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> | |
282 | ||
ab32ea5d | 283 | int sysctl_tcp_fin_timeout __read_mostly = TCP_FIN_TIMEOUT; |
1da177e4 | 284 | |
dd24c001 | 285 | struct percpu_counter tcp_orphan_count; |
0a5578cf ACM |
286 | EXPORT_SYMBOL_GPL(tcp_orphan_count); |
287 | ||
b8059ead DM |
288 | int sysctl_tcp_wmem[3] __read_mostly; |
289 | int sysctl_tcp_rmem[3] __read_mostly; | |
1da177e4 | 290 | |
1da177e4 LT |
291 | EXPORT_SYMBOL(sysctl_tcp_rmem); |
292 | EXPORT_SYMBOL(sysctl_tcp_wmem); | |
293 | ||
8d987e5c | 294 | atomic_long_t tcp_memory_allocated; /* Current allocated memory. */ |
1da177e4 | 295 | EXPORT_SYMBOL(tcp_memory_allocated); |
1748376b ED |
296 | |
297 | /* | |
298 | * Current number of TCP sockets. | |
299 | */ | |
300 | struct percpu_counter tcp_sockets_allocated; | |
1da177e4 LT |
301 | EXPORT_SYMBOL(tcp_sockets_allocated); |
302 | ||
9c55e01c JA |
303 | /* |
304 | * TCP splice context | |
305 | */ | |
306 | struct tcp_splice_state { | |
307 | struct pipe_inode_info *pipe; | |
308 | size_t len; | |
309 | unsigned int flags; | |
310 | }; | |
311 | ||
1da177e4 LT |
312 | /* |
313 | * Pressure flag: try to collapse. | |
314 | * Technical note: it is used by multiple contexts non atomically. | |
3ab224be | 315 | * All the __sk_mem_schedule() is of this nature: accounting |
1da177e4 LT |
316 | * is strict, actions are advisory and have some latency. |
317 | */ | |
4103f8cd | 318 | int tcp_memory_pressure __read_mostly; |
1da177e4 LT |
319 | EXPORT_SYMBOL(tcp_memory_pressure); |
320 | ||
5c52ba17 | 321 | void tcp_enter_memory_pressure(struct sock *sk) |
1da177e4 LT |
322 | { |
323 | if (!tcp_memory_pressure) { | |
4e673444 | 324 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES); |
1da177e4 LT |
325 | tcp_memory_pressure = 1; |
326 | } | |
327 | } | |
1da177e4 LT |
328 | EXPORT_SYMBOL(tcp_enter_memory_pressure); |
329 | ||
b103cf34 JA |
330 | /* Convert seconds to retransmits based on initial and max timeout */ |
331 | static u8 secs_to_retrans(int seconds, int timeout, int rto_max) | |
332 | { | |
333 | u8 res = 0; | |
334 | ||
335 | if (seconds > 0) { | |
336 | int period = timeout; | |
337 | ||
338 | res = 1; | |
339 | while (seconds > period && res < 255) { | |
340 | res++; | |
341 | timeout <<= 1; | |
342 | if (timeout > rto_max) | |
343 | timeout = rto_max; | |
344 | period += timeout; | |
345 | } | |
346 | } | |
347 | return res; | |
348 | } | |
349 | ||
350 | /* Convert retransmits to seconds based on initial and max timeout */ | |
351 | static int retrans_to_secs(u8 retrans, int timeout, int rto_max) | |
352 | { | |
353 | int period = 0; | |
354 | ||
355 | if (retrans > 0) { | |
356 | period = timeout; | |
357 | while (--retrans) { | |
358 | timeout <<= 1; | |
359 | if (timeout > rto_max) | |
360 | timeout = rto_max; | |
361 | period += timeout; | |
362 | } | |
363 | } | |
364 | return period; | |
365 | } | |
366 | ||
900f65d3 NC |
367 | /* Address-family independent initialization for a tcp_sock. |
368 | * | |
369 | * NOTE: A lot of things set to zero explicitly by call to | |
370 | * sk_alloc() so need not be done here. | |
371 | */ | |
372 | void tcp_init_sock(struct sock *sk) | |
373 | { | |
374 | struct inet_connection_sock *icsk = inet_csk(sk); | |
375 | struct tcp_sock *tp = tcp_sk(sk); | |
376 | ||
377 | skb_queue_head_init(&tp->out_of_order_queue); | |
378 | tcp_init_xmit_timers(sk); | |
379 | tcp_prequeue_init(tp); | |
46d3ceab | 380 | INIT_LIST_HEAD(&tp->tsq_node); |
900f65d3 NC |
381 | |
382 | icsk->icsk_rto = TCP_TIMEOUT_INIT; | |
383 | tp->mdev = TCP_TIMEOUT_INIT; | |
384 | ||
385 | /* So many TCP implementations out there (incorrectly) count the | |
386 | * initial SYN frame in their delayed-ACK and congestion control | |
387 | * algorithms that we must have the following bandaid to talk | |
388 | * efficiently to them. -DaveM | |
389 | */ | |
390 | tp->snd_cwnd = TCP_INIT_CWND; | |
391 | ||
392 | /* See draft-stevens-tcpca-spec-01 for discussion of the | |
393 | * initialization of these values. | |
394 | */ | |
395 | tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; | |
396 | tp->snd_cwnd_clamp = ~0; | |
397 | tp->mss_cache = TCP_MSS_DEFAULT; | |
398 | ||
399 | tp->reordering = sysctl_tcp_reordering; | |
eed530b6 | 400 | tcp_enable_early_retrans(tp); |
900f65d3 NC |
401 | icsk->icsk_ca_ops = &tcp_init_congestion_ops; |
402 | ||
403 | sk->sk_state = TCP_CLOSE; | |
404 | ||
405 | sk->sk_write_space = sk_stream_write_space; | |
406 | sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); | |
407 | ||
408 | icsk->icsk_sync_mss = tcp_sync_mss; | |
409 | ||
410 | /* TCP Cookie Transactions */ | |
411 | if (sysctl_tcp_cookie_size > 0) { | |
412 | /* Default, cookies without s_data_payload. */ | |
413 | tp->cookie_values = | |
414 | kzalloc(sizeof(*tp->cookie_values), | |
415 | sk->sk_allocation); | |
416 | if (tp->cookie_values != NULL) | |
417 | kref_init(&tp->cookie_values->kref); | |
418 | } | |
419 | /* Presumed zeroed, in order of appearance: | |
420 | * cookie_in_always, cookie_out_never, | |
421 | * s_data_constant, s_data_in, s_data_out | |
422 | */ | |
423 | sk->sk_sndbuf = sysctl_tcp_wmem[1]; | |
424 | sk->sk_rcvbuf = sysctl_tcp_rmem[1]; | |
425 | ||
426 | local_bh_disable(); | |
427 | sock_update_memcg(sk); | |
428 | sk_sockets_allocated_inc(sk); | |
429 | local_bh_enable(); | |
430 | } | |
431 | EXPORT_SYMBOL(tcp_init_sock); | |
432 | ||
1da177e4 LT |
433 | /* |
434 | * Wait for a TCP event. | |
435 | * | |
436 | * Note that we don't need to lock the socket, as the upper poll layers | |
437 | * take care of normal races (between the test and the event) and we don't | |
438 | * go look at any of the socket buffers directly. | |
439 | */ | |
440 | unsigned int tcp_poll(struct file *file, struct socket *sock, poll_table *wait) | |
441 | { | |
442 | unsigned int mask; | |
443 | struct sock *sk = sock->sk; | |
cf533ea5 | 444 | const struct tcp_sock *tp = tcp_sk(sk); |
1da177e4 | 445 | |
aa395145 | 446 | sock_poll_wait(file, sk_sleep(sk), wait); |
1da177e4 | 447 | if (sk->sk_state == TCP_LISTEN) |
dc40c7bc | 448 | return inet_csk_listen_poll(sk); |
1da177e4 LT |
449 | |
450 | /* Socket is not locked. We are protected from async events | |
70efce27 WN |
451 | * by poll logic and correct handling of state changes |
452 | * made by other threads is impossible in any case. | |
1da177e4 LT |
453 | */ |
454 | ||
455 | mask = 0; | |
1da177e4 LT |
456 | |
457 | /* | |
458 | * POLLHUP is certainly not done right. But poll() doesn't | |
459 | * have a notion of HUP in just one direction, and for a | |
460 | * socket the read side is more interesting. | |
461 | * | |
462 | * Some poll() documentation says that POLLHUP is incompatible | |
463 | * with the POLLOUT/POLLWR flags, so somebody should check this | |
464 | * all. But careful, it tends to be safer to return too many | |
465 | * bits than too few, and you can easily break real applications | |
466 | * if you don't tell them that something has hung up! | |
467 | * | |
468 | * Check-me. | |
469 | * | |
470 | * Check number 1. POLLHUP is _UNMASKABLE_ event (see UNIX98 and | |
471 | * our fs/select.c). It means that after we received EOF, | |
472 | * poll always returns immediately, making impossible poll() on write() | |
473 | * in state CLOSE_WAIT. One solution is evident --- to set POLLHUP | |
474 | * if and only if shutdown has been made in both directions. | |
475 | * Actually, it is interesting to look how Solaris and DUX | |
70efce27 | 476 | * solve this dilemma. I would prefer, if POLLHUP were maskable, |
1da177e4 LT |
477 | * then we could set it on SND_SHUTDOWN. BTW examples given |
478 | * in Stevens' books assume exactly this behaviour, it explains | |
70efce27 | 479 | * why POLLHUP is incompatible with POLLOUT. --ANK |
1da177e4 LT |
480 | * |
481 | * NOTE. Check for TCP_CLOSE is added. The goal is to prevent | |
482 | * blocking on fresh not-connected or disconnected socket. --ANK | |
483 | */ | |
484 | if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == TCP_CLOSE) | |
485 | mask |= POLLHUP; | |
486 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
f348d70a | 487 | mask |= POLLIN | POLLRDNORM | POLLRDHUP; |
1da177e4 LT |
488 | |
489 | /* Connected? */ | |
490 | if ((1 << sk->sk_state) & ~(TCPF_SYN_SENT | TCPF_SYN_RECV)) { | |
c7004482 DM |
491 | int target = sock_rcvlowat(sk, 0, INT_MAX); |
492 | ||
493 | if (tp->urg_seq == tp->copied_seq && | |
494 | !sock_flag(sk, SOCK_URGINLINE) && | |
495 | tp->urg_data) | |
b634f875 | 496 | target++; |
c7004482 | 497 | |
1da177e4 LT |
498 | /* Potential race condition. If read of tp below will |
499 | * escape above sk->sk_state, we can be illegally awaken | |
500 | * in SYN_* states. */ | |
c7004482 | 501 | if (tp->rcv_nxt - tp->copied_seq >= target) |
1da177e4 LT |
502 | mask |= POLLIN | POLLRDNORM; |
503 | ||
504 | if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { | |
505 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) { | |
506 | mask |= POLLOUT | POLLWRNORM; | |
507 | } else { /* send SIGIO later */ | |
508 | set_bit(SOCK_ASYNC_NOSPACE, | |
509 | &sk->sk_socket->flags); | |
510 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
511 | ||
512 | /* Race breaker. If space is freed after | |
513 | * wspace test but before the flags are set, | |
514 | * IO signal will be lost. | |
515 | */ | |
516 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) | |
517 | mask |= POLLOUT | POLLWRNORM; | |
518 | } | |
d84ba638 KM |
519 | } else |
520 | mask |= POLLOUT | POLLWRNORM; | |
1da177e4 LT |
521 | |
522 | if (tp->urg_data & TCP_URG_VALID) | |
523 | mask |= POLLPRI; | |
524 | } | |
a4d25803 TM |
525 | /* This barrier is coupled with smp_wmb() in tcp_reset() */ |
526 | smp_rmb(); | |
527 | if (sk->sk_err) | |
528 | mask |= POLLERR; | |
529 | ||
1da177e4 LT |
530 | return mask; |
531 | } | |
4bc2f18b | 532 | EXPORT_SYMBOL(tcp_poll); |
1da177e4 LT |
533 | |
534 | int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg) | |
535 | { | |
536 | struct tcp_sock *tp = tcp_sk(sk); | |
537 | int answ; | |
538 | ||
539 | switch (cmd) { | |
540 | case SIOCINQ: | |
541 | if (sk->sk_state == TCP_LISTEN) | |
542 | return -EINVAL; | |
543 | ||
544 | lock_sock(sk); | |
545 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) | |
546 | answ = 0; | |
547 | else if (sock_flag(sk, SOCK_URGINLINE) || | |
548 | !tp->urg_data || | |
549 | before(tp->urg_seq, tp->copied_seq) || | |
550 | !before(tp->urg_seq, tp->rcv_nxt)) { | |
91521944 DM |
551 | struct sk_buff *skb; |
552 | ||
1da177e4 LT |
553 | answ = tp->rcv_nxt - tp->copied_seq; |
554 | ||
555 | /* Subtract 1, if FIN is in queue. */ | |
91521944 DM |
556 | skb = skb_peek_tail(&sk->sk_receive_queue); |
557 | if (answ && skb) | |
558 | answ -= tcp_hdr(skb)->fin; | |
1da177e4 LT |
559 | } else |
560 | answ = tp->urg_seq - tp->copied_seq; | |
561 | release_sock(sk); | |
562 | break; | |
563 | case SIOCATMARK: | |
564 | answ = tp->urg_data && tp->urg_seq == tp->copied_seq; | |
565 | break; | |
566 | case SIOCOUTQ: | |
567 | if (sk->sk_state == TCP_LISTEN) | |
568 | return -EINVAL; | |
569 | ||
570 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) | |
571 | answ = 0; | |
572 | else | |
573 | answ = tp->write_seq - tp->snd_una; | |
574 | break; | |
2f4e1b39 MS |
575 | case SIOCOUTQNSD: |
576 | if (sk->sk_state == TCP_LISTEN) | |
577 | return -EINVAL; | |
578 | ||
579 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) | |
580 | answ = 0; | |
581 | else | |
582 | answ = tp->write_seq - tp->snd_nxt; | |
583 | break; | |
1da177e4 LT |
584 | default: |
585 | return -ENOIOCTLCMD; | |
3ff50b79 | 586 | } |
1da177e4 LT |
587 | |
588 | return put_user(answ, (int __user *)arg); | |
589 | } | |
4bc2f18b | 590 | EXPORT_SYMBOL(tcp_ioctl); |
1da177e4 | 591 | |
1da177e4 LT |
592 | static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb) |
593 | { | |
4de075e0 | 594 | TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH; |
1da177e4 LT |
595 | tp->pushed_seq = tp->write_seq; |
596 | } | |
597 | ||
a2a385d6 | 598 | static inline bool forced_push(const struct tcp_sock *tp) |
1da177e4 LT |
599 | { |
600 | return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1)); | |
601 | } | |
602 | ||
9e412ba7 | 603 | static inline void skb_entail(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 604 | { |
9e412ba7 | 605 | struct tcp_sock *tp = tcp_sk(sk); |
352d4800 ACM |
606 | struct tcp_skb_cb *tcb = TCP_SKB_CB(skb); |
607 | ||
608 | skb->csum = 0; | |
609 | tcb->seq = tcb->end_seq = tp->write_seq; | |
4de075e0 | 610 | tcb->tcp_flags = TCPHDR_ACK; |
352d4800 | 611 | tcb->sacked = 0; |
1da177e4 | 612 | skb_header_release(skb); |
fe067e8a | 613 | tcp_add_write_queue_tail(sk, skb); |
3ab224be HA |
614 | sk->sk_wmem_queued += skb->truesize; |
615 | sk_mem_charge(sk, skb->truesize); | |
89ebd197 | 616 | if (tp->nonagle & TCP_NAGLE_PUSH) |
e905a9ed | 617 | tp->nonagle &= ~TCP_NAGLE_PUSH; |
1da177e4 LT |
618 | } |
619 | ||
afeca340 | 620 | static inline void tcp_mark_urg(struct tcp_sock *tp, int flags) |
1da177e4 | 621 | { |
33f5f57e | 622 | if (flags & MSG_OOB) |
1da177e4 | 623 | tp->snd_up = tp->write_seq; |
1da177e4 LT |
624 | } |
625 | ||
9e412ba7 IJ |
626 | static inline void tcp_push(struct sock *sk, int flags, int mss_now, |
627 | int nonagle) | |
1da177e4 | 628 | { |
fe067e8a | 629 | if (tcp_send_head(sk)) { |
afeca340 KK |
630 | struct tcp_sock *tp = tcp_sk(sk); |
631 | ||
1da177e4 | 632 | if (!(flags & MSG_MORE) || forced_push(tp)) |
afeca340 KK |
633 | tcp_mark_push(tp, tcp_write_queue_tail(sk)); |
634 | ||
635 | tcp_mark_urg(tp, flags); | |
9e412ba7 | 636 | __tcp_push_pending_frames(sk, mss_now, |
1da177e4 LT |
637 | (flags & MSG_MORE) ? TCP_NAGLE_CORK : nonagle); |
638 | } | |
639 | } | |
640 | ||
6ff7751d AB |
641 | static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, |
642 | unsigned int offset, size_t len) | |
9c55e01c JA |
643 | { |
644 | struct tcp_splice_state *tss = rd_desc->arg.data; | |
33966dd0 | 645 | int ret; |
9c55e01c | 646 | |
9fa5fdf2 DM |
647 | ret = skb_splice_bits(skb, offset, tss->pipe, min(rd_desc->count, len), |
648 | tss->flags); | |
33966dd0 WT |
649 | if (ret > 0) |
650 | rd_desc->count -= ret; | |
651 | return ret; | |
9c55e01c JA |
652 | } |
653 | ||
654 | static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss) | |
655 | { | |
656 | /* Store TCP splice context information in read_descriptor_t. */ | |
657 | read_descriptor_t rd_desc = { | |
658 | .arg.data = tss, | |
33966dd0 | 659 | .count = tss->len, |
9c55e01c JA |
660 | }; |
661 | ||
662 | return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv); | |
663 | } | |
664 | ||
665 | /** | |
666 | * tcp_splice_read - splice data from TCP socket to a pipe | |
667 | * @sock: socket to splice from | |
668 | * @ppos: position (not valid) | |
669 | * @pipe: pipe to splice to | |
670 | * @len: number of bytes to splice | |
671 | * @flags: splice modifier flags | |
672 | * | |
673 | * Description: | |
674 | * Will read pages from given socket and fill them into a pipe. | |
675 | * | |
676 | **/ | |
677 | ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos, | |
678 | struct pipe_inode_info *pipe, size_t len, | |
679 | unsigned int flags) | |
680 | { | |
681 | struct sock *sk = sock->sk; | |
682 | struct tcp_splice_state tss = { | |
683 | .pipe = pipe, | |
684 | .len = len, | |
685 | .flags = flags, | |
686 | }; | |
687 | long timeo; | |
688 | ssize_t spliced; | |
689 | int ret; | |
690 | ||
3a047bf8 | 691 | sock_rps_record_flow(sk); |
