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 | ||
cf80e0e4 | 250 | #include <crypto/hash.h> |
172589cc | 251 | #include <linux/kernel.h> |
1da177e4 LT |
252 | #include <linux/module.h> |
253 | #include <linux/types.h> | |
254 | #include <linux/fcntl.h> | |
255 | #include <linux/poll.h> | |
6e9250f5 | 256 | #include <linux/inet_diag.h> |
1da177e4 | 257 | #include <linux/init.h> |
1da177e4 | 258 | #include <linux/fs.h> |
9c55e01c | 259 | #include <linux/skbuff.h> |
81b23b4a | 260 | #include <linux/scatterlist.h> |
9c55e01c JA |
261 | #include <linux/splice.h> |
262 | #include <linux/net.h> | |
263 | #include <linux/socket.h> | |
1da177e4 LT |
264 | #include <linux/random.h> |
265 | #include <linux/bootmem.h> | |
57413ebc MS |
266 | #include <linux/highmem.h> |
267 | #include <linux/swap.h> | |
b8059ead | 268 | #include <linux/cache.h> |
f4c50d99 | 269 | #include <linux/err.h> |
da5c78c8 | 270 | #include <linux/time.h> |
5a0e3ad6 | 271 | #include <linux/slab.h> |
1da177e4 LT |
272 | |
273 | #include <net/icmp.h> | |
cf60af03 | 274 | #include <net/inet_common.h> |
1da177e4 LT |
275 | #include <net/tcp.h> |
276 | #include <net/xfrm.h> | |
277 | #include <net/ip.h> | |
9c55e01c | 278 | #include <net/sock.h> |
1da177e4 LT |
279 | |
280 | #include <asm/uaccess.h> | |
281 | #include <asm/ioctls.h> | |
ff5d7497 | 282 | #include <asm/unaligned.h> |
076bb0c8 | 283 | #include <net/busy_poll.h> |
1da177e4 | 284 | |
95bd09eb ED |
285 | int sysctl_tcp_min_tso_segs __read_mostly = 2; |
286 | ||
f54b3111 ED |
287 | int sysctl_tcp_autocorking __read_mostly = 1; |
288 | ||
dd24c001 | 289 | struct percpu_counter tcp_orphan_count; |
0a5578cf ACM |
290 | EXPORT_SYMBOL_GPL(tcp_orphan_count); |
291 | ||
a4fe34bf | 292 | long sysctl_tcp_mem[3] __read_mostly; |
b8059ead DM |
293 | int sysctl_tcp_wmem[3] __read_mostly; |
294 | int sysctl_tcp_rmem[3] __read_mostly; | |
1da177e4 | 295 | |
a4fe34bf | 296 | EXPORT_SYMBOL(sysctl_tcp_mem); |
1da177e4 LT |
297 | EXPORT_SYMBOL(sysctl_tcp_rmem); |
298 | EXPORT_SYMBOL(sysctl_tcp_wmem); | |
299 | ||
8d987e5c | 300 | atomic_long_t tcp_memory_allocated; /* Current allocated memory. */ |
1da177e4 | 301 | EXPORT_SYMBOL(tcp_memory_allocated); |
1748376b ED |
302 | |
303 | /* | |
304 | * Current number of TCP sockets. | |
305 | */ | |
306 | struct percpu_counter tcp_sockets_allocated; | |
1da177e4 LT |
307 | EXPORT_SYMBOL(tcp_sockets_allocated); |
308 | ||
9c55e01c JA |
309 | /* |
310 | * TCP splice context | |
311 | */ | |
312 | struct tcp_splice_state { | |
313 | struct pipe_inode_info *pipe; | |
314 | size_t len; | |
315 | unsigned int flags; | |
316 | }; | |
317 | ||
1da177e4 LT |
318 | /* |
319 | * Pressure flag: try to collapse. | |
320 | * Technical note: it is used by multiple contexts non atomically. | |
3ab224be | 321 | * All the __sk_mem_schedule() is of this nature: accounting |
1da177e4 LT |
322 | * is strict, actions are advisory and have some latency. |
323 | */ | |
4103f8cd | 324 | int tcp_memory_pressure __read_mostly; |
1da177e4 LT |
325 | EXPORT_SYMBOL(tcp_memory_pressure); |
326 | ||
5c52ba17 | 327 | void tcp_enter_memory_pressure(struct sock *sk) |
