seq; | |
1484 | u32 offset; | |
1485 | int copied = 0; | |
1486 | ||
1487 | if (sk->sk_state == TCP_LISTEN) | |
1488 | return -ENOTCONN; | |
1489 | while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) { | |
1490 | if (offset < skb->len) { | |
374e7b59 OP |
1491 | int used; |
1492 | size_t len; | |
1da177e4 LT |
1493 | |
1494 | len = skb->len - offset; | |
1495 | /* Stop reading if we hit a patch of urgent data */ | |
1496 | if (tp->urg_data) { | |
1497 | u32 urg_offset = tp->urg_seq - seq; | |
1498 | if (urg_offset < len) | |
1499 | len = urg_offset; | |
1500 | if (!len) | |
1501 | break; | |
1502 | } | |
1503 | used = recv_actor(desc, skb, offset, len); | |
ff905b1e | 1504 | if (used <= 0) { |
ddb61a57 JA |
1505 | if (!copied) |
1506 | copied = used; | |
1507 | break; | |
1508 | } else if (used <= len) { | |
1da177e4 LT |
1509 | seq += used; |
1510 | copied += used; | |
1511 | offset += used; | |
1512 | } | |
02275a2e | 1513 | /* If recv_actor drops the lock (e.g. TCP splice |
293ad604 OP |
1514 | * receive) the skb pointer might be invalid when |
1515 | * getting here: tcp_collapse might have deleted it | |
1516 | * while aggregating skbs from the socket queue. | |
1517 | */ | |
02275a2e WT |
1518 | skb = tcp_recv_skb(sk, seq - 1, &offset); |
1519 | if (!skb) | |
1da177e4 | 1520 | break; |
02275a2e WT |
1521 | /* TCP coalescing might have appended data to the skb. |
1522 | * Try to splice more frags | |
1523 | */ | |
1524 | if (offset + 1 != skb->len) | |
1525 | continue; | |
1da177e4 | 1526 | } |
aa8223c7 | 1527 | if (tcp_hdr(skb)->fin) { |
dc6b9b78 | 1528 | sk_eat_skb(sk, skb, false); |
1da177e4 LT |
1529 | ++seq; |
1530 | break; | |
1531 | } | |
dc6b9b78 | 1532 | sk_eat_skb(sk, skb, false); |
1da177e4 LT |
1533 | if (!desc->count) |
1534 | break; | |
baff42ab | 1535 | tp->copied_seq = seq; |
1da177e4 LT |
1536 | } |
1537 | tp->copied_seq = seq; | |
1538 | ||
1539 | tcp_rcv_space_adjust(sk); | |
1540 | ||
1541 | /* Clean up data we have read: This will do ACK frames. */ | |
f26845b4 ED |
1542 | if (copied > 0) { |
1543 | tcp_recv_skb(sk, seq, &offset); | |
0e4b4992 | 1544 | tcp_cleanup_rbuf(sk, copied); |
f26845b4 | 1545 | } |
1da177e4 LT |
1546 | return copied; |
1547 | } | |
4bc2f18b | 1548 | EXPORT_SYMBOL(tcp_read_sock); |
1da177e4 LT |
1549 | |
1550 | /* | |
1551 | * This routine copies from a sock struct into the user buffer. | |
1552 | * | |
1553 | * Technical note: in 2.3 we work on _locked_ socket, so that | |
1554 | * tricks with *seq access order and skb->users are not required. | |
1555 | * Probably, code can be easily improved even more. | |
1556 | */ | |
1557 | ||
1558 | int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, | |
1559 | size_t len, int nonblock, int flags, int *addr_len) | |
1560 | { | |
1561 | struct tcp_sock *tp = tcp_sk(sk); | |
1562 | int copied = 0; | |
1563 | u32 peek_seq; | |
1564 | u32 *seq; | |
1565 | unsigned long used; | |
1566 | int err; | |
1567 | int target; /* Read at least this many bytes */ | |
1568 | long timeo; | |
1569 | struct task_struct *user_recv = NULL; | |
dc6b9b78 | 1570 | bool copied_early = false; |
2b1244a4 | 1571 | struct sk_buff *skb; |
77527313 | 1572 | u32 urg_hole = 0; |
1da177e4 | 1573 | |
cbf55001 ET |
1574 | if (sk_can_busy_loop(sk) && skb_queue_empty(&sk->sk_receive_queue) && |
1575 | (sk->sk_state == TCP_ESTABLISHED)) | |
1576 | sk_busy_loop(sk, nonblock); | |
d30e383b | 1577 | |
1da177e4 LT |
1578 | lock_sock(sk); |
1579 | ||
1da177e4 LT |
1580 | err = -ENOTCONN; |
1581 | if (sk->sk_state == TCP_LISTEN) | |
1582 | goto out; | |
1583 | ||
1584 | timeo = sock_rcvtimeo(sk, nonblock); | |
1585 | ||
1586 | /* Urgent data needs to be handled specially. */ | |
1587 | if (flags & MSG_OOB) | |
1588 | goto recv_urg; | |
1589 | ||
c0e88ff0 PE |
1590 | if (unlikely(tp->repair)) { |
1591 | err = -EPERM; | |
1592 | if (!(flags & MSG_PEEK)) | |
1593 | goto out; | |
1594 | ||
1595 | if (tp->repair_queue == TCP_SEND_QUEUE) | |
1596 | goto recv_sndq; | |
1597 | ||
1598 | err = -EINVAL; | |
1599 | if (tp->repair_queue == TCP_NO_QUEUE) | |
1600 | goto out; | |
1601 | ||
1602 | /* 'common' recv queue MSG_PEEK-ing */ | |
1603 | } | |
1604 | ||
1da177e4 LT |
1605 | seq = &tp->copied_seq; |
1606 | if (flags & MSG_PEEK) { | |
1607 | peek_seq = tp->copied_seq; | |
1608 | seq = &peek_seq; | |
1609 | } | |
1610 | ||
1611 | target = sock_rcvlowat(sk, flags & MSG_WAITALL, len); | |
1612 | ||
1a2449a8 CL |
1613 | #ifdef CONFIG_NET_DMA |
1614 | tp->ucopy.dma_chan = NULL; | |
1615 | preempt_disable(); | |
2b1244a4 | 1616 | skb = skb_peek_tail(&sk->sk_receive_queue); |
e00c5d8b AM |
1617 | { |
1618 | int available = 0; | |
1619 | ||
1620 | if (skb) | |
1621 | available = TCP_SKB_CB(skb)->seq + skb->len - (*seq); | |
1622 | if ((available < target) && | |
1623 | (len > sysctl_tcp_dma_copybreak) && !(flags & MSG_PEEK) && | |
1624 | !sysctl_tcp_low_latency && | |
a2bd1140 | 1625 | net_dma_find_channel()) { |
e00c5d8b AM |
1626 | preempt_enable_no_resched(); |
1627 | tp->ucopy.pinned_list = | |
1628 | dma_pin_iovec_pages(msg->msg_iov, len); | |
1629 | } else { | |
1630 | preempt_enable_no_resched(); | |
1631 | } | |
1632 | } | |
1a2449a8 CL |
1633 | #endif |
1634 | ||
1da177e4 | 1635 | do { |
1da177e4 LT |
1636 | u32 offset; |
1637 | ||
1638 | /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */ | |
1639 | if (tp->urg_data && tp->urg_seq == *seq) { | |
1640 | if (copied) | |
1641 | break; | |
1642 | if (signal_pending(current)) { | |
1643 | copied = timeo ? sock_intr_errno(timeo) : -EAGAIN; | |
1644 | break; | |
1645 | } | |
1646 | } | |
1647 | ||
1648 | /* Next get a buffer. */ | |
1649 | ||
91521944 | 1650 | skb_queue_walk(&sk->sk_receive_queue, skb) { |
1da177e4 LT |
1651 | /* Now that we have two receive queues this |
1652 | * shouldn't happen. | |
1653 | */ | |
d792c100 | 1654 | if (WARN(before(*seq, TCP_SKB_CB(skb)->seq), |
2af6fd8b JP |
1655 | "recvmsg bug: copied %X seq %X rcvnxt %X fl %X\n", |
1656 | *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, | |
1657 | flags)) | |
1da177e4 | 1658 | break; |
d792c100 | 1659 | |
1da177e4 | 1660 | offset = *seq - TCP_SKB_CB(skb)->seq; |
aa8223c7 | 1661 | if (tcp_hdr(skb)->syn) |
1da177e4 LT |
1662 | offset--; |
1663 | if (offset < skb->len) | |
1664 | goto found_ok_skb; | |
aa8223c7 | 1665 | if (tcp_hdr(skb)->fin) |
1da177e4 | 1666 | goto found_fin_ok; |
2af6fd8b JP |
1667 | WARN(!(flags & MSG_PEEK), |
1668 | "recvmsg bug 2: copied %X seq %X rcvnxt %X fl %X\n", | |
1669 | *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags); | |
91521944 | 1670 | } |
1da177e4 LT |
1671 | |
1672 | /* Well, if we have backlog, try to process it now yet. */ | |
1673 | ||
1674 | if (copied >= target && !sk->sk_backlog.tail) | |
1675 | break; | |
1676 | ||
1677 | if (copied) { | |
1678 | if (sk->sk_err || | |
1679 | sk->sk_state == TCP_CLOSE || | |
1680 | (sk->sk_shutdown & RCV_SHUTDOWN) || | |
1681 | !timeo || | |
518a09ef | 1682 | signal_pending(current)) |
1da177e4 LT |
1683 | break; |
1684 | } else { | |
1685 | if (sock_flag(sk, SOCK_DONE)) | |
1686 | break; | |
1687 | ||
1688 | if (sk->sk_err) { | |
1689 | copied = sock_error(sk); | |
1690 | break; | |
1691 | } | |
1692 | ||
1693 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
1694 | break; | |
1695 | ||
1696 | if (sk->sk_state == TCP_CLOSE) { | |
1697 | if (!sock_flag(sk, SOCK_DONE)) { | |
1698 | /* This occurs when user tries to read | |
1699 | * from never connected socket. | |
1700 | */ | |
1701 | copied = -ENOTCONN; | |
1702 | break; | |
1703 | } | |
1704 | break; | |
1705 | } | |
1706 | ||
1707 | if (!timeo) { | |
1708 | copied = -EAGAIN; | |
1709 | break; | |
1710 | } | |
1711 | ||
1712 | if (signal_pending(current)) { | |
1713 | copied = sock_intr_errno(timeo); | |
1714 | break; | |
1715 | } | |
1716 | } | |
1717 | ||
0e4b4992 | 1718 | tcp_cleanup_rbuf(sk, copied); |
1da177e4 | 1719 | |