9c55e01c JA |
692 | /* |
693 | * We can't seek on a socket input | |
694 | */ | |
695 | if (unlikely(*ppos)) | |
696 | return -ESPIPE; | |
697 | ||
698 | ret = spliced = 0; | |
699 | ||
700 | lock_sock(sk); | |
701 | ||
42324c62 | 702 | timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK); |
9c55e01c JA |
703 | while (tss.len) { |
704 | ret = __tcp_splice_read(sk, &tss); | |
705 | if (ret < 0) | |
706 | break; | |
707 | else if (!ret) { | |
708 | if (spliced) | |
709 | break; | |
9c55e01c JA |
710 | if (sock_flag(sk, SOCK_DONE)) |
711 | break; | |
712 | if (sk->sk_err) { | |
713 | ret = sock_error(sk); | |
714 | break; | |
715 | } | |
716 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
717 | break; | |
718 | if (sk->sk_state == TCP_CLOSE) { | |
719 | /* | |
720 | * This occurs when user tries to read | |
721 | * from never connected socket. | |
722 | */ | |
723 | if (!sock_flag(sk, SOCK_DONE)) | |
724 | ret = -ENOTCONN; | |
725 | break; | |
726 | } | |
727 | if (!timeo) { | |
728 | ret = -EAGAIN; | |
729 | break; | |
730 | } | |
731 | sk_wait_data(sk, &timeo); | |
732 | if (signal_pending(current)) { | |
733 | ret = sock_intr_errno(timeo); | |
734 | break; | |
735 | } | |
736 | continue; | |
737 | } | |
738 | tss.len -= ret; | |
739 | spliced += ret; | |
740 | ||
33966dd0 WT |
741 | if (!timeo) |
742 | break; | |
9c55e01c JA |
743 | release_sock(sk); |
744 | lock_sock(sk); | |
745 | ||
746 | if (sk->sk_err || sk->sk_state == TCP_CLOSE || | |
33966dd0 | 747 | (sk->sk_shutdown & RCV_SHUTDOWN) || |
9c55e01c JA |
748 | signal_pending(current)) |
749 | break; | |
750 | } | |
751 | ||
752 | release_sock(sk); | |
753 | ||
754 | if (spliced) | |
755 | return spliced; | |
756 | ||
757 | return ret; | |
758 | } | |
4bc2f18b | 759 | EXPORT_SYMBOL(tcp_splice_read); |
9c55e01c | 760 | |
df97c708 | 761 | struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp) |
f561d0f2 PE |
762 | { |
763 | struct sk_buff *skb; | |
764 | ||
765 | /* The TCP header must be at least 32-bit aligned. */ | |
766 | size = ALIGN(size, 4); | |
767 | ||
768 | skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp); | |
769 | if (skb) { | |
3ab224be | 770 | if (sk_wmem_schedule(sk, skb->truesize)) { |
a21d4572 | 771 | skb_reserve(skb, sk->sk_prot->max_header); |
f561d0f2 PE |
772 | /* |
773 | * Make sure that we have exactly size bytes | |
774 | * available to the caller, no more, no less. | |
775 | */ | |
a21d4572 | 776 | skb->avail_size = size; |
f561d0f2 PE |
777 | return skb; |
778 | } | |
779 | __kfree_skb(skb); | |
780 | } else { | |
5c52ba17 | 781 | sk->sk_prot->enter_memory_pressure(sk); |
f561d0f2 PE |
782 | sk_stream_moderate_sndbuf(sk); |
783 | } | |
784 | return NULL; | |
785 | } | |
786 | ||
0c54b85f IJ |
787 | static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now, |
788 | int large_allowed) | |
789 | { | |
790 | struct tcp_sock *tp = tcp_sk(sk); | |
2a3a041c | 791 | u32 xmit_size_goal, old_size_goal; |
0c54b85f IJ |
792 | |
793 | xmit_size_goal = mss_now; | |
794 | ||
795 | if (large_allowed && sk_can_gso(sk)) { | |
796 | xmit_size_goal = ((sk->sk_gso_max_size - 1) - | |
797 | inet_csk(sk)->icsk_af_ops->net_header_len - | |
798 | inet_csk(sk)->icsk_ext_hdr_len - | |
799 | tp->tcp_header_len); | |
800 | ||
46d3ceab ED |
801 | /* TSQ : try to have two TSO segments in flight */ |
802 | xmit_size_goal = min_t(u32, xmit_size_goal, | |
803 | sysctl_tcp_limit_output_bytes >> 1); | |
804 | ||
0c54b85f | 805 | xmit_size_goal = tcp_bound_to_half_wnd(tp, xmit_size_goal); |
2a3a041c IJ |
806 | |
807 | /* We try hard to avoid divides here */ | |
808 | old_size_goal = tp->xmit_size_goal_segs * mss_now; | |
809 | ||
810 | if (likely(old_size_goal <= xmit_size_goal && | |
811 | old_size_goal + mss_now > xmit_size_goal)) { | |
812 | xmit_size_goal = old_size_goal; | |
813 | } else { | |
1485348d BH |
814 | tp->xmit_size_goal_segs = |
815 | min_t(u16, xmit_size_goal / mss_now, | |
816 | sk->sk_gso_max_segs); | |
2a3a041c IJ |
817 | xmit_size_goal = tp->xmit_size_goal_segs * mss_now; |
818 | } | |
0c54b85f IJ |
819 | } |
820 | ||
afece1c6 | 821 | return max(xmit_size_goal, mss_now); |
0c54b85f IJ |
822 | } |
823 | ||
824 | static int tcp_send_mss(struct sock *sk, int *size_goal, int flags) | |
825 | { | |
826 | int mss_now; | |
827 | ||
828 | mss_now = tcp_current_mss(sk); | |
829 | *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB)); | |
830 | ||
831 | return mss_now; | |
832 | } | |
833 | ||
1da177e4 LT |
834 | static ssize_t do_tcp_sendpages(struct sock *sk, struct page **pages, int poffset, |
835 | size_t psize, int flags) | |
836 | { | |
837 | struct tcp_sock *tp = tcp_sk(sk); | |
c1b4a7e6 | 838 | int mss_now, size_goal; |
1da177e4 LT |
839 | int err; |
840 | ssize_t copied; | |
841 | long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); | |
842 | ||
843 | /* Wait for a connection to finish. */ | |
844 | if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) | |
845 | if ((err = sk_stream_wait_connect(sk, &timeo)) != 0) | |
846 | goto out_err; | |
847 | ||
848 | clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); | |
849 | ||
0c54b85f | 850 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
851 | copied = 0; |
852 | ||
853 | err = -EPIPE; | |
854 | if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) | |
0d6a775e | 855 | goto out_err; |
1da177e4 LT |
856 | |
857 | while (psize > 0) { | |
fe067e8a | 858 | struct sk_buff *skb = tcp_write_queue_tail(sk); |
1da177e4 | 859 | struct page *page = pages[poffset / PAGE_SIZE]; |
38ba0a65 | 860 | int copy, i; |
1da177e4 LT |
861 | int offset = poffset % PAGE_SIZE; |
862 | int size = min_t(size_t, psize, PAGE_SIZE - offset); | |
38ba0a65 | 863 | bool can_coalesce; |
1da177e4 | 864 | |
fe067e8a | 865 | if (!tcp_send_head(sk) || (copy = size_goal - skb->len) <= 0) { |
1da177e4 LT |
866 | new_segment: |
867 | if (!sk_stream_memory_free(sk)) | |
868 | goto wait_for_sndbuf; | |
869 | ||
df97c708 | 870 | skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation); |
1da177e4 LT |
871 | if (!skb) |
872 | goto wait_for_memory; | |
873 | ||
9e412ba7 | 874 | skb_entail(sk, skb); |
c1b4a7e6 | 875 | copy = size_goal; |
1da177e4 LT |
876 | } |
877 | ||
878 | if (copy > size) | |
879 | copy = size; | |
880 | ||
881 | i = skb_shinfo(skb)->nr_frags; | |
882 | can_coalesce = skb_can_coalesce(skb, i, page, offset); | |
883 | if (!can_coalesce && i >= MAX_SKB_FRAGS) { | |
884 | tcp_mark_push(tp, skb); | |
885 | goto new_segment; | |
886 | } | |
3ab224be | 887 | if (!sk_wmem_schedule(sk, copy)) |
1da177e4 | 888 | goto wait_for_memory; |
e905a9ed | 889 | |
1da177e4 | 890 | if (can_coalesce) { |
9e903e08 | 891 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); |
1da177e4 LT |
892 | } else { |
893 | get_page(page); | |
894 | skb_fill_page_desc(skb, i, page, offset, copy); | |
895 | } | |
896 | ||
897 | skb->len += copy; | |
898 | skb->data_len += copy; | |
899 | skb->truesize += copy; | |
900 | sk->sk_wmem_queued += copy; | |
3ab224be | 901 | sk_mem_charge(sk, copy); |
84fa7933 | 902 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 LT |
903 | tp->write_seq += copy; |
904 | TCP_SKB_CB(skb)->end_seq += copy; | |
7967168c | 905 | skb_shinfo(skb)->gso_segs = 0; |
1da177e4 LT |
906 | |
907 | if (!copied) | |
4de075e0 | 908 | TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH; |
1da177e4 LT |
909 | |
910 | copied += copy; | |
911 | poffset += copy; | |
912 | if (!(psize -= copy)) | |
913 | goto out; | |
914 | ||
69d15067 | 915 | if (skb->len < size_goal || (flags & MSG_OOB)) |
1da177e4 LT |
916 | continue; |
917 | ||
918 | if (forced_push(tp)) { | |
919 | tcp_mark_push(tp, skb); | |
9e412ba7 | 920 | __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH); |
fe067e8a | 921 | } else if (skb == tcp_send_head(sk)) |
1da177e4 LT |
922 | tcp_push_one(sk, mss_now); |
923 | continue; | |
924 | ||
925 | wait_for_sndbuf: | |
926 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
927 | wait_for_memory: | |
bad115cf | 928 | tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH); |
1da177e4 LT |
929 | |
930 | if ((err = sk_stream_wait_memory(sk, &timeo)) != 0) | |
931 | goto do_error; | |
932 | ||
0c54b85f | 933 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
934 | } |
935 | ||
936 | out: | |
35f9c09f | 937 | if (copied && !(flags & MSG_SENDPAGE_NOTLAST)) |
9e412ba7 | 938 | tcp_push(sk, flags, mss_now, tp->nonagle); |
1da177e4 LT |
939 | return copied; |
940 | ||
941 | do_error: | |
942 | if (copied) | |
943 | goto out; | |
944 | out_err: | |
945 | return sk_stream_error(sk, flags, err); | |
946 | } | |
947 | ||
7ba42910 CG |
948 | int tcp_sendpage(struct sock *sk, struct page *page, int offset, |
949 | size_t size, int flags) | |
1da177e4 LT |
950 | { |
951 | ssize_t res; | |
1da177e4 | 952 | |
1da177e4 | 953 | if (!(sk->sk_route_caps & NETIF_F_SG) || |
8648b305 | 954 | !(sk->sk_route_caps & NETIF_F_ALL_CSUM)) |
7ba42910 CG |
955 | return sock_no_sendpage(sk->sk_socket, page, offset, size, |
956 | flags); | |
1da177e4 | 957 | |
1da177e4 | 958 | lock_sock(sk); |
1da177e4 | 959 | res = do_tcp_sendpages(sk, &page, offset, size, flags); |
1da177e4 LT |
960 | release_sock(sk); |
961 | return res; | |
962 | } | |
4bc2f18b | 963 | EXPORT_SYMBOL(tcp_sendpage); |
1da177e4 | 964 | |
690e99c4 | 965 | static inline int select_size(const struct sock *sk, bool sg) |
1da177e4 | 966 | { |
cf533ea5 | 967 | const struct tcp_sock *tp = tcp_sk(sk); |
c1b4a7e6 | 968 | int tmp = tp->mss_cache; |
1da177e4 | 969 | |
def87cf4 | 970 | if (sg) { |
f07d960d ED |
971 | if (sk_can_gso(sk)) { |
972 | /* Small frames wont use a full page: | |
973 | * Payload will immediately follow tcp header. | |
974 | */ | |
975 | tmp = SKB_WITH_OVERHEAD(2048 - MAX_TCP_HEADER); | |
976 | } else { | |
b4e26f5e DM |
977 | int pgbreak = SKB_MAX_HEAD(MAX_TCP_HEADER); |
978 | ||
979 | if (tmp >= pgbreak && | |
980 | tmp <= pgbreak + (MAX_SKB_FRAGS - 1) * PAGE_SIZE) | |
981 | tmp = pgbreak; | |
982 | } | |
983 | } | |
1da177e4 | 984 | |
1da177e4 LT |
985 | return tmp; |
986 | } | |
987 | ||
cf60af03 YC |
988 | void tcp_free_fastopen_req(struct tcp_sock *tp) |
989 | { | |
990 | if (tp->fastopen_req != NULL) { | |
991 | kfree(tp->fastopen_req); | |
992 | tp->fastopen_req = NULL; | |
993 | } | |
994 | } | |
995 | ||
996 | static int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *size) | |
997 | { | |
998 | struct tcp_sock *tp = tcp_sk(sk); | |
999 | int err, flags; | |
1000 | ||
1001 | if (!(sysctl_tcp_fastopen & TFO_CLIENT_ENABLE)) | |
1002 | return -EOPNOTSUPP; | |
1003 | if (tp->fastopen_req != NULL) | |
1004 | return -EALREADY; /* Another Fast Open is in progress */ | |
1005 | ||
1006 | tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request), | |
1007 | sk->sk_allocation); | |
1008 | if (unlikely(tp->fastopen_req == NULL)) | |
1009 | return -ENOBUFS; | |
1010 | tp->fastopen_req->data = msg; | |
1011 | ||
1012 | flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0; | |
1013 | err = __inet_stream_connect(sk->sk_socket, msg->msg_name, | |
1014 | msg->msg_namelen, flags); | |
1015 | *size = tp->fastopen_req->copied; | |
1016 | tcp_free_fastopen_req(tp); | |
1017 | return err; | |
1018 | } | |
1019 | ||
7ba42910 | 1020 | int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, |
1da177e4 LT |
1021 | size_t size) |
1022 | { | |
1023 | struct iovec *iov; | |
1024 | struct tcp_sock *tp = tcp_sk(sk); | |
1025 | struct sk_buff *skb; | |
cf60af03 YC |
1026 | int iovlen, flags, err, copied = 0; |
1027 | int mss_now = 0, size_goal, copied_syn = 0, offset = 0; | |
690e99c4 | 1028 | bool sg; |
1da177e4 LT |
1029 | long timeo; |
1030 | ||
1031 | lock_sock(sk); | |
1da177e4 LT |
1032 | |
1033 | flags = msg->msg_flags; | |
cf60af03 YC |
1034 | if (flags & MSG_FASTOPEN) { |
1035 | err = tcp_sendmsg_fastopen(sk, msg, &copied_syn); | |
1036 | if (err == -EINPROGRESS && copied_syn > 0) | |
1037 | goto out; | |
1038 | else if (err) | |
1039 | goto out_err; | |
1040 | offset = copied_syn; | |
1041 | } | |
1042 | ||
1da177e4 LT |
1043 | timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); |
1044 | ||
1045 | /* Wait for a connection to finish. */ | |
1046 | if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) | |
1047 | if ((err = sk_stream_wait_connect(sk, &timeo)) != 0) | |
cf60af03 | 1048 | goto do_error; |
1da177e4 | 1049 | |
c0e88ff0 PE |
1050 | if (unlikely(tp->repair)) { |
1051 | if (tp->repair_queue == TCP_RECV_QUEUE) { | |
1052 | copied = tcp_send_rcvq(sk, msg, size); | |
1053 | goto out; | |
1054 | } | |
1055 | ||
1056 | err = -EINVAL; | |
1057 | if (tp->repair_queue == TCP_NO_QUEUE) | |
1058 | goto out_err; | |
1059 | ||
1060 | /* 'common' sending to sendq */ | |
1061 | } | |
1062 | ||
1da177e4 LT |
1063 | /* This should be in poll */ |
1064 | clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); | |
1065 | ||
0c54b85f | 1066 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
1067 | |
1068 | /* Ok commence sending. */ | |
1069 | iovlen = msg->msg_iovlen; | |
1070 | iov = msg->msg_iov; | |
1071 | copied = 0; | |
1072 | ||
1073 | err = -EPIPE; | |
1074 | if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) | |
0d6a775e | 1075 | goto out_err; |
1da177e4 | 1076 | |
690e99c4 | 1077 | sg = !!(sk->sk_route_caps & NETIF_F_SG); |
def87cf4 | 1078 | |
1da177e4 | 1079 | while (--iovlen >= 0) { |
01db403c | 1080 | size_t seglen = iov->iov_len; |
1da177e4 LT |
1081 | unsigned char __user *from = iov->iov_base; |
1082 | ||
1083 | iov++; | |
cf60af03 YC |
1084 | if (unlikely(offset > 0)) { /* Skip bytes copied in SYN */ |
1085 | if (offset >= seglen) { | |
1086 | offset -= seglen; | |
1087 | continue; | |
1088 | } | |
1089 | seglen -= offset; | |
1090 | from += offset; | |
1091 | offset = 0; | |
1092 | } | |
1da177e4 LT |
1093 | |
1094 | while (seglen > 0) { | |
6828b92b HX |
1095 | int copy = 0; |
1096 | int max = size_goal; | |
1da177e4 | 1097 | |
fe067e8a | 1098 | skb = tcp_write_queue_tail(sk); |
6828b92b HX |
1099 | if (tcp_send_head(sk)) { |
1100 | if (skb->ip_summed == CHECKSUM_NONE) | |
1101 | max = mss_now; | |
1102 | copy = max - skb->len; | |
1103 | } | |
1da177e4 | 1104 | |
6828b92b | 1105 | if (copy <= 0) { |
1da177e4 LT |
1106 | new_segment: |
1107 | /* Allocate new segment. If the interface is SG, | |
1108 | * allocate skb fitting to single page. | |
1109 | */ | |
1110 | if (!sk_stream_memory_free(sk)) | |
1111 | goto wait_for_sndbuf; | |
1112 | ||
def87cf4 KK |
1113 | skb = sk_stream_alloc_skb(sk, |
1114 | select_size(sk, sg), | |
1115 | sk->sk_allocation); | |
1da177e4 LT |
1116 | if (!skb) |
1117 | goto wait_for_memory; | |
1118 | ||
1119 | /* | |
1120 | * Check whether we can use HW checksum. | |
1121 | */ | |
8648b305 | 1122 | if (sk->sk_route_caps & NETIF_F_ALL_CSUM) |
84fa7933 | 1123 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 | 1124 | |
9e412ba7 | 1125 | skb_entail(sk, skb); |
c1b4a7e6 | 1126 | copy = size_goal; |
6828b92b | 1127 | max = size_goal; |
1da177e4 LT |
1128 | } |
1129 | ||
1130 | /* Try to append data to the end of skb. */ | |
1131 | if (copy > seglen) | |
1132 | copy = seglen; | |
1133 | ||
1134 | /* Where to copy to? */ | |
a21d4572 | 1135 | if (skb_availroom(skb) > 0) { |
1da177e4 | 1136 | /* We have some space in skb head. Superb! */ |
a21d4572 | 1137 | copy = min_t(int, copy, skb_availroom(skb)); |
c6e1a0d1 TH |
1138 | err = skb_add_data_nocache(sk, skb, from, copy); |
1139 | if (err) | |
1da177e4 LT |
1140 | goto do_fault; |
1141 | } else { | |
a2a385d6 | 1142 | bool merge = false; |
