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 | ||
ceaa1fef AV |
403 | tp->tsoffset = 0; |
404 | ||
900f65d3 NC |
405 | sk->sk_state = TCP_CLOSE; |
406 | ||
407 | sk->sk_write_space = sk_stream_write_space; | |
408 | sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); | |
409 | ||
410 | icsk->icsk_sync_mss = tcp_sync_mss; | |
411 | ||
900f65d3 NC |
412 | /* Presumed zeroed, in order of appearance: |
413 | * cookie_in_always, cookie_out_never, | |
414 | * s_data_constant, s_data_in, s_data_out | |
415 | */ | |
416 | sk->sk_sndbuf = sysctl_tcp_wmem[1]; | |
417 | sk->sk_rcvbuf = sysctl_tcp_rmem[1]; | |
418 | ||
419 | local_bh_disable(); | |
420 | sock_update_memcg(sk); | |
421 | sk_sockets_allocated_inc(sk); | |
422 | local_bh_enable(); | |
423 | } | |
424 | EXPORT_SYMBOL(tcp_init_sock); | |
425 | ||
1da177e4 LT |
426 | /* |
427 | * Wait for a TCP event. | |
428 | * | |
429 | * Note that we don't need to lock the socket, as the upper poll layers | |
430 | * take care of normal races (between the test and the event) and we don't | |
431 | * go look at any of the socket buffers directly. | |
432 | */ | |
433 | unsigned int tcp_poll(struct file *file, struct socket *sock, poll_table *wait) | |
434 | { | |
435 | unsigned int mask; | |
436 | struct sock *sk = sock->sk; | |
cf533ea5 | 437 | const struct tcp_sock *tp = tcp_sk(sk); |
1da177e4 | 438 | |
aa395145 | 439 | sock_poll_wait(file, sk_sleep(sk), wait); |
1da177e4 | 440 | if (sk->sk_state == TCP_LISTEN) |
dc40c7bc | 441 | return inet_csk_listen_poll(sk); |
1da177e4 LT |
442 | |
443 | /* Socket is not locked. We are protected from async events | |
70efce27 WN |
444 | * by poll logic and correct handling of state changes |
445 | * made by other threads is impossible in any case. | |
1da177e4 LT |
446 | */ |
447 | ||
448 | mask = 0; | |
1da177e4 LT |
449 | |
450 | /* | |
451 | * POLLHUP is certainly not done right. But poll() doesn't | |
452 | * have a notion of HUP in just one direction, and for a | |
453 | * socket the read side is more interesting. | |
454 | * | |
455 | * Some poll() documentation says that POLLHUP is incompatible | |
456 | * with the POLLOUT/POLLWR flags, so somebody should check this | |
457 | * all. But careful, it tends to be safer to return too many | |
458 | * bits than too few, and you can easily break real applications | |
459 | * if you don't tell them that something has hung up! | |
460 | * | |
461 | * Check-me. | |
462 | * | |
463 | * Check number 1. POLLHUP is _UNMASKABLE_ event (see UNIX98 and | |
464 | * our fs/select.c). It means that after we received EOF, | |
465 | * poll always returns immediately, making impossible poll() on write() | |
466 | * in state CLOSE_WAIT. One solution is evident --- to set POLLHUP | |
467 | * if and only if shutdown has been made in both directions. | |
468 | * Actually, it is interesting to look how Solaris and DUX | |
70efce27 | 469 | * solve this dilemma. I would prefer, if POLLHUP were maskable, |
1da177e4 LT |
470 | * then we could set it on SND_SHUTDOWN. BTW examples given |
471 | * in Stevens' books assume exactly this behaviour, it explains | |
70efce27 | 472 | * why POLLHUP is incompatible with POLLOUT. --ANK |
1da177e4 LT |
473 | * |
474 | * NOTE. Check for TCP_CLOSE is added. The goal is to prevent | |
475 | * blocking on fresh not-connected or disconnected socket. --ANK | |
476 | */ | |
477 | if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == TCP_CLOSE) | |
478 | mask |= POLLHUP; | |
479 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
f348d70a | 480 | mask |= POLLIN | POLLRDNORM | POLLRDHUP; |
1da177e4 | 481 | |
8336886f JC |
482 | /* Connected or passive Fast Open socket? */ |
483 | if (sk->sk_state != TCP_SYN_SENT && | |
484 | (sk->sk_state != TCP_SYN_RECV || tp->fastopen_rsk != NULL)) { | |
c7004482 DM |
485 | int target = sock_rcvlowat(sk, 0, INT_MAX); |
486 | ||
487 | if (tp->urg_seq == tp->copied_seq && | |
488 | !sock_flag(sk, SOCK_URGINLINE) && | |
489 | tp->urg_data) | |
b634f875 | 490 | target++; |
c7004482 | 491 | |
1da177e4 LT |
492 | /* Potential race condition. If read of tp below will |
493 | * escape above sk->sk_state, we can be illegally awaken | |
494 | * in SYN_* states. */ | |
c7004482 | 495 | if (tp->rcv_nxt - tp->copied_seq >= target) |
1da177e4 LT |
496 | mask |= POLLIN | POLLRDNORM; |
497 | ||
498 | if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { | |
499 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) { | |
500 | mask |= POLLOUT | POLLWRNORM; | |
501 | } else { /* send SIGIO later */ | |
502 | set_bit(SOCK_ASYNC_NOSPACE, | |
503 | &sk->sk_socket->flags); | |
504 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
505 | ||
506 | /* Race breaker. If space is freed after | |
507 | * wspace test but before the flags are set, | |
508 | * IO signal will be lost. | |
509 | */ | |
510 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) | |
511 | mask |= POLLOUT | POLLWRNORM; | |
512 | } | |
d84ba638 KM |
513 | } else |
514 | mask |= POLLOUT | POLLWRNORM; | |
1da177e4 LT |
515 | |
516 | if (tp->urg_data & TCP_URG_VALID) | |
517 | mask |= POLLPRI; | |
518 | } | |
a4d25803 TM |
519 | /* This barrier is coupled with smp_wmb() in tcp_reset() */ |
520 | smp_rmb(); | |
521 | if (sk->sk_err) | |
522 | mask |= POLLERR; | |
523 | ||
1da177e4 LT |
524 | return mask; |
525 | } | |
4bc2f18b | 526 | EXPORT_SYMBOL(tcp_poll); |
1da177e4 LT |
527 | |
528 | int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg) | |
529 | { | |
530 | struct tcp_sock *tp = tcp_sk(sk); | |
531 | int answ; | |
0e71c55c | 532 | bool slow; |
1da177e4 LT |
533 | |
534 | switch (cmd) { | |
535 | case SIOCINQ: | |
536 | if (sk->sk_state == TCP_LISTEN) | |
537 | return -EINVAL; | |
538 | ||
0e71c55c | 539 | slow = lock_sock_fast(sk); |
1da177e4 LT |
540 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) |
541 | answ = 0; | |
542 | else if (sock_flag(sk, SOCK_URGINLINE) || | |
543 | !tp->urg_data || | |
544 | before(tp->urg_seq, tp->copied_seq) || | |
545 | !before(tp->urg_seq, tp->rcv_nxt)) { | |
91521944 | 546 | |
1da177e4 LT |
547 | answ = tp->rcv_nxt - tp->copied_seq; |
548 | ||
a3374c42 ED |
549 | /* Subtract 1, if FIN was received */ |
550 | if (answ && sock_flag(sk, SOCK_DONE)) | |
551 | answ--; | |
1da177e4 LT |
552 | } else |
553 | answ = tp->urg_seq - tp->copied_seq; | |
0e71c55c | 554 | unlock_sock_fast(sk, slow); |
1da177e4 LT |
555 | break; |
556 | case SIOCATMARK: | |
557 | answ = tp->urg_data && tp->urg_seq == tp->copied_seq; | |
558 | break; | |
559 | case SIOCOUTQ: | |
560 | if (sk->sk_state == TCP_LISTEN) | |
561 | return -EINVAL; | |
562 | ||
563 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) | |
564 | answ = 0; | |
565 | else | |
566 | answ = tp->write_seq - tp->snd_una; | |
567 | break; | |
2f4e1b39 MS |
568 | case SIOCOUTQNSD: |
569 | if (sk->sk_state == TCP_LISTEN) | |
570 | return -EINVAL; | |
571 | ||
572 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) | |
573 | answ = 0; | |
574 | else | |
575 | answ = tp->write_seq - tp->snd_nxt; | |
576 | break; | |
1da177e4 LT |
577 | default: |
578 | return -ENOIOCTLCMD; | |
3ff50b79 | 579 | } |
1da177e4 LT |
580 | |
581 | return put_user(answ, (int __user *)arg); | |
582 | } | |
4bc2f18b | 583 | EXPORT_SYMBOL(tcp_ioctl); |
1da177e4 | 584 | |
1da177e4 LT |
585 | static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb) |
586 | { | |
4de075e0 | 587 | TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH; |
1da177e4 LT |
588 | tp->pushed_seq = tp->write_seq; |
589 | } | |
590 | ||
a2a385d6 | 591 | static inline bool forced_push(const struct tcp_sock *tp) |
1da177e4 LT |
592 | { |
593 | return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1)); | |
594 | } | |
595 | ||
9e412ba7 | 596 | static inline void skb_entail(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 597 | { |
9e412ba7 | 598 | struct tcp_sock *tp = tcp_sk(sk); |
352d4800 ACM |
599 | struct tcp_skb_cb *tcb = TCP_SKB_CB(skb); |
600 | ||
601 | skb->csum = 0; | |
602 | tcb->seq = tcb->end_seq = tp->write_seq; | |
4de075e0 | 603 | tcb->tcp_flags = TCPHDR_ACK; |
352d4800 | 604 | tcb->sacked = 0; |
1da177e4 | 605 | skb_header_release(skb); |
fe067e8a | 606 | tcp_add_write_queue_tail(sk, skb); |
3ab224be HA |
607 | sk->sk_wmem_queued += skb->truesize; |
608 | sk_mem_charge(sk, skb->truesize); | |
89ebd197 | 609 | if (tp->nonagle & TCP_NAGLE_PUSH) |
e905a9ed | 610 | tp->nonagle &= ~TCP_NAGLE_PUSH; |
1da177e4 LT |
611 | } |
612 | ||
afeca340 | 613 | static inline void tcp_mark_urg(struct tcp_sock *tp, int flags) |
1da177e4 | 614 | { |
33f5f57e | 615 | if (flags & MSG_OOB) |
1da177e4 | 616 | tp->snd_up = tp->write_seq; |
1da177e4 LT |
617 | } |
618 | ||
9e412ba7 IJ |
619 | static inline void tcp_push(struct sock *sk, int flags, int mss_now, |
620 | int nonagle) | |
1da177e4 | 621 | { |
fe067e8a | 622 | if (tcp_send_head(sk)) { |
afeca340 KK |
623 | struct tcp_sock *tp = tcp_sk(sk); |
624 | ||
1da177e4 | 625 | if (!(flags & MSG_MORE) || forced_push(tp)) |
afeca340 KK |
626 | tcp_mark_push(tp, tcp_write_queue_tail(sk)); |
627 | ||
628 | tcp_mark_urg(tp, flags); | |
9e412ba7 | 629 | __tcp_push_pending_frames(sk, mss_now, |
1da177e4 LT |
630 | (flags & MSG_MORE) ? TCP_NAGLE_CORK : nonagle); |
631 | } | |
632 | } | |
633 | ||
6ff7751d AB |
634 | static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, |
635 | unsigned int offset, size_t len) | |
9c55e01c JA |
636 | { |
637 | struct tcp_splice_state *tss = rd_desc->arg.data; | |
33966dd0 | 638 | int ret; |
9c55e01c | 639 | |
9fa5fdf2 DM |
640 | ret = skb_splice_bits(skb, offset, tss->pipe, min(rd_desc->count, len), |
641 | tss->flags); | |
33966dd0 WT |
642 | if (ret > 0) |
643 | rd_desc->count -= ret; | |
644 | return ret; | |
9c55e01c JA |
645 | } |
646 | ||
647 | static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss) | |
648 | { | |
649 | /* Store TCP splice context information in read_descriptor_t. */ | |
650 | read_descriptor_t rd_desc = { | |
651 | .arg.data = tss, | |
33966dd0 | 652 | .count = tss->len, |
9c55e01c JA |
653 | }; |
654 | ||
655 | return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv); | |
656 | } | |
657 | ||
658 | /** | |
659 | * tcp_splice_read - splice data from TCP socket to a pipe | |
660 | * @sock: socket to splice from | |
661 | * @ppos: position (not valid) | |
662 | * @pipe: pipe to splice to | |
663 | * @len: number of bytes to splice | |
664 | * @flags: splice modifier flags | |
665 | * | |
666 | * Description: | |
667 | * Will read pages from given socket and fill them into a pipe. | |
668 | * | |
669 | **/ | |
670 | ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos, | |
671 | struct pipe_inode_info *pipe, size_t len, | |
672 | unsigned int flags) | |
673 | { | |
674 | struct sock *sk = sock->sk; | |
675 | struct tcp_splice_state tss = { | |
676 | .pipe = pipe, | |
677 | .len = len, | |
678 | .flags = flags, | |
679 | }; | |
680 | long timeo; | |
681 | ssize_t spliced; | |
682 | int ret; | |
683 | ||
3a047bf8 | 684 | sock_rps_record_flow(sk); |
9c55e01c JA |
685 | /* |
686 | * We can't seek on a socket input | |
687 | */ | |
688 | if (unlikely(*ppos)) | |
689 | return -ESPIPE; | |
690 | ||
691 | ret = spliced = 0; | |
692 | ||
693 | lock_sock(sk); | |
694 | ||
42324c62 | 695 | timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK); |
9c55e01c JA |
696 | while (tss.len) { |
697 | ret = __tcp_splice_read(sk, &tss); | |
698 | if (ret < 0) | |
699 | break; | |
700 | else if (!ret) { | |
701 | if (spliced) | |
702 | break; | |
9c55e01c JA |
703 | if (sock_flag(sk, SOCK_DONE)) |
704 | break; | |
705 | if (sk->sk_err) { | |
706 | ret = sock_error(sk); | |
707 | break; | |
708 | } | |
709 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
710 | break; | |
711 | if (sk->sk_state == TCP_CLOSE) { | |
712 | /* | |
713 | * This occurs when user tries to read | |
714 | * from never connected socket. | |
715 | */ | |
716 | if (!sock_flag(sk, SOCK_DONE)) | |
717 | ret = -ENOTCONN; | |
718 | break; | |
719 | } | |
720 | if (!timeo) { | |
721 | ret = -EAGAIN; | |
722 | break; | |
723 | } | |
724 | sk_wait_data(sk, &timeo); | |
725 | if (signal_pending(current)) { | |
726 | ret = sock_intr_errno(timeo); | |
727 | break; | |
728 | } | |
729 | continue; | |
730 | } | |
731 | tss.len -= ret; | |
732 | spliced += ret; | |
733 | ||
33966dd0 WT |
734 | if (!timeo) |
735 | break; | |
9c55e01c JA |
736 | release_sock(sk); |
737 | lock_sock(sk); | |
738 | ||
739 | if (sk->sk_err || sk->sk_state == TCP_CLOSE || | |
33966dd0 | 740 | (sk->sk_shutdown & RCV_SHUTDOWN) || |
9c55e01c JA |
741 | signal_pending(current)) |
742 | break; | |
743 | } | |
744 | ||
745 | release_sock(sk); | |
746 | ||
747 | if (spliced) | |
748 | return spliced; | |
749 | ||
750 | return ret; | |
751 | } | |
4bc2f18b | 752 | EXPORT_SYMBOL(tcp_splice_read); |
9c55e01c | 753 | |
df97c708 | 754 | struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp) |
f561d0f2 PE |
755 | { |
756 | struct sk_buff *skb; | |
757 | ||
758 | /* The TCP header must be at least 32-bit aligned. */ | |
759 | size = ALIGN(size, 4); | |
760 | ||
761 | skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp); | |
762 | if (skb) { | |
3ab224be | 763 | if (sk_wmem_schedule(sk, skb->truesize)) { |
a21d4572 | 764 | skb_reserve(skb, sk->sk_prot->max_header); |
f561d0f2 PE |
765 | /* |
766 | * Make sure that we have exactly size bytes | |
767 | * available to the caller, no more, no less. | |
768 | */ | |
16fad69c | 769 | skb->reserved_tailroom = skb->end - skb->tail - size; |
f561d0f2 PE |
770 | return skb; |
771 | } | |
772 | __kfree_skb(skb); | |
773 | } else { | |
5c52ba17 | 774 | sk->sk_prot->enter_memory_pressure(sk); |
f561d0f2 PE |
775 | sk_stream_moderate_sndbuf(sk); |
776 | } | |
777 | return NULL; | |
778 | } | |
779 | ||
0c54b85f IJ |
780 | static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now, |
781 | int large_allowed) | |
782 | { | |
783 | struct tcp_sock *tp = tcp_sk(sk); | |
2a3a041c | 784 | u32 xmit_size_goal, old_size_goal; |
0c54b85f IJ |
785 | |
786 | xmit_size_goal = mss_now; | |
787 | ||
788 | if (large_allowed && sk_can_gso(sk)) { | |