7df55125 | 1720 | if (!sysctl_tcp_low_latency && tp->ucopy.task == user_recv) { |
1da177e4 LT |
1721 | /* Install new reader */ |
1722 | if (!user_recv && !(flags & (MSG_TRUNC | MSG_PEEK))) { | |
1723 | user_recv = current; | |
1724 | tp->ucopy.task = user_recv; | |
1725 | tp->ucopy.iov = msg->msg_iov; | |
1726 | } | |
1727 | ||
1728 | tp->ucopy.len = len; | |
1729 | ||
547b792c IJ |
1730 | WARN_ON(tp->copied_seq != tp->rcv_nxt && |
1731 | !(flags & (MSG_PEEK | MSG_TRUNC))); | |
1da177e4 LT |
1732 | |
1733 | /* Ugly... If prequeue is not empty, we have to | |
1734 | * process it before releasing socket, otherwise | |
1735 | * order will be broken at second iteration. | |
1736 | * More elegant solution is required!!! | |
1737 | * | |
1738 | * Look: we have the following (pseudo)queues: | |
1739 | * | |
1740 | * 1. packets in flight | |
1741 | * 2. backlog | |
1742 | * 3. prequeue | |
1743 | * 4. receive_queue | |
1744 | * | |
1745 | * Each queue can be processed only if the next ones | |
1746 | * are empty. At this point we have empty receive_queue. | |
1747 | * But prequeue _can_ be not empty after 2nd iteration, | |
1748 | * when we jumped to start of loop because backlog | |
1749 | * processing added something to receive_queue. | |
1750 | * We cannot release_sock(), because backlog contains | |
1751 | * packets arrived _after_ prequeued ones. | |
1752 | * | |
1753 | * Shortly, algorithm is clear --- to process all | |
1754 | * the queues in order. We could make it more directly, | |
1755 | * requeueing packets from backlog to prequeue, if | |
1756 | * is not empty. It is more elegant, but eats cycles, | |
1757 | * unfortunately. | |
1758 | */ | |
b03efcfb | 1759 | if (!skb_queue_empty(&tp->ucopy.prequeue)) |
1da177e4 LT |
1760 | goto do_prequeue; |
1761 | ||
1762 | /* __ Set realtime policy in scheduler __ */ | |
1763 | } | |
1764 | ||
73852e81 | 1765 | #ifdef CONFIG_NET_DMA |
15c04175 MK |
1766 | if (tp->ucopy.dma_chan) { |
1767 | if (tp->rcv_wnd == 0 && | |
1768 | !skb_queue_empty(&sk->sk_async_wait_queue)) { | |
1769 | tcp_service_net_dma(sk, true); | |
1770 | tcp_cleanup_rbuf(sk, copied); | |
1771 | } else | |
b9ee8683 | 1772 | dma_async_issue_pending(tp->ucopy.dma_chan); |
15c04175 | 1773 | } |
73852e81 | 1774 | #endif |
1da177e4 LT |
1775 | if (copied >= target) { |
1776 | /* Do not sleep, just process backlog. */ | |
1777 | release_sock(sk); | |
1778 | lock_sock(sk); | |
1779 | } else | |
1780 | sk_wait_data(sk, &timeo); | |
1781 | ||
1a2449a8 | 1782 | #ifdef CONFIG_NET_DMA |
73852e81 | 1783 | tcp_service_net_dma(sk, false); /* Don't block */ |
1a2449a8 CL |
1784 | tp->ucopy.wakeup = 0; |
1785 | #endif | |
1786 | ||
1da177e4 LT |
1787 | if (user_recv) { |
1788 | int chunk; | |
1789 | ||
1790 | /* __ Restore normal policy in scheduler __ */ | |
1791 | ||
1792 | if ((chunk = len - tp->ucopy.len) != 0) { | |
ed88098e | 1793 | NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMBACKLOG, chunk); |
1da177e4 LT |
1794 | len -= chunk; |
1795 | copied += chunk; | |
1796 | } | |
1797 | ||
1798 | if (tp->rcv_nxt == tp->copied_seq && | |
b03efcfb | 1799 | !skb_queue_empty(&tp->ucopy.prequeue)) { |
1da177e4 LT |
1800 | do_prequeue: |
1801 | tcp_prequeue_process(sk); | |
1802 | ||
1803 | if ((chunk = len - tp->ucopy.len) != 0) { | |
ed88098e | 1804 | NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk); |
1da177e4 LT |
1805 | len -= chunk; |
1806 | copied += chunk; | |
1807 | } | |
1808 | } | |
1809 | } | |
77527313 IJ |
1810 | if ((flags & MSG_PEEK) && |
1811 | (peek_seq - copied - urg_hole != tp->copied_seq)) { | |
e87cc472 JP |
1812 | net_dbg_ratelimited("TCP(%s:%d): Application bug, race in MSG_PEEK\n", |
1813 | current->comm, | |
1814 | task_pid_nr(current)); | |
1da177e4 LT |
1815 | peek_seq = tp->copied_seq; |
1816 | } | |
1817 | continue; | |
1818 | ||
1819 | found_ok_skb: | |
1820 | /* Ok so how much can we use? */ | |
1821 | used = skb->len - offset; | |
1822 | if (len < used) | |
1823 | used = len; | |
1824 | ||
1825 | /* Do we have urgent data here? */ | |
1826 | if (tp->urg_data) { | |
1827 | u32 urg_offset = tp->urg_seq - *seq; | |
1828 | if (urg_offset < used) { | |
1829 | if (!urg_offset) { | |
1830 | if (!sock_flag(sk, SOCK_URGINLINE)) { | |
1831 | ++*seq; | |
77527313 | 1832 | urg_hole++; |
1da177e4 LT |
1833 | offset++; |
1834 | used--; | |
1835 | if (!used) | |
1836 | goto skip_copy; | |
1837 | } | |
1838 | } else | |
1839 | used = urg_offset; | |
1840 | } | |
1841 | } | |
1842 | ||
1843 | if (!(flags & MSG_TRUNC)) { | |
1a2449a8 CL |
1844 | #ifdef CONFIG_NET_DMA |
1845 | if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list) | |
a2bd1140 | 1846 | tp->ucopy.dma_chan = net_dma_find_channel(); |
1a2449a8 CL |
1847 | |
1848 | if (tp->ucopy.dma_chan) { | |
1849 | tp->ucopy.dma_cookie = dma_skb_copy_datagram_iovec( | |
1850 | tp->ucopy.dma_chan, skb, offset, | |
1851 | msg->msg_iov, used, | |
1852 | tp->ucopy.pinned_list); | |
1853 | ||
1854 | if (tp->ucopy.dma_cookie < 0) { | |
1855 | ||
afd46503 JP |
1856 | pr_alert("%s: dma_cookie < 0\n", |
1857 | __func__); | |
1a2449a8 CL |
1858 | |
1859 | /* Exception. Bailout! */ | |
1860 | if (!copied) | |
1861 | copied = -EFAULT; | |
1862 | break; | |
1863 | } | |
73852e81 | 1864 | |
b9ee8683 | 1865 | dma_async_issue_pending(tp->ucopy.dma_chan); |
73852e81 | 1866 | |
1a2449a8 | 1867 | if ((offset + used) == skb->len) |
dc6b9b78 | 1868 | copied_early = true; |
1a2449a8 CL |
1869 | |
1870 | } else | |
1871 | #endif | |
1872 | { | |
1873 | err = skb_copy_datagram_iovec(skb, offset, | |
1874 | msg->msg_iov, used); | |
1875 | if (err) { | |
1876 | /* Exception. Bailout! */ | |
1877 | if (!copied) | |
1878 | copied = -EFAULT; | |
1879 | break; | |
1880 | } | |
1da177e4 LT |
1881 | } |
1882 | } | |
1883 | ||
1884 | *seq += used; | |
1885 | copied += used; | |
1886 | len -= used; | |
1887 | ||
1888 | tcp_rcv_space_adjust(sk); | |
1889 | ||
1890 | skip_copy: | |
1891 | if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) { | |
1892 | tp->urg_data = 0; | |
9e412ba7 | 1893 | tcp_fast_path_check(sk); |
1da177e4 LT |
1894 | } |
1895 | if (used + offset < skb->len) | |
1896 | continue; | |
1897 | ||
aa8223c7 | 1898 | if (tcp_hdr(skb)->fin) |
1da177e4 | 1899 | goto found_fin_ok; |
1a2449a8 CL |
1900 | if (!(flags & MSG_PEEK)) { |
1901 | sk_eat_skb(sk, skb, copied_early); | |
dc6b9b78 | 1902 | copied_early = false; |
1a2449a8 | 1903 | } |
1da177e4 LT |
1904 | continue; |
1905 | ||
1906 | found_fin_ok: | |
1907 | /* Process the FIN. */ | |
1908 | ++*seq; | |
1a2449a8 CL |
1909 | if (!(flags & MSG_PEEK)) { |
1910 | sk_eat_skb(sk, skb, copied_early); | |
dc6b9b78 | 1911 | copied_early = false; |
1a2449a8 | 1912 | } |
1da177e4 LT |
1913 | break; |
1914 | } while (len > 0); | |
1915 | ||
1916 | if (user_recv) { | |
b03efcfb | 1917 | if (!skb_queue_empty(&tp->ucopy.prequeue)) { |
1da177e4 LT |
1918 | int chunk; |
1919 | ||
1920 | tp->ucopy.len = copied > 0 ? len : 0; | |
1921 | ||
1922 | tcp_prequeue_process(sk); | |
1923 | ||
1924 | if (copied > 0 && (chunk = len - tp->ucopy.len) != 0) { | |
ed88098e | 1925 | NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk); |
1da177e4 LT |
1926 | len -= chunk; |
1927 | copied += chunk; | |
1928 | } | |
1929 | } | |
1930 | ||
1931 | tp->ucopy.task = NULL; | |
1932 | tp->ucopy.len = 0; | |
1933 | } | |
1934 | ||
1a2449a8 | 1935 | #ifdef CONFIG_NET_DMA |
73852e81 SM |
1936 | tcp_service_net_dma(sk, true); /* Wait for queue to drain */ |
1937 | tp->ucopy.dma_chan = NULL; | |
1a2449a8 | 1938 | |
1a2449a8 CL |
1939 | if (tp->ucopy.pinned_list) { |
1940 | dma_unpin_iovec_pages(tp->ucopy.pinned_list); | |
1941 | tp->ucopy.pinned_list = NULL; | |
1942 | } | |
1943 | #endif | |
1944 | ||
1da177e4 LT |
1945 | /* According to UNIX98, msg_name/msg_namelen are ignored |
1946 | * on connected socket. I was just happy when found this 8) --ANK | |
1947 | */ | |
1948 | ||
1949 | /* Clean up data we have read: This will do ACK frames. */ | |