1da177e4 | 1143 | int i = skb_shinfo(skb)->nr_frags; |
0a5912db | 1144 | struct page *page = sk->sk_sndmsg_page; |
761965ea ED |
1145 | int off; |
1146 | ||
1147 | if (page && page_count(page) == 1) | |
0a5912db | 1148 | sk->sk_sndmsg_off = 0; |
761965ea | 1149 | |
0a5912db | 1150 | off = sk->sk_sndmsg_off; |
1da177e4 LT |
1151 | |
1152 | if (skb_can_coalesce(skb, i, page, off) && | |
1153 | off != PAGE_SIZE) { | |
1154 | /* We can extend the last page | |
1155 | * fragment. */ | |
a2a385d6 | 1156 | merge = true; |
def87cf4 | 1157 | } else if (i == MAX_SKB_FRAGS || !sg) { |
1da177e4 LT |
1158 | /* Need to add new fragment and cannot |
1159 | * do this because interface is non-SG, | |
1160 | * or because all the page slots are | |
1161 | * busy. */ | |
1162 | tcp_mark_push(tp, skb); | |
1163 | goto new_segment; | |
1164 | } else if (page) { | |
1da177e4 LT |
1165 | if (off == PAGE_SIZE) { |
1166 | put_page(page); | |
0a5912db | 1167 | sk->sk_sndmsg_page = page = NULL; |
fb5f5e6e | 1168 | off = 0; |
1da177e4 | 1169 | } |
ef015786 | 1170 | } else |
fb5f5e6e | 1171 | off = 0; |
ef015786 HX |
1172 | |
1173 | if (copy > PAGE_SIZE - off) | |
1174 | copy = PAGE_SIZE - off; | |
1175 | ||
3ab224be | 1176 | if (!sk_wmem_schedule(sk, copy)) |
ef015786 | 1177 | goto wait_for_memory; |
1da177e4 LT |
1178 | |
1179 | if (!page) { | |
1180 | /* Allocate new cache page. */ | |
1181 | if (!(page = sk_stream_alloc_page(sk))) | |
1182 | goto wait_for_memory; | |
1da177e4 LT |
1183 | } |
1184 | ||
1da177e4 LT |
1185 | /* Time to copy data. We are close to |
1186 | * the end! */ | |
c6e1a0d1 TH |
1187 | err = skb_copy_to_page_nocache(sk, from, skb, |
1188 | page, off, copy); | |
1da177e4 LT |
1189 | if (err) { |
1190 | /* If this page was new, give it to the | |
1191 | * socket so it does not get leaked. | |
1192 | */ | |
0a5912db ED |
1193 | if (!sk->sk_sndmsg_page) { |
1194 | sk->sk_sndmsg_page = page; | |
1195 | sk->sk_sndmsg_off = 0; | |
1da177e4 LT |
1196 | } |
1197 | goto do_error; | |
1198 | } | |
1199 | ||
1200 | /* Update the skb. */ | |
1201 | if (merge) { | |
9e903e08 | 1202 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); |
1da177e4 LT |
1203 | } else { |
1204 | skb_fill_page_desc(skb, i, page, off, copy); | |
0a5912db | 1205 | if (sk->sk_sndmsg_page) { |
1da177e4 LT |
1206 | get_page(page); |
1207 | } else if (off + copy < PAGE_SIZE) { | |
1208 | get_page(page); | |
0a5912db | 1209 | sk->sk_sndmsg_page = page; |
1da177e4 LT |
1210 | } |
1211 | } | |
1212 | ||
0a5912db | 1213 | sk->sk_sndmsg_off = off + copy; |
1da177e4 LT |
1214 | } |
1215 | ||
1216 | if (!copied) | |
4de075e0 | 1217 | TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH; |
1da177e4 LT |
1218 | |
1219 | tp->write_seq += copy; | |
1220 | TCP_SKB_CB(skb)->end_seq += copy; | |
7967168c | 1221 | skb_shinfo(skb)->gso_segs = 0; |
1da177e4 LT |
1222 | |
1223 | from += copy; | |
1224 | copied += copy; | |
1225 | if ((seglen -= copy) == 0 && iovlen == 0) | |
1226 | goto out; | |
1227 | ||
c0e88ff0 | 1228 | if (skb->len < max || (flags & MSG_OOB) || unlikely(tp->repair)) |
1da177e4 LT |
1229 | continue; |
1230 | ||
1231 | if (forced_push(tp)) { | |
1232 | tcp_mark_push(tp, skb); | |
9e412ba7 | 1233 | __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH); |
fe067e8a | 1234 | } else if (skb == tcp_send_head(sk)) |
1da177e4 LT |
1235 | tcp_push_one(sk, mss_now); |
1236 | continue; | |
1237 | ||
1238 | wait_for_sndbuf: | |
1239 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
1240 | wait_for_memory: | |
c0e88ff0 | 1241 | if (copied && likely(!tp->repair)) |
9e412ba7 | 1242 | tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH); |
1da177e4 LT |
1243 | |
1244 | if ((err = sk_stream_wait_memory(sk, &timeo)) != 0) | |
1245 | goto do_error; | |
1246 | ||
0c54b85f | 1247 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
1248 | } |
1249 | } | |
1250 | ||
1251 | out: | |
c0e88ff0 | 1252 | if (copied && likely(!tp->repair)) |
9e412ba7 | 1253 | tcp_push(sk, flags, mss_now, tp->nonagle); |
1da177e4 | 1254 | release_sock(sk); |
cf60af03 | 1255 | return copied + copied_syn; |
1da177e4 LT |
1256 | |
1257 | do_fault: | |
1258 | if (!skb->len) { | |
fe067e8a DM |
1259 | tcp_unlink_write_queue(skb, sk); |
1260 | /* It is the one place in all of TCP, except connection | |
1261 | * reset, where we can be unlinking the send_head. | |
1262 | */ | |
1263 | tcp_check_send_head(sk, skb); | |
3ab224be | 1264 | sk_wmem_free_skb(sk, skb); |
1da177e4 LT |
1265 | } |
1266 | ||
1267 | do_error: | |
cf60af03 | 1268 | if (copied + copied_syn) |
1da177e4 LT |
1269 | goto out; |
1270 | out_err: | |
1271 | err = sk_stream_error(sk, flags, err); | |
1da177e4 LT |
1272 | release_sock(sk); |
1273 | return err; | |
1274 | } | |
4bc2f18b | 1275 | EXPORT_SYMBOL(tcp_sendmsg); |
1da177e4 LT |
1276 | |
1277 | /* | |
1278 | * Handle reading urgent data. BSD has very simple semantics for | |
1279 | * this, no blocking and very strange errors 8) | |
1280 | */ | |
1281 | ||
377f0a08 | 1282 | static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags) |
1da177e4 LT |
1283 | { |
1284 | struct tcp_sock *tp = tcp_sk(sk); | |
1285 | ||
1286 | /* No URG data to read. */ | |
1287 | if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data || | |
1288 | tp->urg_data == TCP_URG_READ) | |
1289 | return -EINVAL; /* Yes this is right ! */ | |
1290 | ||
1291 | if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE)) | |
1292 | return -ENOTCONN; | |
1293 | ||
1294 | if (tp->urg_data & TCP_URG_VALID) { | |
1295 | int err = 0; | |
1296 | char c = tp->urg_data; | |
1297 | ||
1298 | if (!(flags & MSG_PEEK)) | |
1299 | tp->urg_data = TCP_URG_READ; | |
1300 | ||
1301 | /* Read urgent data. */ | |
1302 | msg->msg_flags |= MSG_OOB; | |
1303 | ||
1304 | if (len > 0) { | |
1305 | if (!(flags & MSG_TRUNC)) | |
1306 | err = memcpy_toiovec(msg->msg_iov, &c, 1); | |
1307 | len = 1; | |
1308 | } else | |
1309 | msg->msg_flags |= MSG_TRUNC; | |
1310 | ||
1311 | return err ? -EFAULT : len; | |
1312 | } | |
1313 | ||
1314 | if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN)) | |
1315 | return 0; | |
1316 | ||
1317 | /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and | |
1318 | * the available implementations agree in this case: | |
1319 | * this call should never block, independent of the | |
1320 | * blocking state of the socket. | |
1321 | * Mike <pall@rz.uni-karlsruhe.de> | |
1322 | */ | |
1323 | return -EAGAIN; | |
1324 | } | |
1325 | ||
c0e88ff0 PE |
1326 | static int tcp_peek_sndq(struct sock *sk, struct msghdr *msg, int len) |
1327 | { | |
1328 | struct sk_buff *skb; | |
1329 | int copied = 0, err = 0; | |
1330 | ||
1331 | /* XXX -- need to support SO_PEEK_OFF */ | |
1332 | ||
1333 | skb_queue_walk(&sk->sk_write_queue, skb) { | |
1334 | err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, skb->len); | |
1335 | if (err) | |
1336 | break; | |
1337 | ||
1338 | copied += skb->len; | |
1339 | } | |
1340 | ||
1341 | return err ?: copied; | |
1342 | } | |
1343 | ||
1da177e4 LT |
1344 | /* Clean up the receive buffer for full frames taken by the user, |
1345 | * then send an ACK if necessary. COPIED is the number of bytes | |
1346 | * tcp_recvmsg has given to the user so far, it speeds up the | |
1347 | * calculation of whether or not we must ACK for the sake of | |
1348 | * a window update. | |
1349 | */ | |
0e4b4992 | 1350 | void tcp_cleanup_rbuf(struct sock *sk, int copied) |
1da177e4 LT |
1351 | { |
1352 | struct tcp_sock *tp = tcp_sk(sk); | |
a2a385d6 | 1353 | bool time_to_ack = false; |
1da177e4 | 1354 | |
1da177e4 LT |
1355 | struct sk_buff *skb = skb_peek(&sk->sk_receive_queue); |
1356 | ||
d792c100 | 1357 | WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq), |
2af6fd8b | 1358 | "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n", |
d792c100 | 1359 | tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt); |
1da177e4 | 1360 | |
463c84b9 ACM |
1361 | if (inet_csk_ack_scheduled(sk)) { |
1362 | const struct inet_connection_sock *icsk = inet_csk(sk); | |
1da177e4 LT |
1363 | /* Delayed ACKs frequently hit locked sockets during bulk |
1364 | * receive. */ | |
463c84b9 | 1365 | if (icsk->icsk_ack.blocked || |
1da177e4 | 1366 | /* Once-per-two-segments ACK was not sent by tcp_input.c */ |
463c84b9 | 1367 | tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss || |
1da177e4 LT |
1368 | /* |
1369 | * If this read emptied read buffer, we send ACK, if | |
1370 | * connection is not bidirectional, user drained | |
1371 | * receive buffer and there was a small segment | |
1372 | * in queue. | |
1373 | */ | |
1ef9696c AK |
1374 | (copied > 0 && |
1375 | ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) || | |
1376 | ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) && | |
1377 | !icsk->icsk_ack.pingpong)) && | |
1378 | !atomic_read(&sk->sk_rmem_alloc))) | |
a2a385d6 | 1379 | time_to_ack = true; |
1da177e4 LT |
1380 | } |
1381 | ||
1382 | /* We send an ACK if we can now advertise a non-zero window | |
1383 | * which has been raised "significantly". | |
1384 | * | |
1385 | * Even if window raised up to infinity, do not send window open ACK | |
1386 | * in states, where we will not receive more. It is useless. | |
1387 | */ | |
1388 | if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) { | |
1389 | __u32 rcv_window_now = tcp_receive_window(tp); | |
1390 | ||
1391 | /* Optimize, __tcp_select_window() is not cheap. */ | |
1392 | if (2*rcv_window_now <= tp->window_clamp) { | |
1393 | __u32 new_window = __tcp_select_window(sk); | |
1394 | ||
1395 | /* Send ACK now, if this read freed lots of space | |
1396 | * in our buffer. Certainly, new_window is new window. | |
1397 | * We can advertise it now, if it is not less than current one. | |
1398 | * "Lots" means "at least twice" here. | |
1399 | */ | |
1400 | if (new_window && new_window >= 2 * rcv_window_now) | |
a2a385d6 | 1401 | time_to_ack = true; |
1da177e4 LT |
1402 | } |
1403 | } | |
1404 | if (time_to_ack) | |
1405 | tcp_send_ack(sk); | |
1406 | } | |
1407 | ||
1408 | static void tcp_prequeue_process(struct sock *sk) | |
1409 | { | |
1410 | struct sk_buff *skb; | |
1411 | struct tcp_sock *tp = tcp_sk(sk); | |
1412 | ||
6f67c817 | 1413 | NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPPREQUEUED); |
1da177e4 LT |
1414 | |
1415 | /* RX process wants to run with disabled BHs, though it is not | |
1416 | * necessary */ | |
1417 | local_bh_disable(); | |
1418 | while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) | |
c57943a1 | 1419 | sk_backlog_rcv(sk, skb); |
1da177e4 LT |
1420 | local_bh_enable(); |
1421 | ||
1422 | /* Clear memory counter. */ | |
1423 | tp->ucopy.memory = 0; | |
1424 | } | |
1425 | ||
73852e81 SM |
1426 | #ifdef CONFIG_NET_DMA |
1427 | static void tcp_service_net_dma(struct sock *sk, bool wait) | |
1428 | { | |
1429 | dma_cookie_t done, used; | |
1430 | dma_cookie_t last_issued; | |
1431 | struct tcp_sock *tp = tcp_sk(sk); | |
1432 | ||
1433 | if (!tp->ucopy.dma_chan) | |
1434 | return; | |
1435 | ||
1436 | last_issued = tp->ucopy.dma_cookie; | |
1437 | dma_async_memcpy_issue_pending(tp->ucopy.dma_chan); | |
1438 | ||
1439 | do { | |
1440 | if (dma_async_memcpy_complete(tp->ucopy.dma_chan, | |
1441 | last_issued, &done, | |
1442 | &used) == DMA_SUCCESS) { | |
1443 | /* Safe to free early-copied skbs now */ | |
1444 | __skb_queue_purge(&sk->sk_async_wait_queue); | |
1445 | break; | |
1446 | } else { | |
1447 | struct sk_buff *skb; | |
1448 | while ((skb = skb_peek(&sk->sk_async_wait_queue)) && | |
1449 | (dma_async_is_complete(skb->dma_cookie, done, | |
1450 | used) == DMA_SUCCESS)) { | |
1451 | __skb_dequeue(&sk->sk_async_wait_queue); | |
1452 | kfree_skb(skb); | |
1453 | } | |
1454 | } | |
1455 | } while (wait); | |
1456 | } | |
1457 | #endif | |
1458 | ||
1da177e4 LT |
1459 | static inline struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off) |
1460 | { | |
1461 | struct sk_buff *skb; | |
1462 | u32 offset; | |
1463 | ||
1464 | skb_queue_walk(&sk->sk_receive_queue, skb) { | |
1465 | offset = seq - TCP_SKB_CB(skb)->seq; | |
aa8223c7 | 1466 | if (tcp_hdr(skb)->syn) |
1da177e4 | 1467 | offset--; |
aa8223c7 | 1468 | if (offset < skb->len || tcp_hdr(skb)->fin) { |
1da177e4 LT |
1469 | *off = offset; |
1470 | return skb; | |
1471 | } | |
1472 | } | |
1473 | return NULL; | |
1474 | } | |
1475 | ||
1476 | /* | |
1477 | * This routine provides an alternative to tcp_recvmsg() for routines | |
1478 | * that would like to handle copying from skbuffs directly in 'sendfile' | |
1479 | * fashion. | |
1480 | * Note: | |
1481 | * - It is assumed that the socket was locked by the caller. | |
1482 | * - The routine does not block. | |
1483 | * - At present, there is no support for reading OOB data | |
1484 | * or for 'peeking' the socket using this routine | |
1485 | * (although both would be easy to implement). | |
1486 | */ | |
1487 | int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, | |
1488 | sk_read_actor_t recv_actor) | |
1489 | { | |
1490 | struct sk_buff *skb; | |
1491 | struct tcp_sock *tp = tcp_sk(sk); | |
1492 | u32 seq = tp->copied_seq; | |
1493 | u32 offset; | |
1494 | int copied = 0; | |
1495 | ||
1496 | if (sk->sk_state == TCP_LISTEN) | |
1497 | return -ENOTCONN; | |
1498 | while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) { | |
1499 | if (offset < skb->len) { | |
374e7b59 OP |
1500 | int used; |
1501 | size_t len; | |
1da177e4 LT |
1502 | |
1503 | len = skb->len - offset; | |
1504 | /* Stop reading if we hit a patch of urgent data */ | |
1505 | if (tp->urg_data) { | |
1506 | u32 urg_offset = tp->urg_seq - seq; | |
1507 | if (urg_offset < len) | |
1508 | len = urg_offset; | |
1509 | if (!len) | |
1510 | break; | |
1511 | } | |
1512 | used = recv_actor(desc, skb, offset, len); | |
ddb61a57 JA |
1513 | if (used < 0) { |
1514 | if (!copied) | |
1515 | copied = used; | |
1516 | break; | |
1517 | } else if (used <= len) { | |
1da177e4 LT |
1518 | seq += used; |
1519 | copied += used; | |
1520 | offset += used; | |
1521 | } | |
293ad604 OP |
1522 | /* |
1523 | * If recv_actor drops the lock (e.g. TCP splice | |
1524 | * receive) the skb pointer might be invalid when | |
1525 | * getting here: tcp_collapse might have deleted it | |
1526 | * while aggregating skbs from the socket queue. | |
1527 | */ | |
1528 | skb = tcp_recv_skb(sk, seq-1, &offset); | |
1529 | if (!skb || (offset+1 != skb->len)) | |
1da177e4 LT |
1530 | break; |
1531 | } | |
aa8223c7 | 1532 | if (tcp_hdr(skb)->fin) { |
dc6b9b78 | 1533 | sk_eat_skb(sk, skb, false); |
1da177e4 LT |
1534 | ++seq; |
1535 | break; | |
1536 | } | |
dc6b9b78 | 1537 | sk_eat_skb(sk, skb, false); |
1da177e4 LT |
1538 | if (!desc->count) |
1539 | break; | |
baff42ab | 1540 | tp->copied_seq = seq; |
1da177e4 LT |
1541 | } |
1542 | tp->copied_seq = seq; | |
1543 | ||
1544 | tcp_rcv_space_adjust(sk); | |
1545 | ||
1546 | /* Clean up data we have read: This will do ACK frames. */ | |
ddb61a57 | 1547 | if (copied > 0) |
0e4b4992 | 1548 | tcp_cleanup_rbuf(sk, copied); |
1da177e4 LT |
1549 | return copied; |
1550 | } | |
4bc2f18b | 1551 | EXPORT_SYMBOL(tcp_read_sock); |
1da177e4 LT |
1552 | |
1553 | /* | |
1554 | * This routine copies from a sock struct into the user buffer. | |
1555 | * | |
1556 | * Technical note: in 2.3 we work on _locked_ socket, so that | |
1557 | * tricks with *seq access order and skb->users are not required. | |
1558 | * Probably, code can be easily improved even more. | |
1559 | */ | |
1560 | ||