1da177e4 LT |
328 | { |
329 | if (!tcp_memory_pressure) { | |
4e673444 | 330 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES); |
1da177e4 LT |
331 | tcp_memory_pressure = 1; |
332 | } | |
333 | } | |
1da177e4 LT |
334 | EXPORT_SYMBOL(tcp_enter_memory_pressure); |
335 | ||
b103cf34 JA |
336 | /* Convert seconds to retransmits based on initial and max timeout */ |
337 | static u8 secs_to_retrans(int seconds, int timeout, int rto_max) | |
338 | { | |
339 | u8 res = 0; | |
340 | ||
341 | if (seconds > 0) { | |
342 | int period = timeout; | |
343 | ||
344 | res = 1; | |
345 | while (seconds > period && res < 255) { | |
346 | res++; | |
347 | timeout <<= 1; | |
348 | if (timeout > rto_max) | |
349 | timeout = rto_max; | |
350 | period += timeout; | |
351 | } | |
352 | } | |
353 | return res; | |
354 | } | |
355 | ||
356 | /* Convert retransmits to seconds based on initial and max timeout */ | |
357 | static int retrans_to_secs(u8 retrans, int timeout, int rto_max) | |
358 | { | |
359 | int period = 0; | |
360 | ||
361 | if (retrans > 0) { | |
362 | period = timeout; | |
363 | while (--retrans) { | |
364 | timeout <<= 1; | |
365 | if (timeout > rto_max) | |
366 | timeout = rto_max; | |
367 | period += timeout; | |
368 | } | |
369 | } | |
370 | return period; | |
371 | } | |
372 | ||
900f65d3 NC |
373 | /* Address-family independent initialization for a tcp_sock. |
374 | * | |
375 | * NOTE: A lot of things set to zero explicitly by call to | |
376 | * sk_alloc() so need not be done here. | |
377 | */ | |
378 | void tcp_init_sock(struct sock *sk) | |
379 | { | |
380 | struct inet_connection_sock *icsk = inet_csk(sk); | |
381 | struct tcp_sock *tp = tcp_sk(sk); | |
382 | ||
996b175e | 383 | __skb_queue_head_init(&tp->out_of_order_queue); |
900f65d3 NC |
384 | tcp_init_xmit_timers(sk); |
385 | tcp_prequeue_init(tp); | |
46d3ceab | 386 | INIT_LIST_HEAD(&tp->tsq_node); |
900f65d3 NC |
387 | |
388 | icsk->icsk_rto = TCP_TIMEOUT_INIT; | |
740b0f18 | 389 | tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT); |
f6722583 | 390 | tp->rtt_min[0].rtt = ~0U; |
900f65d3 NC |
391 | |
392 | /* So many TCP implementations out there (incorrectly) count the | |
393 | * initial SYN frame in their delayed-ACK and congestion control | |
394 | * algorithms that we must have the following bandaid to talk | |
395 | * efficiently to them. -DaveM | |
396 | */ | |
397 | tp->snd_cwnd = TCP_INIT_CWND; | |
398 | ||
399 | /* See draft-stevens-tcpca-spec-01 for discussion of the | |
400 | * initialization of these values. | |
401 | */ | |
402 | tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; | |
403 | tp->snd_cwnd_clamp = ~0; | |
404 | tp->mss_cache = TCP_MSS_DEFAULT; | |
d654976c | 405 | u64_stats_init(&tp->syncp); |
900f65d3 | 406 | |
1043e25f | 407 | tp->reordering = sock_net(sk)->ipv4.sysctl_tcp_reordering; |
eed530b6 | 408 | tcp_enable_early_retrans(tp); |
55d8694f | 409 | tcp_assign_congestion_control(sk); |
900f65d3 | 410 | |
ceaa1fef AV |
411 | tp->tsoffset = 0; |
412 | ||
900f65d3 NC |
413 | sk->sk_state = TCP_CLOSE; |
414 | ||
415 | sk->sk_write_space = sk_stream_write_space; | |