0e4b4992 | 1950 | tcp_cleanup_rbuf(sk, copied); |
1da177e4 | 1951 | |
1da177e4 LT |
1952 | release_sock(sk); |
1953 | return copied; | |
1954 | ||
1955 | out: | |
1da177e4 LT |
1956 | release_sock(sk); |
1957 | return err; | |
1958 | ||
1959 | recv_urg: | |
377f0a08 | 1960 | err = tcp_recv_urg(sk, msg, len, flags); |
1da177e4 | 1961 | goto out; |
c0e88ff0 PE |
1962 | |
1963 | recv_sndq: | |
1964 | err = tcp_peek_sndq(sk, msg, len); | |
1965 | goto out; | |
1da177e4 | 1966 | } |
4bc2f18b | 1967 | EXPORT_SYMBOL(tcp_recvmsg); |
1da177e4 | 1968 | |
490d5046 IJ |
1969 | void tcp_set_state(struct sock *sk, int state) |
1970 | { | |
1971 | int oldstate = sk->sk_state; | |
1972 | ||
1973 | switch (state) { | |
1974 | case TCP_ESTABLISHED: | |
1975 | if (oldstate != TCP_ESTABLISHED) | |
81cc8a75 | 1976 | TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB); |
490d5046 IJ |
1977 | break; |
1978 | ||
1979 | case TCP_CLOSE: | |
1980 | if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED) | |
81cc8a75 | 1981 | TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS); |
490d5046 IJ |
1982 | |
1983 | sk->sk_prot->unhash(sk); | |
1984 | if (inet_csk(sk)->icsk_bind_hash && | |
1985 | !(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) | |
ab1e0a13 | 1986 | inet_put_port(sk); |
490d5046 IJ |
1987 | /* fall through */ |
1988 | default: | |
5a5f3a8d | 1989 | if (oldstate == TCP_ESTABLISHED) |
74688e48 | 1990 | TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB); |
490d5046 IJ |
1991 | } |
1992 | ||
1993 | /* Change state AFTER socket is unhashed to avoid closed | |
1994 | * socket sitting in hash tables. | |
1995 | */ | |
1996 | sk->sk_state = state; | |
1997 | ||
1998 | #ifdef STATE_TRACE | |
5a5f3a8d | 1999 | SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n", sk, statename[oldstate], statename[state]); |
490d5046 IJ |
2000 | #endif |
2001 | } | |
2002 | EXPORT_SYMBOL_GPL(tcp_set_state); | |
2003 | ||
1da177e4 LT |
2004 | /* |
2005 | * State processing on a close. This implements the state shift for | |
2006 | * sending our FIN frame. Note that we only send a FIN for some | |
2007 | * states. A shutdown() may have already sent the FIN, or we may be | |
2008 | * closed. | |
2009 | */ | |
2010 | ||
9b5b5cff | 2011 | static const unsigned char new_state[16] = { |
1da177e4 LT |
2012 | /* current state: new state: action: */ |
2013 | /* (Invalid) */ TCP_CLOSE, | |
2014 | /* TCP_ESTABLISHED */ TCP_FIN_WAIT1 | TCP_ACTION_FIN, | |
2015 | /* TCP_SYN_SENT */ TCP_CLOSE, | |
2016 | /* TCP_SYN_RECV */ TCP_FIN_WAIT1 | TCP_ACTION_FIN, | |
2017 | /* TCP_FIN_WAIT1 */ TCP_FIN_WAIT1, | |
2018 | /* TCP_FIN_WAIT2 */ TCP_FIN_WAIT2, | |
2019 | /* TCP_TIME_WAIT */ TCP_CLOSE, | |
2020 | /* TCP_CLOSE */ TCP_CLOSE, | |
2021 | /* TCP_CLOSE_WAIT */ TCP_LAST_ACK | TCP_ACTION_FIN, | |
2022 | /* TCP_LAST_ACK */ TCP_LAST_ACK, | |
2023 | /* TCP_LISTEN */ TCP_CLOSE, | |
2024 | /* TCP_CLOSING */ TCP_CLOSING, | |
2025 | }; | |
2026 | ||
2027 | static int tcp_close_state(struct sock *sk) | |
2028 | { | |
2029 | int next = (int)new_state[sk->sk_state]; | |
2030 | int ns = next & TCP_STATE_MASK; | |
2031 | ||
2032 | tcp_set_state(sk, ns); | |
2033 | ||
2034 | return next & TCP_ACTION_FIN; | |
2035 | } | |
2036 | ||
2037 | /* | |
2038 | * Shutdown the sending side of a connection. Much like close except | |
1f29b058 | 2039 | * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD). |
1da177e4 LT |
2040 | */ |
2041 | ||
2042 | void tcp_shutdown(struct sock *sk, int how) | |
2043 | { | |
2044 | /* We need to grab some memory, and put together a FIN, | |
2045 | * and then put it into the queue to be sent. | |
2046 | * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92. | |
2047 | */ | |
2048 | if (!(how & SEND_SHUTDOWN)) | |
2049 | return; | |
2050 | ||
2051 | /* If we've already sent a FIN, or it's a closed state, skip this. */ | |
2052 | if ((1 << sk->sk_state) & | |
2053 | (TCPF_ESTABLISHED | TCPF_SYN_SENT | | |
2054 | TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) { | |
2055 | /* Clear out any half completed packets. FIN if needed. */ | |
2056 | if (tcp_close_state(sk)) | |
2057 | tcp_send_fin(sk); | |
2058 | } | |
2059 | } | |
4bc2f18b | 2060 | EXPORT_SYMBOL(tcp_shutdown); |
1da177e4 | 2061 | |
efcdbf24 AS |
2062 | bool tcp_check_oom(struct sock *sk, int shift) |
2063 | { | |
2064 | bool too_many_orphans, out_of_socket_memory; | |
2065 | ||
2066 | too_many_orphans = tcp_too_many_orphans(sk, shift); | |
2067 | out_of_socket_memory = tcp_out_of_memory(sk); | |
2068 | ||
e87cc472 JP |
2069 | if (too_many_orphans) |
2070 | net_info_ratelimited("too many orphaned sockets\n"); | |
2071 | if (out_of_socket_memory) | |
2072 | net_info_ratelimited("out of memory -- consider tuning tcp_mem\n"); | |
efcdbf24 AS |
2073 | return too_many_orphans || out_of_socket_memory; |
2074 | } | |
2075 | ||
1da177e4 LT |
2076 | void tcp_close(struct sock *sk, long timeout) |
2077 | { | |
2078 | struct sk_buff *skb; | |
2079 | int data_was_unread = 0; | |
75c2d907 | 2080 | int state; |
1da177e4 LT |
2081 | |
2082 | lock_sock(sk); | |
2083 | sk->sk_shutdown = SHUTDOWN_MASK; | |
2084 | ||
2085 | if (sk->sk_state == TCP_LISTEN) { | |
2086 | tcp_set_state(sk, TCP_CLOSE); | |
2087 | ||
2088 | /* Special case. */ | |
0a5578cf | 2089 | inet_csk_listen_stop(sk); |
1da177e4 LT |
2090 | |
2091 | goto adjudge_to_death; | |
2092 | } | |
2093 | ||
2094 | /* We need to flush the recv. buffs. We do this only on the | |
2095 | * descriptor close, not protocol-sourced closes, because the | |
2096 | * reader process may not have drained the data yet! | |
2097 | */ | |
2098 | while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) { | |
2099 | u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq - | |
aa8223c7 | 2100 | tcp_hdr(skb)->fin; |
1da177e4 LT |
2101 | data_was_unread += len; |
2102 | __kfree_skb(skb); | |
2103 | } | |
2104 | ||
3ab224be | 2105 | sk_mem_reclaim(sk); |
1da177e4 | 2106 | |
565b7b2d KK |
2107 | /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */ |
2108 | if (sk->sk_state == TCP_CLOSE) | |
2109 | goto adjudge_to_death; | |
2110 | ||
65bb723c GR |
2111 | /* As outlined in RFC 2525, section 2.17, we send a RST here because |
2112 | * data was lost. To witness the awful effects of the old behavior of | |
2113 | * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk | |
2114 | * GET in an FTP client, suspend the process, wait for the client to | |
2115 | * advertise a zero window, then kill -9 the FTP client, wheee... | |
2116 | * Note: timeout is always zero in such a case. | |
1da177e4 | 2117 | */ |
ee995283 PE |
2118 | if (unlikely(tcp_sk(sk)->repair)) { |
2119 | sk->sk_prot->disconnect(sk, 0); | |
2120 | } else if (data_was_unread) { | |
1da177e4 | 2121 | /* Unread data was tossed, zap the connection. */ |
6f67c817 | 2122 | NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE); |
1da177e4 | 2123 | tcp_set_state(sk, TCP_CLOSE); |
aa133076 | 2124 | tcp_send_active_reset(sk, sk->sk_allocation); |
1da177e4 LT |
2125 | } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) { |
2126 | /* Check zero linger _after_ checking for unread data. */ | |
2127 | sk->sk_prot->disconnect(sk, 0); | |
6f67c817 | 2128 | NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONDATA); |
1da177e4 LT |
2129 | } else if (tcp_close_state(sk)) { |
2130 | /* We FIN if the application ate all the data before | |
2131 | * zapping the connection. | |
2132 | */ | |
2133 | ||
2134 | /* RED-PEN. Formally speaking, we have broken TCP state | |
2135 | * machine. State transitions: | |
2136 | * | |
2137 | * TCP_ESTABLISHED -> TCP_FIN_WAIT1 | |
2138 | * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible) | |
2139 | * TCP_CLOSE_WAIT -> TCP_LAST_ACK | |
2140 | * | |
2141 | * are legal only when FIN has been sent (i.e. in window), | |
2142 | * rather than queued out of window. Purists blame. | |
2143 | * | |
2144 | * F.e. "RFC state" is ESTABLISHED, | |