789 | xmit_size_goal = ((sk->sk_gso_max_size - 1) - | |
790 | inet_csk(sk)->icsk_af_ops->net_header_len - | |
791 | inet_csk(sk)->icsk_ext_hdr_len - | |
792 | tp->tcp_header_len); | |
793 | ||
46d3ceab ED |
794 | /* TSQ : try to have two TSO segments in flight */ |
795 | xmit_size_goal = min_t(u32, xmit_size_goal, | |
796 | sysctl_tcp_limit_output_bytes >> 1); | |
797 | ||
0c54b85f | 798 | xmit_size_goal = tcp_bound_to_half_wnd(tp, xmit_size_goal); |
2a3a041c IJ |
799 | |
800 | /* We try hard to avoid divides here */ | |
801 | old_size_goal = tp->xmit_size_goal_segs * mss_now; | |
802 | ||
803 | if (likely(old_size_goal <= xmit_size_goal && | |
804 | old_size_goal + mss_now > xmit_size_goal)) { | |
805 | xmit_size_goal = old_size_goal; | |
806 | } else { | |
1485348d BH |
807 | tp->xmit_size_goal_segs = |
808 | min_t(u16, xmit_size_goal / mss_now, | |
809 | sk->sk_gso_max_segs); | |
2a3a041c IJ |
810 | xmit_size_goal = tp->xmit_size_goal_segs * mss_now; |
811 | } | |
0c54b85f IJ |
812 | } |
813 | ||
afece1c6 | 814 | return max(xmit_size_goal, mss_now); |
0c54b85f IJ |
815 | } |
816 | ||
817 | static int tcp_send_mss(struct sock *sk, int *size_goal, int flags) | |
818 | { | |
819 | int mss_now; | |
820 | ||
821 | mss_now = tcp_current_mss(sk); | |
822 | *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB)); | |
823 | ||
824 | return mss_now; | |
825 | } | |
826 | ||
64022d0b ED |
827 | static ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, |
828 | size_t size, int flags) | |
1da177e4 LT |
829 | { |
830 | struct tcp_sock *tp = tcp_sk(sk); | |
c1b4a7e6 | 831 | int mss_now, size_goal; |
1da177e4 LT |
832 | int err; |
833 | ssize_t copied; | |
834 | long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); | |
835 | ||
8336886f JC |
836 | /* Wait for a connection to finish. One exception is TCP Fast Open |
837 | * (passive side) where data is allowed to be sent before a connection | |
838 | * is fully established. | |
839 | */ | |
840 | if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) && | |
841 | !tcp_passive_fastopen(sk)) { | |
1da177e4 LT |
842 | if ((err = sk_stream_wait_connect(sk, &timeo)) != 0) |
843 | goto out_err; | |
8336886f | 844 | } |
1da177e4 LT |
845 | |
846 | clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); | |
847 | ||
0c54b85f | 848 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
849 | copied = 0; |
850 | ||
851 | err = -EPIPE; | |
852 | if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) | |
0d6a775e | 853 | goto out_err; |
1da177e4 | 854 | |
64022d0b | 855 | while (size > 0) { |
fe067e8a | 856 | struct sk_buff *skb = tcp_write_queue_tail(sk); |
38ba0a65 | 857 | int copy, i; |
38ba0a65 | 858 | bool can_coalesce; |
1da177e4 | 859 | |
fe067e8a | 860 | if (!tcp_send_head(sk) || (copy = size_goal - skb->len) <= 0) { |
1da177e4 LT |
861 | new_segment: |
862 | if (!sk_stream_memory_free(sk)) | |
863 | goto wait_for_sndbuf; | |
864 | ||
df97c708 | 865 | skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation); |
1da177e4 LT |
866 | if (!skb) |
867 | goto wait_for_memory; | |
868 | ||
9e412ba7 | 869 | skb_entail(sk, skb); |
c1b4a7e6 | 870 | copy = size_goal; |
1da177e4 LT |
871 | } |
872 | ||
873 | if (copy > size) | |
874 | copy = size; | |
875 | ||
876 | i = skb_shinfo(skb)->nr_frags; | |
877 | can_coalesce = skb_can_coalesce(skb, i, page, offset); | |
878 | if (!can_coalesce && i >= MAX_SKB_FRAGS) { | |
879 | tcp_mark_push(tp, skb); | |
880 | goto new_segment; | |
881 | } | |
3ab224be | 882 | if (!sk_wmem_schedule(sk, copy)) |
1da177e4 | 883 | goto wait_for_memory; |
e905a9ed | 884 | |
1da177e4 | 885 | if (can_coalesce) { |
9e903e08 | 886 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); |
1da177e4 LT |
887 | } else { |
888 | get_page(page); | |
889 | skb_fill_page_desc(skb, i, page, offset, copy); | |
890 | } | |
c9af6db4 | 891 | skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; |
cef401de | 892 | |
1da177e4 LT |
893 | skb->len += copy; |
894 | skb->data_len += copy; | |
895 | skb->truesize += copy; | |
896 | sk->sk_wmem_queued += copy; | |
3ab224be | 897 | sk_mem_charge(sk, copy); |
84fa7933 | 898 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 LT |
899 | tp->write_seq += copy; |
900 | TCP_SKB_CB(skb)->end_seq += copy; | |
7967168c | 901 | skb_shinfo(skb)->gso_segs = 0; |
1da177e4 LT |
902 | |
903 | if (!copied) | |
4de075e0 | 904 | TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH; |
1da177e4 LT |
905 | |
906 | copied += copy; | |
64022d0b ED |
907 | offset += copy; |
908 | if (!(size -= copy)) | |
1da177e4 LT |
909 | goto out; |
910 | ||
69d15067 | 911 | if (skb->len < size_goal || (flags & MSG_OOB)) |
1da177e4 LT |
912 | continue; |
913 | ||
914 | if (forced_push(tp)) { | |
915 | tcp_mark_push(tp, skb); | |
9e412ba7 | 916 | __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH); |
fe067e8a | 917 | } else if (skb == tcp_send_head(sk)) |
1da177e4 LT |
918 | tcp_push_one(sk, mss_now); |
919 | continue; | |
920 | ||
921 | wait_for_sndbuf: | |
922 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
923 | wait_for_memory: | |
bad115cf | 924 | tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH); |
1da177e4 LT |
925 | |
926 | if ((err = sk_stream_wait_memory(sk, &timeo)) != 0) | |
927 | goto do_error; | |
928 | ||
0c54b85f | 929 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
930 | } |
931 | ||
932 | out: | |
35f9c09f | 933 | if (copied && !(flags & MSG_SENDPAGE_NOTLAST)) |
9e412ba7 | 934 | tcp_push(sk, flags, mss_now, tp->nonagle); |
1da177e4 LT |
935 | return copied; |
936 | ||
937 | do_error: | |
938 | if (copied) | |
939 | goto out; | |
940 | out_err: | |
941 | return sk_stream_error(sk, flags, err); | |
942 | } | |
943 | ||
7ba42910 CG |
944 | int tcp_sendpage(struct sock *sk, struct page *page, int offset, |
945 | size_t size, int flags) | |
1da177e4 LT |
946 | { |
947 | ssize_t res; | |
1da177e4 | 948 | |
1da177e4 | 949 | if (!(sk->sk_route_caps & NETIF_F_SG) || |
8648b305 | 950 | !(sk->sk_route_caps & NETIF_F_ALL_CSUM)) |
7ba42910 CG |
951 | return sock_no_sendpage(sk->sk_socket, page, offset, size, |
952 | flags); | |
1da177e4 | 953 | |
1da177e4 | 954 | lock_sock(sk); |
64022d0b | 955 | res = do_tcp_sendpages(sk, page, offset, size, flags); |
1da177e4 LT |
956 | release_sock(sk); |
957 | return res; | |
958 | } | |
4bc2f18b | 959 | EXPORT_SYMBOL(tcp_sendpage); |
1da177e4 | 960 | |
690e99c4 | 961 | static inline int select_size(const struct sock *sk, bool sg) |
1da177e4 | 962 | { |
cf533ea5 | 963 | const struct tcp_sock *tp = tcp_sk(sk); |
c1b4a7e6 | 964 | int tmp = tp->mss_cache; |
1da177e4 | 965 | |
def87cf4 | 966 | if (sg) { |
f07d960d ED |
967 | if (sk_can_gso(sk)) { |
968 | /* Small frames wont use a full page: | |
969 | * Payload will immediately follow tcp header. | |
970 | */ | |
971 | tmp = SKB_WITH_OVERHEAD(2048 - MAX_TCP_HEADER); | |
972 | } else { | |
b4e26f5e DM |
973 | int pgbreak = SKB_MAX_HEAD(MAX_TCP_HEADER); |
974 | ||
975 | if (tmp >= pgbreak && | |
976 | tmp <= pgbreak + (MAX_SKB_FRAGS - 1) * PAGE_SIZE) | |
977 | tmp = pgbreak; | |
978 | } | |
979 | } | |
1da177e4 | 980 | |
1da177e4 LT |
981 | return tmp; |
982 | } | |
983 | ||
cf60af03 YC |
984 | void tcp_free_fastopen_req(struct tcp_sock *tp) |
985 | { | |
986 | if (tp->fastopen_req != NULL) { | |
987 | kfree(tp->fastopen_req); | |
988 | tp->fastopen_req = NULL; | |
989 | } | |
990 | } | |
991 | ||
992 | static int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *size) | |
993 | { | |
994 | struct tcp_sock *tp = tcp_sk(sk); | |
995 | int err, flags; | |
996 | ||
997 | if (!(sysctl_tcp_fastopen & TFO_CLIENT_ENABLE)) | |
998 | return -EOPNOTSUPP; | |
999 | if (tp->fastopen_req != NULL) | |
1000 | return -EALREADY; /* Another Fast Open is in progress */ | |
1001 | ||
1002 | tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request), | |
1003 | sk->sk_allocation); | |
1004 | if (unlikely(tp->fastopen_req == NULL)) | |
1005 | return -ENOBUFS; | |
1006 | tp->fastopen_req->data = msg; | |
1007 | ||
1008 | flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0; | |
1009 | err = __inet_stream_connect(sk->sk_socket, msg->msg_name, | |
1010 | msg->msg_namelen, flags); | |
1011 | *size = tp->fastopen_req->copied; | |
1012 | tcp_free_fastopen_req(tp); | |
1013 | return err; | |
1014 | } | |
1015 | ||
7ba42910 | 1016 | int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, |
1da177e4 LT |
1017 | size_t size) |
1018 | { | |
1019 | struct iovec *iov; | |
1020 | struct tcp_sock *tp = tcp_sk(sk); | |
1021 | struct sk_buff *skb; | |
cf60af03 YC |
1022 | int iovlen, flags, err, copied = 0; |
1023 | int mss_now = 0, size_goal, copied_syn = 0, offset = 0; | |
690e99c4 | 1024 | bool sg; |
1da177e4 LT |
1025 | long timeo; |
1026 | ||
1027 | lock_sock(sk); | |
1da177e4 LT |
1028 | |
1029 | flags = msg->msg_flags; | |
cf60af03 YC |
1030 | if (flags & MSG_FASTOPEN) { |
1031 | err = tcp_sendmsg_fastopen(sk, msg, &copied_syn); | |
1032 | if (err == -EINPROGRESS && copied_syn > 0) | |
1033 | goto out; | |
1034 | else if (err) | |
1035 | goto out_err; | |
1036 | offset = copied_syn; | |
1037 | } | |
1038 | ||
1da177e4 LT |
1039 | timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); |
1040 | ||
8336886f JC |
1041 | /* Wait for a connection to finish. One exception is TCP Fast Open |
1042 | * (passive side) where data is allowed to be sent before a connection | |
1043 | * is fully established. | |
1044 | */ | |
1045 | if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) && | |
1046 | !tcp_passive_fastopen(sk)) { | |
1da177e4 | 1047 | if ((err = sk_stream_wait_connect(sk, &timeo)) != 0) |
cf60af03 | 1048 | goto do_error; |
8336886f | 1049 | } |
1da177e4 | 1050 | |
c0e88ff0 PE |
1051 | if (unlikely(tp->repair)) { |
1052 | if (tp->repair_queue == TCP_RECV_QUEUE) { | |
1053 | copied = tcp_send_rcvq(sk, msg, size); | |
1054 | goto out; | |
1055 | } | |
1056 | ||
1057 | err = -EINVAL; | |
1058 | if (tp->repair_queue == TCP_NO_QUEUE) | |
1059 | goto out_err; | |
1060 | ||
1061 | /* 'common' sending to sendq */ | |
1062 | } | |
1063 | ||
1da177e4 LT |
1064 | /* This should be in poll */ |
1065 | clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); | |
1066 | ||
0c54b85f | 1067 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
1068 | |
1069 | /* Ok commence sending. */ | |
1070 | iovlen = msg->msg_iovlen; | |
1071 | iov = msg->msg_iov; | |
1072 | copied = 0; | |
1073 | ||
1074 | err = -EPIPE; | |
1075 | if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) | |
0d6a775e | 1076 | goto out_err; |
1da177e4 | 1077 | |
690e99c4 | 1078 | sg = !!(sk->sk_route_caps & NETIF_F_SG); |
def87cf4 | 1079 | |
1da177e4 | 1080 | while (--iovlen >= 0) { |
01db403c | 1081 | size_t seglen = iov->iov_len; |
1da177e4 LT |
1082 | unsigned char __user *from = iov->iov_base; |
1083 | ||
1084 | iov++; | |
cf60af03 YC |
1085 | if (unlikely(offset > 0)) { /* Skip bytes copied in SYN */ |
1086 | if (offset >= seglen) { | |
1087 | offset -= seglen; | |
1088 | continue; | |
1089 | } | |
1090 | seglen -= offset; | |
1091 | from += offset; | |
1092 | offset = 0; | |
1093 | } | |
1da177e4 LT |
1094 | |
1095 | while (seglen > 0) { | |
6828b92b HX |
1096 | int copy = 0; |
1097 | int max = size_goal; | |
1da177e4 | 1098 | |
fe067e8a | 1099 | skb = tcp_write_queue_tail(sk); |
6828b92b HX |
1100 | if (tcp_send_head(sk)) { |
1101 | if (skb->ip_summed == CHECKSUM_NONE) | |
1102 | max = mss_now; | |
1103 | copy = max - skb->len; | |
1104 | } | |
1da177e4 | 1105 | |
6828b92b | 1106 | if (copy <= 0) { |
1da177e4 LT |
1107 | new_segment: |
1108 | /* Allocate new segment. If the interface is SG, | |
1109 | * allocate skb fitting to single page. | |
1110 | */ | |
1111 | if (!sk_stream_memory_free(sk)) | |
1112 | goto wait_for_sndbuf; | |
1113 | ||
def87cf4 KK |
1114 | skb = sk_stream_alloc_skb(sk, |
1115 | select_size(sk, sg), | |
1116 | sk->sk_allocation); | |
1da177e4 LT |
1117 | if (!skb) |
1118 | goto wait_for_memory; | |
1119 | ||
f3f90538 AV |
1120 | /* |
1121 | * All packets are restored as if they have | |
1122 | * already been sent. | |
1123 | */ | |
1124 | if (tp->repair) | |
1125 | TCP_SKB_CB(skb)->when = tcp_time_stamp; | |
1126 | ||
1da177e4 LT |
1127 | /* |
1128 | * Check whether we can use HW checksum. | |
1129 | */ | |
8648b305 | 1130 | if (sk->sk_route_caps & NETIF_F_ALL_CSUM) |
84fa7933 | 1131 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 | 1132 | |
9e412ba7 | 1133 | skb_entail(sk, skb); |
c1b4a7e6 | 1134 | copy = size_goal; |
6828b92b | 1135 | max = size_goal; |
1da177e4 LT |
1136 | } |
1137 | ||
1138 | /* Try to append data to the end of skb. */ | |
1139 | if (copy > seglen) | |
1140 | copy = seglen; | |
1141 | ||
1142 | /* Where to copy to? */ | |
a21d4572 | 1143 | if (skb_availroom(skb) > 0) { |
1da177e4 | 1144 | /* We have some space in skb head. Superb! */ |
a21d4572 | 1145 | copy = min_t(int, copy, skb_availroom(skb)); |
c6e1a0d1 TH |
1146 | err = skb_add_data_nocache(sk, skb, from, copy); |
1147 | if (err) | |
1da177e4 LT |
1148 | goto do_fault; |
1149 | } else { | |
5640f768 | 1150 | bool merge = true; |
1da177e4 | 1151 | int i = skb_shinfo(skb)->nr_frags; |
5640f768 ED |
1152 | struct page_frag *pfrag = sk_page_frag(sk); |
1153 | ||
1154 | if (!sk_page_frag_refill(sk, pfrag)) | |
1155 | goto wait_for_memory; | |
1156 | ||
1157 | if (!skb_can_coalesce(skb, i, pfrag->page, | |
1158 | pfrag->offset)) { | |
1159 | if (i == MAX_SKB_FRAGS || !sg) { | |
1160 | tcp_mark_push(tp, skb); | |
1161 | goto new_segment; | |
1da177e4 | 1162 | } |
5640f768 ED |
1163 | merge = false; |
1164 | } | |
ef015786 | 1165 | |
5640f768 | 1166 | copy = min_t(int, copy, pfrag->size - pfrag->offset); |
ef015786 | 1167 | |
3ab224be | 1168 | if (!sk_wmem_schedule(sk, copy)) |
ef015786 | 1169 | goto wait_for_memory; |
1da177e4 | 1170 | |
c6e1a0d1 | 1171 | err = skb_copy_to_page_nocache(sk, from, skb, |
5640f768 ED |
1172 | pfrag->page, |
1173 | pfrag->offset, | |
1174 | copy); | |
1175 | if (err) | |
1da177e4 | 1176 | goto do_error; |
1da177e4 LT |
1177 | |
1178 | /* Update the skb. */ | |
1179 | if (merge) { | |
9e903e08 | 1180 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); |