416 | sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); | |
417 | ||
418 | icsk->icsk_sync_mss = tcp_sync_mss; | |
419 | ||
900f65d3 NC |
420 | sk->sk_sndbuf = sysctl_tcp_wmem[1]; |
421 | sk->sk_rcvbuf = sysctl_tcp_rmem[1]; | |
422 | ||
423 | local_bh_disable(); | |
3d596f7b JW |
424 | if (mem_cgroup_sockets_enabled) |
425 | sock_update_memcg(sk); | |
900f65d3 NC |
426 | sk_sockets_allocated_inc(sk); |
427 | local_bh_enable(); | |
428 | } | |
429 | EXPORT_SYMBOL(tcp_init_sock); | |
430 | ||
f066e2b0 | 431 | static void tcp_tx_timestamp(struct sock *sk, struct sk_buff *skb) |
4ed2d765 | 432 | { |
f066e2b0 WB |
433 | if (sk->sk_tsflags) { |
434 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
4ed2d765 | 435 | |
f066e2b0 WB |
436 | sock_tx_timestamp(sk, &shinfo->tx_flags); |
437 | if (shinfo->tx_flags & SKBTX_ANY_TSTAMP) | |
438 | shinfo->tskey = TCP_SKB_CB(skb)->seq + skb->len - 1; | |
439 | } | |
4ed2d765 WB |
440 | } |
441 | ||
1da177e4 LT |
442 | /* |
443 | * Wait for a TCP event. | |
444 | * | |
445 | * Note that we don't need to lock the socket, as the upper poll layers | |
446 | * take care of normal races (between the test and the event) and we don't | |
447 | * go look at any of the socket buffers directly. | |
448 | */ | |
449 | unsigned int tcp_poll(struct file *file, struct socket *sock, poll_table *wait) | |
450 | { | |
451 | unsigned int mask; | |
452 | struct sock *sk = sock->sk; | |
cf533ea5 | 453 | const struct tcp_sock *tp = tcp_sk(sk); |
00fd38d9 | 454 | int state; |
1da177e4 | 455 | |
c3f1dbaf DM |
456 | sock_rps_record_flow(sk); |
457 | ||
aa395145 | 458 | sock_poll_wait(file, sk_sleep(sk), wait); |
00fd38d9 ED |
459 | |
460 | state = sk_state_load(sk); | |
461 | if (state == TCP_LISTEN) | |
dc40c7bc | 462 | return inet_csk_listen_poll(sk); |
1da177e4 LT |
463 | |
464 | /* Socket is not locked. We are protected from async events | |
70efce27 WN |
465 | * by poll logic and correct handling of state changes |
466 | * made by other threads is impossible in any case. | |
1da177e4 LT |
467 | */ |
468 | ||
469 | mask = 0; | |
1da177e4 LT |
470 | |
471 | /* | |
472 | * POLLHUP is certainly not done right. But poll() doesn't | |
473 | * have a notion of HUP in just one direction, and for a | |
474 | * socket the read side is more interesting. | |
475 | * | |
476 | * Some poll() documentation says that POLLHUP is incompatible | |
477 | * with the POLLOUT/POLLWR flags, so somebody should check this | |
478 | * all. But careful, it tends to be safer to return too many | |
479 | * bits than too few, and you can easily break real applications | |
480 | * if you don't tell them that something has hung up! | |
481 | * | |
482 | * Check-me. | |
483 | * | |
484 | * Check number 1. POLLHUP is _UNMASKABLE_ event (see UNIX98 and | |
485 | * our fs/select.c). It means that after we received EOF, | |
486 | * poll always returns immediately, making impossible poll() on write() | |
487 | * in state CLOSE_WAIT. One solution is evident --- to set POLLHUP | |
488 | * if and only if shutdown has been made in both directions. | |
489 | * Actually, it is interesting to look how Solaris and DUX | |