2145 | * if Linux state is FIN-WAIT-1, but FIN is still not sent. | |
2146 | * | |
2147 | * The visible declinations are that sometimes | |
2148 | * we enter time-wait state, when it is not required really | |
2149 | * (harmless), do not send active resets, when they are | |
2150 | * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when | |
2151 | * they look as CLOSING or LAST_ACK for Linux) | |
2152 | * Probably, I missed some more holelets. | |
2153 | * --ANK | |
8336886f JC |
2154 | * XXX (TFO) - To start off we don't support SYN+ACK+FIN |
2155 | * in a single packet! (May consider it later but will | |
2156 | * probably need API support or TCP_CORK SYN-ACK until | |
2157 | * data is written and socket is closed.) | |
1da177e4 LT |
2158 | */ |
2159 | tcp_send_fin(sk); | |
2160 | } | |
2161 | ||
2162 | sk_stream_wait_close(sk, timeout); | |
2163 | ||
2164 | adjudge_to_death: | |
75c2d907 HX |
2165 | state = sk->sk_state; |
2166 | sock_hold(sk); | |
2167 | sock_orphan(sk); | |
75c2d907 | 2168 | |
1da177e4 LT |
2169 | /* It is the last release_sock in its life. It will remove backlog. */ |
2170 | release_sock(sk); | |
2171 | ||
2172 | ||
2173 | /* Now socket is owned by kernel and we acquire BH lock | |
2174 | to finish close. No need to check for user refs. | |
2175 | */ | |
2176 | local_bh_disable(); | |
2177 | bh_lock_sock(sk); | |
547b792c | 2178 | WARN_ON(sock_owned_by_user(sk)); |
1da177e4 | 2179 | |
eb4dea58 HX |
2180 | percpu_counter_inc(sk->sk_prot->orphan_count); |
2181 | ||
75c2d907 HX |
2182 | /* Have we already been destroyed by a softirq or backlog? */ |
2183 | if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE) | |
2184 | goto out; | |
1da177e4 LT |
2185 | |
2186 | /* This is a (useful) BSD violating of the RFC. There is a | |
2187 | * problem with TCP as specified in that the other end could | |
2188 | * keep a socket open forever with no application left this end. | |
2189 | * We use a 3 minute timeout (about the same as BSD) then kill | |
2190 | * our end. If they send after that then tough - BUT: long enough | |
2191 | * that we won't make the old 4*rto = almost no time - whoops | |
2192 | * reset mistake. | |
2193 | * | |
2194 | * Nope, it was not mistake. It is really desired behaviour | |
2195 | * f.e. on http servers, when such sockets are useless, but | |
2196 | * consume significant resources. Let's do it with special | |
2197 | * linger2 option. --ANK | |
2198 | */ | |
2199 | ||
2200 | if (sk->sk_state == TCP_FIN_WAIT2) { | |
2201 | struct tcp_sock *tp = tcp_sk(sk); | |
2202 | if (tp->linger2 < 0) { | |
2203 | tcp_set_state(sk, TCP_CLOSE); | |
2204 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
de0744af PE |
2205 | NET_INC_STATS_BH(sock_net(sk), |
2206 | LINUX_MIB_TCPABORTONLINGER); | |
1da177e4 | 2207 | } else { |
463c84b9 | 2208 | const int tmo = tcp_fin_time(sk); |
1da177e4 LT |
2209 | |
2210 | if (tmo > TCP_TIMEWAIT_LEN) { | |
52499afe DM |
2211 | inet_csk_reset_keepalive_timer(sk, |
2212 | tmo - TCP_TIMEWAIT_LEN); | |
1da177e4 | 2213 | } else { |
1da177e4 LT |
2214 | tcp_time_wait(sk, TCP_FIN_WAIT2, tmo); |
2215 | goto out; | |
2216 | } | |
2217 | } | |
2218 | } | |
2219 | if (sk->sk_state != TCP_CLOSE) { | |
3ab224be | 2220 | sk_mem_reclaim(sk); |
efcdbf24 | 2221 | if (tcp_check_oom(sk, 0)) { |
1da177e4 LT |
2222 | tcp_set_state(sk, TCP_CLOSE); |
2223 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
de0744af PE |
2224 | NET_INC_STATS_BH(sock_net(sk), |
2225 | LINUX_MIB_TCPABORTONMEMORY); | |
1da177e4 LT |
2226 | } |
2227 | } | |
1da177e4 | 2228 | |
8336886f JC |
2229 | if (sk->sk_state == TCP_CLOSE) { |
2230 | struct request_sock *req = tcp_sk(sk)->fastopen_rsk; | |
2231 | /* We could get here with a non-NULL req if the socket is | |
2232 | * aborted (e.g., closed with unread data) before 3WHS | |
2233 | * finishes. | |
2234 | */ | |
2235 | if (req != NULL) | |
2236 | reqsk_fastopen_remove(sk, req, false); | |
0a5578cf | 2237 | inet_csk_destroy_sock(sk); |
8336886f | 2238 | } |
1da177e4 LT |
2239 | /* Otherwise, socket is reprieved until protocol close. */ |
2240 | ||
2241 | out: | |
2242 | bh_unlock_sock(sk); | |
2243 | local_bh_enable(); | |
2244 | sock_put(sk); | |
2245 | } | |
4bc2f18b | 2246 | EXPORT_SYMBOL(tcp_close); |
1da177e4 LT |
2247 | |
2248 | /* These states need RST on ABORT according to RFC793 */ | |
2249 | ||
a2a385d6 | 2250 | static inline bool tcp_need_reset(int state) |
1da177e4 LT |
2251 | { |
2252 | return (1 << state) & | |
2253 | (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 | | |
2254 | TCPF_FIN_WAIT2 | TCPF_SYN_RECV); | |
2255 | } | |
2256 | ||
2257 | int tcp_disconnect(struct sock *sk, int flags) | |
2258 | { | |
2259 | struct inet_sock *inet = inet_sk(sk); | |
463c84b9 | 2260 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
2261 | struct tcp_sock *tp = tcp_sk(sk); |
2262 | int err = 0; | |
2263 | int old_state = sk->sk_state; | |
2264 | ||
2265 | if (old_state != TCP_CLOSE) | |
2266 | tcp_set_state(sk, TCP_CLOSE); | |
2267 | ||
2268 | /* ABORT function of RFC793 */ | |
2269 | if (old_state == TCP_LISTEN) { | |
0a5578cf | 2270 | inet_csk_listen_stop(sk); |
ee995283 PE |
2271 | } else if (unlikely(tp->repair)) { |
2272 | sk->sk_err = ECONNABORTED; | |
1da177e4 LT |
2273 | } else if (tcp_need_reset(old_state) || |
2274 | (tp->snd_nxt != tp->write_seq && | |
2275 | (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) { | |
caa20d9a | 2276 | /* The last check adjusts for discrepancy of Linux wrt. RFC |
1da177e4 LT |
2277 | * states |
2278 | */ | |
2279 | tcp_send_active_reset(sk, gfp_any()); | |
2280 | sk->sk_err = ECONNRESET; | |
2281 | } else if (old_state == TCP_SYN_SENT) | |
2282 | sk->sk_err = ECONNRESET; | |
2283 | ||
2284 | tcp_clear_xmit_timers(sk); | |
2285 | __skb_queue_purge(&sk->sk_receive_queue); | |
fe067e8a | 2286 | tcp_write_queue_purge(sk); |
1da177e4 | 2287 | __skb_queue_purge(&tp->out_of_order_queue); |
1a2449a8 CL |
2288 | #ifdef CONFIG_NET_DMA |
2289 | __skb_queue_purge(&sk->sk_async_wait_queue); | |
2290 | #endif | |
1da177e4 | 2291 | |
c720c7e8 | 2292 | inet->inet_dport = 0; |
1da177e4 LT |
2293 | |
2294 | if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) | |
2295 | inet_reset_saddr(sk); | |
2296 | ||
2297 | sk->sk_shutdown = 0; | |
2298 | sock_reset_flag(sk, SOCK_DONE); | |
2299 | tp->srtt = 0; | |
2300 | if ((tp->write_seq += tp->max_window + 2) == 0) | |
2301 | tp->write_seq = 1; | |
463c84b9 | 2302 | icsk->icsk_backoff = 0; |
1da177e4 | 2303 | tp->snd_cwnd = 2; |
6687e988 | 2304 | icsk->icsk_probes_out = 0; |
1da177e4 | 2305 | tp->packets_out = 0; |
0b6a05c1 | 2306 | tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; |
1da177e4 | 2307 | tp->snd_cwnd_cnt = 0; |
1fdf475a | 2308 | tp->window_clamp = 0; |
6687e988 | 2309 | tcp_set_ca_state(sk, TCP_CA_Open); |
1da177e4 | 2310 | tcp_clear_retrans(tp); |
463c84b9 | 2311 | inet_csk_delack_init(sk); |
fe067e8a | 2312 | tcp_init_send_head(sk); |
b40b4f79 | 2313 | memset(&tp->rx_opt, 0, sizeof(tp->rx_opt)); |
1da177e4 LT |
2314 | __sk_dst_reset(sk); |
2315 | ||
c720c7e8 | 2316 | WARN_ON(inet->inet_num && !icsk->icsk_bind_hash); |
1da177e4 LT |
2317 | |
2318 | sk->sk_error_report(sk); | |
2319 | return err; | |
2320 | } | |
4bc2f18b | 2321 | EXPORT_SYMBOL(tcp_disconnect); |
1da177e4 | 2322 | |
bb68b647 CP |
2323 | void tcp_sock_destruct(struct sock *sk) |
2324 | { | |
2325 | inet_sock_destruct(sk); | |
2326 | ||
2327 | kfree(inet_csk(sk)->icsk_accept_queue.fastopenq); | |
2328 | } | |
2329 | ||
a2a385d6 | 2330 | static inline bool tcp_can_repair_sock(const struct sock *sk) |
ee995283 | 2331 | { |
52e804c6 | 2332 | return ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN) && |
ee995283 PE |
2333 | ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_ESTABLISHED)); |
2334 | } | |
2335 | ||
de248a75 PE |
2336 | static int tcp_repair_options_est(struct tcp_sock *tp, |