1da177e4 | 1181 | } else { |
5640f768 ED |
1182 | skb_fill_page_desc(skb, i, pfrag->page, |
1183 | pfrag->offset, copy); | |
1184 | get_page(pfrag->page); | |
1da177e4 | 1185 | } |
5640f768 | 1186 | pfrag->offset += copy; |
1da177e4 LT |
1187 | } |
1188 | ||
1189 | if (!copied) | |
4de075e0 | 1190 | TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH; |
1da177e4 LT |
1191 | |
1192 | tp->write_seq += copy; | |
1193 | TCP_SKB_CB(skb)->end_seq += copy; | |
7967168c | 1194 | skb_shinfo(skb)->gso_segs = 0; |
1da177e4 LT |
1195 | |
1196 | from += copy; | |
1197 | copied += copy; | |
1198 | if ((seglen -= copy) == 0 && iovlen == 0) | |
1199 | goto out; | |
1200 | ||
c0e88ff0 | 1201 | if (skb->len < max || (flags & MSG_OOB) || unlikely(tp->repair)) |
1da177e4 LT |
1202 | continue; |
1203 | ||
1204 | if (forced_push(tp)) { | |
1205 | tcp_mark_push(tp, skb); | |
9e412ba7 | 1206 | __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH); |
fe067e8a | 1207 | } else if (skb == tcp_send_head(sk)) |
1da177e4 LT |
1208 | tcp_push_one(sk, mss_now); |
1209 | continue; | |
1210 | ||
1211 | wait_for_sndbuf: | |
1212 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
1213 | wait_for_memory: | |
ec342325 | 1214 | if (copied) |
9e412ba7 | 1215 | tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH); |
1da177e4 LT |
1216 | |
1217 | if ((err = sk_stream_wait_memory(sk, &timeo)) != 0) | |
1218 | goto do_error; | |
1219 | ||
0c54b85f | 1220 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
1221 | } |
1222 | } | |
1223 | ||
1224 | out: | |
ec342325 | 1225 | if (copied) |
9e412ba7 | 1226 | tcp_push(sk, flags, mss_now, tp->nonagle); |
1da177e4 | 1227 | release_sock(sk); |
cf60af03 | 1228 | return copied + copied_syn; |
1da177e4 LT |
1229 | |
1230 | do_fault: | |
1231 | if (!skb->len) { | |
fe067e8a DM |
1232 | tcp_unlink_write_queue(skb, sk); |
1233 | /* It is the one place in all of TCP, except connection | |
1234 | * reset, where we can be unlinking the send_head. | |
1235 | */ | |
1236 | tcp_check_send_head(sk, skb); | |
3ab224be | 1237 | sk_wmem_free_skb(sk, skb); |
1da177e4 LT |
1238 | } |
1239 | ||
1240 | do_error: | |
cf60af03 | 1241 | if (copied + copied_syn) |
1da177e4 LT |
1242 | goto out; |
1243 | out_err: | |
1244 | err = sk_stream_error(sk, flags, err); | |
1da177e4 LT |
1245 | release_sock(sk); |
1246 | return err; | |
1247 | } | |
4bc2f18b | 1248 | EXPORT_SYMBOL(tcp_sendmsg); |
1da177e4 LT |
1249 | |
1250 | /* | |
1251 | * Handle reading urgent data. BSD has very simple semantics for | |
1252 | * this, no blocking and very strange errors 8) | |
1253 | */ | |
1254 | ||
377f0a08 | 1255 | static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags) |
1da177e4 LT |
1256 | { |
1257 | struct tcp_sock *tp = tcp_sk(sk); | |
1258 | ||
1259 | /* No URG data to read. */ | |
1260 | if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data || | |
1261 | tp->urg_data == TCP_URG_READ) | |
1262 | return -EINVAL; /* Yes this is right ! */ | |
1263 | ||
1264 | if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE)) | |
1265 | return -ENOTCONN; | |
1266 | ||
1267 | if (tp->urg_data & TCP_URG_VALID) { | |
1268 | int err = 0; | |
1269 | char c = tp->urg_data; | |
1270 | ||
1271 | if (!(flags & MSG_PEEK)) | |
1272 | tp->urg_data = TCP_URG_READ; | |
1273 | ||
1274 | /* Read urgent data. */ | |
1275 | msg->msg_flags |= MSG_OOB; | |
1276 | ||
1277 | if (len > 0) { | |
1278 | if (!(flags & MSG_TRUNC)) | |
1279 | err = memcpy_toiovec(msg->msg_iov, &c, 1); | |
1280 | len = 1; | |
1281 | } else | |
1282 | msg->msg_flags |= MSG_TRUNC; | |
1283 | ||
1284 | return err ? -EFAULT : len; | |
1285 | } | |
1286 | ||
1287 | if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN)) | |
1288 | return 0; | |
1289 | ||
1290 | /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and | |
1291 | * the available implementations agree in this case: | |
1292 | * this call should never block, independent of the | |
1293 | * blocking state of the socket. | |
1294 | * Mike <pall@rz.uni-karlsruhe.de> | |
1295 | */ | |
1296 | return -EAGAIN; | |
1297 | } | |
1298 | ||
c0e88ff0 PE |
1299 | static int tcp_peek_sndq(struct sock *sk, struct msghdr *msg, int len) |
1300 | { | |
1301 | struct sk_buff *skb; | |
1302 | int copied = 0, err = 0; | |
1303 | ||
1304 | /* XXX -- need to support SO_PEEK_OFF */ | |
1305 | ||
1306 | skb_queue_walk(&sk->sk_write_queue, skb) { | |
1307 | err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, skb->len); | |
1308 | if (err) | |
1309 | break; | |
1310 | ||
1311 | copied += skb->len; | |
1312 | } | |
1313 | ||
1314 | return err ?: copied; | |
1315 | } | |
1316 | ||
1da177e4 LT |
1317 | /* Clean up the receive buffer for full frames taken by the user, |
1318 | * then send an ACK if necessary. COPIED is the number of bytes | |
1319 | * tcp_recvmsg has given to the user so far, it speeds up the | |
1320 | * calculation of whether or not we must ACK for the sake of | |
1321 | * a window update. | |
1322 | */ | |
0e4b4992 | 1323 | void tcp_cleanup_rbuf(struct sock *sk, int copied) |
1da177e4 LT |
1324 | { |
1325 | struct tcp_sock *tp = tcp_sk(sk); | |
a2a385d6 | 1326 | bool time_to_ack = false; |
1da177e4 | 1327 | |
1da177e4 LT |
1328 | struct sk_buff *skb = skb_peek(&sk->sk_receive_queue); |
1329 | ||
d792c100 | 1330 | WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq), |
2af6fd8b | 1331 | "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n", |
d792c100 | 1332 | tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt); |
1da177e4 | 1333 | |
463c84b9 ACM |
1334 | if (inet_csk_ack_scheduled(sk)) { |
1335 | const struct inet_connection_sock *icsk = inet_csk(sk); | |
1da177e4 LT |
1336 | /* Delayed ACKs frequently hit locked sockets during bulk |
1337 | * receive. */ | |
463c84b9 | 1338 | if (icsk->icsk_ack.blocked || |
1da177e4 | 1339 | /* Once-per-two-segments ACK was not sent by tcp_input.c */ |
463c84b9 | 1340 | tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss || |
1da177e4 LT |
1341 | /* |
1342 | * If this read emptied read buffer, we send ACK, if | |
1343 | * connection is not bidirectional, user drained | |
1344 | * receive buffer and there was a small segment | |
1345 | * in queue. | |
1346 | */ | |
1ef9696c AK |
1347 | (copied > 0 && |
1348 | ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) || | |
1349 | ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) && | |
1350 | !icsk->icsk_ack.pingpong)) && | |
1351 | !atomic_read(&sk->sk_rmem_alloc))) | |
a2a385d6 | 1352 | time_to_ack = true; |
1da177e4 LT |
1353 | } |
1354 | ||
1355 | /* We send an ACK if we can now advertise a non-zero window | |
1356 | * which has been raised "significantly". | |
1357 | * | |
1358 | * Even if window raised up to infinity, do not send window open ACK | |
1359 | * in states, where we will not receive more. It is useless. | |
1360 | */ | |
1361 | if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) { | |
1362 | __u32 rcv_window_now = tcp_receive_window(tp); | |
1363 | ||
1364 | /* Optimize, __tcp_select_window() is not cheap. */ | |
1365 | if (2*rcv_window_now <= tp->window_clamp) { | |
1366 | __u32 new_window = __tcp_select_window(sk); | |
1367 | ||
1368 | /* Send ACK now, if this read freed lots of space | |
1369 | * in our buffer. Certainly, new_window is new window. | |
1370 | * We can advertise it now, if it is not less than current one. | |
1371 | * "Lots" means "at least twice" here. | |
1372 | */ | |
1373 | if (new_window && new_window >= 2 * rcv_window_now) | |
a2a385d6 | 1374 | time_to_ack = true; |
1da177e4 LT |
1375 | } |
1376 | } | |
1377 | if (time_to_ack) | |
1378 | tcp_send_ack(sk); | |
1379 | } | |
1380 | ||
1381 | static void tcp_prequeue_process(struct sock *sk) | |
1382 | { | |
1383 | struct sk_buff *skb; | |
1384 | struct tcp_sock *tp = tcp_sk(sk); | |
1385 | ||
6f67c817 | 1386 | NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPPREQUEUED); |
1da177e4 LT |
1387 | |
1388 | /* RX process wants to run with disabled BHs, though it is not | |
1389 | * necessary */ | |
1390 | local_bh_disable(); | |
1391 | while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) | |
c57943a1 | 1392 | sk_backlog_rcv(sk, skb); |
1da177e4 LT |
1393 | local_bh_enable(); |
1394 | ||
1395 | /* Clear memory counter. */ | |
1396 | tp->ucopy.memory = 0; | |
1397 | } | |
1398 | ||
73852e81 SM |
1399 | #ifdef CONFIG_NET_DMA |
1400 | static void tcp_service_net_dma(struct sock *sk, bool wait) | |
1401 | { | |
1402 | dma_cookie_t done, used; | |
1403 | dma_cookie_t last_issued; | |
1404 | struct tcp_sock *tp = tcp_sk(sk); | |
1405 | ||
1406 | if (!tp->ucopy.dma_chan) | |
1407 | return; | |
1408 | ||
1409 | last_issued = tp->ucopy.dma_cookie; | |
b9ee8683 | 1410 | dma_async_issue_pending(tp->ucopy.dma_chan); |
73852e81 SM |
1411 | |
1412 | do { | |
e239345f | 1413 | if (dma_async_is_tx_complete(tp->ucopy.dma_chan, |
73852e81 SM |
1414 | last_issued, &done, |
1415 | &used) == DMA_SUCCESS) { | |
1416 | /* Safe to free early-copied skbs now */ | |
1417 | __skb_queue_purge(&sk->sk_async_wait_queue); | |
1418 | break; | |
1419 | } else { | |
1420 | struct sk_buff *skb; | |
1421 | while ((skb = skb_peek(&sk->sk_async_wait_queue)) && | |
1422 | (dma_async_is_complete(skb->dma_cookie, done, | |
1423 | used) == DMA_SUCCESS)) { | |
1424 | __skb_dequeue(&sk->sk_async_wait_queue); | |
1425 | kfree_skb(skb); | |
1426 | } | |
1427 | } | |
1428 | } while (wait); | |
1429 | } | |
1430 | #endif | |
1431 | ||
f26845b4 | 1432 | static struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off) |
1da177e4 LT |
1433 | { |
1434 | struct sk_buff *skb; | |
1435 | u32 offset; | |
1436 | ||
f26845b4 | 1437 | while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) { |
1da177e4 | 1438 | offset = seq - TCP_SKB_CB(skb)->seq; |
aa8223c7 | 1439 | if (tcp_hdr(skb)->syn) |
1da177e4 | 1440 | offset--; |
aa8223c7 | 1441 | if (offset < skb->len || tcp_hdr(skb)->fin) { |
1da177e4 LT |
1442 | *off = offset; |
1443 | return skb; | |
1444 | } | |
f26845b4 ED |
1445 | /* This looks weird, but this can happen if TCP collapsing |
1446 | * splitted a fat GRO packet, while we released socket lock | |
1447 | * in skb_splice_bits() | |
1448 | */ | |
1449 | sk_eat_skb(sk, skb, false); | |
1da177e4 LT |
1450 | } |
1451 | return NULL; | |
1452 | } | |
1453 | ||
1454 | /* | |
1455 | * This routine provides an alternative to tcp_recvmsg() for routines | |
1456 | * that would like to handle copying from skbuffs directly in 'sendfile' | |
1457 | * fashion. | |
1458 | * Note: | |
1459 | * - It is assumed that the socket was locked by the caller. | |
1460 | * - The routine does not block. | |
1461 | * - At present, there is no support for reading OOB data | |
1462 | * or for 'peeking' the socket using this routine | |
1463 | * (although both would be easy to implement). | |
1464 | */ | |
1465 | int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, | |
1466 | sk_read_actor_t recv_actor) | |
1467 | { | |
1468 | struct sk_buff *skb; | |
1469 | struct tcp_sock *tp = tcp_sk(sk); | |
1470 | u32 seq = tp->copied_seq; | |
1471 | u32 offset; | |
1472 | int copied = 0; | |
1473 | ||
1474 | if (sk->sk_state == TCP_LISTEN) | |
1475 | return -ENOTCONN; | |
1476 | while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) { | |
1477 | if (offset < skb->len) { | |
374e7b59 OP |
1478 | int used; |
1479 | size_t len; | |
1da177e4 LT |
1480 | |
1481 | len = skb->len - offset; | |
1482 | /* Stop reading if we hit a patch of urgent data */ | |
1483 | if (tp->urg_data) { | |
1484 | u32 urg_offset = tp->urg_seq - seq; | |
1485 | if (urg_offset < len) | |
1486 | len = urg_offset; | |
1487 | if (!len) | |
1488 | break; | |
1489 | } | |
1490 | used = recv_actor(desc, skb, offset, len); | |
ff905b1e | 1491 | if (used <= 0) { |
ddb61a57 JA |
1492 | if (!copied) |
1493 | copied = used; | |
1494 | break; | |
1495 | } else if (used <= len) { | |
1da177e4 LT |
1496 | seq += used; |
1497 | copied += used; | |
1498 | offset += used; | |
1499 | } | |
02275a2e | 1500 | /* If recv_actor drops the lock (e.g. TCP splice |
293ad604 OP |
1501 | * receive) the skb pointer might be invalid when |
1502 | * getting here: tcp_collapse might have deleted it | |
1503 | * while aggregating skbs from the socket queue. | |
1504 | */ | |
02275a2e WT |
1505 | skb = tcp_recv_skb(sk, seq - 1, &offset); |
1506 | if (!skb) | |
1da177e4 | 1507 | break; |
02275a2e WT |
1508 | /* TCP coalescing might have appended data to the skb. |
1509 | * Try to splice more frags | |
1510 | */ | |
1511 | if (offset + 1 != skb->len) | |
1512 | continue; | |
1da177e4 | 1513 | } |
aa8223c7 | 1514 | if (tcp_hdr(skb)->fin) { |
dc6b9b78 | 1515 | sk_eat_skb(sk, skb, false); |
1da177e4 LT |
1516 | ++seq; |
1517 | break; | |
1518 | } | |
dc6b9b78 | 1519 | sk_eat_skb(sk, skb, false); |
1da177e4 LT |
1520 | if (!desc->count) |
1521 | break; | |
baff42ab | 1522 | tp->copied_seq = seq; |
1da177e4 LT |
1523 | } |
1524 | tp->copied_seq = seq; | |
1525 | ||
1526 | tcp_rcv_space_adjust(sk); | |
1527 | ||
1528 | /* Clean up data we have read: This will do ACK frames. */ | |
f26845b4 ED |
1529 | if (copied > 0) { |
1530 | tcp_recv_skb(sk, seq, &offset); | |
0e4b4992 | 1531 | tcp_cleanup_rbuf(sk, copied); |
f26845b4 | 1532 | } |
1da177e4 LT |
1533 | return copied; |
1534 | } | |
4bc2f18b | 1535 | EXPORT_SYMBOL(tcp_read_sock); |
1da177e4 LT |
1536 | |
1537 | /* | |
1538 | * This routine copies from a sock struct into the user buffer. | |
1539 | * | |
1540 | * Technical note: in 2.3 we work on _locked_ socket, so that | |
1541 | * tricks with *seq access order and skb->users are not required. | |
1542 | * Probably, code can be easily improved even more. | |
1543 | */ | |
1544 | ||
1545 | int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, | |
1546 | size_t len, int nonblock, int flags, int *addr_len) | |
1547 | { | |
1548 | struct tcp_sock *tp = tcp_sk(sk); | |
1549 | int copied = 0; | |
1550 | u32 peek_seq; | |
1551 | u32 *seq; | |
1552 | unsigned long used; | |
1553 | int err; | |
1554 | int target; /* Read at least this many bytes */ | |
1555 | long timeo; | |
1556 | struct task_struct *user_recv = NULL; | |