1561 | int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, | |
1562 | size_t len, int nonblock, int flags, int *addr_len) | |
1563 | { | |
1564 | struct tcp_sock *tp = tcp_sk(sk); | |
1565 | int copied = 0; | |
1566 | u32 peek_seq; | |
1567 | u32 *seq; | |
1568 | unsigned long used; | |
1569 | int err; | |
1570 | int target; /* Read at least this many bytes */ | |
1571 | long timeo; | |
1572 | struct task_struct *user_recv = NULL; | |
dc6b9b78 | 1573 | bool copied_early = false; |
2b1244a4 | 1574 | struct sk_buff *skb; |
77527313 | 1575 | u32 urg_hole = 0; |
1da177e4 LT |
1576 | |
1577 | lock_sock(sk); | |
1578 | ||
1da177e4 LT |
1579 | err = -ENOTCONN; |
1580 | if (sk->sk_state == TCP_LISTEN) | |
1581 | goto out; | |
1582 | ||
1583 | timeo = sock_rcvtimeo(sk, nonblock); | |
1584 | ||
1585 | /* Urgent data needs to be handled specially. */ | |
1586 | if (flags & MSG_OOB) | |
1587 | goto recv_urg; | |
1588 | ||
c0e88ff0 PE |
1589 | if (unlikely(tp->repair)) { |
1590 | err = -EPERM; | |
1591 | if (!(flags & MSG_PEEK)) | |
1592 | goto out; | |
1593 | ||
1594 | if (tp->repair_queue == TCP_SEND_QUEUE) | |
1595 | goto recv_sndq; | |
1596 | ||
1597 | err = -EINVAL; | |
1598 | if (tp->repair_queue == TCP_NO_QUEUE) | |
1599 | goto out; | |
1600 | ||
1601 | /* 'common' recv queue MSG_PEEK-ing */ | |
1602 | } | |
1603 | ||
1da177e4 LT |
1604 | seq = &tp->copied_seq; |
1605 | if (flags & MSG_PEEK) { | |
1606 | peek_seq = tp->copied_seq; | |
1607 | seq = &peek_seq; | |
1608 | } | |
1609 | ||
1610 | target = sock_rcvlowat(sk, flags & MSG_WAITALL, len); | |
1611 | ||
1a2449a8 CL |
1612 | #ifdef CONFIG_NET_DMA |
1613 | tp->ucopy.dma_chan = NULL; | |
1614 | preempt_disable(); | |
2b1244a4 | 1615 | skb = skb_peek_tail(&sk->sk_receive_queue); |
e00c5d8b AM |
1616 | { |
1617 | int available = 0; | |
1618 | ||
1619 | if (skb) | |
1620 | available = TCP_SKB_CB(skb)->seq + skb->len - (*seq); | |
1621 | if ((available < target) && | |
1622 | (len > sysctl_tcp_dma_copybreak) && !(flags & MSG_PEEK) && | |
1623 | !sysctl_tcp_low_latency && | |
a2bd1140 | 1624 | net_dma_find_channel()) { |
e00c5d8b AM |
1625 | preempt_enable_no_resched(); |
1626 | tp->ucopy.pinned_list = | |
1627 | dma_pin_iovec_pages(msg->msg_iov, len); | |
1628 | } else { | |
1629 | preempt_enable_no_resched(); | |
1630 | } | |
1631 | } | |
1a2449a8 CL |
1632 | #endif |
1633 | ||
1da177e4 | 1634 | do { |
1da177e4 LT |
1635 | u32 offset; |
1636 | ||
1637 | /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */ | |
1638 | if (tp->urg_data && tp->urg_seq == *seq) { | |
1639 | if (copied) | |
1640 | break; | |
1641 | if (signal_pending(current)) { | |
1642 | copied = timeo ? sock_intr_errno(timeo) : -EAGAIN; | |
1643 | break; | |
1644 | } | |
1645 | } | |
1646 | ||
1647 | /* Next get a buffer. */ | |
1648 | ||
91521944 | 1649 | skb_queue_walk(&sk->sk_receive_queue, skb) { |
1da177e4 LT |
1650 | /* Now that we have two receive queues this |
1651 | * shouldn't happen. | |
1652 | */ | |
d792c100 | 1653 | if (WARN(before(*seq, TCP_SKB_CB(skb)->seq), |
2af6fd8b JP |
1654 | "recvmsg bug: copied %X seq %X rcvnxt %X fl %X\n", |
1655 | *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, | |
1656 | flags)) | |
1da177e4 | 1657 | break; |
d792c100 | 1658 | |
1da177e4 | 1659 | offset = *seq - TCP_SKB_CB(skb)->seq; |
aa8223c7 | 1660 | if (tcp_hdr(skb)->syn) |
1da177e4 LT |
1661 | offset--; |
1662 | if (offset < skb->len) | |
1663 | goto found_ok_skb; | |
aa8223c7 | 1664 | if (tcp_hdr(skb)->fin) |
1da177e4 | 1665 | goto found_fin_ok; |
2af6fd8b JP |
1666 | WARN(!(flags & MSG_PEEK), |
1667 | "recvmsg bug 2: copied %X seq %X rcvnxt %X fl %X\n", | |
1668 | *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags); | |
91521944 | 1669 | } |
1da177e4 LT |
1670 | |
1671 | /* Well, if we have backlog, try to process it now yet. */ | |
1672 | ||
1673 | if (copied >= target && !sk->sk_backlog.tail) | |
1674 | break; | |
1675 | ||
1676 | if (copied) { | |
1677 | if (sk->sk_err || | |
1678 | sk->sk_state == TCP_CLOSE || | |
1679 | (sk->sk_shutdown & RCV_SHUTDOWN) || | |
1680 | !timeo || | |
518a09ef | 1681 | signal_pending(current)) |
1da177e4 LT |
1682 | break; |
1683 | } else { | |
1684 | if (sock_flag(sk, SOCK_DONE)) | |
1685 | break; | |
1686 | ||
1687 | if (sk->sk_err) { | |
1688 | copied = sock_error(sk); | |
1689 | break; | |
1690 | } | |
1691 | ||
1692 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
1693 | break; | |
1694 | ||
1695 | if (sk->sk_state == TCP_CLOSE) { | |
1696 | if (!sock_flag(sk, SOCK_DONE)) { | |
1697 | /* This occurs when user tries to read | |
1698 | * from never connected socket. | |
1699 | */ | |
1700 | copied = -ENOTCONN; | |
1701 | break; | |
1702 | } | |
1703 | break; | |
1704 | } | |
1705 | ||
1706 | if (!timeo) { | |
1707 | copied = -EAGAIN; | |
1708 | break; | |
1709 | } | |
1710 | ||
1711 | if (signal_pending(current)) { | |
1712 | copied = sock_intr_errno(timeo); | |
1713 | break; | |
1714 | } | |
1715 | } | |
1716 | ||
0e4b4992 | 1717 | tcp_cleanup_rbuf(sk, copied); |
1da177e4 | 1718 | |
7df55125 | 1719 | if (!sysctl_tcp_low_latency && tp->ucopy.task == user_recv) { |
1da177e4 LT |
1720 | /* Install new reader */ |
1721 | if (!user_recv && !(flags & (MSG_TRUNC | MSG_PEEK))) { | |
1722 | user_recv = current; | |
1723 | tp->ucopy.task = user_recv; | |
1724 | tp->ucopy.iov = msg->msg_iov; | |
1725 | } | |
1726 | ||
1727 | tp->ucopy.len = len; | |
1728 | ||
547b792c IJ |
1729 | WARN_ON(tp->copied_seq != tp->rcv_nxt && |
1730 | !(flags & (MSG_PEEK | MSG_TRUNC))); | |
1da177e4 LT |
1731 | |
1732 | /* Ugly... If prequeue is not empty, we have to | |
1733 | * process it before releasing socket, otherwise | |
1734 | * order will be broken at second iteration. | |
1735 | * More elegant solution is required!!! | |
1736 | * | |
1737 | * Look: we have the following (pseudo)queues: | |
1738 | * | |
1739 | * 1. packets in flight | |
1740 | * 2. backlog | |
1741 | * 3. prequeue | |
1742 | * 4. receive_queue | |
1743 | * | |
1744 | * Each queue can be processed only if the next ones | |
1745 | * are empty. At this point we have empty receive_queue. | |
1746 | * But prequeue _can_ be not empty after 2nd iteration, | |
1747 | * when we jumped to start of loop because backlog | |
1748 | * processing added something to receive_queue. | |
1749 | * We cannot release_sock(), because backlog contains | |
1750 | * packets arrived _after_ prequeued ones. | |
1751 | * | |
1752 | * Shortly, algorithm is clear --- to process all | |
1753 | * the queues in order. We could make it more directly, | |
1754 | * requeueing packets from backlog to prequeue, if | |
1755 | * is not empty. It is more elegant, but eats cycles, | |
1756 | * unfortunately. | |
1757 | */ | |
b03efcfb | 1758 | if (!skb_queue_empty(&tp->ucopy.prequeue)) |
1da177e4 LT |
1759 | goto do_prequeue; |
1760 | ||
1761 | /* __ Set realtime policy in scheduler __ */ | |
1762 | } | |
1763 | ||
73852e81 SM |
1764 | #ifdef CONFIG_NET_DMA |
1765 | if (tp->ucopy.dma_chan) | |
1766 | dma_async_memcpy_issue_pending(tp->ucopy.dma_chan); | |
1767 | #endif | |
1da177e4 LT |
1768 | if (copied >= target) { |
1769 | /* Do not sleep, just process backlog. */ | |
1770 | release_sock(sk); | |
1771 | lock_sock(sk); | |
1772 | } else | |
1773 | sk_wait_data(sk, &timeo); | |
1774 | ||
1a2449a8 | 1775 | #ifdef CONFIG_NET_DMA |
73852e81 | 1776 | tcp_service_net_dma(sk, false); /* Don't block */ |
1a2449a8 CL |
1777 | tp->ucopy.wakeup = 0; |
1778 | #endif | |
1779 | ||
1da177e4 LT |
1780 | if (user_recv) { |
1781 | int chunk; | |
1782 | ||
1783 | /* __ Restore normal policy in scheduler __ */ | |
1784 | ||
1785 | if ((chunk = len - tp->ucopy.len) != 0) { | |
ed88098e | 1786 | NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMBACKLOG, chunk); |
1da177e4 LT |
1787 | len -= chunk; |
1788 | copied += chunk; | |
1789 | } | |
1790 | ||
1791 | if (tp->rcv_nxt == tp->copied_seq && | |
b03efcfb | 1792 | !skb_queue_empty(&tp->ucopy.prequeue)) { |
1da177e4 LT |
1793 | do_prequeue: |
1794 | tcp_prequeue_process(sk); | |
1795 | ||
1796 | if ((chunk = len - tp->ucopy.len) != 0) { | |
ed88098e | 1797 | NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk); |
1da177e4 LT |
1798 | len -= chunk; |
1799 | copied += chunk; | |
1800 | } | |
1801 | } | |
1802 | } | |
77527313 IJ |
1803 | if ((flags & MSG_PEEK) && |
1804 | (peek_seq - copied - urg_hole != tp->copied_seq)) { | |
e87cc472 JP |
1805 | net_dbg_ratelimited("TCP(%s:%d): Application bug, race in MSG_PEEK\n", |
1806 | current->comm, | |
1807 | task_pid_nr(current)); | |
1da177e4 LT |
1808 | peek_seq = tp->copied_seq; |
1809 | } | |
1810 | continue; | |
1811 | ||
1812 | found_ok_skb: | |
1813 | /* Ok so how much can we use? */ | |
1814 | used = skb->len - offset; | |
1815 | if (len < used) | |
1816 | used = len; | |
1817 | ||
1818 | /* Do we have urgent data here? */ | |
1819 | if (tp->urg_data) { | |
1820 | u32 urg_offset = tp->urg_seq - *seq; | |
1821 | if (urg_offset < used) { | |
1822 | if (!urg_offset) { | |
1823 | if (!sock_flag(sk, SOCK_URGINLINE)) { | |
1824 | ++*seq; | |
77527313 | 1825 | urg_hole++; |
1da177e4 LT |
1826 | offset++; |
1827 | used--; | |
1828 | if (!used) | |
1829 | goto skip_copy; | |
1830 | } | |
1831 | } else | |
1832 | used = urg_offset; | |
1833 | } | |
1834 | } | |
1835 | ||
1836 | if (!(flags & MSG_TRUNC)) { | |
1a2449a8 CL |
1837 | #ifdef CONFIG_NET_DMA |
1838 | if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list) | |
a2bd1140 | 1839 | tp->ucopy.dma_chan = net_dma_find_channel(); |
1a2449a8 CL |
1840 | |
1841 | if (tp->ucopy.dma_chan) { | |
1842 | tp->ucopy.dma_cookie = dma_skb_copy_datagram_iovec( | |
1843 | tp->ucopy.dma_chan, skb, offset, | |
1844 | msg->msg_iov, used, | |
1845 | tp->ucopy.pinned_list); | |
1846 | ||
1847 | if (tp->ucopy.dma_cookie < 0) { | |
1848 | ||
afd46503 JP |
1849 | pr_alert("%s: dma_cookie < 0\n", |
1850 | __func__); | |
1a2449a8 CL |
1851 | |
1852 | /* Exception. Bailout! */ | |
1853 | if (!copied) | |
1854 | copied = -EFAULT; | |
1855 | break; | |
1856 | } | |
73852e81 SM |
1857 | |
1858 | dma_async_memcpy_issue_pending(tp->ucopy.dma_chan); | |
1859 | ||
1a2449a8 | 1860 | if ((offset + used) == skb->len) |
dc6b9b78 | 1861 | copied_early = true; |
1a2449a8 CL |
1862 | |
1863 | } else | |
1864 | #endif | |
1865 | { | |
1866 | err = skb_copy_datagram_iovec(skb, offset, | |
1867 | msg->msg_iov, used); | |
1868 | if (err) { | |
1869 | /* Exception. Bailout! */ | |
1870 | if (!copied) | |
1871 | copied = -EFAULT; | |
1872 | break; | |
1873 | } | |
1da177e4 LT |
1874 | } |
1875 | } | |
1876 | ||
1877 | *seq += used; | |
1878 | copied += used; | |
1879 | len -= used; | |
1880 | ||
1881 | tcp_rcv_space_adjust(sk); | |
1882 | ||
1883 | skip_copy: | |
1884 | if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) { | |
1885 | tp->urg_data = 0; | |
9e412ba7 | 1886 | tcp_fast_path_check(sk); |
1da177e4 LT |
1887 | } |
1888 | if (used + offset < skb->len) | |
1889 | continue; | |
1890 | ||
aa8223c7 | 1891 | if (tcp_hdr(skb)->fin) |
1da177e4 | 1892 | goto found_fin_ok; |
1a2449a8 CL |
1893 | if (!(flags & MSG_PEEK)) { |
1894 | sk_eat_skb(sk, skb, copied_early); | |
dc6b9b78 | 1895 | copied_early = false; |
1a2449a8 | 1896 | } |
1da177e4 LT |
1897 | continue; |
1898 | ||
1899 | found_fin_ok: | |
1900 | /* Process the FIN. */ | |
1901 | ++*seq; | |
1a2449a8 CL |
1902 | if (!(flags & MSG_PEEK)) { |
1903 | sk_eat_skb(sk, skb, copied_early); | |
dc6b9b78 | 1904 | copied_early = false; |
1a2449a8 | 1905 | } |
1da177e4 LT |
1906 | break; |
1907 | } while (len > 0); | |
1908 | ||
1909 | if (user_recv) { | |
b03efcfb | 1910 | if (!skb_queue_empty(&tp->ucopy.prequeue)) { |
1da177e4 LT |
1911 | int chunk; |
1912 | ||
1913 | tp->ucopy.len = copied > 0 ? len : 0; | |
1914 | ||
1915 | tcp_prequeue_process(sk); | |
1916 | ||
1917 | if (copied > 0 && (chunk = len - tp->ucopy.len) != 0) { | |
ed88098e | 1918 | NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk); |
1da177e4 LT |
1919 | len -= chunk; |
1920 | copied += chunk; | |
1921 | } | |
1922 | } | |
1923 | ||
1924 | tp->ucopy.task = NULL; | |
1925 | tp->ucopy.len = 0; | |
1926 | } | |
1927 | ||
1a2449a8 | 1928 | #ifdef CONFIG_NET_DMA |
73852e81 SM |
1929 | tcp_service_net_dma(sk, true); /* Wait for queue to drain */ |
1930 | tp->ucopy.dma_chan = NULL; | |
1a2449a8 | 1931 | |
1a2449a8 CL |
1932 | if (tp->ucopy.pinned_list) { |
1933 | dma_unpin_iovec_pages(tp->ucopy.pinned_list); | |
1934 | tp->ucopy.pinned_list = NULL; | |
1935 | } | |
1936 | #endif | |
1937 | ||
1da177e4 LT |
1938 | /* According to UNIX98, msg_name/msg_namelen are ignored |
1939 | * on connected socket. I was just happy when found this 8) --ANK | |
1940 | */ | |
1941 | ||
1942 | /* Clean up data we have read: This will do ACK frames. */ | |
0e4b4992 | 1943 | tcp_cleanup_rbuf(sk, copied); |
1da177e4 | 1944 | |
1da177e4 LT |
1945 | release_sock(sk); |
1946 | return copied; | |
1947 | ||
1948 | out: | |
1da177e4 LT |
1949 | release_sock(sk); |
1950 | return err; | |
1951 | ||
1952 | recv_urg: | |
377f0a08 | 1953 | err = tcp_recv_urg(sk, msg, len, flags); |
1da177e4 | 1954 | goto out; |
c0e88ff0 PE |
1955 | |
1956 | recv_sndq: | |
1957 | err = tcp_peek_sndq(sk, msg, len); | |
1958 | goto out; | |
1da177e4 | 1959 | } |
4bc2f18b | 1960 | EXPORT_SYMBOL(tcp_recvmsg); |
1da177e4 | 1961 | |
490d5046 IJ |
1962 | void tcp_set_state(struct sock *sk, int state) |
1963 | { | |
1964 | int oldstate = sk->sk_state; | |
1965 | ||
1966 | switch (state) { | |
1967 | case TCP_ESTABLISHED: | |
1968 | if (oldstate != TCP_ESTABLISHED) | |
81cc8a75 | 1969 | TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB); |
490d5046 IJ |
1970 | break; |
1971 | ||
1972 | case TCP_CLOSE: | |
1973 | if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED) | |
81cc8a75 | 1974 | TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS); |
490d5046 IJ |
1975 | |
1976 | sk->sk_prot->unhash(sk); | |
1977 | if (inet_csk(sk)->icsk_bind_hash && | |
1978 | !(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) | |
ab1e0a13 | 1979 | inet_put_port(sk); |
490d5046 IJ |
1980 | /* fall through */ |
1981 | default: | |
5a5f3a8d | 1982 | if (oldstate == TCP_ESTABLISHED) |
74688e48 | 1983 | TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB); |
490d5046 IJ |
1984 | } |
1985 | ||
1986 | /* Change state AFTER socket is unhashed to avoid closed | |
1987 | * socket sitting in hash tables. | |
1988 | */ | |
1989 | sk->sk_state = state; | |
1990 | ||
1991 | #ifdef STATE_TRACE | |
5a5f3a8d | 1992 | SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n", sk, statename[oldstate], statename[state]); |
490d5046 IJ |
1993 | #endif |
1994 | } | |
1995 | EXPORT_SYMBOL_GPL(tcp_set_state); | |
1996 | ||
1da177e4 LT |
1997 | /* |
1998 | * State processing on a close. This implements the state shift for | |
1999 | * sending our FIN frame. Note that we only send a FIN for some | |
2000 | * states. A shutdown() may have already sent the FIN, or we may be | |
2001 | * closed. | |
2002 | */ | |
2003 | ||
9b5b5cff | 2004 | static const unsigned char new_state[16] = { |
1da177e4 LT |
2005 | /* current state: new state: action: */ |
2006 | /* (Invalid) */ TCP_CLOSE, | |
2007 | /* TCP_ESTABLISHED */ TCP_FIN_WAIT1 | TCP_ACTION_FIN, | |
2008 | /* TCP_SYN_SENT */ TCP_CLOSE, | |
2009 | /* TCP_SYN_RECV */ TCP_FIN_WAIT1 | TCP_ACTION_FIN, | |
2010 | /* TCP_FIN_WAIT1 */ TCP_FIN_WAIT1, | |
2011 | /* TCP_FIN_WAIT2 */ TCP_FIN_WAIT2, | |
2012 | /* TCP_TIME_WAIT */ TCP_CLOSE, | |
2013 | /* TCP_CLOSE */ TCP_CLOSE, | |