2337 | struct tcp_repair_opt __user *optbuf, unsigned int len) | |
b139ba4e | 2338 | { |
de248a75 | 2339 | struct tcp_repair_opt opt; |
b139ba4e | 2340 | |
de248a75 PE |
2341 | while (len >= sizeof(opt)) { |
2342 | if (copy_from_user(&opt, optbuf, sizeof(opt))) | |
b139ba4e PE |
2343 | return -EFAULT; |
2344 | ||
2345 | optbuf++; | |
de248a75 | 2346 | len -= sizeof(opt); |
b139ba4e | 2347 | |
de248a75 PE |
2348 | switch (opt.opt_code) { |
2349 | case TCPOPT_MSS: | |
2350 | tp->rx_opt.mss_clamp = opt.opt_val; | |
b139ba4e | 2351 | break; |
de248a75 | 2352 | case TCPOPT_WINDOW: |
bc26ccd8 AV |
2353 | { |
2354 | u16 snd_wscale = opt.opt_val & 0xFFFF; | |
2355 | u16 rcv_wscale = opt.opt_val >> 16; | |
2356 | ||
2357 | if (snd_wscale > 14 || rcv_wscale > 14) | |
2358 | return -EFBIG; | |
b139ba4e | 2359 | |
bc26ccd8 AV |
2360 | tp->rx_opt.snd_wscale = snd_wscale; |
2361 | tp->rx_opt.rcv_wscale = rcv_wscale; | |
2362 | tp->rx_opt.wscale_ok = 1; | |
2363 | } | |
b139ba4e | 2364 | break; |
b139ba4e | 2365 | case TCPOPT_SACK_PERM: |
de248a75 PE |
2366 | if (opt.opt_val != 0) |
2367 | return -EINVAL; | |
2368 | ||
b139ba4e PE |
2369 | tp->rx_opt.sack_ok |= TCP_SACK_SEEN; |
2370 | if (sysctl_tcp_fack) | |
2371 | tcp_enable_fack(tp); | |
2372 | break; | |
2373 | case TCPOPT_TIMESTAMP: | |
de248a75 PE |
2374 | if (opt.opt_val != 0) |
2375 | return -EINVAL; | |
2376 | ||
b139ba4e PE |
2377 | tp->rx_opt.tstamp_ok = 1; |
2378 | break; | |
2379 | } | |
2380 | } | |
2381 | ||
2382 | return 0; | |
2383 | } | |
2384 | ||
1da177e4 LT |
2385 | /* |
2386 | * Socket option code for TCP. | |
2387 | */ | |
3fdadf7d | 2388 | static int do_tcp_setsockopt(struct sock *sk, int level, |
b7058842 | 2389 | int optname, char __user *optval, unsigned int optlen) |
1da177e4 LT |
2390 | { |
2391 | struct tcp_sock *tp = tcp_sk(sk); | |
463c84b9 | 2392 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
2393 | int val; |
2394 | int err = 0; | |
2395 | ||
e56fb50f WAS |
2396 | /* These are data/string values, all the others are ints */ |
2397 | switch (optname) { | |
2398 | case TCP_CONGESTION: { | |
5f8ef48d SH |
2399 | char name[TCP_CA_NAME_MAX]; |
2400 | ||
2401 | if (optlen < 1) | |
2402 | return -EINVAL; | |
2403 | ||
2404 | val = strncpy_from_user(name, optval, | |
4fdb78d3 | 2405 | min_t(long, TCP_CA_NAME_MAX-1, optlen)); |
5f8ef48d SH |
2406 | if (val < 0) |
2407 | return -EFAULT; | |
2408 | name[val] = 0; | |
2409 | ||
2410 | lock_sock(sk); | |
6687e988 | 2411 | err = tcp_set_congestion_control(sk, name); |
5f8ef48d SH |
2412 | release_sock(sk); |
2413 | return err; | |
2414 | } | |
e56fb50f WAS |
2415 | default: |
2416 | /* fallthru */ | |
2417 | break; | |
ccbd6a5a | 2418 | } |
5f8ef48d | 2419 | |
1da177e4 LT |
2420 | if (optlen < sizeof(int)) |
2421 | return -EINVAL; | |
2422 | ||
2423 | if (get_user(val, (int __user *)optval)) | |
2424 | return -EFAULT; | |
2425 | ||
2426 | lock_sock(sk); | |
2427 | ||
2428 | switch (optname) { | |
2429 | case TCP_MAXSEG: | |
2430 | /* Values greater than interface MTU won't take effect. However | |
2431 | * at the point when this call is done we typically don't yet | |
2432 | * know which interface is going to be used */ | |
c39508d6 | 2433 | if (val < TCP_MIN_MSS || val > MAX_TCP_WINDOW) { |
1da177e4 LT |
2434 | err = -EINVAL; |
2435 | break; | |
2436 | } | |
2437 | tp->rx_opt.user_mss = val; | |
2438 | break; | |
2439 | ||
2440 | case TCP_NODELAY: | |
2441 | if (val) { | |
2442 | /* TCP_NODELAY is weaker than TCP_CORK, so that | |
2443 | * this option on corked socket is remembered, but | |
2444 | * it is not activated until cork is cleared. | |
2445 | * | |
2446 | * However, when TCP_NODELAY is set we make | |
2447 | * an explicit push, which overrides even TCP_CORK | |
2448 | * for currently queued segments. | |
2449 | */ | |
2450 | tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH; | |
9e412ba7 | 2451 | tcp_push_pending_frames(sk); |
1da177e4 LT |
2452 | } else { |
2453 | tp->nonagle &= ~TCP_NAGLE_OFF; | |
2454 | } | |
2455 | break; | |
2456 | ||
36e31b0a AP |
2457 | case TCP_THIN_LINEAR_TIMEOUTS: |
2458 | if (val < 0 || val > 1) | |
2459 | err = -EINVAL; | |
2460 | else | |
2461 | tp->thin_lto = val; | |
2462 | break; | |
2463 | ||
7e380175 AP |
2464 | case TCP_THIN_DUPACK: |
2465 | if (val < 0 || val > 1) | |
2466 | err = -EINVAL; | |
e2e5c4c0 | 2467 | else { |
7e380175 | 2468 | tp->thin_dupack = val; |
eed530b6 YC |
2469 | if (tp->thin_dupack) |
2470 | tcp_disable_early_retrans(tp); | |
e2e5c4c0 | 2471 | } |
7e380175 AP |
2472 | break; |
2473 | ||
ee995283 PE |
2474 | case TCP_REPAIR: |
2475 | if (!tcp_can_repair_sock(sk)) | |
2476 | err = -EPERM; | |
2477 | else if (val == 1) { | |
2478 | tp->repair = 1; | |
2479 | sk->sk_reuse = SK_FORCE_REUSE; | |
2480 | tp->repair_queue = TCP_NO_QUEUE; | |
2481 | } else if (val == 0) { | |
2482 | tp->repair = 0; | |
2483 | sk->sk_reuse = SK_NO_REUSE; | |
2484 | tcp_send_window_probe(sk); | |
2485 | } else | |
2486 | err = -EINVAL; | |
2487 | ||
2488 | break; | |
2489 | ||
2490 | case TCP_REPAIR_QUEUE: | |
2491 | if (!tp->repair) | |
2492 | err = -EPERM; | |
2493 | else if (val < TCP_QUEUES_NR) | |
2494 | tp->repair_queue = val; | |
2495 | else | |
2496 | err = -EINVAL; | |
2497 | break; | |
2498 | ||
2499 | case TCP_QUEUE_SEQ: | |
2500 | if (sk->sk_state != TCP_CLOSE) | |
2501 | err = -EPERM; | |
2502 | else if (tp->repair_queue == TCP_SEND_QUEUE) | |
2503 | tp->write_seq = val; | |
2504 | else if (tp->repair_queue == TCP_RECV_QUEUE) | |
2505 | tp->rcv_nxt = val; | |
2506 | else | |
2507 | err = -EINVAL; | |
2508 | break; | |
2509 | ||
b139ba4e PE |
2510 | case TCP_REPAIR_OPTIONS: |
2511 | if (!tp->repair) | |
2512 | err = -EINVAL; | |
2513 | else if (sk->sk_state == TCP_ESTABLISHED) | |
de248a75 PE |
2514 | err = tcp_repair_options_est(tp, |
2515 | (struct tcp_repair_opt __user *)optval, | |
2516 | optlen); | |
b139ba4e PE |
2517 | else |
2518 | err = -EPERM; | |
2519 | break; | |
2520 | ||
1da177e4 LT |
2521 | case TCP_CORK: |
2522 | /* When set indicates to always queue non-full frames. | |
2523 | * Later the user clears this option and we transmit | |
2524 | * any pending partial frames in the queue. This is | |
2525 | * meant to be used alongside sendfile() to get properly | |
2526 | * filled frames when the user (for example) must write | |
2527 | * out headers with a write() call first and then use | |
2528 | * sendfile to send out the data parts. | |
2529 | * | |
2530 | * TCP_CORK can be set together with TCP_NODELAY and it is | |
2531 | * stronger than TCP_NODELAY. | |
2532 | */ | |
2533 | if (val) { | |
2534 | tp->nonagle |= TCP_NAGLE_CORK; | |
2535 | } else { | |
2536 | tp->nonagle &= ~TCP_NAGLE_CORK; | |
2537 | if (tp->nonagle&TCP_NAGLE_OFF) | |
2538 | tp->nonagle |= TCP_NAGLE_PUSH; | |
9e412ba7 | 2539 | tcp_push_pending_frames(sk); |
1da177e4 LT |
2540 | } |
2541 | break; | |
2542 | ||
2543 | case TCP_KEEPIDLE: | |
2544 | if (val < 1 || val > MAX_TCP_KEEPIDLE) | |
2545 | err = -EINVAL; | |
2546 | else { | |
2547 | tp->keepalive_time = val * HZ; | |
2548 | if (sock_flag(sk, SOCK_KEEPOPEN) && | |
2549 | !((1 << sk->sk_state) & | |
2550 | (TCPF_CLOSE | TCPF_LISTEN))) { | |
6c37e5de | 2551 | u32 elapsed = keepalive_time_elapsed(tp); |
1da177e4 LT |
2552 | if (tp->keepalive_time > elapsed) |
2553 | elapsed = tp->keepalive_time - elapsed; | |
2554 | else | |
2555 | elapsed = 0; | |
463c84b9 | 2556 | inet_csk_reset_keepalive_timer(sk, elapsed); |
1da177e4 LT |
2557 | } |
2558 | } | |
2559 | break; | |
2560 | case TCP_KEEPINTVL: | |
2561 | if (val < 1 || val > MAX_TCP_KEEPINTVL) | |
2562 | err = -EINVAL; | |
2563 | else | |
2564 | tp->keepalive_intvl = val * HZ; | |
2565 | break; | |
2566 | case TCP_KEEPCNT: | |
2567 | if (val < 1 || val > MAX_TCP_KEEPCNT) | |
2568 | err = -EINVAL; | |
2569 | else | |
2570 | tp->keepalive_probes = val; | |
2571 | break; | |
2572 | case TCP_SYNCNT: | |
2573 | if (val < 1 || val > MAX_TCP_SYNCNT) | |