dc6b9b78 | 1557 | bool copied_early = false; |
2b1244a4 | 1558 | struct sk_buff *skb; |
77527313 | 1559 | u32 urg_hole = 0; |
1da177e4 LT |
1560 | |
1561 | lock_sock(sk); | |
1562 | ||
1da177e4 LT |
1563 | err = -ENOTCONN; |
1564 | if (sk->sk_state == TCP_LISTEN) | |
1565 | goto out; | |
1566 | ||
1567 | timeo = sock_rcvtimeo(sk, nonblock); | |
1568 | ||
1569 | /* Urgent data needs to be handled specially. */ | |
1570 | if (flags & MSG_OOB) | |
1571 | goto recv_urg; | |
1572 | ||
c0e88ff0 PE |
1573 | if (unlikely(tp->repair)) { |
1574 | err = -EPERM; | |
1575 | if (!(flags & MSG_PEEK)) | |
1576 | goto out; | |
1577 | ||
1578 | if (tp->repair_queue == TCP_SEND_QUEUE) | |
1579 | goto recv_sndq; | |
1580 | ||
1581 | err = -EINVAL; | |
1582 | if (tp->repair_queue == TCP_NO_QUEUE) | |
1583 | goto out; | |
1584 | ||
1585 | /* 'common' recv queue MSG_PEEK-ing */ | |
1586 | } | |
1587 | ||
1da177e4 LT |
1588 | seq = &tp->copied_seq; |
1589 | if (flags & MSG_PEEK) { | |
1590 | peek_seq = tp->copied_seq; | |
1591 | seq = &peek_seq; | |
1592 | } | |
1593 | ||
1594 | target = sock_rcvlowat(sk, flags & MSG_WAITALL, len); | |
1595 | ||
1a2449a8 CL |
1596 | #ifdef CONFIG_NET_DMA |
1597 | tp->ucopy.dma_chan = NULL; | |
1598 | preempt_disable(); | |
2b1244a4 | 1599 | skb = skb_peek_tail(&sk->sk_receive_queue); |
e00c5d8b AM |
1600 | { |
1601 | int available = 0; | |
1602 | ||
1603 | if (skb) | |
1604 | available = TCP_SKB_CB(skb)->seq + skb->len - (*seq); | |
1605 | if ((available < target) && | |
1606 | (len > sysctl_tcp_dma_copybreak) && !(flags & MSG_PEEK) && | |
1607 | !sysctl_tcp_low_latency && | |
a2bd1140 | 1608 | net_dma_find_channel()) { |
e00c5d8b AM |
1609 | preempt_enable_no_resched(); |
1610 | tp->ucopy.pinned_list = | |
1611 | dma_pin_iovec_pages(msg->msg_iov, len); | |
1612 | } else { | |
1613 | preempt_enable_no_resched(); | |
1614 | } | |
1615 | } | |
1a2449a8 CL |
1616 | #endif |
1617 | ||
1da177e4 | 1618 | do { |
1da177e4 LT |
1619 | u32 offset; |
1620 | ||
1621 | /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */ | |
1622 | if (tp->urg_data && tp->urg_seq == *seq) { | |
1623 | if (copied) | |
1624 | break; | |
1625 | if (signal_pending(current)) { | |
1626 | copied = timeo ? sock_intr_errno(timeo) : -EAGAIN; | |
1627 | break; | |
1628 | } | |
1629 | } | |
1630 | ||
1631 | /* Next get a buffer. */ | |
1632 | ||
91521944 | 1633 | skb_queue_walk(&sk->sk_receive_queue, skb) { |
1da177e4 LT |
1634 | /* Now that we have two receive queues this |
1635 | * shouldn't happen. | |
1636 | */ | |
d792c100 | 1637 | if (WARN(before(*seq, TCP_SKB_CB(skb)->seq), |
2af6fd8b JP |
1638 | "recvmsg bug: copied %X seq %X rcvnxt %X fl %X\n", |
1639 | *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, | |
1640 | flags)) | |
1da177e4 | 1641 | break; |
d792c100 | 1642 | |
1da177e4 | 1643 | offset = *seq - TCP_SKB_CB(skb)->seq; |
aa8223c7 | 1644 | if (tcp_hdr(skb)->syn) |
1da177e4 LT |
1645 | offset--; |
1646 | if (offset < skb->len) | |
1647 | goto found_ok_skb; | |
aa8223c7 | 1648 | if (tcp_hdr(skb)->fin) |
1da177e4 | 1649 | goto found_fin_ok; |
2af6fd8b JP |
1650 | WARN(!(flags & MSG_PEEK), |
1651 | "recvmsg bug 2: copied %X seq %X rcvnxt %X fl %X\n", | |
1652 | *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags); | |
91521944 | 1653 | } |
1da177e4 LT |
1654 | |
1655 | /* Well, if we have backlog, try to process it now yet. */ | |
1656 | ||
1657 | if (copied >= target && !sk->sk_backlog.tail) | |
1658 | break; | |
1659 | ||
1660 | if (copied) { | |
1661 | if (sk->sk_err || | |
1662 | sk->sk_state == TCP_CLOSE || | |
1663 | (sk->sk_shutdown & RCV_SHUTDOWN) || | |
1664 | !timeo || | |
518a09ef | 1665 | signal_pending(current)) |
1da177e4 LT |
1666 | break; |
1667 | } else { | |
1668 | if (sock_flag(sk, SOCK_DONE)) | |
1669 | break; | |
1670 | ||
1671 | if (sk->sk_err) { | |
1672 | copied = sock_error(sk); | |
1673 | break; | |
1674 | } | |
1675 | ||
1676 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
1677 | break; | |
1678 | ||
1679 | if (sk->sk_state == TCP_CLOSE) { | |
1680 | if (!sock_flag(sk, SOCK_DONE)) { | |
1681 | /* This occurs when user tries to read | |
1682 | * from never connected socket. | |
1683 | */ | |
1684 | copied = -ENOTCONN; | |
1685 | break; | |
1686 | } | |
1687 | break; | |
1688 | } | |
1689 | ||
1690 | if (!timeo) { | |
1691 | copied = -EAGAIN; | |
1692 | break; | |
1693 | } | |
1694 | ||
1695 | if (signal_pending(current)) { | |
1696 | copied = sock_intr_errno(timeo); | |
1697 | break; | |
1698 | } | |
1699 | } | |
1700 | ||
0e4b4992 | 1701 | tcp_cleanup_rbuf(sk, copied); |
1da177e4 | 1702 | |
7df55125 | 1703 | if (!sysctl_tcp_low_latency && tp->ucopy.task == user_recv) { |
1da177e4 LT |
1704 | /* Install new reader */ |
1705 | if (!user_recv && !(flags & (MSG_TRUNC | MSG_PEEK))) { | |
1706 | user_recv = current; | |
1707 | tp->ucopy.task = user_recv; | |
1708 | tp->ucopy.iov = msg->msg_iov; | |
1709 | } | |
1710 | ||
1711 | tp->ucopy.len = len; | |
1712 | ||
547b792c IJ |
1713 | WARN_ON(tp->copied_seq != tp->rcv_nxt && |
1714 | !(flags & (MSG_PEEK | MSG_TRUNC))); | |
1da177e4 LT |
1715 | |
1716 | /* Ugly... If prequeue is not empty, we have to | |
1717 | * process it before releasing socket, otherwise | |
1718 | * order will be broken at second iteration. | |
1719 | * More elegant solution is required!!! | |
1720 | * | |
1721 | * Look: we have the following (pseudo)queues: | |
1722 | * | |
1723 | * 1. packets in flight | |
1724 | * 2. backlog | |
1725 | * 3. prequeue | |
1726 | * 4. receive_queue | |
1727 | * | |
1728 | * Each queue can be processed only if the next ones | |
1729 | * are empty. At this point we have empty receive_queue. | |
1730 | * But prequeue _can_ be not empty after 2nd iteration, | |
1731 | * when we jumped to start of loop because backlog | |
1732 | * processing added something to receive_queue. | |
1733 | * We cannot release_sock(), because backlog contains | |
1734 | * packets arrived _after_ prequeued ones. | |
1735 | * | |
1736 | * Shortly, algorithm is clear --- to process all | |
1737 | * the queues in order. We could make it more directly, | |
1738 | * requeueing packets from backlog to prequeue, if | |
1739 | * is not empty. It is more elegant, but eats cycles, | |
1740 | * unfortunately. | |
1741 | */ | |
b03efcfb | 1742 | if (!skb_queue_empty(&tp->ucopy.prequeue)) |
1da177e4 LT |
1743 | goto do_prequeue; |
1744 | ||
1745 | /* __ Set realtime policy in scheduler __ */ | |
1746 | } | |
1747 | ||
73852e81 | 1748 | #ifdef CONFIG_NET_DMA |
15c04175 MK |
1749 | if (tp->ucopy.dma_chan) { |
1750 | if (tp->rcv_wnd == 0 && | |
1751 | !skb_queue_empty(&sk->sk_async_wait_queue)) { | |
1752 | tcp_service_net_dma(sk, true); | |
1753 | tcp_cleanup_rbuf(sk, copied); | |
1754 | } else | |
b9ee8683 | 1755 | dma_async_issue_pending(tp->ucopy.dma_chan); |
15c04175 | 1756 | } |
73852e81 | 1757 | #endif |
1da177e4 LT |
1758 | if (copied >= target) { |
1759 | /* Do not sleep, just process backlog. */ | |
1760 | release_sock(sk); | |
1761 | lock_sock(sk); | |
1762 | } else | |
1763 | sk_wait_data(sk, &timeo); | |
1764 | ||
1a2449a8 | 1765 | #ifdef CONFIG_NET_DMA |
73852e81 | 1766 | tcp_service_net_dma(sk, false); /* Don't block */ |
1a2449a8 CL |
1767 | tp->ucopy.wakeup = 0; |
1768 | #endif | |
1769 | ||
1da177e4 LT |
1770 | if (user_recv) { |
1771 | int chunk; | |
1772 | ||
1773 | /* __ Restore normal policy in scheduler __ */ | |
1774 | ||
1775 | if ((chunk = len - tp->ucopy.len) != 0) { | |
ed88098e | 1776 | NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMBACKLOG, chunk); |
1da177e4 LT |
1777 | len -= chunk; |
1778 | copied += chunk; | |
1779 | } | |
1780 | ||
1781 | if (tp->rcv_nxt == tp->copied_seq && | |
b03efcfb | 1782 | !skb_queue_empty(&tp->ucopy.prequeue)) { |
1da177e4 LT |
1783 | do_prequeue: |
1784 | tcp_prequeue_process(sk); | |
1785 | ||
1786 | if ((chunk = len - tp->ucopy.len) != 0) { | |
ed88098e | 1787 | NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk); |
1da177e4 LT |
1788 | len -= chunk; |
1789 | copied += chunk; | |
1790 | } | |
1791 | } | |
1792 | } | |
77527313 IJ |
1793 | if ((flags & MSG_PEEK) && |
1794 | (peek_seq - copied - urg_hole != tp->copied_seq)) { | |
e87cc472 JP |
1795 | net_dbg_ratelimited("TCP(%s:%d): Application bug, race in MSG_PEEK\n", |
1796 | current->comm, | |
1797 | task_pid_nr(current)); | |
1da177e4 LT |
1798 | peek_seq = tp->copied_seq; |
1799 | } | |
1800 | continue; | |
1801 | ||
1802 | found_ok_skb: | |
1803 | /* Ok so how much can we use? */ | |
1804 | used = skb->len - offset; | |
1805 | if (len < used) | |
1806 | used = len; | |
1807 | ||
1808 | /* Do we have urgent data here? */ | |
1809 | if (tp->urg_data) { | |
1810 | u32 urg_offset = tp->urg_seq - *seq; | |
1811 | if (urg_offset < used) { | |
1812 | if (!urg_offset) { | |
1813 | if (!sock_flag(sk, SOCK_URGINLINE)) { | |
1814 | ++*seq; | |
77527313 | 1815 | urg_hole++; |
1da177e4 LT |
1816 | offset++; |
1817 | used--; | |
1818 | if (!used) | |
1819 | goto skip_copy; | |
1820 | } | |
1821 | } else | |
1822 | used = urg_offset; | |
1823 | } | |
1824 | } | |
1825 | ||
1826 | if (!(flags & MSG_TRUNC)) { | |
1a2449a8 CL |
1827 | #ifdef CONFIG_NET_DMA |
1828 | if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list) | |
a2bd1140 | 1829 | tp->ucopy.dma_chan = net_dma_find_channel(); |
1a2449a8 CL |
1830 | |
1831 | if (tp->ucopy.dma_chan) { | |
1832 | tp->ucopy.dma_cookie = dma_skb_copy_datagram_iovec( | |
1833 | tp->ucopy.dma_chan, skb, offset, | |
1834 | msg->msg_iov, used, | |
1835 | tp->ucopy.pinned_list); | |
1836 | ||
1837 | if (tp->ucopy.dma_cookie < 0) { | |
1838 | ||
afd46503 JP |
1839 | pr_alert("%s: dma_cookie < 0\n", |
1840 | __func__); | |
1a2449a8 CL |
1841 | |
1842 | /* Exception. Bailout! */ | |
1843 | if (!copied) | |
1844 | copied = -EFAULT; | |
1845 | break; | |
1846 | } | |
73852e81 | 1847 | |
b9ee8683 | 1848 | dma_async_issue_pending(tp->ucopy.dma_chan); |
73852e81 | 1849 | |
1a2449a8 | 1850 | if ((offset + used) == skb->len) |
dc6b9b78 | 1851 | copied_early = true; |
1a2449a8 CL |
1852 | |
1853 | } else | |
1854 | #endif | |
1855 | { | |
1856 | err = skb_copy_datagram_iovec(skb, offset, | |
1857 | msg->msg_iov, used); | |
1858 | if (err) { | |
1859 | /* Exception. Bailout! */ | |
1860 | if (!copied) | |
1861 | copied = -EFAULT; | |
1862 | break; | |
1863 | } | |
1da177e4 LT |
1864 | } |
1865 | } | |
1866 | ||
1867 | *seq += used; | |
1868 | copied += used; | |
1869 | len -= used; | |
1870 | ||
1871 | tcp_rcv_space_adjust(sk); | |
1872 | ||
1873 | skip_copy: | |
1874 | if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) { | |
1875 | tp->urg_data = 0; | |
9e412ba7 | 1876 | tcp_fast_path_check(sk); |
1da177e4 LT |
1877 | } |
1878 | if (used + offset < skb->len) | |
1879 | continue; | |
1880 | ||
aa8223c7 | 1881 | if (tcp_hdr(skb)->fin) |
1da177e4 | 1882 | goto found_fin_ok; |
1a2449a8 CL |
1883 | if (!(flags & MSG_PEEK)) { |
1884 | sk_eat_skb(sk, skb, copied_early); | |
dc6b9b78 | 1885 | copied_early = false; |
1a2449a8 | 1886 | } |
1da177e4 LT |
1887 | continue; |
1888 | ||
1889 | found_fin_ok: | |
1890 | /* Process the FIN. */ | |
1891 | ++*seq; | |
1a2449a8 CL |
1892 | if (!(flags & MSG_PEEK)) { |
1893 | sk_eat_skb(sk, skb, copied_early); | |
dc6b9b78 | 1894 | copied_early = false; |
1a2449a8 | 1895 | } |
1da177e4 LT |
1896 | break; |
1897 | } while (len > 0); | |
1898 | ||
1899 | if (user_recv) { | |
b03efcfb | 1900 | if (!skb_queue_empty(&tp->ucopy.prequeue)) { |
1da177e4 LT |
1901 | int chunk; |
1902 | ||
1903 | tp->ucopy.len = copied > 0 ? len : 0; | |
1904 | ||
1905 | tcp_prequeue_process(sk); | |
1906 | ||
1907 | if (copied > 0 && (chunk = len - tp->ucopy.len) != 0) { | |
ed88098e | 1908 | NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk); |
1da177e4 LT |
1909 | len -= chunk; |
1910 | copied += chunk; | |
1911 | } | |
1912 | } | |
1913 | ||
1914 | tp->ucopy.task = NULL; | |
1915 | tp->ucopy.len = 0; | |
1916 | } | |
1917 | ||
1a2449a8 | 1918 | #ifdef CONFIG_NET_DMA |
73852e81 SM |
1919 | tcp_service_net_dma(sk, true); /* Wait for queue to drain */ |
1920 | tp->ucopy.dma_chan = NULL; | |
1a2449a8 | 1921 | |
1a2449a8 CL |
1922 | if (tp->ucopy.pinned_list) { |
1923 | dma_unpin_iovec_pages(tp->ucopy.pinned_list); | |
1924 | tp->ucopy.pinned_list = NULL; | |
1925 | } | |
1926 | #endif | |
1927 | ||
1da177e4 LT |
1928 | /* According to UNIX98, msg_name/msg_namelen are ignored |
1929 | * on connected socket. I was just happy when found this 8) --ANK | |
1930 | */ | |
1931 | ||
1932 | /* Clean up data we have read: This will do ACK frames. */ | |
0e4b4992 | 1933 | tcp_cleanup_rbuf(sk, copied); |
1da177e4 | 1934 | |
1da177e4 LT |
1935 | release_sock(sk); |
1936 | return copied; | |
1937 | ||
1938 | out: | |
1da177e4 LT |
1939 | release_sock(sk); |
1940 | return err; | |
1941 | ||
1942 | recv_urg: | |
377f0a08 | 1943 | err = tcp_recv_urg(sk, msg, len, flags); |
1da177e4 | 1944 | goto out; |
c0e88ff0 PE |
1945 | |
1946 | recv_sndq: | |
1947 | err = tcp_peek_sndq(sk, msg, len); | |
1948 | goto out; | |
1da177e4 | 1949 | } |
4bc2f18b | 1950 | EXPORT_SYMBOL(tcp_recvmsg); |
1da177e4 | 1951 | |
490d5046 IJ |
1952 | void tcp_set_state(struct sock *sk, int state) |
1953 | { | |
1954 | int oldstate = sk->sk_state; | |
1955 | ||
1956 | switch (state) { | |
1957 | case TCP_ESTABLISHED: | |
1958 | if (oldstate != TCP_ESTABLISHED) | |
81cc8a75 | 1959 | TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB); |
490d5046 IJ |
1960 | break; |
1961 | ||
1962 | case TCP_CLOSE: | |
1963 | if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED) | |
81cc8a75 | 1964 | TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS); |
490d5046 IJ |
1965 | |
1966 | sk->sk_prot->unhash(sk); | |
1967 | if (inet_csk(sk)->icsk_bind_hash && | |
1968 | !(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) | |
ab1e0a13 | 1969 | inet_put_port(sk); |
490d5046 IJ |
1970 | /* fall through */ |
1971 | default: | |
5a5f3a8d | 1972 | if (oldstate == TCP_ESTABLISHED) |
74688e48 | 1973 | TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB); |
490d5046 IJ |
1974 | } |
1975 | ||
1976 | /* Change state AFTER socket is unhashed to avoid closed | |