2014 | /* TCP_CLOSE_WAIT */ TCP_LAST_ACK | TCP_ACTION_FIN, | |
2015 | /* TCP_LAST_ACK */ TCP_LAST_ACK, | |
2016 | /* TCP_LISTEN */ TCP_CLOSE, | |
2017 | /* TCP_CLOSING */ TCP_CLOSING, | |
2018 | }; | |
2019 | ||
2020 | static int tcp_close_state(struct sock *sk) | |
2021 | { | |
2022 | int next = (int)new_state[sk->sk_state]; | |
2023 | int ns = next & TCP_STATE_MASK; | |
2024 | ||
2025 | tcp_set_state(sk, ns); | |
2026 | ||
2027 | return next & TCP_ACTION_FIN; | |
2028 | } | |
2029 | ||
2030 | /* | |
2031 | * Shutdown the sending side of a connection. Much like close except | |
1f29b058 | 2032 | * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD). |
1da177e4 LT |
2033 | */ |
2034 | ||
2035 | void tcp_shutdown(struct sock *sk, int how) | |
2036 | { | |
2037 | /* We need to grab some memory, and put together a FIN, | |
2038 | * and then put it into the queue to be sent. | |
2039 | * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92. | |
2040 | */ | |
2041 | if (!(how & SEND_SHUTDOWN)) | |
2042 | return; | |
2043 | ||
2044 | /* If we've already sent a FIN, or it's a closed state, skip this. */ | |
2045 | if ((1 << sk->sk_state) & | |
2046 | (TCPF_ESTABLISHED | TCPF_SYN_SENT | | |
2047 | TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) { | |
2048 | /* Clear out any half completed packets. FIN if needed. */ | |
2049 | if (tcp_close_state(sk)) | |
2050 | tcp_send_fin(sk); | |
2051 | } | |
2052 | } | |
4bc2f18b | 2053 | EXPORT_SYMBOL(tcp_shutdown); |
1da177e4 | 2054 | |
efcdbf24 AS |
2055 | bool tcp_check_oom(struct sock *sk, int shift) |
2056 | { | |
2057 | bool too_many_orphans, out_of_socket_memory; | |
2058 | ||
2059 | too_many_orphans = tcp_too_many_orphans(sk, shift); | |
2060 | out_of_socket_memory = tcp_out_of_memory(sk); | |
2061 | ||
e87cc472 JP |
2062 | if (too_many_orphans) |
2063 | net_info_ratelimited("too many orphaned sockets\n"); | |
2064 | if (out_of_socket_memory) | |
2065 | net_info_ratelimited("out of memory -- consider tuning tcp_mem\n"); | |
efcdbf24 AS |
2066 | return too_many_orphans || out_of_socket_memory; |
2067 | } | |
2068 | ||
1da177e4 LT |
2069 | void tcp_close(struct sock *sk, long timeout) |
2070 | { | |
2071 | struct sk_buff *skb; | |
2072 | int data_was_unread = 0; | |
75c2d907 | 2073 | int state; |
1da177e4 LT |
2074 | |
2075 | lock_sock(sk); | |
2076 | sk->sk_shutdown = SHUTDOWN_MASK; | |
2077 | ||
2078 | if (sk->sk_state == TCP_LISTEN) { | |
2079 | tcp_set_state(sk, TCP_CLOSE); | |
2080 | ||
2081 | /* Special case. */ | |
0a5578cf | 2082 | inet_csk_listen_stop(sk); |
1da177e4 LT |
2083 | |
2084 | goto adjudge_to_death; | |
2085 | } | |
2086 | ||
2087 | /* We need to flush the recv. buffs. We do this only on the | |
2088 | * descriptor close, not protocol-sourced closes, because the | |
2089 | * reader process may not have drained the data yet! | |
2090 | */ | |
2091 | while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) { | |
2092 | u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq - | |
aa8223c7 | 2093 | tcp_hdr(skb)->fin; |
1da177e4 LT |
2094 | data_was_unread += len; |
2095 | __kfree_skb(skb); | |
2096 | } | |
2097 | ||
3ab224be | 2098 | sk_mem_reclaim(sk); |
1da177e4 | 2099 | |
565b7b2d KK |
2100 | /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */ |
2101 | if (sk->sk_state == TCP_CLOSE) | |
2102 | goto adjudge_to_death; | |
2103 | ||
65bb723c GR |
2104 | /* As outlined in RFC 2525, section 2.17, we send a RST here because |
2105 | * data was lost. To witness the awful effects of the old behavior of | |
2106 | * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk | |
2107 | * GET in an FTP client, suspend the process, wait for the client to | |
2108 | * advertise a zero window, then kill -9 the FTP client, wheee... | |
2109 | * Note: timeout is always zero in such a case. | |
1da177e4 | 2110 | */ |
ee995283 PE |
2111 | if (unlikely(tcp_sk(sk)->repair)) { |
2112 | sk->sk_prot->disconnect(sk, 0); | |
2113 | } else if (data_was_unread) { | |
1da177e4 | 2114 | /* Unread data was tossed, zap the connection. */ |
6f67c817 | 2115 | NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE); |
1da177e4 | 2116 | tcp_set_state(sk, TCP_CLOSE); |
aa133076 | 2117 | tcp_send_active_reset(sk, sk->sk_allocation); |
1da177e4 LT |
2118 | } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) { |
2119 | /* Check zero linger _after_ checking for unread data. */ | |
2120 | sk->sk_prot->disconnect(sk, 0); | |
6f67c817 | 2121 | NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONDATA); |
1da177e4 LT |
2122 | } else if (tcp_close_state(sk)) { |
2123 | /* We FIN if the application ate all the data before | |
2124 | * zapping the connection. | |
2125 | */ | |
2126 | ||
2127 | /* RED-PEN. Formally speaking, we have broken TCP state | |
2128 | * machine. State transitions: | |
2129 | * | |
2130 | * TCP_ESTABLISHED -> TCP_FIN_WAIT1 | |
2131 | * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible) | |
2132 | * TCP_CLOSE_WAIT -> TCP_LAST_ACK | |
2133 | * | |
2134 | * are legal only when FIN has been sent (i.e. in window), | |
2135 | * rather than queued out of window. Purists blame. | |
2136 | * | |
2137 | * F.e. "RFC state" is ESTABLISHED, | |
2138 | * if Linux state is FIN-WAIT-1, but FIN is still not sent. | |
2139 | * | |
2140 | * The visible declinations are that sometimes | |
2141 | * we enter time-wait state, when it is not required really | |
2142 | * (harmless), do not send active resets, when they are | |
2143 | * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when | |
2144 | * they look as CLOSING or LAST_ACK for Linux) | |
2145 | * Probably, I missed some more holelets. | |
2146 | * --ANK | |
2147 | */ | |
2148 | tcp_send_fin(sk); | |
2149 | } | |
2150 | ||
2151 | sk_stream_wait_close(sk, timeout); | |
2152 | ||
2153 | adjudge_to_death: | |
75c2d907 HX |
2154 | state = sk->sk_state; |
2155 | sock_hold(sk); | |
2156 | sock_orphan(sk); | |
75c2d907 | 2157 | |
1da177e4 LT |
2158 | /* It is the last release_sock in its life. It will remove backlog. */ |
2159 | release_sock(sk); | |
2160 | ||
2161 | ||
2162 | /* Now socket is owned by kernel and we acquire BH lock | |
2163 | to finish close. No need to check for user refs. | |
2164 | */ | |
2165 | local_bh_disable(); | |
2166 | bh_lock_sock(sk); | |
547b792c | 2167 | WARN_ON(sock_owned_by_user(sk)); |
1da177e4 | 2168 | |
eb4dea58 HX |
2169 | percpu_counter_inc(sk->sk_prot->orphan_count); |
2170 | ||
75c2d907 HX |
2171 | /* Have we already been destroyed by a softirq or backlog? */ |
2172 | if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE) | |
2173 | goto out; | |
1da177e4 LT |
2174 | |
2175 | /* This is a (useful) BSD violating of the RFC. There is a | |
2176 | * problem with TCP as specified in that the other end could | |
2177 | * keep a socket open forever with no application left this end. | |
2178 | * We use a 3 minute timeout (about the same as BSD) then kill | |
2179 | * our end. If they send after that then tough - BUT: long enough | |
2180 | * that we won't make the old 4*rto = almost no time - whoops | |
2181 | * reset mistake. | |
2182 | * | |
2183 | * Nope, it was not mistake. It is really desired behaviour | |
2184 | * f.e. on http servers, when such sockets are useless, but | |
2185 | * consume significant resources. Let's do it with special | |
2186 | * linger2 option. --ANK | |
2187 | */ | |
2188 | ||
2189 | if (sk->sk_state == TCP_FIN_WAIT2) { | |
2190 | struct tcp_sock *tp = tcp_sk(sk); | |
2191 | if (tp->linger2 < 0) { | |
2192 | tcp_set_state(sk, TCP_CLOSE); | |
2193 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
de0744af PE |
2194 | NET_INC_STATS_BH(sock_net(sk), |
2195 | LINUX_MIB_TCPABORTONLINGER); | |
1da177e4 | 2196 | } else { |
463c84b9 | 2197 | const int tmo = tcp_fin_time(sk); |
1da177e4 LT |
2198 | |
2199 | if (tmo > TCP_TIMEWAIT_LEN) { | |
52499afe DM |
2200 | inet_csk_reset_keepalive_timer(sk, |
2201 | tmo - TCP_TIMEWAIT_LEN); | |
1da177e4 | 2202 | } else { |
1da177e4 LT |
2203 | tcp_time_wait(sk, TCP_FIN_WAIT2, tmo); |
2204 | goto out; | |
2205 | } | |
2206 | } | |
2207 | } | |
2208 | if (sk->sk_state != TCP_CLOSE) { | |
3ab224be | 2209 | sk_mem_reclaim(sk); |
efcdbf24 | 2210 | if (tcp_check_oom(sk, 0)) { |
1da177e4 LT |
2211 | tcp_set_state(sk, TCP_CLOSE); |
2212 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
de0744af PE |
2213 | NET_INC_STATS_BH(sock_net(sk), |
2214 | LINUX_MIB_TCPABORTONMEMORY); | |
1da177e4 LT |
2215 | } |
2216 | } | |
1da177e4 LT |
2217 | |
2218 | if (sk->sk_state == TCP_CLOSE) | |
0a5578cf | 2219 | inet_csk_destroy_sock(sk); |
1da177e4 LT |
2220 | /* Otherwise, socket is reprieved until protocol close. */ |
2221 | ||
2222 | out: | |
2223 | bh_unlock_sock(sk); | |
2224 | local_bh_enable(); | |
2225 | sock_put(sk); | |
2226 | } | |
4bc2f18b | 2227 | EXPORT_SYMBOL(tcp_close); |
1da177e4 LT |
2228 | |
2229 | /* These states need RST on ABORT according to RFC793 */ | |
2230 | ||
a2a385d6 | 2231 | static inline bool tcp_need_reset(int state) |
1da177e4 LT |
2232 | { |
2233 | return (1 << state) & | |
2234 | (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 | | |
2235 | TCPF_FIN_WAIT2 | TCPF_SYN_RECV); | |
2236 | } | |
2237 | ||
2238 | int tcp_disconnect(struct sock *sk, int flags) | |
2239 | { | |
2240 | struct inet_sock *inet = inet_sk(sk); | |
463c84b9 | 2241 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
2242 | struct tcp_sock *tp = tcp_sk(sk); |
2243 | int err = 0; | |
2244 | int old_state = sk->sk_state; | |
2245 | ||
2246 | if (old_state != TCP_CLOSE) | |
2247 | tcp_set_state(sk, TCP_CLOSE); | |
2248 | ||
2249 | /* ABORT function of RFC793 */ | |
2250 | if (old_state == TCP_LISTEN) { | |
0a5578cf | 2251 | inet_csk_listen_stop(sk); |
ee995283 PE |
2252 | } else if (unlikely(tp->repair)) { |
2253 | sk->sk_err = ECONNABORTED; | |
1da177e4 LT |
2254 | } else if (tcp_need_reset(old_state) || |
2255 | (tp->snd_nxt != tp->write_seq && | |
2256 | (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) { | |
caa20d9a | 2257 | /* The last check adjusts for discrepancy of Linux wrt. RFC |
1da177e4 LT |
2258 | * states |
2259 | */ | |
2260 | tcp_send_active_reset(sk, gfp_any()); | |
2261 | sk->sk_err = ECONNRESET; | |
2262 | } else if (old_state == TCP_SYN_SENT) | |
2263 | sk->sk_err = ECONNRESET; | |
2264 | ||
2265 | tcp_clear_xmit_timers(sk); | |
2266 | __skb_queue_purge(&sk->sk_receive_queue); | |
fe067e8a | 2267 | tcp_write_queue_purge(sk); |
1da177e4 | 2268 | __skb_queue_purge(&tp->out_of_order_queue); |
1a2449a8 CL |
2269 | #ifdef CONFIG_NET_DMA |
2270 | __skb_queue_purge(&sk->sk_async_wait_queue); | |
2271 | #endif | |
1da177e4 | 2272 | |
c720c7e8 | 2273 | inet->inet_dport = 0; |
1da177e4 LT |
2274 | |
2275 | if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) | |
2276 | inet_reset_saddr(sk); | |
2277 | ||
2278 | sk->sk_shutdown = 0; | |
2279 | sock_reset_flag(sk, SOCK_DONE); | |
2280 | tp->srtt = 0; | |
2281 | if ((tp->write_seq += tp->max_window + 2) == 0) | |
2282 | tp->write_seq = 1; | |
463c84b9 | 2283 | icsk->icsk_backoff = 0; |
1da177e4 | 2284 | tp->snd_cwnd = 2; |
6687e988 | 2285 | icsk->icsk_probes_out = 0; |
1da177e4 | 2286 | tp->packets_out = 0; |
0b6a05c1 | 2287 | tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; |
1da177e4 | 2288 | tp->snd_cwnd_cnt = 0; |
9772efb9 | 2289 | tp->bytes_acked = 0; |
1fdf475a | 2290 | tp->window_clamp = 0; |
6687e988 | 2291 | tcp_set_ca_state(sk, TCP_CA_Open); |
1da177e4 | 2292 | tcp_clear_retrans(tp); |
463c84b9 | 2293 | inet_csk_delack_init(sk); |
fe067e8a | 2294 | tcp_init_send_head(sk); |
b40b4f79 | 2295 | memset(&tp->rx_opt, 0, sizeof(tp->rx_opt)); |
1da177e4 LT |
2296 | __sk_dst_reset(sk); |
2297 | ||
c720c7e8 | 2298 | WARN_ON(inet->inet_num && !icsk->icsk_bind_hash); |
1da177e4 LT |
2299 | |
2300 | sk->sk_error_report(sk); | |
2301 | return err; | |
2302 | } | |
4bc2f18b | 2303 | EXPORT_SYMBOL(tcp_disconnect); |
1da177e4 | 2304 | |
a2a385d6 | 2305 | static inline bool tcp_can_repair_sock(const struct sock *sk) |
ee995283 PE |
2306 | { |
2307 | return capable(CAP_NET_ADMIN) && | |
2308 | ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_ESTABLISHED)); | |
2309 | } | |
2310 | ||
de248a75 PE |
2311 | static int tcp_repair_options_est(struct tcp_sock *tp, |
2312 | struct tcp_repair_opt __user *optbuf, unsigned int len) | |
b139ba4e | 2313 | { |
de248a75 | 2314 | struct tcp_repair_opt opt; |
b139ba4e | 2315 | |
de248a75 PE |
2316 | while (len >= sizeof(opt)) { |
2317 | if (copy_from_user(&opt, optbuf, sizeof(opt))) | |
b139ba4e PE |
2318 | return -EFAULT; |
2319 | ||
2320 | optbuf++; | |
de248a75 | 2321 | len -= sizeof(opt); |
b139ba4e | 2322 | |
de248a75 PE |
2323 | switch (opt.opt_code) { |
2324 | case TCPOPT_MSS: | |
2325 | tp->rx_opt.mss_clamp = opt.opt_val; | |
b139ba4e | 2326 | break; |
de248a75 PE |
2327 | case TCPOPT_WINDOW: |
2328 | if (opt.opt_val > 14) | |
b139ba4e PE |
2329 | return -EFBIG; |
2330 | ||
de248a75 | 2331 | tp->rx_opt.snd_wscale = opt.opt_val; |
b139ba4e | 2332 | break; |
b139ba4e | 2333 | case TCPOPT_SACK_PERM: |
de248a75 PE |
2334 | if (opt.opt_val != 0) |
2335 | return -EINVAL; | |
2336 | ||
b139ba4e PE |
2337 | tp->rx_opt.sack_ok |= TCP_SACK_SEEN; |
2338 | if (sysctl_tcp_fack) | |
2339 | tcp_enable_fack(tp); | |
2340 | break; | |
2341 | case TCPOPT_TIMESTAMP: | |
de248a75 PE |
2342 | if (opt.opt_val != 0) |
2343 | return -EINVAL; | |
2344 | ||
b139ba4e PE |
2345 | tp->rx_opt.tstamp_ok = 1; |
2346 | break; | |
2347 | } | |
2348 | } | |
2349 | ||
2350 | return 0; | |
2351 | } | |
2352 | ||
1da177e4 LT |
2353 | /* |
2354 | * Socket option code for TCP. | |
2355 | */ | |
3fdadf7d | 2356 | static int do_tcp_setsockopt(struct sock *sk, int level, |
b7058842 | 2357 | int optname, char __user *optval, unsigned int optlen) |
1da177e4 LT |
2358 | { |
2359 | struct tcp_sock *tp = tcp_sk(sk); | |
463c84b9 | 2360 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
2361 | int val; |
2362 | int err = 0; | |
2363 | ||
e56fb50f WAS |
2364 | /* These are data/string values, all the others are ints */ |
2365 | switch (optname) { | |
2366 | case TCP_CONGESTION: { | |
5f8ef48d SH |
2367 | char name[TCP_CA_NAME_MAX]; |
2368 | ||
2369 | if (optlen < 1) | |
2370 | return -EINVAL; | |
2371 | ||
2372 | val = strncpy_from_user(name, optval, | |
4fdb78d3 | 2373 | min_t(long, TCP_CA_NAME_MAX-1, optlen)); |
5f8ef48d SH |
2374 | if (val < 0) |
2375 | return -EFAULT; | |
2376 | name[val] = 0; | |
2377 | ||
2378 | lock_sock(sk); | |
6687e988 | 2379 | err = tcp_set_congestion_control(sk, name); |
5f8ef48d SH |
2380 | release_sock(sk); |
2381 | return err; | |
2382 | } | |
e56fb50f WAS |
2383 | case TCP_COOKIE_TRANSACTIONS: { |
2384 | struct tcp_cookie_transactions ctd; | |
2385 | struct tcp_cookie_values *cvp = NULL; | |
2386 | ||
2387 | if (sizeof(ctd) > optlen) | |
2388 | return -EINVAL; | |
2389 | if (copy_from_user(&ctd, optval, sizeof(ctd))) | |
2390 | return -EFAULT; | |
2391 | ||
2392 | if (ctd.tcpct_used > sizeof(ctd.tcpct_value) || | |
2393 | ctd.tcpct_s_data_desired > TCP_MSS_DESIRED) | |
2394 | return -EINVAL; | |
2395 | ||
2396 | if (ctd.tcpct_cookie_desired == 0) { | |
2397 | /* default to global value */ | |
2398 | } else if ((0x1 & ctd.tcpct_cookie_desired) || | |
2399 | ctd.tcpct_cookie_desired > TCP_COOKIE_MAX || | |
2400 | ctd.tcpct_cookie_desired < TCP_COOKIE_MIN) { | |
2401 | return -EINVAL; | |
2402 | } | |
2403 | ||
2404 | if (TCP_COOKIE_OUT_NEVER & ctd.tcpct_flags) { | |
2405 | /* Supercedes all other values */ | |
2406 | lock_sock(sk); | |
2407 | if (tp->cookie_values != NULL) { | |