2574 | err = -EINVAL; | |
2575 | else | |
463c84b9 | 2576 | icsk->icsk_syn_retries = val; |
1da177e4 LT |
2577 | break; |
2578 | ||
2579 | case TCP_LINGER2: | |
2580 | if (val < 0) | |
2581 | tp->linger2 = -1; | |
2582 | else if (val > sysctl_tcp_fin_timeout / HZ) | |
2583 | tp->linger2 = 0; | |
2584 | else | |
2585 | tp->linger2 = val * HZ; | |
2586 | break; | |
2587 | ||
2588 | case TCP_DEFER_ACCEPT: | |
b103cf34 JA |
2589 | /* Translate value in seconds to number of retransmits */ |
2590 | icsk->icsk_accept_queue.rskq_defer_accept = | |
2591 | secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ, | |
2592 | TCP_RTO_MAX / HZ); | |
1da177e4 LT |
2593 | break; |
2594 | ||
2595 | case TCP_WINDOW_CLAMP: | |
2596 | if (!val) { | |
2597 | if (sk->sk_state != TCP_CLOSE) { | |
2598 | err = -EINVAL; | |
2599 | break; | |
2600 | } | |
2601 | tp->window_clamp = 0; | |
2602 | } else | |
2603 | tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ? | |
2604 | SOCK_MIN_RCVBUF / 2 : val; | |
2605 | break; | |
2606 | ||
2607 | case TCP_QUICKACK: | |
2608 | if (!val) { | |
463c84b9 | 2609 | icsk->icsk_ack.pingpong = 1; |
1da177e4 | 2610 | } else { |
463c84b9 | 2611 | icsk->icsk_ack.pingpong = 0; |
1da177e4 LT |
2612 | if ((1 << sk->sk_state) & |
2613 | (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) && | |
463c84b9 ACM |
2614 | inet_csk_ack_scheduled(sk)) { |
2615 | icsk->icsk_ack.pending |= ICSK_ACK_PUSHED; | |
0e4b4992 | 2616 | tcp_cleanup_rbuf(sk, 1); |
1da177e4 | 2617 | if (!(val & 1)) |
463c84b9 | 2618 | icsk->icsk_ack.pingpong = 1; |
1da177e4 LT |
2619 | } |
2620 | } | |
2621 | break; | |
2622 | ||
cfb6eeb4 YH |
2623 | #ifdef CONFIG_TCP_MD5SIG |
2624 | case TCP_MD5SIG: | |
2625 | /* Read the IP->Key mappings from userspace */ | |
2626 | err = tp->af_specific->md5_parse(sk, optval, optlen); | |
2627 | break; | |
2628 | #endif | |
dca43c75 JC |
2629 | case TCP_USER_TIMEOUT: |
2630 | /* Cap the max timeout in ms TCP will retry/retrans | |
2631 | * before giving up and aborting (ETIMEDOUT) a connection. | |
2632 | */ | |
42493570 HL |
2633 | if (val < 0) |
2634 | err = -EINVAL; | |
2635 | else | |
2636 | icsk->icsk_user_timeout = msecs_to_jiffies(val); | |
dca43c75 | 2637 | break; |
8336886f JC |
2638 | |
2639 | case TCP_FASTOPEN: | |
2640 | if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE | | |
2641 | TCPF_LISTEN))) | |
2642 | err = fastopen_init_queue(sk, val); | |
2643 | else | |
2644 | err = -EINVAL; | |
2645 | break; | |
93be6ce0 AV |
2646 | case TCP_TIMESTAMP: |
2647 | if (!tp->repair) | |
2648 | err = -EPERM; | |
2649 | else | |
2650 | tp->tsoffset = val - tcp_time_stamp; | |
2651 | break; | |
c9bee3b7 ED |
2652 | case TCP_NOTSENT_LOWAT: |
2653 | tp->notsent_lowat = val; | |
2654 | sk->sk_write_space(sk); | |
2655 | break; | |
1da177e4 LT |
2656 | default: |
2657 | err = -ENOPROTOOPT; | |
2658 | break; | |
3ff50b79 SH |
2659 | } |
2660 | ||
1da177e4 LT |
2661 | release_sock(sk); |
2662 | return err; | |
2663 | } | |
2664 | ||
3fdadf7d | 2665 | int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval, |
b7058842 | 2666 | unsigned int optlen) |
3fdadf7d | 2667 | { |
cf533ea5 | 2668 | const struct inet_connection_sock *icsk = inet_csk(sk); |
3fdadf7d DM |
2669 | |
2670 | if (level != SOL_TCP) | |
2671 | return icsk->icsk_af_ops->setsockopt(sk, level, optname, | |
2672 | optval, optlen); | |
2673 | return do_tcp_setsockopt(sk, level, optname, optval, optlen); | |
2674 | } | |
4bc2f18b | 2675 | EXPORT_SYMBOL(tcp_setsockopt); |
3fdadf7d DM |
2676 | |
2677 | #ifdef CONFIG_COMPAT | |
543d9cfe | 2678 | int compat_tcp_setsockopt(struct sock *sk, int level, int optname, |
b7058842 | 2679 | char __user *optval, unsigned int optlen) |
3fdadf7d | 2680 | { |
dec73ff0 ACM |
2681 | if (level != SOL_TCP) |
2682 | return inet_csk_compat_setsockopt(sk, level, optname, | |
2683 | optval, optlen); | |
3fdadf7d DM |
2684 | return do_tcp_setsockopt(sk, level, optname, optval, optlen); |
2685 | } | |
543d9cfe | 2686 | EXPORT_SYMBOL(compat_tcp_setsockopt); |
3fdadf7d DM |
2687 | #endif |
2688 | ||
1da177e4 | 2689 | /* Return information about state of tcp endpoint in API format. */ |
cf533ea5 | 2690 | void tcp_get_info(const struct sock *sk, struct tcp_info *info) |
1da177e4 | 2691 | { |
cf533ea5 | 2692 | const struct tcp_sock *tp = tcp_sk(sk); |
463c84b9 | 2693 | const struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
2694 | u32 now = tcp_time_stamp; |
2695 | ||
2696 | memset(info, 0, sizeof(*info)); | |
2697 | ||
2698 | info->tcpi_state = sk->sk_state; | |
6687e988 | 2699 | info->tcpi_ca_state = icsk->icsk_ca_state; |
463c84b9 | 2700 | info->tcpi_retransmits = icsk->icsk_retransmits; |
6687e988 | 2701 | info->tcpi_probes = icsk->icsk_probes_out; |
463c84b9 | 2702 | info->tcpi_backoff = icsk->icsk_backoff; |
1da177e4 LT |
2703 | |
2704 | if (tp->rx_opt.tstamp_ok) | |
2705 | info->tcpi_options |= TCPI_OPT_TIMESTAMPS; | |
e60402d0 | 2706 | if (tcp_is_sack(tp)) |
1da177e4 LT |
2707 | info->tcpi_options |= TCPI_OPT_SACK; |
2708 | if (tp->rx_opt.wscale_ok) { | |
2709 | info->tcpi_options |= TCPI_OPT_WSCALE; | |
2710 | info->tcpi_snd_wscale = tp->rx_opt.snd_wscale; | |
2711 | info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale; | |
e905a9ed | 2712 | } |
1da177e4 | 2713 | |
b5c5693b | 2714 | if (tp->ecn_flags & TCP_ECN_OK) |
1da177e4 | 2715 | info->tcpi_options |= TCPI_OPT_ECN; |
b5c5693b ED |
2716 | if (tp->ecn_flags & TCP_ECN_SEEN) |
2717 | info->tcpi_options |= TCPI_OPT_ECN_SEEN; | |
6f73601e YC |
2718 | if (tp->syn_data_acked) |
2719 | info->tcpi_options |= TCPI_OPT_SYN_DATA; | |
1da177e4 | 2720 | |
463c84b9 ACM |
2721 | info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto); |
2722 | info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato); | |
c1b4a7e6 | 2723 | info->tcpi_snd_mss = tp->mss_cache; |
463c84b9 | 2724 | info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss; |
1da177e4 | 2725 | |
5ee3afba RJ |
2726 | if (sk->sk_state == TCP_LISTEN) { |
2727 | info->tcpi_unacked = sk->sk_ack_backlog; | |
2728 | info->tcpi_sacked = sk->sk_max_ack_backlog; | |
2729 | } else { | |
2730 | info->tcpi_unacked = tp->packets_out; | |
2731 | info->tcpi_sacked = tp->sacked_out; | |
2732 | } | |
1da177e4 LT |
2733 | info->tcpi_lost = tp->lost_out; |
2734 | info->tcpi_retrans = tp->retrans_out; | |
2735 | info->tcpi_fackets = tp->fackets_out; | |
2736 | ||
2737 | info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime); | |
463c84b9 | 2738 | info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime); |
1da177e4 LT |
2739 | info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp); |
2740 | ||
d83d8461 | 2741 | info->tcpi_pmtu = icsk->icsk_pmtu_cookie; |
1da177e4 LT |
2742 | info->tcpi_rcv_ssthresh = tp->rcv_ssthresh; |
2743 | info->tcpi_rtt = jiffies_to_usecs(tp->srtt)>>3; | |
2744 | info->tcpi_rttvar = jiffies_to_usecs(tp->mdev)>>2; | |
2745 | info->tcpi_snd_ssthresh = tp->snd_ssthresh; | |
2746 | info->tcpi_snd_cwnd = tp->snd_cwnd; | |
2747 | info->tcpi_advmss = tp->advmss; | |
2748 | info->tcpi_reordering = tp->reordering; | |
2749 | ||
2750 | info->tcpi_rcv_rtt = jiffies_to_usecs(tp->rcv_rtt_est.rtt)>>3; | |
2751 | info->tcpi_rcv_space = tp->rcvq_space.space; | |
2752 | ||
2753 | info->tcpi_total_retrans = tp->total_retrans; | |
2754 | } | |
1da177e4 LT |
2755 | EXPORT_SYMBOL_GPL(tcp_get_info); |
2756 | ||
3fdadf7d DM |
2757 | static int do_tcp_getsockopt(struct sock *sk, int level, |
2758 | int optname, char __user *optval, int __user *optlen) | |
1da177e4 | 2759 | { |
295f7324 | 2760 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
2761 | struct tcp_sock *tp = tcp_sk(sk); |
2762 | int val, len; | |
2763 | ||
1da177e4 LT |
2764 | if (get_user(len, optlen)) |
2765 | return -EFAULT; | |
2766 | ||
2767 | len = min_t(unsigned int, len, sizeof(int)); | |
2768 | ||
2769 | if (len < 0) | |
2770 | return -EINVAL; | |
2771 | ||
2772 | switch (optname) { | |