1977 | * socket sitting in hash tables. | |
1978 | */ | |
1979 | sk->sk_state = state; | |
1980 | ||
1981 | #ifdef STATE_TRACE | |
5a5f3a8d | 1982 | SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n", sk, statename[oldstate], statename[state]); |
490d5046 IJ |
1983 | #endif |
1984 | } | |
1985 | EXPORT_SYMBOL_GPL(tcp_set_state); | |
1986 | ||
1da177e4 LT |
1987 | /* |
1988 | * State processing on a close. This implements the state shift for | |
1989 | * sending our FIN frame. Note that we only send a FIN for some | |
1990 | * states. A shutdown() may have already sent the FIN, or we may be | |
1991 | * closed. | |
1992 | */ | |
1993 | ||
9b5b5cff | 1994 | static const unsigned char new_state[16] = { |
1da177e4 LT |
1995 | /* current state: new state: action: */ |
1996 | /* (Invalid) */ TCP_CLOSE, | |
1997 | /* TCP_ESTABLISHED */ TCP_FIN_WAIT1 | TCP_ACTION_FIN, | |
1998 | /* TCP_SYN_SENT */ TCP_CLOSE, | |
1999 | /* TCP_SYN_RECV */ TCP_FIN_WAIT1 | TCP_ACTION_FIN, | |
2000 | /* TCP_FIN_WAIT1 */ TCP_FIN_WAIT1, | |
2001 | /* TCP_FIN_WAIT2 */ TCP_FIN_WAIT2, | |
2002 | /* TCP_TIME_WAIT */ TCP_CLOSE, | |
2003 | /* TCP_CLOSE */ TCP_CLOSE, | |
2004 | /* TCP_CLOSE_WAIT */ TCP_LAST_ACK | TCP_ACTION_FIN, | |
2005 | /* TCP_LAST_ACK */ TCP_LAST_ACK, | |
2006 | /* TCP_LISTEN */ TCP_CLOSE, | |
2007 | /* TCP_CLOSING */ TCP_CLOSING, | |
2008 | }; | |
2009 | ||
2010 | static int tcp_close_state(struct sock *sk) | |
2011 | { | |
2012 | int next = (int)new_state[sk->sk_state]; | |
2013 | int ns = next & TCP_STATE_MASK; | |
2014 | ||
2015 | tcp_set_state(sk, ns); | |
2016 | ||
2017 | return next & TCP_ACTION_FIN; | |
2018 | } | |
2019 | ||
2020 | /* | |
2021 | * Shutdown the sending side of a connection. Much like close except | |
1f29b058 | 2022 | * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD). |
1da177e4 LT |
2023 | */ |
2024 | ||
2025 | void tcp_shutdown(struct sock *sk, int how) | |
2026 | { | |
2027 | /* We need to grab some memory, and put together a FIN, | |
2028 | * and then put it into the queue to be sent. | |
2029 | * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92. | |
2030 | */ | |
2031 | if (!(how & SEND_SHUTDOWN)) | |
2032 | return; | |
2033 | ||
2034 | /* If we've already sent a FIN, or it's a closed state, skip this. */ | |
2035 | if ((1 << sk->sk_state) & | |
2036 | (TCPF_ESTABLISHED | TCPF_SYN_SENT | | |
2037 | TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) { | |
2038 | /* Clear out any half completed packets. FIN if needed. */ | |
2039 | if (tcp_close_state(sk)) | |
2040 | tcp_send_fin(sk); | |
2041 | } | |
2042 | } | |
4bc2f18b | 2043 | EXPORT_SYMBOL(tcp_shutdown); |
1da177e4 | 2044 | |
efcdbf24 AS |
2045 | bool tcp_check_oom(struct sock *sk, int shift) |
2046 | { | |
2047 | bool too_many_orphans, out_of_socket_memory; | |
2048 | ||
2049 | too_many_orphans = tcp_too_many_orphans(sk, shift); | |
2050 | out_of_socket_memory = tcp_out_of_memory(sk); | |
2051 | ||
e87cc472 JP |
2052 | if (too_many_orphans) |
2053 | net_info_ratelimited("too many orphaned sockets\n"); | |
2054 | if (out_of_socket_memory) | |
2055 | net_info_ratelimited("out of memory -- consider tuning tcp_mem\n"); | |
efcdbf24 AS |
2056 | return too_many_orphans || out_of_socket_memory; |
2057 | } | |
2058 | ||
1da177e4 LT |
2059 | void tcp_close(struct sock *sk, long timeout) |
2060 | { | |
2061 | struct sk_buff *skb; | |
2062 | int data_was_unread = 0; | |
75c2d907 | 2063 | int state; |
1da177e4 LT |
2064 | |
2065 | lock_sock(sk); | |
2066 | sk->sk_shutdown = SHUTDOWN_MASK; | |
2067 | ||
2068 | if (sk->sk_state == TCP_LISTEN) { | |
2069 | tcp_set_state(sk, TCP_CLOSE); | |
2070 | ||
2071 | /* Special case. */ | |
0a5578cf | 2072 | inet_csk_listen_stop(sk); |
1da177e4 LT |
2073 | |
2074 | goto adjudge_to_death; | |
2075 | } | |
2076 | ||
2077 | /* We need to flush the recv. buffs. We do this only on the | |
2078 | * descriptor close, not protocol-sourced closes, because the | |
2079 | * reader process may not have drained the data yet! | |
2080 | */ | |
2081 | while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) { | |
2082 | u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq - | |
aa8223c7 | 2083 | tcp_hdr(skb)->fin; |
1da177e4 LT |
2084 | data_was_unread += len; |
2085 | __kfree_skb(skb); | |
2086 | } | |
2087 | ||
3ab224be | 2088 | sk_mem_reclaim(sk); |
1da177e4 | 2089 | |
565b7b2d KK |
2090 | /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */ |
2091 | if (sk->sk_state == TCP_CLOSE) | |
2092 | goto adjudge_to_death; | |
2093 | ||
65bb723c GR |
2094 | /* As outlined in RFC 2525, section 2.17, we send a RST here because |
2095 | * data was lost. To witness the awful effects of the old behavior of | |
2096 | * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk | |
2097 | * GET in an FTP client, suspend the process, wait for the client to | |
2098 | * advertise a zero window, then kill -9 the FTP client, wheee... | |
2099 | * Note: timeout is always zero in such a case. | |
1da177e4 | 2100 | */ |
ee995283 PE |
2101 | if (unlikely(tcp_sk(sk)->repair)) { |
2102 | sk->sk_prot->disconnect(sk, 0); | |
2103 | } else if (data_was_unread) { | |
1da177e4 | 2104 | /* Unread data was tossed, zap the connection. */ |
6f67c817 | 2105 | NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE); |
1da177e4 | 2106 | tcp_set_state(sk, TCP_CLOSE); |
aa133076 | 2107 | tcp_send_active_reset(sk, sk->sk_allocation); |
1da177e4 LT |
2108 | } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) { |
2109 | /* Check zero linger _after_ checking for unread data. */ | |
2110 | sk->sk_prot->disconnect(sk, 0); | |
6f67c817 | 2111 | NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONDATA); |
1da177e4 LT |
2112 | } else if (tcp_close_state(sk)) { |
2113 | /* We FIN if the application ate all the data before | |
2114 | * zapping the connection. | |
2115 | */ | |
2116 | ||
2117 | /* RED-PEN. Formally speaking, we have broken TCP state | |
2118 | * machine. State transitions: | |
2119 | * | |
2120 | * TCP_ESTABLISHED -> TCP_FIN_WAIT1 | |
2121 | * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible) | |
2122 | * TCP_CLOSE_WAIT -> TCP_LAST_ACK | |
2123 | * | |
2124 | * are legal only when FIN has been sent (i.e. in window), | |
2125 | * rather than queued out of window. Purists blame. | |
2126 | * | |
2127 | * F.e. "RFC state" is ESTABLISHED, | |
2128 | * if Linux state is FIN-WAIT-1, but FIN is still not sent. | |
2129 | * | |
2130 | * The visible declinations are that sometimes | |
2131 | * we enter time-wait state, when it is not required really | |
2132 | * (harmless), do not send active resets, when they are | |
2133 | * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when | |
2134 | * they look as CLOSING or LAST_ACK for Linux) | |
2135 | * Probably, I missed some more holelets. | |
2136 | * --ANK | |
8336886f JC |
2137 | * XXX (TFO) - To start off we don't support SYN+ACK+FIN |
2138 | * in a single packet! (May consider it later but will | |
2139 | * probably need API support or TCP_CORK SYN-ACK until | |
2140 | * data is written and socket is closed.) | |
1da177e4 LT |
2141 | */ |
2142 | tcp_send_fin(sk); | |
2143 | } | |
2144 | ||
2145 | sk_stream_wait_close(sk, timeout); | |
2146 | ||
2147 | adjudge_to_death: | |
75c2d907 HX |
2148 | state = sk->sk_state; |
2149 | sock_hold(sk); | |
2150 | sock_orphan(sk); | |
75c2d907 | 2151 | |
1da177e4 LT |
2152 | /* It is the last release_sock in its life. It will remove backlog. */ |
2153 | release_sock(sk); | |
2154 | ||
2155 | ||
2156 | /* Now socket is owned by kernel and we acquire BH lock | |
2157 | to finish close. No need to check for user refs. | |
2158 | */ | |
2159 | local_bh_disable(); | |
2160 | bh_lock_sock(sk); | |
547b792c | 2161 | WARN_ON(sock_owned_by_user(sk)); |
1da177e4 | 2162 | |
eb4dea58 HX |
2163 | percpu_counter_inc(sk->sk_prot->orphan_count); |
2164 | ||
75c2d907 HX |
2165 | /* Have we already been destroyed by a softirq or backlog? */ |
2166 | if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE) | |
2167 | goto out; | |
1da177e4 LT |
2168 | |
2169 | /* This is a (useful) BSD violating of the RFC. There is a | |
2170 | * problem with TCP as specified in that the other end could | |
2171 | * keep a socket open forever with no application left this end. | |
2172 | * We use a 3 minute timeout (about the same as BSD) then kill | |
2173 | * our end. If they send after that then tough - BUT: long enough | |
2174 | * that we won't make the old 4*rto = almost no time - whoops | |
2175 | * reset mistake. | |
2176 | * | |
2177 | * Nope, it was not mistake. It is really desired behaviour | |
2178 | * f.e. on http servers, when such sockets are useless, but | |
2179 | * consume significant resources. Let's do it with special | |
2180 | * linger2 option. --ANK | |
2181 | */ | |
2182 | ||
2183 | if (sk->sk_state == TCP_FIN_WAIT2) { | |
2184 | struct tcp_sock *tp = tcp_sk(sk); | |
2185 | if (tp->linger2 < 0) { | |
2186 | tcp_set_state(sk, TCP_CLOSE); | |
2187 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
de0744af PE |
2188 | NET_INC_STATS_BH(sock_net(sk), |
2189 | LINUX_MIB_TCPABORTONLINGER); | |
1da177e4 | 2190 | } else { |
463c84b9 | 2191 | const int tmo = tcp_fin_time(sk); |
1da177e4 LT |
2192 | |
2193 | if (tmo > TCP_TIMEWAIT_LEN) { | |
52499afe DM |
2194 | inet_csk_reset_keepalive_timer(sk, |
2195 | tmo - TCP_TIMEWAIT_LEN); | |
1da177e4 | 2196 | } else { |
1da177e4 LT |
2197 | tcp_time_wait(sk, TCP_FIN_WAIT2, tmo); |
2198 | goto out; | |
2199 | } | |
2200 | } | |
2201 | } | |
2202 | if (sk->sk_state != TCP_CLOSE) { | |
3ab224be | 2203 | sk_mem_reclaim(sk); |
efcdbf24 | 2204 | if (tcp_check_oom(sk, 0)) { |
1da177e4 LT |
2205 | tcp_set_state(sk, TCP_CLOSE); |
2206 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
de0744af PE |
2207 | NET_INC_STATS_BH(sock_net(sk), |
2208 | LINUX_MIB_TCPABORTONMEMORY); | |
1da177e4 LT |
2209 | } |
2210 | } | |
1da177e4 | 2211 | |
8336886f JC |
2212 | if (sk->sk_state == TCP_CLOSE) { |
2213 | struct request_sock *req = tcp_sk(sk)->fastopen_rsk; | |
2214 | /* We could get here with a non-NULL req if the socket is | |
2215 | * aborted (e.g., closed with unread data) before 3WHS | |
2216 | * finishes. | |
2217 | */ | |
2218 | if (req != NULL) | |
2219 | reqsk_fastopen_remove(sk, req, false); | |
0a5578cf | 2220 | inet_csk_destroy_sock(sk); |
8336886f | 2221 | } |
1da177e4 LT |
2222 | /* Otherwise, socket is reprieved until protocol close. */ |
2223 | ||
2224 | out: | |
2225 | bh_unlock_sock(sk); | |
2226 | local_bh_enable(); | |
2227 | sock_put(sk); | |
2228 | } | |
4bc2f18b | 2229 | EXPORT_SYMBOL(tcp_close); |
1da177e4 LT |
2230 | |
2231 | /* These states need RST on ABORT according to RFC793 */ | |
2232 | ||
a2a385d6 | 2233 | static inline bool tcp_need_reset(int state) |
1da177e4 LT |
2234 | { |
2235 | return (1 << state) & | |
2236 | (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 | | |
2237 | TCPF_FIN_WAIT2 | TCPF_SYN_RECV); | |
2238 | } | |
2239 | ||
2240 | int tcp_disconnect(struct sock *sk, int flags) | |
2241 | { | |
2242 | struct inet_sock *inet = inet_sk(sk); | |
463c84b9 | 2243 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
2244 | struct tcp_sock *tp = tcp_sk(sk); |
2245 | int err = 0; | |
2246 | int old_state = sk->sk_state; | |
2247 | ||
2248 | if (old_state != TCP_CLOSE) | |
2249 | tcp_set_state(sk, TCP_CLOSE); | |
2250 | ||
2251 | /* ABORT function of RFC793 */ | |
2252 | if (old_state == TCP_LISTEN) { | |
0a5578cf | 2253 | inet_csk_listen_stop(sk); |
ee995283 PE |
2254 | } else if (unlikely(tp->repair)) { |
2255 | sk->sk_err = ECONNABORTED; | |
1da177e4 LT |
2256 | } else if (tcp_need_reset(old_state) || |
2257 | (tp->snd_nxt != tp->write_seq && | |
2258 | (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) { | |
caa20d9a | 2259 | /* The last check adjusts for discrepancy of Linux wrt. RFC |
1da177e4 LT |
2260 | * states |
2261 | */ | |
2262 | tcp_send_active_reset(sk, gfp_any()); | |
2263 | sk->sk_err = ECONNRESET; | |
2264 | } else if (old_state == TCP_SYN_SENT) | |
2265 | sk->sk_err = ECONNRESET; | |
2266 | ||
2267 | tcp_clear_xmit_timers(sk); | |
2268 | __skb_queue_purge(&sk->sk_receive_queue); | |
fe067e8a | 2269 | tcp_write_queue_purge(sk); |
1da177e4 | 2270 | __skb_queue_purge(&tp->out_of_order_queue); |
1a2449a8 CL |
2271 | #ifdef CONFIG_NET_DMA |
2272 | __skb_queue_purge(&sk->sk_async_wait_queue); | |
2273 | #endif | |
1da177e4 | 2274 | |
c720c7e8 | 2275 | inet->inet_dport = 0; |
1da177e4 LT |
2276 | |
2277 | if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) | |
2278 | inet_reset_saddr(sk); | |
2279 | ||
2280 | sk->sk_shutdown = 0; | |
2281 | sock_reset_flag(sk, SOCK_DONE); | |
2282 | tp->srtt = 0; | |
2283 | if ((tp->write_seq += tp->max_window + 2) == 0) | |
2284 | tp->write_seq = 1; | |
463c84b9 | 2285 | icsk->icsk_backoff = 0; |
1da177e4 | 2286 | tp->snd_cwnd = 2; |
6687e988 | 2287 | icsk->icsk_probes_out = 0; |
1da177e4 | 2288 | tp->packets_out = 0; |
0b6a05c1 | 2289 | tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; |
1da177e4 | 2290 | tp->snd_cwnd_cnt = 0; |
1fdf475a | 2291 | tp->window_clamp = 0; |
6687e988 | 2292 | tcp_set_ca_state(sk, TCP_CA_Open); |
1da177e4 | 2293 | tcp_clear_retrans(tp); |
463c84b9 | 2294 | inet_csk_delack_init(sk); |
fe067e8a | 2295 | tcp_init_send_head(sk); |
b40b4f79 | 2296 | memset(&tp->rx_opt, 0, sizeof(tp->rx_opt)); |
1da177e4 LT |
2297 | __sk_dst_reset(sk); |
2298 | ||
c720c7e8 | 2299 | WARN_ON(inet->inet_num && !icsk->icsk_bind_hash); |
1da177e4 LT |
2300 | |
2301 | sk->sk_error_report(sk); | |
2302 | return err; | |
2303 | } | |
4bc2f18b | 2304 | EXPORT_SYMBOL(tcp_disconnect); |
1da177e4 | 2305 | |
bb68b647 CP |
2306 | void tcp_sock_destruct(struct sock *sk) |
2307 | { | |
2308 | inet_sock_destruct(sk); | |
2309 | ||
2310 | kfree(inet_csk(sk)->icsk_accept_queue.fastopenq); | |
2311 | } | |
2312 | ||
a2a385d6 | 2313 | static inline bool tcp_can_repair_sock(const struct sock *sk) |
ee995283 | 2314 | { |
52e804c6 | 2315 | return ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN) && |
ee995283 PE |
2316 | ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_ESTABLISHED)); |