70efce27 | 490 | * solve this dilemma. I would prefer, if POLLHUP were maskable, |
1da177e4 LT |
491 | * then we could set it on SND_SHUTDOWN. BTW examples given |
492 | * in Stevens' books assume exactly this behaviour, it explains | |
70efce27 | 493 | * why POLLHUP is incompatible with POLLOUT. --ANK |
1da177e4 LT |
494 | * |
495 | * NOTE. Check for TCP_CLOSE is added. The goal is to prevent | |
496 | * blocking on fresh not-connected or disconnected socket. --ANK | |
497 | */ | |
00fd38d9 | 498 | if (sk->sk_shutdown == SHUTDOWN_MASK || state == TCP_CLOSE) |
1da177e4 LT |
499 | mask |= POLLHUP; |
500 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
f348d70a | 501 | mask |= POLLIN | POLLRDNORM | POLLRDHUP; |
1da177e4 | 502 | |
8336886f | 503 | /* Connected or passive Fast Open socket? */ |
00fd38d9 ED |
504 | if (state != TCP_SYN_SENT && |
505 | (state != TCP_SYN_RECV || tp->fastopen_rsk)) { | |
c7004482 DM |
506 | int target = sock_rcvlowat(sk, 0, INT_MAX); |
507 | ||
508 | if (tp->urg_seq == tp->copied_seq && | |
509 | !sock_flag(sk, SOCK_URGINLINE) && | |
510 | tp->urg_data) | |
b634f875 | 511 | target++; |
c7004482 | 512 | |
c7004482 | 513 | if (tp->rcv_nxt - tp->copied_seq >= target) |
1da177e4 LT |
514 | mask |= POLLIN | POLLRDNORM; |
515 | ||
516 | if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { | |
64dc6130 | 517 | if (sk_stream_is_writeable(sk)) { |
1da177e4 LT |
518 | mask |= POLLOUT | POLLWRNORM; |
519 | } else { /* send SIGIO later */ | |
9cd3e072 | 520 | sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 LT |
521 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
522 | ||
523 | /* Race breaker. If space is freed after | |
524 | * wspace test but before the flags are set, | |
3c715127 | 525 | * IO signal will be lost. Memory barrier |
526 | * pairs with the input side. | |
1da177e4 | 527 | */ |
3c715127 | 528 | smp_mb__after_atomic(); |
64dc6130 | 529 | if (sk_stream_is_writeable(sk)) |
1da177e4 LT |
530 | mask |= POLLOUT | POLLWRNORM; |
531 | } | |
d84ba638 KM |
532 | } else |
533 | mask |= POLLOUT | POLLWRNORM; | |
1da177e4 LT |
534 | |
535 | if (tp->urg_data & TCP_URG_VALID) | |
536 | mask |= POLLPRI; | |
537 | } | |
a4d25803 TM |
538 | /* This barrier is coupled with smp_wmb() in tcp_reset() */ |
539 | smp_rmb(); | |
4ed2d765 | 540 | if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue)) |
a4d25803 TM |
541 | mask |= POLLERR; |
542 | ||
1da177e4 LT |
543 | return mask; |
544 | } | |
4bc2f18b | 545 | EXPORT_SYMBOL(tcp_poll); |
1da177e4 LT |
546 | |
547 | int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg) | |
548 | { | |
549 | struct tcp_sock *tp = tcp_sk(sk); | |
550 | int answ; | |
0e71c55c | 551 | bool slow; |
1da177e4 LT |
552 | |
553 | switch (cmd) { | |
554 | case SIOCINQ: | |
555 | if (sk->sk_state == TCP_LISTEN) | |
556 | return -EINVAL; | |
557 | ||
0e71c55c | 558 | slow = lock_sock_fast(sk); |
473bd239 | 559 | answ = tcp_inq(sk); |
0e71c55c | 560 | unlock_sock_fast(sk, slow); |
1da177e4 LT |
561 | break; |
562 | case SIOCATMARK: | |
563 | answ = tp->urg_data && tp->urg_seq == tp->copied_seq; | |