2773 | case TCP_MAXSEG: | |
c1b4a7e6 | 2774 | val = tp->mss_cache; |
1da177e4 LT |
2775 | if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) |
2776 | val = tp->rx_opt.user_mss; | |
5e6a3ce6 PE |
2777 | if (tp->repair) |
2778 | val = tp->rx_opt.mss_clamp; | |
1da177e4 LT |
2779 | break; |
2780 | case TCP_NODELAY: | |
2781 | val = !!(tp->nonagle&TCP_NAGLE_OFF); | |
2782 | break; | |
2783 | case TCP_CORK: | |
2784 | val = !!(tp->nonagle&TCP_NAGLE_CORK); | |
2785 | break; | |
2786 | case TCP_KEEPIDLE: | |
df19a626 | 2787 | val = keepalive_time_when(tp) / HZ; |
1da177e4 LT |
2788 | break; |
2789 | case TCP_KEEPINTVL: | |
df19a626 | 2790 | val = keepalive_intvl_when(tp) / HZ; |
1da177e4 LT |
2791 | break; |
2792 | case TCP_KEEPCNT: | |
df19a626 | 2793 | val = keepalive_probes(tp); |
1da177e4 LT |
2794 | break; |
2795 | case TCP_SYNCNT: | |
295f7324 | 2796 | val = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries; |
1da177e4 LT |
2797 | break; |
2798 | case TCP_LINGER2: | |
2799 | val = tp->linger2; | |
2800 | if (val >= 0) | |
2801 | val = (val ? : sysctl_tcp_fin_timeout) / HZ; | |
2802 | break; | |
2803 | case TCP_DEFER_ACCEPT: | |
b103cf34 JA |
2804 | val = retrans_to_secs(icsk->icsk_accept_queue.rskq_defer_accept, |
2805 | TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ); | |
1da177e4 LT |
2806 | break; |
2807 | case TCP_WINDOW_CLAMP: | |
2808 | val = tp->window_clamp; | |
2809 | break; | |
2810 | case TCP_INFO: { | |
2811 | struct tcp_info info; | |
2812 | ||
2813 | if (get_user(len, optlen)) | |
2814 | return -EFAULT; | |
2815 | ||
2816 | tcp_get_info(sk, &info); | |
2817 | ||
2818 | len = min_t(unsigned int, len, sizeof(info)); | |
2819 | if (put_user(len, optlen)) | |
2820 | return -EFAULT; | |
2821 | if (copy_to_user(optval, &info, len)) | |
2822 | return -EFAULT; | |
2823 | return 0; | |
2824 | } | |
2825 | case TCP_QUICKACK: | |
295f7324 | 2826 | val = !icsk->icsk_ack.pingpong; |
1da177e4 | 2827 | break; |
5f8ef48d SH |
2828 | |
2829 | case TCP_CONGESTION: | |
2830 | if (get_user(len, optlen)) | |
2831 | return -EFAULT; | |
2832 | len = min_t(unsigned int, len, TCP_CA_NAME_MAX); | |
2833 | if (put_user(len, optlen)) | |
2834 | return -EFAULT; | |
6687e988 | 2835 | if (copy_to_user(optval, icsk->icsk_ca_ops->name, len)) |
5f8ef48d SH |
2836 | return -EFAULT; |
2837 | return 0; | |
e56fb50f | 2838 | |
3c0fef0b JH |
2839 | case TCP_THIN_LINEAR_TIMEOUTS: |
2840 | val = tp->thin_lto; | |
2841 | break; | |
2842 | case TCP_THIN_DUPACK: | |
2843 | val = tp->thin_dupack; | |
2844 | break; | |
dca43c75 | 2845 | |
ee995283 PE |
2846 | case TCP_REPAIR: |
2847 | val = tp->repair; | |
2848 | break; | |
2849 | ||
2850 | case TCP_REPAIR_QUEUE: | |
2851 | if (tp->repair) | |
2852 | val = tp->repair_queue; | |
2853 | else | |
2854 | return -EINVAL; | |
2855 | break; | |
2856 | ||
2857 | case TCP_QUEUE_SEQ: | |
2858 | if (tp->repair_queue == TCP_SEND_QUEUE) | |
2859 | val = tp->write_seq; | |
2860 | else if (tp->repair_queue == TCP_RECV_QUEUE) | |
2861 | val = tp->rcv_nxt; | |
2862 | else | |
2863 | return -EINVAL; | |
2864 | break; | |
2865 | ||
dca43c75 JC |
2866 | case TCP_USER_TIMEOUT: |
2867 | val = jiffies_to_msecs(icsk->icsk_user_timeout); | |
2868 | break; | |
93be6ce0 AV |
2869 | case TCP_TIMESTAMP: |
2870 | val = tcp_time_stamp + tp->tsoffset; | |
2871 | break; | |
c9bee3b7 ED |
2872 | case TCP_NOTSENT_LOWAT: |
2873 | val = tp->notsent_lowat; | |
2874 | break; | |
1da177e4 LT |
2875 | default: |
2876 | return -ENOPROTOOPT; | |
3ff50b79 | 2877 | } |
1da177e4 LT |
2878 | |
2879 | if (put_user(len, optlen)) | |
2880 | return -EFAULT; | |
2881 | if (copy_to_user(optval, &val, len)) | |
2882 | return -EFAULT; | |
2883 | return 0; | |
2884 | } | |
2885 | ||
3fdadf7d DM |
2886 | int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval, |
2887 | int __user *optlen) | |
2888 | { | |
2889 | struct inet_connection_sock *icsk = inet_csk(sk); | |
2890 | ||
2891 | if (level != SOL_TCP) | |
2892 | return icsk->icsk_af_ops->getsockopt(sk, level, optname, | |
2893 | optval, optlen); | |
2894 | return do_tcp_getsockopt(sk, level, optname, optval, optlen); | |
2895 | } | |
4bc2f18b | 2896 | EXPORT_SYMBOL(tcp_getsockopt); |
3fdadf7d DM |
2897 | |
2898 | #ifdef CONFIG_COMPAT | |
543d9cfe ACM |
2899 | int compat_tcp_getsockopt(struct sock *sk, int level, int optname, |
2900 | char __user *optval, int __user *optlen) | |
3fdadf7d | 2901 | { |
dec73ff0 ACM |
2902 | if (level != SOL_TCP) |
2903 | return inet_csk_compat_getsockopt(sk, level, optname, | |
2904 | optval, optlen); | |
3fdadf7d DM |
2905 | return do_tcp_getsockopt(sk, level, optname, optval, optlen); |
2906 | } | |
543d9cfe | 2907 | EXPORT_SYMBOL(compat_tcp_getsockopt); |
3fdadf7d | 2908 | #endif |
1da177e4 | 2909 | |
cfb6eeb4 | 2910 | #ifdef CONFIG_TCP_MD5SIG |
71cea17e ED |
2911 | static struct tcp_md5sig_pool __percpu *tcp_md5sig_pool __read_mostly; |
2912 | static DEFINE_MUTEX(tcp_md5sig_mutex); | |
cfb6eeb4 | 2913 | |
765cf997 | 2914 | static void __tcp_free_md5sig_pool(struct tcp_md5sig_pool __percpu *pool) |
cfb6eeb4 YH |
2915 | { |
2916 | int cpu; | |
765cf997 | 2917 | |
cfb6eeb4 | 2918 | for_each_possible_cpu(cpu) { |
765cf997 ED |
2919 | struct tcp_md5sig_pool *p = per_cpu_ptr(pool, cpu); |
2920 | ||
2921 | if (p->md5_desc.tfm) | |
2922 | crypto_free_hash(p->md5_desc.tfm); | |
cfb6eeb4 YH |
2923 | } |
2924 | free_percpu(pool); | |
2925 | } | |
2926 | ||
71cea17e | 2927 | static void __tcp_alloc_md5sig_pool(void) |
cfb6eeb4 YH |
2928 | { |
2929 | int cpu; | |
765cf997 | 2930 | struct tcp_md5sig_pool __percpu *pool; |
cfb6eeb4 | 2931 | |
765cf997 | 2932 | pool = alloc_percpu(struct tcp_md5sig_pool); |
cfb6eeb4 | 2933 | if (!pool) |
71cea17e | 2934 | return; |
cfb6eeb4 YH |
2935 | |
2936 | for_each_possible_cpu(cpu) { | |
cfb6eeb4 YH |
2937 | struct crypto_hash *hash; |
2938 | ||
cfb6eeb4 | 2939 | hash = crypto_alloc_hash("md5", 0, CRYPTO_ALG_ASYNC); |
50c3a487 | 2940 | if (IS_ERR_OR_NULL(hash)) |
cfb6eeb4 YH |
2941 | goto out_free; |
2942 | ||
765cf997 | 2943 | per_cpu_ptr(pool, cpu)->md5_desc.tfm = hash; |
cfb6eeb4 | 2944 | } |
71cea17e ED |
2945 | /* before setting tcp_md5sig_pool, we must commit all writes |
2946 | * to memory. See ACCESS_ONCE() in tcp_get_md5sig_pool() | |
2947 | */ | |
2948 | smp_wmb(); | |
2949 | tcp_md5sig_pool = pool; | |
2950 | return; | |
cfb6eeb4 YH |
2951 | out_free: |
2952 | __tcp_free_md5sig_pool(pool); | |
cfb6eeb4 YH |
2953 | } |
2954 | ||
71cea17e | 2955 | bool tcp_alloc_md5sig_pool(void) |
cfb6eeb4 | 2956 | { |
71cea17e ED |
2957 | if (unlikely(!tcp_md5sig_pool)) { |
2958 | mutex_lock(&tcp_md5sig_mutex); | |
2959 | ||
2960 | if (!tcp_md5sig_pool) | |
2961 | __tcp_alloc_md5sig_pool(); | |
2962 | ||
2963 | mutex_unlock(&tcp_md5sig_mutex); | |
cfb6eeb4 | 2964 | } |
71cea17e | 2965 | return tcp_md5sig_pool != NULL; |
cfb6eeb4 | 2966 | } |
cfb6eeb4 YH |
2967 | EXPORT_SYMBOL(tcp_alloc_md5sig_pool); |
2968 | ||
35790c04 ED |
2969 | |
2970 | /** | |
2971 | * tcp_get_md5sig_pool - get md5sig_pool for this user | |
2972 | * | |
2973 | * We use percpu structure, so if we succeed, we exit with preemption | |
2974 | * and BH disabled, to make sure another thread or softirq handling | |
2975 | * wont try to get same context. | |
2976 | */ | |
2977 | struct tcp_md5sig_pool *tcp_get_md5sig_pool(void) | |
cfb6eeb4 | 2978 | { |
765cf997 | 2979 | struct tcp_md5sig_pool __percpu *p; |
35790c04 ED |
2980 | |
2981 | local_bh_disable(); | |
71cea17e | 2982 | p = ACCESS_ONCE(tcp_md5sig_pool); |
35790c04 | 2983 | if (p) |
71cea17e | 2984 | return __this_cpu_ptr(p); |
cfb6eeb4 | 2985 | |
35790c04 ED |
2986 | local_bh_enable(); |
2987 | return NULL; | |
2988 | } | |
2989 | EXPORT_SYMBOL(tcp_get_md5sig_pool); | |
cfb6eeb4 | 2990 | |
49a72dfb | 2991 | int tcp_md5_hash_header(struct tcp_md5sig_pool *hp, |
ca35a0ef | 2992 | const struct tcphdr *th) |
49a72dfb AL |
2993 | { |
2994 | struct scatterlist sg; | |