2317 | } | |
2318 | ||
de248a75 PE |
2319 | static int tcp_repair_options_est(struct tcp_sock *tp, |
2320 | struct tcp_repair_opt __user *optbuf, unsigned int len) | |
b139ba4e | 2321 | { |
de248a75 | 2322 | struct tcp_repair_opt opt; |
b139ba4e | 2323 | |
de248a75 PE |
2324 | while (len >= sizeof(opt)) { |
2325 | if (copy_from_user(&opt, optbuf, sizeof(opt))) | |
b139ba4e PE |
2326 | return -EFAULT; |
2327 | ||
2328 | optbuf++; | |
de248a75 | 2329 | len -= sizeof(opt); |
b139ba4e | 2330 | |
de248a75 PE |
2331 | switch (opt.opt_code) { |
2332 | case TCPOPT_MSS: | |
2333 | tp->rx_opt.mss_clamp = opt.opt_val; | |
b139ba4e | 2334 | break; |
de248a75 | 2335 | case TCPOPT_WINDOW: |
bc26ccd8 AV |
2336 | { |
2337 | u16 snd_wscale = opt.opt_val & 0xFFFF; | |
2338 | u16 rcv_wscale = opt.opt_val >> 16; | |
2339 | ||
2340 | if (snd_wscale > 14 || rcv_wscale > 14) | |
2341 | return -EFBIG; | |
b139ba4e | 2342 | |
bc26ccd8 AV |
2343 | tp->rx_opt.snd_wscale = snd_wscale; |
2344 | tp->rx_opt.rcv_wscale = rcv_wscale; | |
2345 | tp->rx_opt.wscale_ok = 1; | |
2346 | } | |
b139ba4e | 2347 | break; |
b139ba4e | 2348 | case TCPOPT_SACK_PERM: |
de248a75 PE |
2349 | if (opt.opt_val != 0) |
2350 | return -EINVAL; | |
2351 | ||
b139ba4e PE |
2352 | tp->rx_opt.sack_ok |= TCP_SACK_SEEN; |
2353 | if (sysctl_tcp_fack) | |
2354 | tcp_enable_fack(tp); | |
2355 | break; | |
2356 | case TCPOPT_TIMESTAMP: | |
de248a75 PE |
2357 | if (opt.opt_val != 0) |
2358 | return -EINVAL; | |
2359 | ||
b139ba4e PE |
2360 | tp->rx_opt.tstamp_ok = 1; |
2361 | break; | |
2362 | } | |
2363 | } | |
2364 | ||
2365 | return 0; | |
2366 | } | |
2367 | ||
1da177e4 LT |
2368 | /* |
2369 | * Socket option code for TCP. | |
2370 | */ | |
3fdadf7d | 2371 | static int do_tcp_setsockopt(struct sock *sk, int level, |
b7058842 | 2372 | int optname, char __user *optval, unsigned int optlen) |
1da177e4 LT |
2373 | { |
2374 | struct tcp_sock *tp = tcp_sk(sk); | |
463c84b9 | 2375 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
2376 | int val; |
2377 | int err = 0; | |
2378 | ||
e56fb50f WAS |
2379 | /* These are data/string values, all the others are ints */ |
2380 | switch (optname) { | |
2381 | case TCP_CONGESTION: { | |
5f8ef48d SH |
2382 | char name[TCP_CA_NAME_MAX]; |
2383 | ||
2384 | if (optlen < 1) | |
2385 | return -EINVAL; | |
2386 | ||
2387 | val = strncpy_from_user(name, optval, | |
4fdb78d3 | 2388 | min_t(long, TCP_CA_NAME_MAX-1, optlen)); |
5f8ef48d SH |
2389 | if (val < 0) |
2390 | return -EFAULT; | |
2391 | name[val] = 0; | |
2392 | ||
2393 | lock_sock(sk); | |
6687e988 | 2394 | err = tcp_set_congestion_control(sk, name); |
5f8ef48d SH |
2395 | release_sock(sk); |
2396 | return err; | |
2397 | } | |
e56fb50f WAS |
2398 | default: |
2399 | /* fallthru */ | |
2400 | break; | |
ccbd6a5a | 2401 | } |
5f8ef48d | 2402 | |
1da177e4 LT |
2403 | if (optlen < sizeof(int)) |
2404 | return -EINVAL; | |
2405 | ||
2406 | if (get_user(val, (int __user *)optval)) | |
2407 | return -EFAULT; | |
2408 | ||
2409 | lock_sock(sk); | |
2410 | ||
2411 | switch (optname) { | |
2412 | case TCP_MAXSEG: | |
2413 | /* Values greater than interface MTU won't take effect. However | |
2414 | * at the point when this call is done we typically don't yet | |
2415 | * know which interface is going to be used */ | |
c39508d6 | 2416 | if (val < TCP_MIN_MSS || val > MAX_TCP_WINDOW) { |
1da177e4 LT |
2417 | err = -EINVAL; |
2418 | break; | |
2419 | } | |
2420 | tp->rx_opt.user_mss = val; | |
2421 | break; | |
2422 | ||
2423 | case TCP_NODELAY: | |
2424 | if (val) { | |
2425 | /* TCP_NODELAY is weaker than TCP_CORK, so that | |
2426 | * this option on corked socket is remembered, but | |
2427 | * it is not activated until cork is cleared. | |
2428 | * | |
2429 | * However, when TCP_NODELAY is set we make | |
2430 | * an explicit push, which overrides even TCP_CORK | |
2431 | * for currently queued segments. | |
2432 | */ | |
2433 | tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH; | |
9e412ba7 | 2434 | tcp_push_pending_frames(sk); |
1da177e4 LT |
2435 | } else { |
2436 | tp->nonagle &= ~TCP_NAGLE_OFF; | |
2437 | } | |
2438 | break; | |
2439 | ||
36e31b0a AP |
2440 | case TCP_THIN_LINEAR_TIMEOUTS: |
2441 | if (val < 0 || val > 1) | |
2442 | err = -EINVAL; | |
2443 | else | |
2444 | tp->thin_lto = val; | |
2445 | break; | |
2446 | ||
7e380175 AP |
2447 | case TCP_THIN_DUPACK: |
2448 | if (val < 0 || val > 1) | |
2449 | err = -EINVAL; | |
a4ae4c61 | 2450 | else { |
7e380175 | 2451 | tp->thin_dupack = val; |
eed530b6 YC |
2452 | if (tp->thin_dupack) |
2453 | tcp_disable_early_retrans(tp); | |
a4ae4c61 | 2454 | } |
7e380175 AP |
2455 | break; |
2456 | ||
ee995283 PE |
2457 | case TCP_REPAIR: |
2458 | if (!tcp_can_repair_sock(sk)) | |
2459 | err = -EPERM; | |
2460 | else if (val == 1) { | |
2461 | tp->repair = 1; | |
2462 | sk->sk_reuse = SK_FORCE_REUSE; | |
2463 | tp->repair_queue = TCP_NO_QUEUE; | |
2464 | } else if (val == 0) { | |
2465 | tp->repair = 0; | |
2466 | sk->sk_reuse = SK_NO_REUSE; | |
2467 | tcp_send_window_probe(sk); | |
2468 | } else | |
2469 | err = -EINVAL; | |
2470 | ||
2471 | break; | |
2472 | ||
2473 | case TCP_REPAIR_QUEUE: | |
2474 | if (!tp->repair) | |
2475 | err = -EPERM; | |
2476 | else if (val < TCP_QUEUES_NR) | |
2477 | tp->repair_queue = val; | |
2478 | else | |
2479 | err = -EINVAL; | |
2480 | break; | |
2481 | ||
2482 | case TCP_QUEUE_SEQ: | |
2483 | if (sk->sk_state != TCP_CLOSE) | |
2484 | err = -EPERM; | |
2485 | else if (tp->repair_queue == TCP_SEND_QUEUE) | |
2486 | tp->write_seq = val; | |
2487 | else if (tp->repair_queue == TCP_RECV_QUEUE) | |
2488 | tp->rcv_nxt = val; | |
2489 | else | |
2490 | err = -EINVAL; | |
2491 | break; | |
2492 | ||
b139ba4e PE |
2493 | case TCP_REPAIR_OPTIONS: |
2494 | if (!tp->repair) | |
2495 | err = -EINVAL; | |
2496 | else if (sk->sk_state == TCP_ESTABLISHED) | |
de248a75 PE |
2497 | err = tcp_repair_options_est(tp, |
2498 | (struct tcp_repair_opt __user *)optval, | |
2499 | optlen); | |
b139ba4e PE |
2500 | else |
2501 | err = -EPERM; | |
2502 | break; | |
2503 | ||
1da177e4 LT |
2504 | case TCP_CORK: |
2505 | /* When set indicates to always queue non-full frames. | |
2506 | * Later the user clears this option and we transmit | |
2507 | * any pending partial frames in the queue. This is | |
2508 | * meant to be used alongside sendfile() to get properly | |
2509 | * filled frames when the user (for example) must write | |
2510 | * out headers with a write() call first and then use | |
2511 | * sendfile to send out the data parts. | |
2512 | * | |
2513 | * TCP_CORK can be set together with TCP_NODELAY and it is | |
2514 | * stronger than TCP_NODELAY. | |
2515 | */ | |
2516 | if (val) { | |
2517 | tp->nonagle |= TCP_NAGLE_CORK; | |
2518 | } else { | |
2519 | tp->nonagle &= ~TCP_NAGLE_CORK; | |
2520 | if (tp->nonagle&TCP_NAGLE_OFF) | |
2521 | tp->nonagle |= TCP_NAGLE_PUSH; | |
9e412ba7 | 2522 | tcp_push_pending_frames(sk); |
1da177e4 LT |
2523 | } |
2524 | break; | |
2525 | ||
2526 | case TCP_KEEPIDLE: | |
2527 | if (val < 1 || val > MAX_TCP_KEEPIDLE) | |
2528 | err = -EINVAL; | |
2529 | else { | |
2530 | tp->keepalive_time = val * HZ; | |
2531 | if (sock_flag(sk, SOCK_KEEPOPEN) && | |
2532 | !((1 << sk->sk_state) & | |
2533 | (TCPF_CLOSE | TCPF_LISTEN))) { | |
6c37e5de | 2534 | u32 elapsed = keepalive_time_elapsed(tp); |
1da177e4 LT |
2535 | if (tp->keepalive_time > elapsed) |
2536 | elapsed = tp->keepalive_time - elapsed; | |
2537 | else | |
2538 | elapsed = 0; | |
463c84b9 | 2539 | inet_csk_reset_keepalive_timer(sk, elapsed); |
1da177e4 LT |
2540 | } |
2541 | } | |
2542 | break; | |
2543 | case TCP_KEEPINTVL: | |
2544 | if (val < 1 || val > MAX_TCP_KEEPINTVL) | |
2545 | err = -EINVAL; | |
2546 | else | |
2547 | tp->keepalive_intvl = val * HZ; | |
2548 | break; | |
2549 | case TCP_KEEPCNT: | |
2550 | if (val < 1 || val > MAX_TCP_KEEPCNT) | |
2551 | err = -EINVAL; | |
2552 | else | |
2553 | tp->keepalive_probes = val; | |
2554 | break; | |
2555 | case TCP_SYNCNT: | |
2556 | if (val < 1 || val > MAX_TCP_SYNCNT) | |
2557 | err = -EINVAL; | |
2558 | else | |
463c84b9 | 2559 | icsk->icsk_syn_retries = val; |
1da177e4 LT |
2560 | break; |
2561 | ||
2562 | case TCP_LINGER2: | |
2563 | if (val < 0) | |
2564 | tp->linger2 = -1; | |
2565 | else if (val > sysctl_tcp_fin_timeout / HZ) | |
2566 | tp->linger2 = 0; | |
2567 | else | |
2568 | tp->linger2 = val * HZ; | |
2569 | break; | |
2570 | ||
2571 | case TCP_DEFER_ACCEPT: | |
b103cf34 JA |
2572 | /* Translate value in seconds to number of retransmits */ |
2573 | icsk->icsk_accept_queue.rskq_defer_accept = | |
2574 | secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ, | |
2575 | TCP_RTO_MAX / HZ); | |
1da177e4 LT |
2576 | break; |
2577 | ||
2578 | case TCP_WINDOW_CLAMP: | |
2579 | if (!val) { | |
2580 | if (sk->sk_state != TCP_CLOSE) { | |
2581 | err = -EINVAL; | |
2582 | break; | |
2583 | } | |
2584 | tp->window_clamp = 0; | |
2585 | } else | |
2586 | tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ? | |
2587 | SOCK_MIN_RCVBUF / 2 : val; | |
2588 | break; | |
2589 | ||
2590 | case TCP_QUICKACK: | |
2591 | if (!val) { | |
463c84b9 | 2592 | icsk->icsk_ack.pingpong = 1; |
1da177e4 | 2593 | } else { |
463c84b9 | 2594 | icsk->icsk_ack.pingpong = 0; |
1da177e4 LT |
2595 | if ((1 << sk->sk_state) & |
2596 | (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) && | |
463c84b9 ACM |
2597 | inet_csk_ack_scheduled(sk)) { |
2598 | icsk->icsk_ack.pending |= ICSK_ACK_PUSHED; | |
0e4b4992 | 2599 | tcp_cleanup_rbuf(sk, 1); |
1da177e4 | 2600 | if (!(val & 1)) |
463c84b9 | 2601 | icsk->icsk_ack.pingpong = 1; |
1da177e4 LT |
2602 | } |
2603 | } | |
2604 | break; | |
2605 | ||
cfb6eeb4 YH |
2606 | #ifdef CONFIG_TCP_MD5SIG |
2607 | case TCP_MD5SIG: | |
2608 | /* Read the IP->Key mappings from userspace */ | |
2609 | err = tp->af_specific->md5_parse(sk, optval, optlen); | |
2610 | break; | |
2611 | #endif | |
dca43c75 JC |
2612 | case TCP_USER_TIMEOUT: |
2613 | /* Cap the max timeout in ms TCP will retry/retrans | |
2614 | * before giving up and aborting (ETIMEDOUT) a connection. | |
2615 | */ | |
42493570 HL |
2616 | if (val < 0) |
2617 | err = -EINVAL; | |
2618 | else | |
2619 | icsk->icsk_user_timeout = msecs_to_jiffies(val); | |
dca43c75 | 2620 | break; |
8336886f JC |
2621 | |
2622 | case TCP_FASTOPEN: | |
2623 | if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE | | |
2624 | TCPF_LISTEN))) | |
2625 | err = fastopen_init_queue(sk, val); | |
2626 | else | |
2627 | err = -EINVAL; | |
2628 | break; | |
93be6ce0 AV |
2629 | case TCP_TIMESTAMP: |
2630 | if (!tp->repair) | |
2631 | err = -EPERM; | |
2632 | else | |
2633 | tp->tsoffset = val - tcp_time_stamp; | |
2634 | break; | |
1da177e4 LT |
2635 | default: |
2636 | err = -ENOPROTOOPT; | |
2637 | break; | |
3ff50b79 SH |
2638 | } |
2639 | ||
1da177e4 LT |
2640 | release_sock(sk); |
2641 | return err; | |
2642 | } | |
2643 | ||
3fdadf7d | 2644 | int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval, |
b7058842 | 2645 | unsigned int optlen) |
3fdadf7d | 2646 | { |
cf533ea5 | 2647 | const struct inet_connection_sock *icsk = inet_csk(sk); |
3fdadf7d DM |
2648 | |
2649 | if (level != SOL_TCP) | |
2650 | return icsk->icsk_af_ops->setsockopt(sk, level, optname, | |
2651 | optval, optlen); | |
2652 | return do_tcp_setsockopt(sk, level, optname, optval, optlen); | |
2653 | } | |
4bc2f18b | 2654 | EXPORT_SYMBOL(tcp_setsockopt); |
3fdadf7d DM |
2655 | |
2656 | #ifdef CONFIG_COMPAT | |
543d9cfe | 2657 | int compat_tcp_setsockopt(struct sock *sk, int level, int optname, |
b7058842 | 2658 | char __user *optval, unsigned int optlen) |
3fdadf7d | 2659 | { |
dec73ff0 ACM |
2660 | if (level != SOL_TCP) |
2661 | return inet_csk_compat_setsockopt(sk, level, optname, | |
2662 | optval, optlen); | |
3fdadf7d DM |
2663 | return do_tcp_setsockopt(sk, level, optname, optval, optlen); |
2664 | } | |
543d9cfe | 2665 | EXPORT_SYMBOL(compat_tcp_setsockopt); |
3fdadf7d DM |
2666 | #endif |
2667 | ||
1da177e4 | 2668 | /* Return information about state of tcp endpoint in API format. */ |
cf533ea5 | 2669 | void tcp_get_info(const struct sock *sk, struct tcp_info *info) |
1da177e4 | 2670 | { |
cf533ea5 | 2671 | const struct tcp_sock *tp = tcp_sk(sk); |
463c84b9 | 2672 | const struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
2673 | u32 now = tcp_time_stamp; |
2674 | ||
2675 | memset(info, 0, sizeof(*info)); | |
2676 | ||
2677 | info->tcpi_state = sk->sk_state; | |
6687e988 | 2678 | info->tcpi_ca_state = icsk->icsk_ca_state; |
463c84b9 | 2679 | info->tcpi_retransmits = icsk->icsk_retransmits; |
6687e988 | 2680 | info->tcpi_probes = icsk->icsk_probes_out; |
463c84b9 | 2681 | info->tcpi_backoff = icsk->icsk_backoff; |
1da177e4 LT |
2682 | |
2683 | if (tp->rx_opt.tstamp_ok) | |
2684 | info->tcpi_options |= TCPI_OPT_TIMESTAMPS; | |
e60402d0 | 2685 | if (tcp_is_sack(tp)) |
1da177e4 LT |
2686 | info->tcpi_options |= TCPI_OPT_SACK; |
2687 | if (tp->rx_opt.wscale_ok) { | |
2688 | info->tcpi_options |= TCPI_OPT_WSCALE; | |
2689 | info->tcpi_snd_wscale = tp->rx_opt.snd_wscale; | |
2690 | info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale; | |
e905a9ed | 2691 | } |
1da177e4 | 2692 | |
b5c5693b | 2693 | if (tp->ecn_flags & TCP_ECN_OK) |
1da177e4 | 2694 | info->tcpi_options |= TCPI_OPT_ECN; |
b5c5693b ED |
2695 | if (tp->ecn_flags & TCP_ECN_SEEN) |
2696 | info->tcpi_options |= TCPI_OPT_ECN_SEEN; | |
6f73601e YC |
2697 | if (tp->syn_data_acked) |
2698 | info->tcpi_options |= TCPI_OPT_SYN_DATA; | |
1da177e4 | 2699 | |
463c84b9 ACM |
2700 | info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto); |
2701 | info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato); | |
c1b4a7e6 | 2702 | info->tcpi_snd_mss = tp->mss_cache; |
463c84b9 | 2703 | info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss; |
1da177e4 | 2704 | |
5ee3afba RJ |
2705 | if (sk->sk_state == TCP_LISTEN) { |
2706 | info->tcpi_unacked = sk->sk_ack_backlog; | |
2707 | info->tcpi_sacked = sk->sk_max_ack_backlog; | |
2708 | } else { | |
2709 | info->tcpi_unacked = tp->packets_out; | |
2710 | info->tcpi_sacked = tp->sacked_out; | |
2711 | } | |
1da177e4 LT |