564 | break; | |
565 | case SIOCOUTQ: | |
566 | if (sk->sk_state == TCP_LISTEN) | |
567 | return -EINVAL; | |
568 | ||
569 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) | |
570 | answ = 0; | |
571 | else | |
572 | answ = tp->write_seq - tp->snd_una; | |
573 | break; | |
2f4e1b39 MS |
574 | case SIOCOUTQNSD: |
575 | if (sk->sk_state == TCP_LISTEN) | |
576 | return -EINVAL; | |
577 | ||
578 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) | |
579 | answ = 0; | |
580 | else | |
581 | answ = tp->write_seq - tp->snd_nxt; | |
582 | break; | |
1da177e4 LT |
583 | default: |
584 | return -ENOIOCTLCMD; | |
3ff50b79 | 585 | } |
1da177e4 LT |
586 | |
587 | return put_user(answ, (int __user *)arg); | |
588 | } | |
4bc2f18b | 589 | EXPORT_SYMBOL(tcp_ioctl); |
1da177e4 | 590 | |
1da177e4 LT |
591 | static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb) |
592 | { | |
4de075e0 | 593 | TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH; |
1da177e4 LT |
594 | tp->pushed_seq = tp->write_seq; |
595 | } | |
596 | ||
a2a385d6 | 597 | static inline bool forced_push(const struct tcp_sock *tp) |
1da177e4 LT |
598 | { |
599 | return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1)); | |
600 | } | |
601 | ||
f4a775d1 | 602 | static void skb_entail(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 603 | { |
9e412ba7 | 604 | struct tcp_sock *tp = tcp_sk(sk); |
352d4800 ACM |
605 | struct tcp_skb_cb *tcb = TCP_SKB_CB(skb); |
606 | ||
607 | skb->csum = 0; | |
608 | tcb->seq = tcb->end_seq = tp->write_seq; | |
4de075e0 | 609 | tcb->tcp_flags = TCPHDR_ACK; |
352d4800 | 610 | tcb->sacked = 0; |
f4a775d1 | 611 | __skb_header_release(skb); |
fe067e8a | 612 | tcp_add_write_queue_tail(sk, skb); |
3ab224be HA |
613 | sk->sk_wmem_queued += skb->truesize; |
614 | sk_mem_charge(sk, skb->truesize); | |
89ebd197 | 615 | if (tp->nonagle & TCP_NAGLE_PUSH) |
e905a9ed | 616 | tp->nonagle &= ~TCP_NAGLE_PUSH; |
6f021c62 ED |
617 | |
618 | tcp_slow_start_after_idle_check(sk); | |
1da177e4 LT |
619 | } |
620 | ||
afeca340 | 621 | static inline void tcp_mark_urg(struct tcp_sock *tp, int flags) |
1da177e4 | 622 | { |
33f5f57e | 623 | if (flags & MSG_OOB) |
1da177e4 | 624 | tp->snd_up = tp->write_seq; |
1da177e4 LT |
625 | } |
626 | ||
f54b3111 | 627 | /* If a not yet filled skb is pushed, do not send it if |
a181ceb5 | 628 | * we have data packets in Qdisc or NIC queues : |
f54b3111 ED |
629 | * Because TX completion will happen shortly, it gives a chance |
630 | * to coalesce future sendmsg() payload into this skb, without | |
631 | * need for a timer, and with no latency trade off. | |
632 | * As packets containing data payload have a bigger truesize | |
a181ceb5 ED |
633 | * than pure acks (dataless) packets, the last checks prevent |
634 | * autocorking if we only have an ACK in Qdisc/NIC queues, | |
635 | * or if TX completion was delayed after we processed ACK packet. | |
f54b3111 ED |
636 | */ |
637 | static bool tcp_should_autocork(struct sock *sk, struct sk_buff *skb, | |
638 | int size_goal) | |
1da177e4 | 639 | { |
f54b3111 ED |
640 | return skb->len < size_goal && |