ca35a0ef | 2995 | struct tcphdr hdr; |
49a72dfb AL |
2996 | int err; |
2997 | ||
ca35a0ef ED |
2998 | /* We are not allowed to change tcphdr, make a local copy */ |
2999 | memcpy(&hdr, th, sizeof(hdr)); | |
3000 | hdr.check = 0; | |
3001 | ||
49a72dfb | 3002 | /* options aren't included in the hash */ |
ca35a0ef ED |
3003 | sg_init_one(&sg, &hdr, sizeof(hdr)); |
3004 | err = crypto_hash_update(&hp->md5_desc, &sg, sizeof(hdr)); | |
49a72dfb AL |
3005 | return err; |
3006 | } | |
49a72dfb AL |
3007 | EXPORT_SYMBOL(tcp_md5_hash_header); |
3008 | ||
3009 | int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp, | |
cf533ea5 | 3010 | const struct sk_buff *skb, unsigned int header_len) |
49a72dfb AL |
3011 | { |
3012 | struct scatterlist sg; | |
3013 | const struct tcphdr *tp = tcp_hdr(skb); | |
3014 | struct hash_desc *desc = &hp->md5_desc; | |
95c96174 ED |
3015 | unsigned int i; |
3016 | const unsigned int head_data_len = skb_headlen(skb) > header_len ? | |
3017 | skb_headlen(skb) - header_len : 0; | |
49a72dfb | 3018 | const struct skb_shared_info *shi = skb_shinfo(skb); |
d7fd1b57 | 3019 | struct sk_buff *frag_iter; |
49a72dfb AL |
3020 | |
3021 | sg_init_table(&sg, 1); | |
3022 | ||
3023 | sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len); | |
3024 | if (crypto_hash_update(desc, &sg, head_data_len)) | |
3025 | return 1; | |
3026 | ||
3027 | for (i = 0; i < shi->nr_frags; ++i) { | |
3028 | const struct skb_frag_struct *f = &shi->frags[i]; | |
54d27fcb ED |
3029 | unsigned int offset = f->page_offset; |
3030 | struct page *page = skb_frag_page(f) + (offset >> PAGE_SHIFT); | |
3031 | ||
3032 | sg_set_page(&sg, page, skb_frag_size(f), | |
3033 | offset_in_page(offset)); | |
9e903e08 | 3034 | if (crypto_hash_update(desc, &sg, skb_frag_size(f))) |
49a72dfb AL |
3035 | return 1; |
3036 | } | |
3037 | ||
d7fd1b57 ED |
3038 | skb_walk_frags(skb, frag_iter) |
3039 | if (tcp_md5_hash_skb_data(hp, frag_iter, 0)) | |
3040 | return 1; | |
3041 | ||
49a72dfb AL |
3042 | return 0; |
3043 | } | |
49a72dfb AL |
3044 | EXPORT_SYMBOL(tcp_md5_hash_skb_data); |
3045 | ||
cf533ea5 | 3046 | int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, const struct tcp_md5sig_key *key) |
49a72dfb AL |
3047 | { |
3048 | struct scatterlist sg; | |
3049 | ||
3050 | sg_init_one(&sg, key->key, key->keylen); | |
3051 | return crypto_hash_update(&hp->md5_desc, &sg, key->keylen); | |
3052 | } | |
49a72dfb AL |
3053 | EXPORT_SYMBOL(tcp_md5_hash_key); |
3054 | ||
cfb6eeb4 YH |
3055 | #endif |
3056 | ||
4ac02bab AK |
3057 | void tcp_done(struct sock *sk) |
3058 | { | |
8336886f JC |
3059 | struct request_sock *req = tcp_sk(sk)->fastopen_rsk; |
3060 | ||
5a5f3a8d | 3061 | if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV) |
63231bdd | 3062 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS); |
4ac02bab AK |
3063 | |
3064 | tcp_set_state(sk, TCP_CLOSE); | |
3065 | tcp_clear_xmit_timers(sk); | |
8336886f JC |
3066 | if (req != NULL) |
3067 | reqsk_fastopen_remove(sk, req, false); | |
4ac02bab AK |
3068 | |
3069 | sk->sk_shutdown = SHUTDOWN_MASK; | |
3070 | ||
3071 | if (!sock_flag(sk, SOCK_DEAD)) | |
3072 | sk->sk_state_change(sk); | |
3073 | else | |
3074 | inet_csk_destroy_sock(sk); | |
3075 | } | |
3076 | EXPORT_SYMBOL_GPL(tcp_done); | |
3077 | ||
5f8ef48d | 3078 | extern struct tcp_congestion_ops tcp_reno; |
1da177e4 LT |
3079 | |
3080 | static __initdata unsigned long thash_entries; | |
3081 | static int __init set_thash_entries(char *str) | |
3082 | { | |
413c27d8 EZ |
3083 | ssize_t ret; |
3084 | ||
1da177e4 LT |
3085 | if (!str) |
3086 | return 0; | |
413c27d8 EZ |
3087 | |
3088 | ret = kstrtoul(str, 0, &thash_entries); | |
3089 | if (ret) | |
3090 | return 0; | |
3091 | ||
1da177e4 LT |
3092 | return 1; |
3093 | } | |
3094 | __setup("thash_entries=", set_thash_entries); | |
3095 | ||
a4fe34bf | 3096 | static void tcp_init_mem(void) |
4acb4190 | 3097 | { |
4acb4190 GC |
3098 | unsigned long limit = nr_free_buffer_pages() / 8; |
3099 | limit = max(limit, 128UL); | |
a4fe34bf EB |
3100 | sysctl_tcp_mem[0] = limit / 4 * 3; |
3101 | sysctl_tcp_mem[1] = limit; | |
3102 | sysctl_tcp_mem[2] = sysctl_tcp_mem[0] * 2; | |
4acb4190 GC |
3103 | } |
3104 | ||
1da177e4 LT |
3105 | void __init tcp_init(void) |
3106 | { | |
3107 | struct sk_buff *skb = NULL; | |
f03d78db | 3108 | unsigned long limit; |
b49960a0 | 3109 | int max_rshare, max_wshare, cnt; |
074b8517 | 3110 | unsigned int i; |
1da177e4 | 3111 | |
1f9e636e | 3112 | BUILD_BUG_ON(sizeof(struct tcp_skb_cb) > sizeof(skb->cb)); |
1da177e4 | 3113 | |
1748376b | 3114 | percpu_counter_init(&tcp_sockets_allocated, 0); |
dd24c001 | 3115 | percpu_counter_init(&tcp_orphan_count, 0); |
6e04e021 ACM |
3116 | tcp_hashinfo.bind_bucket_cachep = |
3117 | kmem_cache_create("tcp_bind_bucket", | |
3118 | sizeof(struct inet_bind_bucket), 0, | |
20c2df83 | 3119 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3120 | |
1da177e4 LT |
3121 | /* Size and allocate the main established and bind bucket |
3122 | * hash tables. | |
3123 | * | |
3124 | * The methodology is similar to that of the buffer cache. | |
3125 | */ | |
6e04e021 | 3126 | tcp_hashinfo.ehash = |
1da177e4 | 3127 | alloc_large_system_hash("TCP established", |
0f7ff927 | 3128 | sizeof(struct inet_ehash_bucket), |
1da177e4 | 3129 | thash_entries, |
fd90b29d | 3130 | 17, /* one slot per 128 KB of memory */ |
9e950efa | 3131 | 0, |
1da177e4 | 3132 | NULL, |
f373b53b | 3133 | &tcp_hashinfo.ehash_mask, |
31fe62b9 | 3134 | 0, |
0ccfe618 | 3135 | thash_entries ? 0 : 512 * 1024); |
05dbc7b5 | 3136 | for (i = 0; i <= tcp_hashinfo.ehash_mask; i++) |
3ab5aee7 | 3137 | INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i); |
05dbc7b5 | 3138 | |
230140cf ED |
3139 | if (inet_ehash_locks_alloc(&tcp_hashinfo)) |
3140 | panic("TCP: failed to alloc ehash_locks"); | |
6e04e021 | 3141 | tcp_hashinfo.bhash = |
1da177e4 | 3142 | alloc_large_system_hash("TCP bind", |
0f7ff927 | 3143 | sizeof(struct inet_bind_hashbucket), |
f373b53b | 3144 | tcp_hashinfo.ehash_mask + 1, |
fd90b29d | 3145 | 17, /* one slot per 128 KB of memory */ |
9e950efa | 3146 | 0, |
6e04e021 | 3147 | &tcp_hashinfo.bhash_size, |
1da177e4 | 3148 | NULL, |
31fe62b9 | 3149 | 0, |
1da177e4 | 3150 | 64 * 1024); |
074b8517 | 3151 | tcp_hashinfo.bhash_size = 1U << tcp_hashinfo.bhash_size; |
6e04e021 ACM |
3152 | for (i = 0; i < tcp_hashinfo.bhash_size; i++) { |
3153 | spin_lock_init(&tcp_hashinfo.bhash[i].lock); | |
3154 | INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain); | |
1da177e4 LT |
3155 | } |
3156 | ||
c5ed63d6 ED |
3157 | |
3158 | cnt = tcp_hashinfo.ehash_mask + 1; | |
3159 | ||
3160 | tcp_death_row.sysctl_max_tw_buckets = cnt / 2; | |
3161 | sysctl_tcp_max_orphans = cnt / 2; | |
3162 | sysctl_max_syn_backlog = max(128, cnt / 256); | |
1da177e4 | 3163 | |
a4fe34bf | 3164 | tcp_init_mem(); |
c43b874d | 3165 | /* Set per-socket limits to no more than 1/128 the pressure threshold */ |
5fb84b14 | 3166 | limit = nr_free_buffer_pages() << (PAGE_SHIFT - 7); |
b49960a0 ED |
3167 | max_wshare = min(4UL*1024*1024, limit); |
3168 | max_rshare = min(6UL*1024*1024, limit); | |
7b4f4b5e | 3169 | |
3ab224be | 3170 | sysctl_tcp_wmem[0] = SK_MEM_QUANTUM; |
7b4f4b5e | 3171 | sysctl_tcp_wmem[1] = 16*1024; |
b49960a0 | 3172 | sysctl_tcp_wmem[2] = max(64*1024, max_wshare); |
7b4f4b5e | 3173 | |
3ab224be | 3174 | sysctl_tcp_rmem[0] = SK_MEM_QUANTUM; |
7b4f4b5e | 3175 | sysctl_tcp_rmem[1] = 87380; |
b49960a0 | 3176 | sysctl_tcp_rmem[2] = max(87380, max_rshare); |
1da177e4 | 3177 | |
afd46503 | 3178 | pr_info("Hash tables configured (established %u bind %u)\n", |
058bd4d2 | 3179 | tcp_hashinfo.ehash_mask + 1, tcp_hashinfo.bhash_size); |
317a76f9 | 3180 | |
51c5d0c4 DM |
3181 | tcp_metrics_init(); |
3182 | ||
317a76f9 | 3183 | tcp_register_congestion_control(&tcp_reno); |
da5c78c8 | 3184 | |
46d3ceab | 3185 | tcp_tasklet_init(); |
1da177e4 | 3186 | } |