2712 | info->tcpi_lost = tp->lost_out; |
2713 | info->tcpi_retrans = tp->retrans_out; | |
2714 | info->tcpi_fackets = tp->fackets_out; | |
2715 | ||
2716 | info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime); | |
463c84b9 | 2717 | info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime); |
1da177e4 LT |
2718 | info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp); |
2719 | ||
d83d8461 | 2720 | info->tcpi_pmtu = icsk->icsk_pmtu_cookie; |
1da177e4 LT |
2721 | info->tcpi_rcv_ssthresh = tp->rcv_ssthresh; |
2722 | info->tcpi_rtt = jiffies_to_usecs(tp->srtt)>>3; | |
2723 | info->tcpi_rttvar = jiffies_to_usecs(tp->mdev)>>2; | |
2724 | info->tcpi_snd_ssthresh = tp->snd_ssthresh; | |
2725 | info->tcpi_snd_cwnd = tp->snd_cwnd; | |
2726 | info->tcpi_advmss = tp->advmss; | |
2727 | info->tcpi_reordering = tp->reordering; | |
2728 | ||
2729 | info->tcpi_rcv_rtt = jiffies_to_usecs(tp->rcv_rtt_est.rtt)>>3; | |
2730 | info->tcpi_rcv_space = tp->rcvq_space.space; | |
2731 | ||
2732 | info->tcpi_total_retrans = tp->total_retrans; | |
2733 | } | |
1da177e4 LT |
2734 | EXPORT_SYMBOL_GPL(tcp_get_info); |
2735 | ||
3fdadf7d DM |
2736 | static int do_tcp_getsockopt(struct sock *sk, int level, |
2737 | int optname, char __user *optval, int __user *optlen) | |
1da177e4 | 2738 | { |
295f7324 | 2739 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
2740 | struct tcp_sock *tp = tcp_sk(sk); |
2741 | int val, len; | |
2742 | ||
1da177e4 LT |
2743 | if (get_user(len, optlen)) |
2744 | return -EFAULT; | |
2745 | ||
2746 | len = min_t(unsigned int, len, sizeof(int)); | |
2747 | ||
2748 | if (len < 0) | |
2749 | return -EINVAL; | |
2750 | ||
2751 | switch (optname) { | |
2752 | case TCP_MAXSEG: | |
c1b4a7e6 | 2753 | val = tp->mss_cache; |
1da177e4 LT |
2754 | if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) |
2755 | val = tp->rx_opt.user_mss; | |
5e6a3ce6 PE |
2756 | if (tp->repair) |
2757 | val = tp->rx_opt.mss_clamp; | |
1da177e4 LT |
2758 | break; |
2759 | case TCP_NODELAY: | |
2760 | val = !!(tp->nonagle&TCP_NAGLE_OFF); | |
2761 | break; | |
2762 | case TCP_CORK: | |
2763 | val = !!(tp->nonagle&TCP_NAGLE_CORK); | |
2764 | break; | |
2765 | case TCP_KEEPIDLE: | |
df19a626 | 2766 | val = keepalive_time_when(tp) / HZ; |
1da177e4 LT |
2767 | break; |
2768 | case TCP_KEEPINTVL: | |
df19a626 | 2769 | val = keepalive_intvl_when(tp) / HZ; |
1da177e4 LT |
2770 | break; |
2771 | case TCP_KEEPCNT: | |
df19a626 | 2772 | val = keepalive_probes(tp); |
1da177e4 LT |
2773 | break; |
2774 | case TCP_SYNCNT: | |
295f7324 | 2775 | val = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries; |
1da177e4 LT |
2776 | break; |
2777 | case TCP_LINGER2: | |
2778 | val = tp->linger2; | |
2779 | if (val >= 0) | |
2780 | val = (val ? : sysctl_tcp_fin_timeout) / HZ; | |
2781 | break; | |
2782 | case TCP_DEFER_ACCEPT: | |
b103cf34 JA |
2783 | val = retrans_to_secs(icsk->icsk_accept_queue.rskq_defer_accept, |
2784 | TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ); | |
1da177e4 LT |
2785 | break; |
2786 | case TCP_WINDOW_CLAMP: | |
2787 | val = tp->window_clamp; | |
2788 | break; | |
2789 | case TCP_INFO: { | |
2790 | struct tcp_info info; | |
2791 | ||
2792 | if (get_user(len, optlen)) | |
2793 | return -EFAULT; | |
2794 | ||
2795 | tcp_get_info(sk, &info); | |
2796 | ||
2797 | len = min_t(unsigned int, len, sizeof(info)); | |
2798 | if (put_user(len, optlen)) | |
2799 | return -EFAULT; | |
2800 | if (copy_to_user(optval, &info, len)) | |
2801 | return -EFAULT; | |
2802 | return 0; | |
2803 | } | |
2804 | case TCP_QUICKACK: | |
295f7324 | 2805 | val = !icsk->icsk_ack.pingpong; |
1da177e4 | 2806 | break; |
5f8ef48d SH |
2807 | |
2808 | case TCP_CONGESTION: | |
2809 | if (get_user(len, optlen)) | |
2810 | return -EFAULT; | |
2811 | len = min_t(unsigned int, len, TCP_CA_NAME_MAX); | |
2812 | if (put_user(len, optlen)) | |
2813 | return -EFAULT; | |
6687e988 | 2814 | if (copy_to_user(optval, icsk->icsk_ca_ops->name, len)) |
5f8ef48d SH |
2815 | return -EFAULT; |
2816 | return 0; | |
e56fb50f | 2817 | |
3c0fef0b JH |
2818 | case TCP_THIN_LINEAR_TIMEOUTS: |
2819 | val = tp->thin_lto; | |
2820 | break; | |
2821 | case TCP_THIN_DUPACK: | |
2822 | val = tp->thin_dupack; | |
2823 | break; | |
dca43c75 | 2824 | |
ee995283 PE |
2825 | case TCP_REPAIR: |
2826 | val = tp->repair; | |
2827 | break; | |
2828 | ||
2829 | case TCP_REPAIR_QUEUE: | |
2830 | if (tp->repair) | |
2831 | val = tp->repair_queue; | |
2832 | else | |
2833 | return -EINVAL; | |
2834 | break; | |
2835 | ||
2836 | case TCP_QUEUE_SEQ: | |
2837 | if (tp->repair_queue == TCP_SEND_QUEUE) | |
2838 | val = tp->write_seq; | |
2839 | else if (tp->repair_queue == TCP_RECV_QUEUE) | |
2840 | val = tp->rcv_nxt; | |
2841 | else | |
2842 | return -EINVAL; | |
2843 | break; | |
2844 | ||
dca43c75 JC |
2845 | case TCP_USER_TIMEOUT: |
2846 | val = jiffies_to_msecs(icsk->icsk_user_timeout); | |
2847 | break; | |
93be6ce0 AV |
2848 | case TCP_TIMESTAMP: |
2849 | val = tcp_time_stamp + tp->tsoffset; | |
2850 | break; | |
1da177e4 LT |
2851 | default: |
2852 | return -ENOPROTOOPT; | |
3ff50b79 | 2853 | } |
1da177e4 LT |
2854 | |
2855 | if (put_user(len, optlen)) | |
2856 | return -EFAULT; | |
2857 | if (copy_to_user(optval, &val, len)) | |
2858 | return -EFAULT; | |
2859 | return 0; | |
2860 | } | |
2861 | ||
3fdadf7d DM |
2862 | int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval, |
2863 | int __user *optlen) | |
2864 | { | |
2865 | struct inet_connection_sock *icsk = inet_csk(sk); | |
2866 | ||
2867 | if (level != SOL_TCP) | |
2868 | return icsk->icsk_af_ops->getsockopt(sk, level, optname, | |
2869 | optval, optlen); | |
2870 | return do_tcp_getsockopt(sk, level, optname, optval, optlen); | |
2871 | } | |
4bc2f18b | 2872 | EXPORT_SYMBOL(tcp_getsockopt); |
3fdadf7d DM |
2873 | |
2874 | #ifdef CONFIG_COMPAT | |
543d9cfe ACM |
2875 | int compat_tcp_getsockopt(struct sock *sk, int level, int optname, |
2876 | char __user *optval, int __user *optlen) | |
3fdadf7d | 2877 | { |
dec73ff0 ACM |
2878 | if (level != SOL_TCP) |
2879 | return inet_csk_compat_getsockopt(sk, level, optname, | |
2880 | optval, optlen); | |
3fdadf7d DM |
2881 | return do_tcp_getsockopt(sk, level, optname, optval, optlen); |
2882 | } | |
543d9cfe | 2883 | EXPORT_SYMBOL(compat_tcp_getsockopt); |
3fdadf7d | 2884 | #endif |
1da177e4 | 2885 | |
c8f44aff MM |
2886 | struct sk_buff *tcp_tso_segment(struct sk_buff *skb, |
2887 | netdev_features_t features) | |
f4c50d99 HX |
2888 | { |
2889 | struct sk_buff *segs = ERR_PTR(-EINVAL); | |
2890 | struct tcphdr *th; | |
95c96174 | 2891 | unsigned int thlen; |
f4c50d99 | 2892 | unsigned int seq; |
d3bc23e7 | 2893 | __be32 delta; |
f4c50d99 | 2894 | unsigned int oldlen; |
4e704ee3 | 2895 | unsigned int mss; |
d6a4a104 | 2896 | struct sk_buff *gso_skb = skb; |
bece1b97 | 2897 | __sum16 newcheck; |
6ff50cd5 | 2898 | bool ooo_okay, copy_destructor; |
f4c50d99 HX |
2899 | |
2900 | if (!pskb_may_pull(skb, sizeof(*th))) | |
2901 | goto out; | |
2902 | ||
aa8223c7 | 2903 | th = tcp_hdr(skb); |
f4c50d99 HX |
2904 | thlen = th->doff * 4; |
2905 | if (thlen < sizeof(*th)) | |
2906 | goto out; | |
2907 | ||
2908 | if (!pskb_may_pull(skb, thlen)) | |
2909 | goto out; | |
2910 | ||
0718bcc0 | 2911 | oldlen = (u16)~skb->len; |
f4c50d99 HX |
2912 | __skb_pull(skb, thlen); |
2913 | ||
4e704ee3 HX |
2914 | mss = skb_shinfo(skb)->gso_size; |
2915 | if (unlikely(skb->len <= mss)) | |
2916 | goto out; | |
2917 | ||
3820c3f3 HX |
2918 | if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) { |
2919 | /* Packet is from an untrusted source, reset gso_segs. */ | |
bbcf467d | 2920 | int type = skb_shinfo(skb)->gso_type; |
bbcf467d HX |
2921 | |
2922 | if (unlikely(type & | |
2923 | ~(SKB_GSO_TCPV4 | | |
2924 | SKB_GSO_DODGY | | |
2925 | SKB_GSO_TCP_ECN | | |
2926 | SKB_GSO_TCPV6 | | |
68c33163 | 2927 | SKB_GSO_GRE | |
73136267 | 2928 | SKB_GSO_UDP_TUNNEL | |
bbcf467d HX |
2929 | 0) || |
2930 | !(type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))) | |
2931 | goto out; | |
3820c3f3 | 2932 | |
172589cc | 2933 | skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss); |
3820c3f3 HX |
2934 | |
2935 | segs = NULL; | |
2936 | goto out; | |
2937 | } | |
2938 | ||
6ff50cd5 ED |
2939 | copy_destructor = gso_skb->destructor == tcp_wfree; |
2940 | ooo_okay = gso_skb->ooo_okay; | |
2941 | /* All segments but the first should have ooo_okay cleared */ | |
2942 | skb->ooo_okay = 0; | |
2943 | ||
576a30eb | 2944 | segs = skb_segment(skb, features); |
f4c50d99 HX |
2945 | if (IS_ERR(segs)) |
2946 | goto out; | |
2947 | ||
6ff50cd5 ED |
2948 | /* Only first segment might have ooo_okay set */ |
2949 | segs->ooo_okay = ooo_okay; | |
2950 | ||
4e704ee3 | 2951 | delta = htonl(oldlen + (thlen + mss)); |
f4c50d99 HX |
2952 | |
2953 | skb = segs; | |
aa8223c7 | 2954 | th = tcp_hdr(skb); |
f4c50d99 HX |
2955 | seq = ntohl(th->seq); |
2956 | ||
bece1b97 ED |
2957 | newcheck = ~csum_fold((__force __wsum)((__force u32)th->check + |
2958 | (__force u32)delta)); | |
2959 | ||
f4c50d99 HX |
2960 | do { |
2961 | th->fin = th->psh = 0; | |
bece1b97 | 2962 | th->check = newcheck; |
f4c50d99 | 2963 | |
84fa7933 | 2964 | if (skb->ip_summed != CHECKSUM_PARTIAL) |
9c70220b ACM |
2965 | th->check = |
2966 | csum_fold(csum_partial(skb_transport_header(skb), | |
2967 | thlen, skb->csum)); | |
f4c50d99 | 2968 | |
4e704ee3 | 2969 | seq += mss; |
6ff50cd5 ED |
2970 | if (copy_destructor) { |
2971 | skb->destructor = gso_skb->destructor; | |
2972 | skb->sk = gso_skb->sk; | |
2973 | /* {tcp|sock}_wfree() use exact truesize accounting : | |
2974 | * sum(skb->truesize) MUST be exactly be gso_skb->truesize | |
2975 | * So we account mss bytes of 'true size' for each segment. | |
2976 | * The last segment will contain the remaining. | |
2977 | */ | |
2978 | skb->truesize = mss; | |
2979 | gso_skb->truesize -= mss; | |
2980 | } | |
f4c50d99 | 2981 | skb = skb->next; |
aa8223c7 | 2982 | th = tcp_hdr(skb); |
f4c50d99 HX |
2983 | |
2984 | th->seq = htonl(seq); | |
2985 | th->cwr = 0; | |
2986 | } while (skb->next); | |
2987 | ||
d6a4a104 ED |
2988 | /* Following permits TCP Small Queues to work well with GSO : |
2989 | * The callback to TCP stack will be called at the time last frag | |
2990 | * is freed at TX completion, and not right now when gso_skb | |
2991 | * is freed by GSO engine | |
2992 | */ | |
6ff50cd5 | 2993 | if (copy_destructor) { |
d6a4a104 ED |
2994 | swap(gso_skb->sk, skb->sk); |
2995 | swap(gso_skb->destructor, skb->destructor); | |
2996 | swap(gso_skb->truesize, skb->truesize); | |
2997 | } | |
2998 | ||
27a884dc | 2999 | delta = htonl(oldlen + (skb->tail - skb->transport_header) + |
9c70220b | 3000 | skb->data_len); |
d3bc23e7 AV |
3001 | th->check = ~csum_fold((__force __wsum)((__force u32)th->check + |
3002 | (__force u32)delta)); | |
84fa7933 | 3003 | if (skb->ip_summed != CHECKSUM_PARTIAL) |
9c70220b ACM |
3004 | th->check = csum_fold(csum_partial(skb_transport_header(skb), |
3005 | thlen, skb->csum)); | |
f4c50d99 HX |
3006 | |
3007 | out: | |
3008 | return segs; | |
3009 | } | |
adcfc7d0 | 3010 | EXPORT_SYMBOL(tcp_tso_segment); |
f4c50d99 | 3011 | |
bf296b12 HX |
3012 | struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb) |
3013 | { | |
3014 | struct sk_buff **pp = NULL; | |
3015 | struct sk_buff *p; | |
3016 | struct tcphdr *th; | |
3017 | struct tcphdr *th2; | |
a0a69a01 | 3018 | unsigned int len; |
bf296b12 | 3019 | unsigned int thlen; |
0eae88f3 | 3020 | __be32 flags; |
bf296b12 | 3021 | unsigned int mss = 1; |
a5b1cf28 HX |
3022 | unsigned int hlen; |
3023 | unsigned int off; | |
bf296b12 | 3024 | int flush = 1; |
aa6320d3 | 3025 | int i; |
bf296b12 | 3026 | |
a5b1cf28 HX |
3027 | off = skb_gro_offset(skb); |
3028 | hlen = off + sizeof(*th); | |
3029 | th = skb_gro_header_fast(skb, off); | |
3030 | if (skb_gro_header_hard(skb, hlen)) { | |
3031 | th = skb_gro_header_slow(skb, hlen, off); | |
3032 | if (unlikely(!th)) | |
3033 | goto out; | |
3034 | } | |
bf296b12 | 3035 | |
bf296b12 HX |
3036 | thlen = th->doff * 4; |
3037 | if (thlen < sizeof(*th)) | |
3038 | goto out; | |
3039 | ||
a5b1cf28 HX |
3040 | hlen = off + thlen; |
3041 | if (skb_gro_header_hard(skb, hlen)) { | |
3042 | th = skb_gro_header_slow(skb, hlen, off); | |
3043 | if (unlikely(!th)) | |
3044 | goto out; | |
3045 | } | |
bf296b12 | 3046 | |
86911732 | 3047 | skb_gro_pull(skb, thlen); |
bf296b12 | 3048 | |
a0a69a01 | 3049 | len = skb_gro_len(skb); |
bf296b12 HX |
3050 | flags = tcp_flag_word(th); |
3051 | ||
3052 | for (; (p = *head); head = &p->next) { | |
3053 | if (!NAPI_GRO_CB(p)->same_flow) | |
3054 | continue; | |
3055 | ||
3056 | th2 = tcp_hdr(p); | |
3057 | ||
745898ea | 3058 | if (*(u32 *)&th->source ^ *(u32 *)&th2->source) { |
bf296b12 HX |
3059 | NAPI_GRO_CB(p)->same_flow = 0; |
3060 | continue; | |
3061 | } | |
3062 | ||
3063 | goto found; | |
3064 | } | |
3065 | ||
3066 | goto out_check_final; | |
3067 | ||
3068 | found: | |
3069 | flush = NAPI_GRO_CB(p)->flush; | |
0eae88f3 ED |
3070 | flush |= (__force int)(flags & TCP_FLAG_CWR); |
3071 | flush |= (__force int)((flags ^ tcp_flag_word(th2)) & | |
3072 | ~(TCP_FLAG_CWR | TCP_FLAG_FIN | TCP_FLAG_PSH)); | |
3073 | flush |= (__force int)(th->ack_seq ^ th2->ack_seq); | |
a2a804cd | 3074 | for (i = sizeof(*th); i < thlen; i += 4) |
aa6320d3 HX |
3075 | flush |= *(u32 *)((u8 *)th + i) ^ |
3076 | *(u32 *)((u8 *)th2 + i); | |
bf296b12 | 3077 | |
b530256d | 3078 | mss = skb_shinfo(p)->gso_size; |
bf296b12 | 3079 | |
30a3ae30 | 3080 | flush |= (len - 1) >= mss; |
aa6320d3 | 3081 | flush |= (ntohl(th2->seq) + skb_gro_len(p)) ^ ntohl(th->seq); |
bf296b12 HX |
3082 | |
3083 | if (flush || skb_gro_receive(head, skb)) { | |
3084 | mss = 1; | |
3085 | goto out_check_final; | |
3086 | } | |
3087 | ||
3088 | p = *head; | |
3089 | th2 = tcp_hdr(p); | |
3090 | tcp_flag_word(th2) |= flags & (TCP_FLAG_FIN | TCP_FLAG_PSH); | |
3091 | ||
3092 | out_check_final: | |
a0a69a01 | 3093 | flush = len < mss; |
0eae88f3 ED |
3094 | flush |= (__force int)(flags & (TCP_FLAG_URG | TCP_FLAG_PSH | |
3095 | TCP_FLAG_RST | TCP_FLAG_SYN | | |
3096 | TCP_FLAG_FIN)); | |
bf296b12 HX |