641 | sysctl_tcp_autocorking && | |
a181ceb5 | 642 | skb != tcp_write_queue_head(sk) && |
f54b3111 ED |
643 | atomic_read(&sk->sk_wmem_alloc) > skb->truesize; |
644 | } | |
645 | ||
646 | static void tcp_push(struct sock *sk, int flags, int mss_now, | |
647 | int nonagle, int size_goal) | |
648 | { | |
649 | struct tcp_sock *tp = tcp_sk(sk); | |
650 | struct sk_buff *skb; | |
afeca340 | 651 | |
f54b3111 ED |
652 | if (!tcp_send_head(sk)) |
653 | return; | |
afeca340 | 654 | |
f54b3111 ED |
655 | skb = tcp_write_queue_tail(sk); |
656 | if (!(flags & MSG_MORE) || forced_push(tp)) | |
657 | tcp_mark_push(tp, skb); | |
658 | ||
659 | tcp_mark_urg(tp, flags); | |
660 | ||
661 | if (tcp_should_autocork(sk, skb, size_goal)) { | |
662 | ||
663 | /* avoid atomic op if TSQ_THROTTLED bit is already set */ | |
664 | if (!test_bit(TSQ_THROTTLED, &tp->tsq_flags)) { | |
665 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAUTOCORKING); | |
666 | set_bit(TSQ_THROTTLED, &tp->tsq_flags); | |
667 | } | |
a181ceb5 ED |
668 | /* It is possible TX completion already happened |
669 | * before we set TSQ_THROTTLED. | |
670 | */ | |
671 | if (atomic_read(&sk->sk_wmem_alloc) > skb->truesize) | |
672 | return; | |
1da177e4 | 673 | } |
f54b3111 ED |
674 | |
675 | if (flags & MSG_MORE) | |
676 | nonagle = TCP_NAGLE_CORK; | |
677 | ||
678 | __tcp_push_pending_frames(sk, mss_now, nonagle); | |
1da177e4 LT |
679 | } |
680 | ||
6ff7751d AB |
681 | static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, |
682 | unsigned int offset, size_t len) | |
9c55e01c JA |
683 | { |
684 | struct tcp_splice_state *tss = rd_desc->arg.data; | |
33966dd0 | 685 | int ret; |
9c55e01c | 686 | |
a60e3cc7 HFS |
687 | ret = skb_splice_bits(skb, skb->sk, offset, tss->pipe, |
688 | min(rd_desc->count, len), tss->flags, | |
689 | skb_socket_splice); | |
33966dd0 WT |
690 | if (ret > 0) |
691 | rd_desc->count -= ret; | |
692 | return ret; | |
9c55e01c JA |
693 | } |
694 | ||
695 | static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss) | |
696 | { | |
697 | /* Store TCP splice context information in read_descriptor_t. */ | |
698 | read_descriptor_t rd_desc = { | |
699 | .arg.data = tss, | |
33966dd0 | 700 | .count = tss->len, |
9c55e01c JA |
701 | }; |
702 | ||
703 | return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv); | |
704 | } | |
705 | ||
706 | /** | |
707 | * tcp_splice_read - splice data from TCP socket to a pipe | |
708 | * @sock: socket to splice from | |
709 | * @ppos: position (not valid) | |
710 | * @pipe: pipe to splice to | |
711 | * @len: number of bytes to splice | |
712 | * @flags: splice modifier flags | |
713 | * | |
714 | * Description: | |
715 | * Will read pages from given socket and fill them into a pipe. | |
716 | * | |
717 | **/ | |
718 | ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos, | |
719 | struct pipe_inode_info *pipe, size_t len, | |
720 | unsigned int flags) | |
721 | { | |
722 | struct sock *sk = sock->sk; | |
723 | struct tcp_splice_state tss = { | |
724 | .pipe = pipe, | |
725 | .len = len, | |
726 | .flags = flags, | |
727 | }; | |
728 | long timeo; | |
729 | ssize_t spliced; | |