2408 | kref_put(&tp->cookie_values->kref, | |
2409 | tcp_cookie_values_release); | |
2410 | tp->cookie_values = NULL; | |
2411 | } | |
2412 | tp->rx_opt.cookie_in_always = 0; /* false */ | |
2413 | tp->rx_opt.cookie_out_never = 1; /* true */ | |
2414 | release_sock(sk); | |
2415 | return err; | |
2416 | } | |
2417 | ||
2418 | /* Allocate ancillary memory before locking. | |
2419 | */ | |
2420 | if (ctd.tcpct_used > 0 || | |
2421 | (tp->cookie_values == NULL && | |
2422 | (sysctl_tcp_cookie_size > 0 || | |
2423 | ctd.tcpct_cookie_desired > 0 || | |
2424 | ctd.tcpct_s_data_desired > 0))) { | |
2425 | cvp = kzalloc(sizeof(*cvp) + ctd.tcpct_used, | |
2426 | GFP_KERNEL); | |
2427 | if (cvp == NULL) | |
2428 | return -ENOMEM; | |
a3bdb549 DP |
2429 | |
2430 | kref_init(&cvp->kref); | |
e56fb50f WAS |
2431 | } |
2432 | lock_sock(sk); | |
2433 | tp->rx_opt.cookie_in_always = | |
2434 | (TCP_COOKIE_IN_ALWAYS & ctd.tcpct_flags); | |
2435 | tp->rx_opt.cookie_out_never = 0; /* false */ | |
2436 | ||
2437 | if (tp->cookie_values != NULL) { | |
2438 | if (cvp != NULL) { | |
2439 | /* Changed values are recorded by a changed | |
2440 | * pointer, ensuring the cookie will differ, | |
2441 | * without separately hashing each value later. | |
2442 | */ | |
2443 | kref_put(&tp->cookie_values->kref, | |
2444 | tcp_cookie_values_release); | |
e56fb50f WAS |
2445 | } else { |
2446 | cvp = tp->cookie_values; | |
2447 | } | |
2448 | } | |
a3bdb549 | 2449 | |
e56fb50f WAS |
2450 | if (cvp != NULL) { |
2451 | cvp->cookie_desired = ctd.tcpct_cookie_desired; | |
2452 | ||
2453 | if (ctd.tcpct_used > 0) { | |
2454 | memcpy(cvp->s_data_payload, ctd.tcpct_value, | |
2455 | ctd.tcpct_used); | |
2456 | cvp->s_data_desired = ctd.tcpct_used; | |
2457 | cvp->s_data_constant = 1; /* true */ | |
2458 | } else { | |
2459 | /* No constant payload data. */ | |
2460 | cvp->s_data_desired = ctd.tcpct_s_data_desired; | |
2461 | cvp->s_data_constant = 0; /* false */ | |
2462 | } | |
a3bdb549 DP |
2463 | |
2464 | tp->cookie_values = cvp; | |
e56fb50f WAS |
2465 | } |
2466 | release_sock(sk); | |
2467 | return err; | |
2468 | } | |
2469 | default: | |
2470 | /* fallthru */ | |
2471 | break; | |
ccbd6a5a | 2472 | } |
5f8ef48d | 2473 | |
1da177e4 LT |
2474 | if (optlen < sizeof(int)) |
2475 | return -EINVAL; | |
2476 | ||
2477 | if (get_user(val, (int __user *)optval)) | |
2478 | return -EFAULT; | |
2479 | ||
2480 | lock_sock(sk); | |
2481 | ||
2482 | switch (optname) { | |
2483 | case TCP_MAXSEG: | |
2484 | /* Values greater than interface MTU won't take effect. However | |
2485 | * at the point when this call is done we typically don't yet | |
2486 | * know which interface is going to be used */ | |
c39508d6 | 2487 | if (val < TCP_MIN_MSS || val > MAX_TCP_WINDOW) { |
1da177e4 LT |
2488 | err = -EINVAL; |
2489 | break; | |
2490 | } | |
2491 | tp->rx_opt.user_mss = val; | |
2492 | break; | |
2493 | ||
2494 | case TCP_NODELAY: | |
2495 | if (val) { | |
2496 | /* TCP_NODELAY is weaker than TCP_CORK, so that | |
2497 | * this option on corked socket is remembered, but | |
2498 | * it is not activated until cork is cleared. | |
2499 | * | |
2500 | * However, when TCP_NODELAY is set we make | |
2501 | * an explicit push, which overrides even TCP_CORK | |
2502 | * for currently queued segments. | |
2503 | */ | |
2504 | tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH; | |
9e412ba7 | 2505 | tcp_push_pending_frames(sk); |
1da177e4 LT |
2506 | } else { |
2507 | tp->nonagle &= ~TCP_NAGLE_OFF; | |
2508 | } | |
2509 | break; | |
2510 | ||
36e31b0a AP |
2511 | case TCP_THIN_LINEAR_TIMEOUTS: |
2512 | if (val < 0 || val > 1) | |
2513 | err = -EINVAL; | |
2514 | else | |
2515 | tp->thin_lto = val; | |
2516 | break; | |
2517 | ||
7e380175 AP |
2518 | case TCP_THIN_DUPACK: |
2519 | if (val < 0 || val > 1) | |
2520 | err = -EINVAL; | |
2521 | else | |
2522 | tp->thin_dupack = val; | |
eed530b6 YC |
2523 | if (tp->thin_dupack) |
2524 | tcp_disable_early_retrans(tp); | |
7e380175 AP |
2525 | break; |
2526 | ||
ee995283 PE |
2527 | case TCP_REPAIR: |
2528 | if (!tcp_can_repair_sock(sk)) | |
2529 | err = -EPERM; | |
2530 | else if (val == 1) { | |
2531 | tp->repair = 1; | |
2532 | sk->sk_reuse = SK_FORCE_REUSE; | |
2533 | tp->repair_queue = TCP_NO_QUEUE; | |
2534 | } else if (val == 0) { | |
2535 | tp->repair = 0; | |
2536 | sk->sk_reuse = SK_NO_REUSE; | |
2537 | tcp_send_window_probe(sk); | |
2538 | } else | |
2539 | err = -EINVAL; | |
2540 | ||
2541 | break; | |
2542 | ||
2543 | case TCP_REPAIR_QUEUE: | |
2544 | if (!tp->repair) | |
2545 | err = -EPERM; | |
2546 | else if (val < TCP_QUEUES_NR) | |
2547 | tp->repair_queue = val; | |
2548 | else | |
2549 | err = -EINVAL; | |
2550 | break; | |
2551 | ||
2552 | case TCP_QUEUE_SEQ: | |
2553 | if (sk->sk_state != TCP_CLOSE) | |
2554 | err = -EPERM; | |
2555 | else if (tp->repair_queue == TCP_SEND_QUEUE) | |
2556 | tp->write_seq = val; | |
2557 | else if (tp->repair_queue == TCP_RECV_QUEUE) | |
2558 | tp->rcv_nxt = val; | |
2559 | else | |
2560 | err = -EINVAL; | |
2561 | break; | |
2562 | ||
b139ba4e PE |
2563 | case TCP_REPAIR_OPTIONS: |
2564 | if (!tp->repair) | |
2565 | err = -EINVAL; | |
2566 | else if (sk->sk_state == TCP_ESTABLISHED) | |
de248a75 PE |
2567 | err = tcp_repair_options_est(tp, |
2568 | (struct tcp_repair_opt __user *)optval, | |
2569 | optlen); | |
b139ba4e PE |
2570 | else |
2571 | err = -EPERM; | |
2572 | break; | |
2573 | ||
1da177e4 LT |
2574 | case TCP_CORK: |
2575 | /* When set indicates to always queue non-full frames. | |
2576 | * Later the user clears this option and we transmit | |
2577 | * any pending partial frames in the queue. This is | |
2578 | * meant to be used alongside sendfile() to get properly | |
2579 | * filled frames when the user (for example) must write | |
2580 | * out headers with a write() call first and then use | |
2581 | * sendfile to send out the data parts. | |
2582 | * | |
2583 | * TCP_CORK can be set together with TCP_NODELAY and it is | |
2584 | * stronger than TCP_NODELAY. | |
2585 | */ | |
2586 | if (val) { | |
2587 | tp->nonagle |= TCP_NAGLE_CORK; | |
2588 | } else { | |
2589 | tp->nonagle &= ~TCP_NAGLE_CORK; | |
2590 | if (tp->nonagle&TCP_NAGLE_OFF) | |
2591 | tp->nonagle |= TCP_NAGLE_PUSH; | |
9e412ba7 | 2592 | tcp_push_pending_frames(sk); |
1da177e4 LT |
2593 | } |
2594 | break; | |
2595 | ||
2596 | case TCP_KEEPIDLE: | |
2597 | if (val < 1 || val > MAX_TCP_KEEPIDLE) | |
2598 | err = -EINVAL; | |
2599 | else { | |
2600 | tp->keepalive_time = val * HZ; | |
2601 | if (sock_flag(sk, SOCK_KEEPOPEN) && | |
2602 | !((1 << sk->sk_state) & | |
2603 | (TCPF_CLOSE | TCPF_LISTEN))) { | |
6c37e5de | 2604 | u32 elapsed = keepalive_time_elapsed(tp); |
1da177e4 LT |
2605 | if (tp->keepalive_time > elapsed) |
2606 | elapsed = tp->keepalive_time - elapsed; | |
2607 | else | |
2608 | elapsed = 0; | |
463c84b9 | 2609 | inet_csk_reset_keepalive_timer(sk, elapsed); |
1da177e4 LT |
2610 | } |
2611 | } | |
2612 | break; | |
2613 | case TCP_KEEPINTVL: | |
2614 | if (val < 1 || val > MAX_TCP_KEEPINTVL) | |
2615 | err = -EINVAL; | |
2616 | else | |
2617 | tp->keepalive_intvl = val * HZ; | |
2618 | break; | |
2619 | case TCP_KEEPCNT: | |
2620 | if (val < 1 || val > MAX_TCP_KEEPCNT) | |
2621 | err = -EINVAL; | |
2622 | else | |
2623 | tp->keepalive_probes = val; | |
2624 | break; | |
2625 | case TCP_SYNCNT: | |
2626 | if (val < 1 || val > MAX_TCP_SYNCNT) | |
2627 | err = -EINVAL; | |
2628 | else | |
463c84b9 | 2629 | icsk->icsk_syn_retries = val; |
1da177e4 LT |
2630 | break; |
2631 | ||
2632 | case TCP_LINGER2: | |
2633 | if (val < 0) | |
2634 | tp->linger2 = -1; | |
2635 | else if (val > sysctl_tcp_fin_timeout / HZ) | |
2636 | tp->linger2 = 0; | |
2637 | else | |
2638 | tp->linger2 = val * HZ; | |
2639 | break; | |
2640 | ||
2641 | case TCP_DEFER_ACCEPT: | |
b103cf34 JA |
2642 | /* Translate value in seconds to number of retransmits */ |
2643 | icsk->icsk_accept_queue.rskq_defer_accept = | |
2644 | secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ, | |
2645 | TCP_RTO_MAX / HZ); | |
1da177e4 LT |
2646 | break; |
2647 | ||
2648 | case TCP_WINDOW_CLAMP: | |
2649 | if (!val) { | |
2650 | if (sk->sk_state != TCP_CLOSE) { | |
2651 | err = -EINVAL; | |
2652 | break; | |
2653 | } | |
2654 | tp->window_clamp = 0; | |
2655 | } else | |
2656 | tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ? | |
2657 | SOCK_MIN_RCVBUF / 2 : val; | |
2658 | break; | |
2659 | ||
2660 | case TCP_QUICKACK: | |
2661 | if (!val) { | |
463c84b9 | 2662 | icsk->icsk_ack.pingpong = 1; |
1da177e4 | 2663 | } else { |
463c84b9 | 2664 | icsk->icsk_ack.pingpong = 0; |
1da177e4 LT |
2665 | if ((1 << sk->sk_state) & |
2666 | (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) && | |
463c84b9 ACM |
2667 | inet_csk_ack_scheduled(sk)) { |
2668 | icsk->icsk_ack.pending |= ICSK_ACK_PUSHED; | |
0e4b4992 | 2669 | tcp_cleanup_rbuf(sk, 1); |
1da177e4 | 2670 | if (!(val & 1)) |
463c84b9 | 2671 | icsk->icsk_ack.pingpong = 1; |
1da177e4 LT |
2672 | } |
2673 | } | |
2674 | break; | |
2675 | ||
cfb6eeb4 YH |
2676 | #ifdef CONFIG_TCP_MD5SIG |
2677 | case TCP_MD5SIG: | |
2678 | /* Read the IP->Key mappings from userspace */ | |
2679 | err = tp->af_specific->md5_parse(sk, optval, optlen); | |
2680 | break; | |
2681 | #endif | |
dca43c75 JC |
2682 | case TCP_USER_TIMEOUT: |
2683 | /* Cap the max timeout in ms TCP will retry/retrans | |
2684 | * before giving up and aborting (ETIMEDOUT) a connection. | |
2685 | */ | |
42493570 HL |
2686 | if (val < 0) |
2687 | err = -EINVAL; | |
2688 | else | |
2689 | icsk->icsk_user_timeout = msecs_to_jiffies(val); | |
dca43c75 | 2690 | break; |
1da177e4 LT |
2691 | default: |
2692 | err = -ENOPROTOOPT; | |
2693 | break; | |
3ff50b79 SH |
2694 | } |
2695 | ||
1da177e4 LT |
2696 | release_sock(sk); |
2697 | return err; | |
2698 | } | |
2699 | ||
3fdadf7d | 2700 | int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval, |
b7058842 | 2701 | unsigned int optlen) |
3fdadf7d | 2702 | { |
cf533ea5 | 2703 | const struct inet_connection_sock *icsk = inet_csk(sk); |
3fdadf7d DM |
2704 | |
2705 | if (level != SOL_TCP) | |
2706 | return icsk->icsk_af_ops->setsockopt(sk, level, optname, | |
2707 | optval, optlen); | |
2708 | return do_tcp_setsockopt(sk, level, optname, optval, optlen); | |
2709 | } | |
4bc2f18b | 2710 | EXPORT_SYMBOL(tcp_setsockopt); |
3fdadf7d DM |
2711 | |
2712 | #ifdef CONFIG_COMPAT | |
543d9cfe | 2713 | int compat_tcp_setsockopt(struct sock *sk, int level, int optname, |
b7058842 | 2714 | char __user *optval, unsigned int optlen) |
3fdadf7d | 2715 | { |
dec73ff0 ACM |
2716 | if (level != SOL_TCP) |
2717 | return inet_csk_compat_setsockopt(sk, level, optname, | |
2718 | optval, optlen); | |
3fdadf7d DM |
2719 | return do_tcp_setsockopt(sk, level, optname, optval, optlen); |
2720 | } | |
543d9cfe | 2721 | EXPORT_SYMBOL(compat_tcp_setsockopt); |
3fdadf7d DM |
2722 | #endif |
2723 | ||
1da177e4 | 2724 | /* Return information about state of tcp endpoint in API format. */ |
cf533ea5 | 2725 | void tcp_get_info(const struct sock *sk, struct tcp_info *info) |
1da177e4 | 2726 | { |
cf533ea5 | 2727 | const struct tcp_sock *tp = tcp_sk(sk); |
463c84b9 | 2728 | const struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
2729 | u32 now = tcp_time_stamp; |
2730 | ||
2731 | memset(info, 0, sizeof(*info)); | |
2732 | ||
2733 | info->tcpi_state = sk->sk_state; | |
6687e988 | 2734 | info->tcpi_ca_state = icsk->icsk_ca_state; |
463c84b9 | 2735 | info->tcpi_retransmits = icsk->icsk_retransmits; |
6687e988 | 2736 | info->tcpi_probes = icsk->icsk_probes_out; |
463c84b9 | 2737 | info->tcpi_backoff = icsk->icsk_backoff; |
1da177e4 LT |
2738 | |
2739 | if (tp->rx_opt.tstamp_ok) | |
2740 | info->tcpi_options |= TCPI_OPT_TIMESTAMPS; | |
e60402d0 | 2741 | if (tcp_is_sack(tp)) |
1da177e4 LT |
2742 | info->tcpi_options |= TCPI_OPT_SACK; |
2743 | if (tp->rx_opt.wscale_ok) { | |
2744 | info->tcpi_options |= TCPI_OPT_WSCALE; | |
2745 | info->tcpi_snd_wscale = tp->rx_opt.snd_wscale; | |
2746 | info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale; | |
e905a9ed | 2747 | } |
1da177e4 | 2748 | |
b5c5693b | 2749 | if (tp->ecn_flags & TCP_ECN_OK) |
1da177e4 | 2750 | info->tcpi_options |= TCPI_OPT_ECN; |
b5c5693b ED |
2751 | if (tp->ecn_flags & TCP_ECN_SEEN) |
2752 | info->tcpi_options |= TCPI_OPT_ECN_SEEN; | |
1da177e4 | 2753 | |
463c84b9 ACM |
2754 | info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto); |
2755 | info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato); | |
c1b4a7e6 | 2756 | info->tcpi_snd_mss = tp->mss_cache; |
463c84b9 | 2757 | info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss; |
1da177e4 | 2758 | |
5ee3afba RJ |
2759 | if (sk->sk_state == TCP_LISTEN) { |
2760 | info->tcpi_unacked = sk->sk_ack_backlog; | |
2761 | info->tcpi_sacked = sk->sk_max_ack_backlog; | |
2762 | } else { | |
2763 | info->tcpi_unacked = tp->packets_out; | |
2764 | info->tcpi_sacked = tp->sacked_out; | |
2765 | } | |
1da177e4 LT |
2766 | info->tcpi_lost = tp->lost_out; |
2767 | info->tcpi_retrans = tp->retrans_out; | |
2768 | info->tcpi_fackets = tp->fackets_out; | |
2769 | ||
2770 | info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime); | |
463c84b9 | 2771 | info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime); |
1da177e4 LT |
2772 | info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp); |
2773 | ||
d83d8461 | 2774 | info->tcpi_pmtu = icsk->icsk_pmtu_cookie; |
1da177e4 LT |
2775 | info->tcpi_rcv_ssthresh = tp->rcv_ssthresh; |
2776 | info->tcpi_rtt = jiffies_to_usecs(tp->srtt)>>3; | |
2777 | info->tcpi_rttvar = jiffies_to_usecs(tp->mdev)>>2; | |
2778 | info->tcpi_snd_ssthresh = tp->snd_ssthresh; | |
2779 | info->tcpi_snd_cwnd = tp->snd_cwnd; | |
2780 | info->tcpi_advmss = tp->advmss; | |
2781 | info->tcpi_reordering = tp->reordering; | |
2782 | ||
2783 | info->tcpi_rcv_rtt = jiffies_to_usecs(tp->rcv_rtt_est.rtt)>>3; | |
2784 | info->tcpi_rcv_space = tp->rcvq_space.space; | |
2785 | ||
2786 | info->tcpi_total_retrans = tp->total_retrans; | |
2787 | } | |
1da177e4 LT |
2788 | EXPORT_SYMBOL_GPL(tcp_get_info); |
2789 | ||
3fdadf7d DM |
2790 | static int do_tcp_getsockopt(struct sock *sk, int level, |
2791 | int optname, char __user *optval, int __user *optlen) | |
1da177e4 | 2792 | { |
295f7324 | 2793 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
2794 | struct tcp_sock *tp = tcp_sk(sk); |
2795 | int val, len; | |
2796 | ||
1da177e4 LT |
2797 | if (get_user(len, optlen)) |
2798 | return -EFAULT; | |
2799 | ||
2800 | len = min_t(unsigned int, len, sizeof(int)); | |
2801 | ||
2802 | if (len < 0) | |
2803 | return -EINVAL; | |
2804 | ||
2805 | switch (optname) { | |
2806 | case TCP_MAXSEG: | |
c1b4a7e6 | 2807 | val = tp->mss_cache; |
1da177e4 LT |
2808 | if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) |
2809 | val = tp->rx_opt.user_mss; | |
5e6a3ce6 PE |
2810 | if (tp->repair) |
2811 | val = tp->rx_opt.mss_clamp; | |
1da177e4 LT |
2812 | break; |
2813 | case TCP_NODELAY: | |
2814 | val = !!(tp->nonagle&TCP_NAGLE_OFF); | |
2815 | break; | |
2816 | case TCP_CORK: | |
2817 | val = !!(tp->nonagle&TCP_NAGLE_CORK); | |
2818 | break; | |
2819 | case TCP_KEEPIDLE: | |
df19a626 | 2820 | val = keepalive_time_when(tp) / HZ; |
1da177e4 LT |
2821 | break; |
2822 | case TCP_KEEPINTVL: | |
df19a626 | 2823 | val = keepalive_intvl_when(tp) / HZ; |
1da177e4 LT |
2824 | break; |
2825 | case TCP_KEEPCNT: | |
df19a626 | 2826 | val = keepalive_probes(tp); |
1da177e4 LT |
2827 | break; |
2828 | case TCP_SYNCNT: | |