3097 | |
3098 | if (p && (!NAPI_GRO_CB(skb)->same_flow || flush)) | |
3099 | pp = head; | |
3100 | ||
3101 | out: | |
3102 | NAPI_GRO_CB(skb)->flush |= flush; | |
3103 | ||
3104 | return pp; | |
3105 | } | |
684f2176 | 3106 | EXPORT_SYMBOL(tcp_gro_receive); |
bf296b12 HX |
3107 | |
3108 | int tcp_gro_complete(struct sk_buff *skb) | |
3109 | { | |
3110 | struct tcphdr *th = tcp_hdr(skb); | |
3111 | ||
3112 | skb->csum_start = skb_transport_header(skb) - skb->head; | |
3113 | skb->csum_offset = offsetof(struct tcphdr, check); | |
3114 | skb->ip_summed = CHECKSUM_PARTIAL; | |
3115 | ||
bf296b12 HX |
3116 | skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count; |
3117 | ||
3118 | if (th->cwr) | |
3119 | skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; | |
3120 | ||
3121 | return 0; | |
3122 | } | |
684f2176 | 3123 | EXPORT_SYMBOL(tcp_gro_complete); |
bf296b12 | 3124 | |
cfb6eeb4 YH |
3125 | #ifdef CONFIG_TCP_MD5SIG |
3126 | static unsigned long tcp_md5sig_users; | |
765cf997 | 3127 | static struct tcp_md5sig_pool __percpu *tcp_md5sig_pool; |
cfb6eeb4 YH |
3128 | static DEFINE_SPINLOCK(tcp_md5sig_pool_lock); |
3129 | ||
765cf997 | 3130 | static void __tcp_free_md5sig_pool(struct tcp_md5sig_pool __percpu *pool) |
cfb6eeb4 YH |
3131 | { |
3132 | int cpu; | |
765cf997 | 3133 | |
cfb6eeb4 | 3134 | for_each_possible_cpu(cpu) { |
765cf997 ED |
3135 | struct tcp_md5sig_pool *p = per_cpu_ptr(pool, cpu); |
3136 | ||
3137 | if (p->md5_desc.tfm) | |
3138 | crypto_free_hash(p->md5_desc.tfm); | |
cfb6eeb4 YH |
3139 | } |
3140 | free_percpu(pool); | |
3141 | } | |
3142 | ||
3143 | void tcp_free_md5sig_pool(void) | |
3144 | { | |
765cf997 | 3145 | struct tcp_md5sig_pool __percpu *pool = NULL; |
cfb6eeb4 | 3146 | |
2c4f6219 | 3147 | spin_lock_bh(&tcp_md5sig_pool_lock); |
cfb6eeb4 YH |
3148 | if (--tcp_md5sig_users == 0) { |
3149 | pool = tcp_md5sig_pool; | |
3150 | tcp_md5sig_pool = NULL; | |
3151 | } | |
2c4f6219 | 3152 | spin_unlock_bh(&tcp_md5sig_pool_lock); |
cfb6eeb4 YH |
3153 | if (pool) |
3154 | __tcp_free_md5sig_pool(pool); | |
3155 | } | |
cfb6eeb4 YH |
3156 | EXPORT_SYMBOL(tcp_free_md5sig_pool); |
3157 | ||
765cf997 | 3158 | static struct tcp_md5sig_pool __percpu * |
7d720c3e | 3159 | __tcp_alloc_md5sig_pool(struct sock *sk) |
cfb6eeb4 YH |
3160 | { |
3161 | int cpu; | |
765cf997 | 3162 | struct tcp_md5sig_pool __percpu *pool; |
cfb6eeb4 | 3163 | |
765cf997 | 3164 | pool = alloc_percpu(struct tcp_md5sig_pool); |
cfb6eeb4 YH |
3165 | if (!pool) |
3166 | return NULL; | |
3167 | ||
3168 | for_each_possible_cpu(cpu) { | |
cfb6eeb4 YH |
3169 | struct crypto_hash *hash; |
3170 | ||
cfb6eeb4 | 3171 | hash = crypto_alloc_hash("md5", 0, CRYPTO_ALG_ASYNC); |
50c3a487 | 3172 | if (IS_ERR_OR_NULL(hash)) |
cfb6eeb4 YH |
3173 | goto out_free; |
3174 | ||
765cf997 | 3175 | per_cpu_ptr(pool, cpu)->md5_desc.tfm = hash; |
cfb6eeb4 YH |
3176 | } |
3177 | return pool; | |
3178 | out_free: | |
3179 | __tcp_free_md5sig_pool(pool); | |
3180 | return NULL; | |
3181 | } | |
3182 | ||
765cf997 | 3183 | struct tcp_md5sig_pool __percpu *tcp_alloc_md5sig_pool(struct sock *sk) |
cfb6eeb4 | 3184 | { |
765cf997 | 3185 | struct tcp_md5sig_pool __percpu *pool; |
a2a385d6 | 3186 | bool alloc = false; |
cfb6eeb4 YH |
3187 | |
3188 | retry: | |
2c4f6219 | 3189 | spin_lock_bh(&tcp_md5sig_pool_lock); |
cfb6eeb4 YH |
3190 | pool = tcp_md5sig_pool; |
3191 | if (tcp_md5sig_users++ == 0) { | |
a2a385d6 | 3192 | alloc = true; |
2c4f6219 | 3193 | spin_unlock_bh(&tcp_md5sig_pool_lock); |
cfb6eeb4 YH |
3194 | } else if (!pool) { |
3195 | tcp_md5sig_users--; | |
2c4f6219 | 3196 | spin_unlock_bh(&tcp_md5sig_pool_lock); |
cfb6eeb4 YH |
3197 | cpu_relax(); |
3198 | goto retry; | |
3199 | } else | |
2c4f6219 | 3200 | spin_unlock_bh(&tcp_md5sig_pool_lock); |
cfb6eeb4 YH |
3201 | |
3202 | if (alloc) { | |
3203 | /* we cannot hold spinlock here because this may sleep. */ | |
765cf997 | 3204 | struct tcp_md5sig_pool __percpu *p; |
7d720c3e TH |
3205 | |
3206 | p = __tcp_alloc_md5sig_pool(sk); | |
2c4f6219 | 3207 | spin_lock_bh(&tcp_md5sig_pool_lock); |
cfb6eeb4 YH |
3208 | if (!p) { |
3209 | tcp_md5sig_users--; | |
2c4f6219 | 3210 | spin_unlock_bh(&tcp_md5sig_pool_lock); |
cfb6eeb4 YH |
3211 | return NULL; |
3212 | } | |
3213 | pool = tcp_md5sig_pool; | |
3214 | if (pool) { | |
3215 | /* oops, it has already been assigned. */ | |
2c4f6219 | 3216 | spin_unlock_bh(&tcp_md5sig_pool_lock); |
cfb6eeb4 YH |
3217 | __tcp_free_md5sig_pool(p); |
3218 | } else { | |
3219 | tcp_md5sig_pool = pool = p; | |
2c4f6219 | 3220 | spin_unlock_bh(&tcp_md5sig_pool_lock); |
cfb6eeb4 YH |
3221 | } |
3222 | } | |
3223 | return pool; | |
3224 | } | |
cfb6eeb4 YH |
3225 | EXPORT_SYMBOL(tcp_alloc_md5sig_pool); |
3226 | ||
35790c04 ED |
3227 | |
3228 | /** | |
3229 | * tcp_get_md5sig_pool - get md5sig_pool for this user | |
3230 | * | |
3231 | * We use percpu structure, so if we succeed, we exit with preemption | |
3232 | * and BH disabled, to make sure another thread or softirq handling | |
3233 | * wont try to get same context. | |
3234 | */ | |
3235 | struct tcp_md5sig_pool *tcp_get_md5sig_pool(void) | |
cfb6eeb4 | 3236 | { |
765cf997 | 3237 | struct tcp_md5sig_pool __percpu *p; |
35790c04 ED |
3238 | |
3239 | local_bh_disable(); | |
3240 | ||
3241 | spin_lock(&tcp_md5sig_pool_lock); | |
cfb6eeb4 YH |
3242 | p = tcp_md5sig_pool; |
3243 | if (p) | |
3244 | tcp_md5sig_users++; | |
35790c04 ED |
3245 | spin_unlock(&tcp_md5sig_pool_lock); |
3246 | ||
3247 | if (p) | |
765cf997 | 3248 | return this_cpu_ptr(p); |
cfb6eeb4 | 3249 | |
35790c04 ED |
3250 | local_bh_enable(); |
3251 | return NULL; | |
3252 | } | |
3253 | EXPORT_SYMBOL(tcp_get_md5sig_pool); | |
cfb6eeb4 | 3254 | |
35790c04 | 3255 | void tcp_put_md5sig_pool(void) |
6931ba7c | 3256 | { |
35790c04 | 3257 | local_bh_enable(); |
6931ba7c | 3258 | tcp_free_md5sig_pool(); |
cfb6eeb4 | 3259 | } |
35790c04 | 3260 | EXPORT_SYMBOL(tcp_put_md5sig_pool); |
49a72dfb AL |
3261 | |
3262 | int tcp_md5_hash_header(struct tcp_md5sig_pool *hp, | |
ca35a0ef | 3263 | const struct tcphdr *th) |
49a72dfb AL |
3264 | { |
3265 | struct scatterlist sg; | |
ca35a0ef | 3266 | struct tcphdr hdr; |
49a72dfb AL |
3267 | int err; |
3268 | ||
ca35a0ef ED |
3269 | /* We are not allowed to change tcphdr, make a local copy */ |
3270 | memcpy(&hdr, th, sizeof(hdr)); | |
3271 | hdr.check = 0; | |
3272 | ||
49a72dfb | 3273 | /* options aren't included in the hash */ |
ca35a0ef ED |
3274 | sg_init_one(&sg, &hdr, sizeof(hdr)); |
3275 | err = crypto_hash_update(&hp->md5_desc, &sg, sizeof(hdr)); | |
49a72dfb AL |
3276 | return err; |
3277 | } | |
49a72dfb AL |
3278 | EXPORT_SYMBOL(tcp_md5_hash_header); |
3279 | ||
3280 | int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp, | |
cf533ea5 | 3281 | const struct sk_buff *skb, unsigned int header_len) |
49a72dfb AL |
3282 | { |
3283 | struct scatterlist sg; | |
3284 | const struct tcphdr *tp = tcp_hdr(skb); | |
3285 | struct hash_desc *desc = &hp->md5_desc; | |
95c96174 ED |
3286 | unsigned int i; |
3287 | const unsigned int head_data_len = skb_headlen(skb) > header_len ? | |
3288 | skb_headlen(skb) - header_len : 0; | |
49a72dfb | 3289 | const struct skb_shared_info *shi = skb_shinfo(skb); |
d7fd1b57 | 3290 | struct sk_buff *frag_iter; |
49a72dfb AL |
3291 | |
3292 | sg_init_table(&sg, 1); | |
3293 | ||
3294 | sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len); | |
3295 | if (crypto_hash_update(desc, &sg, head_data_len)) | |
3296 | return 1; | |
3297 | ||
3298 | for (i = 0; i < shi->nr_frags; ++i) { | |
3299 | const struct skb_frag_struct *f = &shi->frags[i]; | |
54d27fcb ED |
3300 | unsigned int offset = f->page_offset; |
3301 | struct page *page = skb_frag_page(f) + (offset >> PAGE_SHIFT); | |
3302 | ||
3303 | sg_set_page(&sg, page, skb_frag_size(f), | |
3304 | offset_in_page(offset)); | |
9e903e08 | 3305 | if (crypto_hash_update(desc, &sg, skb_frag_size(f))) |
49a72dfb AL |
3306 | return 1; |
3307 | } | |
3308 | ||
d7fd1b57 ED |
3309 | skb_walk_frags(skb, frag_iter) |
3310 | if (tcp_md5_hash_skb_data(hp, frag_iter, 0)) | |
3311 | return 1; | |
3312 | ||
49a72dfb AL |
3313 | return 0; |
3314 | } | |
49a72dfb AL |
3315 | EXPORT_SYMBOL(tcp_md5_hash_skb_data); |
3316 | ||
cf533ea5 | 3317 | int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, const struct tcp_md5sig_key *key) |
49a72dfb AL |
3318 | { |
3319 | struct scatterlist sg; | |
3320 | ||
3321 | sg_init_one(&sg, key->key, key->keylen); | |
3322 | return crypto_hash_update(&hp->md5_desc, &sg, key->keylen); | |
3323 | } | |
49a72dfb AL |
3324 | EXPORT_SYMBOL(tcp_md5_hash_key); |
3325 | ||
cfb6eeb4 YH |
3326 | #endif |
3327 | ||
4ac02bab AK |
3328 | void tcp_done(struct sock *sk) |
3329 | { | |
8336886f JC |
3330 | struct request_sock *req = tcp_sk(sk)->fastopen_rsk; |
3331 | ||
5a5f3a8d | 3332 | if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV) |
63231bdd | 3333 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS); |
4ac02bab AK |
3334 | |
3335 | tcp_set_state(sk, TCP_CLOSE); | |
3336 | tcp_clear_xmit_timers(sk); | |
8336886f JC |
3337 | if (req != NULL) |
3338 | reqsk_fastopen_remove(sk, req, false); | |
4ac02bab AK |
3339 | |
3340 | sk->sk_shutdown = SHUTDOWN_MASK; | |
3341 | ||
3342 | if (!sock_flag(sk, SOCK_DEAD)) | |
3343 | sk->sk_state_change(sk); | |
3344 | else | |
3345 | inet_csk_destroy_sock(sk); | |
3346 | } | |
3347 | EXPORT_SYMBOL_GPL(tcp_done); | |
3348 | ||
5f8ef48d | 3349 | extern struct tcp_congestion_ops tcp_reno; |
1da177e4 LT |
3350 | |
3351 | static __initdata unsigned long thash_entries; | |
3352 | static int __init set_thash_entries(char *str) | |
3353 | { | |
413c27d8 EZ |
3354 | ssize_t ret; |
3355 | ||
1da177e4 LT |
3356 | if (!str) |
3357 | return 0; | |
413c27d8 EZ |
3358 | |
3359 | ret = kstrtoul(str, 0, &thash_entries); | |
3360 | if (ret) | |
3361 | return 0; | |
3362 | ||
1da177e4 LT |
3363 | return 1; |
3364 | } | |
3365 | __setup("thash_entries=", set_thash_entries); | |
3366 | ||
4acb4190 GC |
3367 | void tcp_init_mem(struct net *net) |
3368 | { | |
4acb4190 GC |
3369 | unsigned long limit = nr_free_buffer_pages() / 8; |
3370 | limit = max(limit, 128UL); | |
3371 | net->ipv4.sysctl_tcp_mem[0] = limit / 4 * 3; | |
3372 | net->ipv4.sysctl_tcp_mem[1] = limit; | |
3373 | net->ipv4.sysctl_tcp_mem[2] = net->ipv4.sysctl_tcp_mem[0] * 2; | |
3374 | } | |
3375 | ||
1da177e4 LT |
3376 | void __init tcp_init(void) |
3377 | { | |
3378 | struct sk_buff *skb = NULL; | |
f03d78db | 3379 | unsigned long limit; |
b49960a0 | 3380 | int max_rshare, max_wshare, cnt; |
074b8517 | 3381 | unsigned int i; |
1da177e4 | 3382 | |
1f9e636e | 3383 | BUILD_BUG_ON(sizeof(struct tcp_skb_cb) > sizeof(skb->cb)); |
1da177e4 | 3384 | |
1748376b | 3385 | percpu_counter_init(&tcp_sockets_allocated, 0); |
dd24c001 | 3386 | percpu_counter_init(&tcp_orphan_count, 0); |
6e04e021 ACM |
3387 | tcp_hashinfo.bind_bucket_cachep = |
3388 | kmem_cache_create("tcp_bind_bucket", | |
3389 | sizeof(struct inet_bind_bucket), 0, | |
20c2df83 | 3390 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3391 | |
1da177e4 LT |
3392 | /* Size and allocate the main established and bind bucket |
3393 | * hash tables. | |
3394 | * | |
3395 | * The methodology is similar to that of the buffer cache. | |
3396 | */ | |
6e04e021 | 3397 | tcp_hashinfo.ehash = |
1da177e4 | 3398 | alloc_large_system_hash("TCP established", |
0f7ff927 | 3399 | sizeof(struct inet_ehash_bucket), |
1da177e4 | 3400 | thash_entries, |
fd90b29d | 3401 | 17, /* one slot per 128 KB of memory */ |
9e950efa | 3402 | 0, |
1da177e4 | 3403 | NULL, |
f373b53b | 3404 | &tcp_hashinfo.ehash_mask, |
31fe62b9 | 3405 | 0, |
0ccfe618 | 3406 | thash_entries ? 0 : 512 * 1024); |
f373b53b | 3407 | for (i = 0; i <= tcp_hashinfo.ehash_mask; i++) { |
3ab5aee7 ED |
3408 | INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i); |
3409 | INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].twchain, i); | |
1da177e4 | 3410 | } |
230140cf ED |
3411 | if (inet_ehash_locks_alloc(&tcp_hashinfo)) |
3412 | panic("TCP: failed to alloc ehash_locks"); | |
6e04e021 | 3413 | tcp_hashinfo.bhash = |
1da177e4 | 3414 | alloc_large_system_hash("TCP bind", |
0f7ff927 | 3415 | sizeof(struct inet_bind_hashbucket), |
f373b53b | 3416 | tcp_hashinfo.ehash_mask + 1, |
fd90b29d | 3417 | 17, /* one slot per 128 KB of memory */ |
9e950efa | 3418 | 0, |
6e04e021 | 3419 | &tcp_hashinfo.bhash_size, |
1da177e4 | 3420 | NULL, |
31fe62b9 | 3421 | 0, |
1da177e4 | 3422 | 64 * 1024); |
074b8517 | 3423 | tcp_hashinfo.bhash_size = 1U << tcp_hashinfo.bhash_size; |
6e04e021 ACM |
3424 | for (i = 0; i < tcp_hashinfo.bhash_size; i++) { |
3425 | spin_lock_init(&tcp_hashinfo.bhash[i].lock); | |
3426 | INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain); | |
1da177e4 LT |
3427 | } |
3428 | ||
c5ed63d6 ED |
3429 | |
3430 | cnt = tcp_hashinfo.ehash_mask + 1; | |
3431 | ||
3432 | tcp_death_row.sysctl_max_tw_buckets = cnt / 2; | |
3433 | sysctl_tcp_max_orphans = cnt / 2; | |
3434 | sysctl_max_syn_backlog = max(128, cnt / 256); | |
1da177e4 | 3435 | |
4acb4190 | 3436 | tcp_init_mem(&init_net); |
c43b874d | 3437 | /* Set per-socket limits to no more than 1/128 the pressure threshold */ |
5fb84b14 | 3438 | limit = nr_free_buffer_pages() << (PAGE_SHIFT - 7); |
b49960a0 ED |
3439 | max_wshare = min(4UL*1024*1024, limit); |
3440 | max_rshare = min(6UL*1024*1024, limit); | |
7b4f4b5e | 3441 | |
3ab224be | 3442 | sysctl_tcp_wmem[0] = SK_MEM_QUANTUM; |
7b4f4b5e | 3443 | sysctl_tcp_wmem[1] = 16*1024; |
b49960a0 | 3444 | sysctl_tcp_wmem[2] = max(64*1024, max_wshare); |
7b4f4b5e | 3445 | |
3ab224be | 3446 | sysctl_tcp_rmem[0] = SK_MEM_QUANTUM; |
7b4f4b5e | 3447 | sysctl_tcp_rmem[1] = 87380; |
b49960a0 | 3448 | sysctl_tcp_rmem[2] = max(87380, max_rshare); |
1da177e4 | 3449 | |
afd46503 | 3450 | pr_info("Hash tables configured (established %u bind %u)\n", |
058bd4d2 | 3451 | tcp_hashinfo.ehash_mask + 1, tcp_hashinfo.bhash_size); |
317a76f9 | 3452 | |
51c5d0c4 DM |
3453 | tcp_metrics_init(); |
3454 | ||
317a76f9 | 3455 | tcp_register_congestion_control(&tcp_reno); |
da5c78c8 | 3456 | |
46d3ceab | 3457 | tcp_tasklet_init(); |
1da177e4 | 3458 | } |