295f7324 | 2829 | val = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries; |
1da177e4 LT |
2830 | break; |
2831 | case TCP_LINGER2: | |
2832 | val = tp->linger2; | |
2833 | if (val >= 0) | |
2834 | val = (val ? : sysctl_tcp_fin_timeout) / HZ; | |
2835 | break; | |
2836 | case TCP_DEFER_ACCEPT: | |
b103cf34 JA |
2837 | val = retrans_to_secs(icsk->icsk_accept_queue.rskq_defer_accept, |
2838 | TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ); | |
1da177e4 LT |
2839 | break; |
2840 | case TCP_WINDOW_CLAMP: | |
2841 | val = tp->window_clamp; | |
2842 | break; | |
2843 | case TCP_INFO: { | |
2844 | struct tcp_info info; | |
2845 | ||
2846 | if (get_user(len, optlen)) | |
2847 | return -EFAULT; | |
2848 | ||
2849 | tcp_get_info(sk, &info); | |
2850 | ||
2851 | len = min_t(unsigned int, len, sizeof(info)); | |
2852 | if (put_user(len, optlen)) | |
2853 | return -EFAULT; | |
2854 | if (copy_to_user(optval, &info, len)) | |
2855 | return -EFAULT; | |
2856 | return 0; | |
2857 | } | |
2858 | case TCP_QUICKACK: | |
295f7324 | 2859 | val = !icsk->icsk_ack.pingpong; |
1da177e4 | 2860 | break; |
5f8ef48d SH |
2861 | |
2862 | case TCP_CONGESTION: | |
2863 | if (get_user(len, optlen)) | |
2864 | return -EFAULT; | |
2865 | len = min_t(unsigned int, len, TCP_CA_NAME_MAX); | |
2866 | if (put_user(len, optlen)) | |
2867 | return -EFAULT; | |
6687e988 | 2868 | if (copy_to_user(optval, icsk->icsk_ca_ops->name, len)) |
5f8ef48d SH |
2869 | return -EFAULT; |
2870 | return 0; | |
e56fb50f WAS |
2871 | |
2872 | case TCP_COOKIE_TRANSACTIONS: { | |
2873 | struct tcp_cookie_transactions ctd; | |
2874 | struct tcp_cookie_values *cvp = tp->cookie_values; | |
2875 | ||
2876 | if (get_user(len, optlen)) | |
2877 | return -EFAULT; | |
2878 | if (len < sizeof(ctd)) | |
2879 | return -EINVAL; | |
2880 | ||
2881 | memset(&ctd, 0, sizeof(ctd)); | |
2882 | ctd.tcpct_flags = (tp->rx_opt.cookie_in_always ? | |
2883 | TCP_COOKIE_IN_ALWAYS : 0) | |
2884 | | (tp->rx_opt.cookie_out_never ? | |
2885 | TCP_COOKIE_OUT_NEVER : 0); | |
2886 | ||
2887 | if (cvp != NULL) { | |
2888 | ctd.tcpct_flags |= (cvp->s_data_in ? | |
2889 | TCP_S_DATA_IN : 0) | |
2890 | | (cvp->s_data_out ? | |
2891 | TCP_S_DATA_OUT : 0); | |
2892 | ||
2893 | ctd.tcpct_cookie_desired = cvp->cookie_desired; | |
2894 | ctd.tcpct_s_data_desired = cvp->s_data_desired; | |
2895 | ||
e56fb50f WAS |
2896 | memcpy(&ctd.tcpct_value[0], &cvp->cookie_pair[0], |
2897 | cvp->cookie_pair_size); | |
2898 | ctd.tcpct_used = cvp->cookie_pair_size; | |
2899 | } | |
2900 | ||
2901 | if (put_user(sizeof(ctd), optlen)) | |
2902 | return -EFAULT; | |
2903 | if (copy_to_user(optval, &ctd, sizeof(ctd))) | |
2904 | return -EFAULT; | |
2905 | return 0; | |
2906 | } | |
3c0fef0b JH |
2907 | case TCP_THIN_LINEAR_TIMEOUTS: |
2908 | val = tp->thin_lto; | |
2909 | break; | |
2910 | case TCP_THIN_DUPACK: | |
2911 | val = tp->thin_dupack; | |
2912 | break; | |
dca43c75 | 2913 | |
ee995283 PE |
2914 | case TCP_REPAIR: |
2915 | val = tp->repair; | |
2916 | break; | |
2917 | ||
2918 | case TCP_REPAIR_QUEUE: | |
2919 | if (tp->repair) | |
2920 | val = tp->repair_queue; | |
2921 | else | |
2922 | return -EINVAL; | |
2923 | break; | |
2924 | ||
2925 | case TCP_QUEUE_SEQ: | |
2926 | if (tp->repair_queue == TCP_SEND_QUEUE) | |
2927 | val = tp->write_seq; | |
2928 | else if (tp->repair_queue == TCP_RECV_QUEUE) | |
2929 | val = tp->rcv_nxt; | |
2930 | else | |
2931 | return -EINVAL; | |
2932 | break; | |
2933 | ||
dca43c75 JC |
2934 | case TCP_USER_TIMEOUT: |
2935 | val = jiffies_to_msecs(icsk->icsk_user_timeout); | |
2936 | break; | |
1da177e4 LT |
2937 | default: |
2938 | return -ENOPROTOOPT; | |
3ff50b79 | 2939 | } |
1da177e4 LT |
2940 | |
2941 | if (put_user(len, optlen)) | |
2942 | return -EFAULT; | |
2943 | if (copy_to_user(optval, &val, len)) | |
2944 | return -EFAULT; | |
2945 | return 0; | |
2946 | } | |
2947 | ||
3fdadf7d DM |
2948 | int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval, |
2949 | int __user *optlen) | |
2950 | { | |
2951 | struct inet_connection_sock *icsk = inet_csk(sk); | |
2952 | ||
2953 | if (level != SOL_TCP) | |
2954 | return icsk->icsk_af_ops->getsockopt(sk, level, optname, | |
2955 | optval, optlen); | |
2956 | return do_tcp_getsockopt(sk, level, optname, optval, optlen); | |
2957 | } | |
4bc2f18b | 2958 | EXPORT_SYMBOL(tcp_getsockopt); |
3fdadf7d DM |
2959 | |
2960 | #ifdef CONFIG_COMPAT | |
543d9cfe ACM |
2961 | int compat_tcp_getsockopt(struct sock *sk, int level, int optname, |
2962 | char __user *optval, int __user *optlen) | |
3fdadf7d | 2963 | { |
dec73ff0 ACM |
2964 | if (level != SOL_TCP) |
2965 | return inet_csk_compat_getsockopt(sk, level, optname, | |
2966 | optval, optlen); | |
3fdadf7d DM |
2967 | return do_tcp_getsockopt(sk, level, optname, optval, optlen); |
2968 | } | |
543d9cfe | 2969 | EXPORT_SYMBOL(compat_tcp_getsockopt); |
3fdadf7d | 2970 | #endif |
1da177e4 | 2971 | |
c8f44aff MM |
2972 | struct sk_buff *tcp_tso_segment(struct sk_buff *skb, |
2973 | netdev_features_t features) | |
f4c50d99 HX |
2974 | { |
2975 | struct sk_buff *segs = ERR_PTR(-EINVAL); | |
2976 | struct tcphdr *th; | |
95c96174 | 2977 | unsigned int thlen; |
f4c50d99 | 2978 | unsigned int seq; |
d3bc23e7 | 2979 | __be32 delta; |
f4c50d99 | 2980 | unsigned int oldlen; |
4e704ee3 | 2981 | unsigned int mss; |
f4c50d99 HX |
2982 | |
2983 | if (!pskb_may_pull(skb, sizeof(*th))) | |
2984 | goto out; | |
2985 | ||
aa8223c7 | 2986 | th = tcp_hdr(skb); |
f4c50d99 HX |
2987 | thlen = th->doff * 4; |
2988 | if (thlen < sizeof(*th)) | |
2989 | goto out; | |
2990 | ||
2991 | if (!pskb_may_pull(skb, thlen)) | |
2992 | goto out; | |
2993 | ||
0718bcc0 | 2994 | oldlen = (u16)~skb->len; |
f4c50d99 HX |
2995 | __skb_pull(skb, thlen); |
2996 | ||
4e704ee3 HX |
2997 | mss = skb_shinfo(skb)->gso_size; |
2998 | if (unlikely(skb->len <= mss)) | |
2999 | goto out; | |
3000 | ||
3820c3f3 HX |
3001 | if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) { |
3002 | /* Packet is from an untrusted source, reset gso_segs. */ | |
bbcf467d | 3003 | int type = skb_shinfo(skb)->gso_type; |
bbcf467d HX |
3004 | |
3005 | if (unlikely(type & | |
3006 | ~(SKB_GSO_TCPV4 | | |
3007 | SKB_GSO_DODGY | | |
3008 | SKB_GSO_TCP_ECN | | |
3009 | SKB_GSO_TCPV6 | | |
3010 | 0) || | |
3011 | !(type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))) | |
3012 | goto out; | |
3820c3f3 | 3013 | |
172589cc | 3014 | skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss); |
3820c3f3 HX |
3015 | |
3016 | segs = NULL; | |
3017 | goto out; | |
3018 | } | |
3019 | ||
576a30eb | 3020 | segs = skb_segment(skb, features); |
f4c50d99 HX |
3021 | if (IS_ERR(segs)) |
3022 | goto out; | |
3023 | ||
4e704ee3 | 3024 | delta = htonl(oldlen + (thlen + mss)); |
f4c50d99 HX |
3025 | |
3026 | skb = segs; | |
aa8223c7 | 3027 | th = tcp_hdr(skb); |
f4c50d99 HX |
3028 | seq = ntohl(th->seq); |
3029 | ||
3030 | do { | |
3031 | th->fin = th->psh = 0; | |
3032 | ||
d3bc23e7 AV |
3033 | th->check = ~csum_fold((__force __wsum)((__force u32)th->check + |
3034 | (__force u32)delta)); | |
84fa7933 | 3035 | if (skb->ip_summed != CHECKSUM_PARTIAL) |
9c70220b ACM |
3036 | th->check = |
3037 | csum_fold(csum_partial(skb_transport_header(skb), | |
3038 | thlen, skb->csum)); | |
f4c50d99 | 3039 | |
4e704ee3 | 3040 | seq += mss; |
f4c50d99 | 3041 | skb = skb->next; |
aa8223c7 | 3042 | th = tcp_hdr(skb); |
f4c50d99 HX |
3043 | |
3044 | th->seq = htonl(seq); | |
3045 | th->cwr = 0; | |
3046 | } while (skb->next); | |
3047 | ||
27a884dc | 3048 | delta = htonl(oldlen + (skb->tail - skb->transport_header) + |
9c70220b | 3049 | skb->data_len); |
d3bc23e7 AV |
3050 | th->check = ~csum_fold((__force __wsum)((__force u32)th->check + |
3051 | (__force u32)delta)); | |
84fa7933 | 3052 | if (skb->ip_summed != CHECKSUM_PARTIAL) |
9c70220b ACM |
3053 | th->check = csum_fold(csum_partial(skb_transport_header(skb), |
3054 | thlen, skb->csum)); | |
f4c50d99 HX |
3055 | |
3056 | out: | |
3057 | return segs; | |
3058 | } | |
adcfc7d0 | 3059 | EXPORT_SYMBOL(tcp_tso_segment); |
f4c50d99 | 3060 | |
bf296b12 HX |
3061 | struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb) |
3062 | { | |
3063 | struct sk_buff **pp = NULL; | |
3064 | struct sk_buff *p; | |
3065 | struct tcphdr *th; | |
3066 | struct tcphdr *th2; | |
a0a69a01 | 3067 | unsigned int len; |
bf296b12 | 3068 | unsigned int thlen; |
0eae88f3 | 3069 | __be32 flags; |
bf296b12 | 3070 | unsigned int mss = 1; |
a5b1cf28 HX |
3071 | unsigned int hlen; |
3072 | unsigned int off; | |
bf296b12 | 3073 | int flush = 1; |
aa6320d3 | 3074 | int i; |
bf296b12 | 3075 | |
a5b1cf28 HX |
3076 | off = skb_gro_offset(skb); |
3077 | hlen = off + sizeof(*th); | |
3078 | th = skb_gro_header_fast(skb, off); | |
3079 | if (skb_gro_header_hard(skb, hlen)) { | |
3080 | th = skb_gro_header_slow(skb, hlen, off); | |
3081 | if (unlikely(!th)) | |
3082 | goto out; | |
3083 | } | |
bf296b12 | 3084 | |
bf296b12 HX |
3085 | thlen = th->doff * 4; |
3086 | if (thlen < sizeof(*th)) | |
3087 | goto out; | |
3088 | ||
a5b1cf28 HX |
3089 | hlen = off + thlen; |
3090 | if (skb_gro_header_hard(skb, hlen)) { | |
3091 | th = skb_gro_header_slow(skb, hlen, off); | |
3092 | if (unlikely(!th)) | |
3093 | goto out; | |
3094 | } | |
bf296b12 | 3095 | |
86911732 | 3096 | skb_gro_pull(skb, thlen); |
bf296b12 | 3097 | |
a0a69a01 | 3098 | len = skb_gro_len(skb); |
bf296b12 HX |
3099 | flags = tcp_flag_word(th); |
3100 | ||
3101 | for (; (p = *head); head = &p->next) { | |
3102 | if (!NAPI_GRO_CB(p)->same_flow) | |
3103 | continue; | |
3104 | ||
3105 | th2 = tcp_hdr(p); | |
3106 | ||
745898ea | 3107 | if (*(u32 *)&th->source ^ *(u32 *)&th2->source) { |
bf296b12 HX |
3108 | NAPI_GRO_CB(p)->same_flow = 0; |
3109 | continue; | |
3110 | } | |
3111 | ||
3112 | goto found; | |
3113 | } | |
3114 | ||
3115 | goto out_check_final; | |
3116 | ||
3117 | found: | |
3118 | flush = NAPI_GRO_CB(p)->flush; | |
0eae88f3 ED |
3119 | flush |= (__force int)(flags & TCP_FLAG_CWR); |
3120 | flush |= (__force int)((flags ^ tcp_flag_word(th2)) & | |
3121 | ~(TCP_FLAG_CWR | TCP_FLAG_FIN | TCP_FLAG_PSH)); | |
3122 | flush |= (__force int)(th->ack_seq ^ th2->ack_seq); | |
a2a804cd | 3123 | for (i = sizeof(*th); i < thlen; i += 4) |
aa6320d3 HX |
3124 | flush |= *(u32 *)((u8 *)th + i) ^ |
3125 | *(u32 *)((u8 *)th2 + i); | |
bf296b12 | 3126 | |
b530256d | 3127 | mss = skb_shinfo(p)->gso_size; |
bf296b12 | 3128 | |
30a3ae30 | 3129 | flush |= (len - 1) >= mss; |
aa6320d3 | 3130 | flush |= (ntohl(th2->seq) + skb_gro_len(p)) ^ ntohl(th->seq); |
bf296b12 HX |
3131 | |
3132 | if (flush || skb_gro_receive(head, skb)) { | |
3133 | mss = 1; | |
3134 | goto out_check_final; | |
3135 | } | |
3136 | ||
3137 | p = *head; | |
3138 | th2 = tcp_hdr(p); | |
3139 | tcp_flag_word(th2) |= flags & (TCP_FLAG_FIN | TCP_FLAG_PSH); | |
3140 | ||
3141 | out_check_final: | |
a0a69a01 | 3142 | flush = len < mss; |
0eae88f3 ED |
3143 | flush |= (__force int)(flags & (TCP_FLAG_URG | TCP_FLAG_PSH | |
3144 | TCP_FLAG_RST | TCP_FLAG_SYN | | |
3145 | TCP_FLAG_FIN)); | |
bf296b12 HX |
3146 | |
3147 | if (p && (!NAPI_GRO_CB(skb)->same_flow || flush)) | |
3148 | pp = head; | |
3149 | ||
3150 | out: | |
3151 | NAPI_GRO_CB(skb)->flush |= flush; | |
3152 | ||
3153 | return pp; | |
3154 | } | |
684f2176 | 3155 | EXPORT_SYMBOL(tcp_gro_receive); |
bf296b12 HX |
3156 | |
3157 | int tcp_gro_complete(struct sk_buff *skb) | |
3158 | { | |
3159 | struct tcphdr *th = tcp_hdr(skb); | |
3160 | ||
3161 | skb->csum_start = skb_transport_header(skb) - skb->head; | |
3162 | skb->csum_offset = offsetof(struct tcphdr, check); | |
3163 | skb->ip_summed = CHECKSUM_PARTIAL; | |
3164 | ||
bf296b12 HX |
3165 | skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count; |
3166 | ||
3167 | if (th->cwr) | |
3168 | skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; | |
3169 | ||
3170 | return 0; | |
3171 | } | |
684f2176 | 3172 | EXPORT_SYMBOL(tcp_gro_complete); |
bf296b12 | 3173 | |
cfb6eeb4 YH |
3174 | #ifdef CONFIG_TCP_MD5SIG |
3175 | static unsigned long tcp_md5sig_users; | |
765cf997 | 3176 | static struct tcp_md5sig_pool __percpu *tcp_md5sig_pool; |
cfb6eeb4 YH |
3177 | static DEFINE_SPINLOCK(tcp_md5sig_pool_lock); |
3178 | ||
765cf997 | 3179 | static void __tcp_free_md5sig_pool(struct tcp_md5sig_pool __percpu *pool) |
cfb6eeb4 YH |
3180 | { |
3181 | int cpu; | |
765cf997 | 3182 | |
cfb6eeb4 | 3183 | for_each_possible_cpu(cpu) { |
765cf997 ED |
3184 | struct tcp_md5sig_pool *p = per_cpu_ptr(pool, cpu); |
3185 | ||
3186 | if (p->md5_desc.tfm) | |
3187 | crypto_free_hash(p->md5_desc.tfm); | |
cfb6eeb4 YH |
3188 | } |
3189 | free_percpu(pool); | |
3190 | } | |
3191 | ||
3192 | void tcp_free_md5sig_pool(void) | |
3193 | { | |
765cf997 | 3194 | struct tcp_md5sig_pool __percpu *pool = NULL; |
cfb6eeb4 | 3195 | |
2c4f6219 | 3196 | spin_lock_bh(&tcp_md5sig_pool_lock); |
cfb6eeb4 YH |
3197 | if (--tcp_md5sig_users == 0) { |
3198 | pool = tcp_md5sig_pool; | |
3199 | tcp_md5sig_pool = NULL; | |
3200 | } | |
2c4f6219 | 3201 | spin_unlock_bh(&tcp_md5sig_pool_lock); |
cfb6eeb4 YH |
3202 | if (pool) |
3203 | __tcp_free_md5sig_pool(pool); | |
3204 | } | |
cfb6eeb4 YH |
3205 | EXPORT_SYMBOL(tcp_free_md5sig_pool); |
3206 | ||
765cf997 | 3207 | static struct tcp_md5sig_pool __percpu * |
7d720c3e | 3208 | __tcp_alloc_md5sig_pool(struct sock *sk) |
cfb6eeb4 YH |
3209 | { |
3210 | int cpu; | |
765cf997 | 3211 | struct tcp_md5sig_pool __percpu *pool; |
cfb6eeb4 | 3212 | |
765cf997 | 3213 | pool = alloc_percpu(struct tcp_md5sig_pool); |
cfb6eeb4 YH |
3214 | if (!pool) |
3215 | return NULL; | |
3216 | ||
3217 | for_each_possible_cpu(cpu) { | |
cfb6eeb4 YH |
3218 | struct crypto_hash *hash; |
3219 | ||
cfb6eeb4 YH |
3220 | hash = crypto_alloc_hash("md5", 0, CRYPTO_ALG_ASYNC); |
3221 | if (!hash || IS_ERR(hash)) | |
3222 | goto out_free; | |
3223 | ||
765cf997 | 3224 | per_cpu_ptr(pool, cpu)->md5_desc.tfm = hash; |
cfb6eeb4 YH |
3225 | } |
3226 | return pool; | |
3227 | out_free: | |
3228 | __tcp_free_md5sig_pool(pool); | |
3229 | return NULL; | |
3230 | } | |
3231 | ||
765cf997 | 3232 | struct tcp_md5sig_pool __percpu *tcp_alloc_md5sig_pool(struct sock *sk) |
cfb6eeb4 | 3233 | { |
765cf997 | 3234 | struct tcp_md5sig_pool __percpu *pool; |
a2a385d6 | 3235 | bool alloc = false; |
cfb6eeb4 YH |
3236 | |
3237 | retry: | |
2c4f6219 | 3238 | spin_lock_bh(&tcp_md5sig_pool_lock); |
cfb6eeb4 YH |
3239 | pool = tcp_md5sig_pool; |
3240 | if (tcp_md5sig_users++ == 0) { | |
a2a385d6 | 3241 | alloc = true; |
2c4f6219 | 3242 | spin_unlock_bh(&tcp_md5sig_pool_lock); |
cfb6eeb4 YH |
3243 | } else if (!pool) { |
3244 | tcp_md5sig_users--; | |
2c4f6219 | 3245 | spin_unlock_bh(&tcp_md5sig_pool_lock); |
cfb6eeb4 YH |
3246 | cpu_relax(); |
3247 | goto retry; | |
3248 | } else | |
2c4f6219 | 3249 | spin_unlock_bh(&tcp_md5sig_pool_lock); |
cfb6eeb4 YH |
3250 | |
3251 | if (alloc) { | |
3252 | /* we cannot hold spinlock here because this may sleep. */ | |
765cf997 | 3253 | struct tcp_md5sig_pool __percpu *p; |
7d720c3e TH |
3254 | |
3255 | p = __tcp_alloc_md5sig_pool(sk); | |
2c4f6219 | 3256 | spin_lock_bh(&tcp_md5sig_pool_lock); |
cfb6eeb4 YH |
3257 | if (!p) { |
3258 | tcp_md5sig_users--; | |
2c4f6219 | 3259 | spin_unlock_bh(&tcp_md5sig_pool_lock); |
cfb6eeb4 YH |
3260 | return NULL; |
3261 | } | |
3262 | pool = tcp_md5sig_pool; | |
3263 | if (pool) { | |
3264 | /* oops, it has already been assigned. */ | |
2c4f6219 | 3265 | spin_unlock_bh(&tcp_md5sig_pool_lock); |
cfb6eeb4 YH |
3266 | __tcp_free_md5sig_pool(p); |
3267 | } else { | |
3268 | tcp_md5sig_pool = pool = p; | |
2c4f6219 | 3269 | spin_unlock_bh(&tcp_md5sig_pool_lock); |
cfb6eeb4 YH |
3270 | } |
3271 | } | |
3272 | return pool; | |
3273 | } | |
cfb6eeb4 YH |
3274 | EXPORT_SYMBOL(tcp_alloc_md5sig_pool); |
3275 | ||
35790c04 ED |
3276 | |
3277 | /** | |
3278 | * tcp_get_md5sig_pool - get md5sig_pool for this user | |
3279 | * | |
3280 | * We use percpu structure, so if we succeed, we exit with preemption | |
3281 | * and BH disabled, to make sure another thread or softirq handling | |
3282 | * wont try to get same context. | |
3283 | */ | |
3284 | struct tcp_md5sig_pool *tcp_get_md5sig_pool(void) | |
cfb6eeb4 | 3285 | { |
765cf997 | 3286 | struct tcp_md5sig_pool __percpu *p; |
35790c04 ED |
3287 | |
3288 | local_bh_disable(); | |
3289 | ||
3290 | spin_lock(&tcp_md5sig_pool_lock); | |
cfb6eeb4 YH |
3291 | p = tcp_md5sig_pool; |
3292 | if (p) | |
3293 | tcp_md5sig_users++; | |
35790c04 ED |
3294 | spin_unlock(&tcp_md5sig_pool_lock); |
3295 | ||
3296 | if (p) | |
765cf997 | 3297 | return this_cpu_ptr(p); |
cfb6eeb4 | 3298 | |
35790c04 ED |
3299 | local_bh_enable(); |
3300 | return NULL; | |
3301 | } | |
3302 | EXPORT_SYMBOL(tcp_get_md5sig_pool); | |
cfb6eeb4 | 3303 | |
35790c04 | 3304 | void tcp_put_md5sig_pool(void) |
6931ba7c | 3305 | { |
35790c04 | 3306 | local_bh_enable(); |
6931ba7c | 3307 | tcp_free_md5sig_pool(); |
cfb6eeb4 | 3308 | } |
35790c04 | 3309 | EXPORT_SYMBOL(tcp_put_md5sig_pool); |
49a72dfb AL |
3310 | |
3311 | int tcp_md5_hash_header(struct tcp_md5sig_pool *hp, | |
ca35a0ef | 3312 | const struct tcphdr *th) |
49a72dfb AL |
3313 | { |
3314 | struct scatterlist sg; | |
ca35a0ef | 3315 | struct tcphdr hdr; |
49a72dfb AL |
3316 | int err; |
3317 | ||
ca35a0ef ED |
3318 | /* We are not allowed to change tcphdr, make a local copy */ |
3319 | memcpy(&hdr, th, sizeof(hdr)); | |
3320 | hdr.check = 0; | |
3321 | ||
49a72dfb | 3322 | /* options aren't included in the hash */ |
ca35a0ef ED |
3323 | sg_init_one(&sg, &hdr, sizeof(hdr)); |
3324 | err = crypto_hash_update(&hp->md5_desc, &sg, sizeof(hdr)); | |
49a72dfb AL |
3325 | return err; |
3326 | } | |
49a72dfb AL |
3327 | EXPORT_SYMBOL(tcp_md5_hash_header); |
3328 | ||
3329 | int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp, | |
cf533ea5 | 3330 | const struct sk_buff *skb, unsigned int header_len) |
49a72dfb AL |
3331 | { |
3332 | struct scatterlist sg; | |
3333 | const struct tcphdr *tp = tcp_hdr(skb); | |
3334 | struct hash_desc *desc = &hp->md5_desc; | |
95c96174 ED |
3335 | unsigned int i; |
3336 | const unsigned int head_data_len = skb_headlen(skb) > header_len ? | |
3337 | skb_headlen(skb) - header_len : 0; | |
49a72dfb | 3338 | const struct skb_shared_info *shi = skb_shinfo(skb); |
d7fd1b57 | 3339 | struct sk_buff *frag_iter; |
49a72dfb AL |
3340 | |
3341 | sg_init_table(&sg, 1); | |
3342 | ||
3343 | sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len); | |
3344 | if (crypto_hash_update(desc, &sg, head_data_len)) | |
3345 | return 1; | |
3346 | ||
3347 | for (i = 0; i < shi->nr_frags; ++i) { | |
3348 | const struct skb_frag_struct *f = &shi->frags[i]; | |
aff65da0 | 3349 | struct page *page = skb_frag_page(f); |
9e903e08 ED |
3350 | sg_set_page(&sg, page, skb_frag_size(f), f->page_offset); |
3351 | if (crypto_hash_update(desc, &sg, skb_frag_size(f))) | |
49a72dfb AL |
3352 | return 1; |
3353 | } | |
3354 | ||
d7fd1b57 ED |
3355 | skb_walk_frags(skb, frag_iter) |
3356 | if (tcp_md5_hash_skb_data(hp, frag_iter, 0)) | |
3357 | return 1; | |
3358 | ||
49a72dfb AL |
3359 | return 0; |
3360 | } | |
49a72dfb AL |
3361 | EXPORT_SYMBOL(tcp_md5_hash_skb_data); |
3362 | ||
cf533ea5 | 3363 | int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, const struct tcp_md5sig_key *key) |
49a72dfb AL |
3364 | { |
3365 | struct scatterlist sg; | |
3366 | ||
3367 | sg_init_one(&sg, key->key, key->keylen); | |
3368 | return crypto_hash_update(&hp->md5_desc, &sg, key->keylen); | |
3369 | } | |
49a72dfb AL |
3370 | EXPORT_SYMBOL(tcp_md5_hash_key); |
3371 | ||
cfb6eeb4 YH |
3372 | #endif |
3373 | ||
ae86b9e3 | 3374 | /* Each Responder maintains up to two secret values concurrently for |
da5c78c8 WAS |
3375 | * efficient secret rollover. Each secret value has 4 states: |
3376 | * | |
3377 | * Generating. (tcp_secret_generating != tcp_secret_primary) | |
3378 | * Generates new Responder-Cookies, but not yet used for primary | |
3379 | * verification. This is a short-term state, typically lasting only | |
3380 | * one round trip time (RTT). | |
3381 | * | |
3382 | * Primary. (tcp_secret_generating == tcp_secret_primary) | |
3383 | * Used both for generation and primary verification. | |
3384 | * | |
3385 | * Retiring. (tcp_secret_retiring != tcp_secret_secondary) | |
3386 | * Used for verification, until the first failure that can be | |
3387 | * verified by the newer Generating secret. At that time, this | |
3388 | * cookie's state is changed to Secondary, and the Generating | |
3389 | * cookie's state is changed to Primary. This is a short-term state, | |
3390 | * typically lasting only one round trip time (RTT). | |
3391 | * | |
3392 | * Secondary. (tcp_secret_retiring == tcp_secret_secondary) | |
3393 | * Used for secondary verification, after primary verification | |
3394 | * failures. This state lasts no more than twice the Maximum Segment | |
3395 | * Lifetime (2MSL). Then, the secret is discarded. | |
3396 | */ | |
3397 | struct tcp_cookie_secret { | |
3398 | /* The secret is divided into two parts. The digest part is the | |
3399 | * equivalent of previously hashing a secret and saving the state, | |
3400 | * and serves as an initialization vector (IV). The message part | |
3401 | * serves as the trailing secret. | |
3402 | */ | |
3403 | u32 secrets[COOKIE_WORKSPACE_WORDS]; | |
3404 | unsigned long expires; | |
3405 | }; | |
3406 | ||
3407 | #define TCP_SECRET_1MSL (HZ * TCP_PAWS_MSL) | |
3408 | #define TCP_SECRET_2MSL (HZ * TCP_PAWS_MSL * 2) | |
3409 | #define TCP_SECRET_LIFE (HZ * 600) | |
3410 | ||
3411 | static struct tcp_cookie_secret tcp_secret_one; | |
3412 | static struct tcp_cookie_secret tcp_secret_two; | |
3413 | ||
3414 | /* Essentially a circular list, without dynamic allocation. */ | |
3415 | static struct tcp_cookie_secret *tcp_secret_generating; | |
3416 | static struct tcp_cookie_secret *tcp_secret_primary; | |
3417 | static struct tcp_cookie_secret *tcp_secret_retiring; | |
3418 | static struct tcp_cookie_secret *tcp_secret_secondary; | |
3419 | ||
3420 | static DEFINE_SPINLOCK(tcp_secret_locker); | |
3421 | ||
3422 | /* Select a pseudo-random word in the cookie workspace. | |
3423 | */ | |
3424 | static inline u32 tcp_cookie_work(const u32 *ws, const int n) | |
3425 | { | |
3426 | return ws[COOKIE_DIGEST_WORDS + ((COOKIE_MESSAGE_WORDS-1) & ws[n])]; | |
3427 | } | |
3428 | ||
3429 | /* Fill bakery[COOKIE_WORKSPACE_WORDS] with generator, updating as needed. | |
3430 | * Called in softirq context. | |
3431 | * Returns: 0 for success. | |
3432 | */ | |
3433 | int tcp_cookie_generator(u32 *bakery) | |
3434 | { | |
3435 | unsigned long jiffy = jiffies; | |
3436 | ||
3437 | if (unlikely(time_after_eq(jiffy, tcp_secret_generating->expires))) { | |
3438 | spin_lock_bh(&tcp_secret_locker); | |
3439 | if (!time_after_eq(jiffy, tcp_secret_generating->expires)) { | |
3440 | /* refreshed by another */ | |
3441 | memcpy(bakery, | |
3442 | &tcp_secret_generating->secrets[0], | |
3443 | COOKIE_WORKSPACE_WORDS); | |
3444 | } else { | |
3445 | /* still needs refreshing */ | |
3446 | get_random_bytes(bakery, COOKIE_WORKSPACE_WORDS); | |
3447 | ||
3448 | /* The first time, paranoia assumes that the | |
3449 | * randomization function isn't as strong. But, | |
3450 | * this secret initialization is delayed until | |
3451 | * the last possible moment (packet arrival). | |
3452 | * Although that time is observable, it is | |
3453 | * unpredictably variable. Mash in the most | |
3454 | * volatile clock bits available, and expire the | |
3455 | * secret extra quickly. | |
3456 | */ | |
3457 | if (unlikely(tcp_secret_primary->expires == | |
3458 | tcp_secret_secondary->expires)) { | |
3459 | struct timespec tv; | |
3460 | ||
3461 | getnstimeofday(&tv); | |
3462 | bakery[COOKIE_DIGEST_WORDS+0] ^= | |
3463 | (u32)tv.tv_nsec; | |
3464 | ||
3465 | tcp_secret_secondary->expires = jiffy | |
3466 | + TCP_SECRET_1MSL | |
3467 | + (0x0f & tcp_cookie_work(bakery, 0)); | |
3468 | } else { | |
3469 | tcp_secret_secondary->expires = jiffy | |
3470 | + TCP_SECRET_LIFE | |
3471 | + (0xff & tcp_cookie_work(bakery, 1)); | |
3472 | tcp_secret_primary->expires = jiffy | |
3473 | + TCP_SECRET_2MSL | |
3474 | + (0x1f & tcp_cookie_work(bakery, 2)); | |
3475 | } | |
3476 | memcpy(&tcp_secret_secondary->secrets[0], | |
3477 | bakery, COOKIE_WORKSPACE_WORDS); | |
3478 | ||
3479 | rcu_assign_pointer(tcp_secret_generating, | |
3480 | tcp_secret_secondary); | |
3481 | rcu_assign_pointer(tcp_secret_retiring, | |
3482 | tcp_secret_primary); | |
3483 | /* | |
3484 | * Neither call_rcu() nor synchronize_rcu() needed. | |
3485 | * Retiring data is not freed. It is replaced after | |
3486 | * further (locked) pointer updates, and a quiet time | |
3487 | * (minimum 1MSL, maximum LIFE - 2MSL). | |
3488 | */ | |
3489 | } | |
3490 | spin_unlock_bh(&tcp_secret_locker); | |
3491 | } else { | |
3492 | rcu_read_lock_bh(); | |
3493 | memcpy(bakery, | |
3494 | &rcu_dereference(tcp_secret_generating)->secrets[0], | |
3495 | COOKIE_WORKSPACE_WORDS); | |
3496 | rcu_read_unlock_bh(); | |
3497 | } | |
3498 | return 0; | |
3499 | } | |
3500 | EXPORT_SYMBOL(tcp_cookie_generator); | |
3501 | ||
4ac02bab AK |
3502 | void tcp_done(struct sock *sk) |
3503 | { | |
5a5f3a8d | 3504 | if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV) |
63231bdd | 3505 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS); |
4ac02bab AK |
3506 | |
3507 | tcp_set_state(sk, TCP_CLOSE); | |
3508 | tcp_clear_xmit_timers(sk); | |
3509 | ||
3510 | sk->sk_shutdown = SHUTDOWN_MASK; | |
3511 | ||
3512 | if (!sock_flag(sk, SOCK_DEAD)) | |
3513 | sk->sk_state_change(sk); | |
3514 | else | |
3515 | inet_csk_destroy_sock(sk); | |
3516 | } | |
3517 | EXPORT_SYMBOL_GPL(tcp_done); | |
3518 | ||
5f8ef48d | 3519 | extern struct tcp_congestion_ops tcp_reno; |
1da177e4 LT |
3520 | |
3521 | static __initdata unsigned long thash_entries; | |
3522 | static int __init set_thash_entries(char *str) | |
3523 | { | |
413c27d8 EZ |
3524 | ssize_t ret; |
3525 | ||
1da177e4 LT |
3526 | if (!str) |
3527 | return 0; | |
413c27d8 EZ |
3528 | |
3529 | ret = kstrtoul(str, 0, &thash_entries); | |
3530 | if (ret) | |
3531 | return 0; | |
3532 | ||
1da177e4 LT |
3533 | return 1; |
3534 | } | |
3535 | __setup("thash_entries=", set_thash_entries); | |
3536 | ||
4acb4190 GC |
3537 | void tcp_init_mem(struct net *net) |
3538 | { | |
4acb4190 GC |
3539 | unsigned long limit = nr_free_buffer_pages() / 8; |
3540 | limit = max(limit, 128UL); | |
3541 | net->ipv4.sysctl_tcp_mem[0] = limit / 4 * 3; | |
3542 | net->ipv4.sysctl_tcp_mem[1] = limit; | |
3543 | net->ipv4.sysctl_tcp_mem[2] = net->ipv4.sysctl_tcp_mem[0] * 2; | |
3544 | } | |
3545 | ||
1da177e4 LT |
3546 | void __init tcp_init(void) |
3547 | { | |
3548 | struct sk_buff *skb = NULL; | |
f03d78db | 3549 | unsigned long limit; |
b49960a0 | 3550 | int max_rshare, max_wshare, cnt; |
074b8517 | 3551 | unsigned int i; |
da5c78c8 | 3552 | unsigned long jiffy = jiffies; |
1da177e4 | 3553 | |
1f9e636e | 3554 | BUILD_BUG_ON(sizeof(struct tcp_skb_cb) > sizeof(skb->cb)); |
1da177e4 | 3555 | |
1748376b | 3556 | percpu_counter_init(&tcp_sockets_allocated, 0); |
dd24c001 | 3557 | percpu_counter_init(&tcp_orphan_count, 0); |
6e04e021 ACM |
3558 | tcp_hashinfo.bind_bucket_cachep = |
3559 | kmem_cache_create("tcp_bind_bucket", | |
3560 | sizeof(struct inet_bind_bucket), 0, | |
20c2df83 | 3561 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3562 | |
1da177e4 LT |
3563 | /* Size and allocate the main established and bind bucket |
3564 | * hash tables. | |
3565 | * | |
3566 | * The methodology is similar to that of the buffer cache. | |
3567 | */ | |
6e04e021 | 3568 | tcp_hashinfo.ehash = |
1da177e4 | 3569 | alloc_large_system_hash("TCP established", |
0f7ff927 | 3570 | sizeof(struct inet_ehash_bucket), |
1da177e4 | 3571 | thash_entries, |
4481374c | 3572 | (totalram_pages >= 128 * 1024) ? |
18955cfc | 3573 | 13 : 15, |
9e950efa | 3574 | 0, |
1da177e4 | 3575 | NULL, |
f373b53b | 3576 | &tcp_hashinfo.ehash_mask, |
31fe62b9 | 3577 | 0, |
0ccfe618 | 3578 | thash_entries ? 0 : 512 * 1024); |
f373b53b | 3579 | for (i = 0; i <= tcp_hashinfo.ehash_mask; i++) { |
3ab5aee7 ED |
3580 | INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i); |
3581 | INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].twchain, i); | |
1da177e4 | 3582 | } |
230140cf ED |
3583 | if (inet_ehash_locks_alloc(&tcp_hashinfo)) |
3584 | panic("TCP: failed to alloc ehash_locks"); | |
6e04e021 | 3585 | tcp_hashinfo.bhash = |
1da177e4 | 3586 | alloc_large_system_hash("TCP bind", |
0f7ff927 | 3587 | sizeof(struct inet_bind_hashbucket), |
f373b53b | 3588 | tcp_hashinfo.ehash_mask + 1, |
4481374c | 3589 | (totalram_pages >= 128 * 1024) ? |
18955cfc | 3590 | 13 : 15, |
9e950efa | 3591 | 0, |
6e04e021 | 3592 | &tcp_hashinfo.bhash_size, |
1da177e4 | 3593 | NULL, |
31fe62b9 | 3594 | 0, |
1da177e4 | 3595 | 64 * 1024); |
074b8517 | 3596 | tcp_hashinfo.bhash_size = 1U << tcp_hashinfo.bhash_size; |
6e04e021 ACM |
3597 | for (i = 0; i < tcp_hashinfo.bhash_size; i++) { |
3598 | spin_lock_init(&tcp_hashinfo.bhash[i].lock); | |
3599 | INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain); | |
1da177e4 LT |
3600 | } |
3601 | ||
c5ed63d6 ED |
3602 | |
3603 | cnt = tcp_hashinfo.ehash_mask + 1; | |
3604 | ||
3605 | tcp_death_row.sysctl_max_tw_buckets = cnt / 2; | |
3606 | sysctl_tcp_max_orphans = cnt / 2; | |
3607 | sysctl_max_syn_backlog = max(128, cnt / 256); | |
1da177e4 | 3608 | |
4acb4190 | 3609 | tcp_init_mem(&init_net); |
c43b874d | 3610 | /* Set per-socket limits to no more than 1/128 the pressure threshold */ |
5fb84b14 | 3611 | limit = nr_free_buffer_pages() << (PAGE_SHIFT - 7); |
b49960a0 ED |
3612 | max_wshare = min(4UL*1024*1024, limit); |
3613 | max_rshare = min(6UL*1024*1024, limit); | |
7b4f4b5e | 3614 | |
3ab224be | 3615 | sysctl_tcp_wmem[0] = SK_MEM_QUANTUM; |
7b4f4b5e | 3616 | sysctl_tcp_wmem[1] = 16*1024; |
b49960a0 | 3617 | sysctl_tcp_wmem[2] = max(64*1024, max_wshare); |
7b4f4b5e | 3618 | |
3ab224be | 3619 | sysctl_tcp_rmem[0] = SK_MEM_QUANTUM; |
7b4f4b5e | 3620 | sysctl_tcp_rmem[1] = 87380; |
b49960a0 | 3621 | sysctl_tcp_rmem[2] = max(87380, max_rshare); |
1da177e4 | 3622 | |
afd46503 | 3623 | pr_info("Hash tables configured (established %u bind %u)\n", |
058bd4d2 | 3624 | tcp_hashinfo.ehash_mask + 1, tcp_hashinfo.bhash_size); |
317a76f9 | 3625 | |
51c5d0c4 DM |
3626 | tcp_metrics_init(); |
3627 | ||
317a76f9 | 3628 | tcp_register_congestion_control(&tcp_reno); |
da5c78c8 WAS |
3629 | |
3630 | memset(&tcp_secret_one.secrets[0], 0, sizeof(tcp_secret_one.secrets)); | |
3631 | memset(&tcp_secret_two.secrets[0], 0, sizeof(tcp_secret_two.secrets)); | |
3632 | tcp_secret_one.expires = jiffy; /* past due */ | |
3633 | tcp_secret_two.expires = jiffy; /* past due */ | |
3634 | tcp_secret_generating = &tcp_secret_one; | |
3635 | tcp_secret_primary = &tcp_secret_one; | |
3636 | tcp_secret_retiring = &tcp_secret_two; | |
3637 | tcp_secret_secondary = &tcp_secret_two; | |
46d3ceab | 3638 | tcp_tasklet_init(); |
1da177e4 | 3639 | } |