730 | int ret; | |
731 | ||
3a047bf8 | 732 | sock_rps_record_flow(sk); |
9c55e01c JA |
733 | /* |
734 | * We can't seek on a socket input | |
735 | */ | |
736 | if (unlikely(*ppos)) | |
737 | return -ESPIPE; | |
738 | ||
739 | ret = spliced = 0; | |
740 | ||
741 | lock_sock(sk); | |
742 | ||
42324c62 | 743 | timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK); |
9c55e01c JA |
744 | while (tss.len) { |
745 | ret = __tcp_splice_read(sk, &tss); | |
746 | if (ret < 0) | |
747 | break; | |
748 | else if (!ret) { | |
749 | if (spliced) | |
750 | break; | |
9c55e01c JA |
751 | if (sock_flag(sk, SOCK_DONE)) |
752 | break; | |
753 | if (sk->sk_err) { | |
754 | ret = sock_error(sk); | |
755 | break; | |
756 | } | |
757 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
758 | break; | |
759 | if (sk->sk_state == TCP_CLOSE) { | |
760 | /* | |
761 | * This occurs when user tries to read | |
762 | * from never connected socket. | |
763 | */ | |
764 | if (!sock_flag(sk, SOCK_DONE)) | |
765 | ret = -ENOTCONN; | |
766 | break; | |
767 | } | |
768 | if (!timeo) { | |
769 | ret = -EAGAIN; | |
770 | break; | |
771 | } | |
dfbafc99 | 772 | sk_wait_data(sk, &timeo, NULL); |
9c55e01c JA |
773 | if (signal_pending(current)) { |
774 | ret = sock_intr_errno(timeo); | |
775 | break; | |
776 | } | |
777 | continue; | |
778 | } | |
779 | tss.len -= ret; | |
780 | spliced += ret; | |
781 | ||
33966dd0 WT |
782 | if (!timeo) |
783 | break; | |
9c55e01c JA |
784 | release_sock(sk); |
785 | lock_sock(sk); | |
786 | ||
787 | if (sk->sk_err || sk->sk_state == TCP_CLOSE || | |
33966dd0 | 788 | (sk->sk_shutdown & RCV_SHUTDOWN) || |
9c55e01c JA |
789 | signal_pending(current)) |
790 | break; | |
791 | } | |
792 | ||
793 | release_sock(sk); | |
794 | ||
795 | if (spliced) | |
796 | return spliced; | |
797 | ||
798 | return ret; | |
799 | } | |
4bc2f18b | 800 | EXPORT_SYMBOL(tcp_splice_read); |
9c55e01c | 801 | |
eb934478 ED |
802 | struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, |
803 | bool force_schedule) | |
f561d0f2 PE |
804 | { |
805 | struct sk_buff *skb; | |
806 | ||
807 | /* The TCP header must be at least 32-bit aligned. */ | |
808 | size = ALIGN(size, 4); | |
809 | ||
8e4d980a ED |
810 | if (unlikely(tcp_under_memory_pressure(sk))) |
811 | sk_mem_reclaim_partial(sk); | |
812 | ||
f561d0f2 | 813 | skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp); |
8e4d980a | 814 | if (likely(skb)) { |
eb934478 | 815 | bool mem_scheduled; |
8e4d980a | 816 | |
eb934478 ED |
817 | if (force_schedule) { |
818 | mem_scheduled = true; | |
8e4d980a ED |
819 | sk_forced_mem_schedule(sk, skb->truesize); |
820 | } else { | |
eb934478 | 821 | mem_scheduled = sk_wmem_schedule(sk, skb->truesize); |
8e4d980a | 822 | } |
eb934478 | 823 | if (likely(mem_scheduled)) { |
a21d4572 | 824 | skb_reserve(skb, sk->sk_prot->max_header); |
f561d0f2 PE |
825 | /* |
826 | * Make sure that we have exactly size bytes | |
827 | * available to the caller, no more, no less. | |
828 | */ | |
16fad69c | 829 | skb->reserved_tailroom = skb->end - skb->tail - size; |
f561d0f2 PE |
830 | return skb; |
831 | } | |
832 | __kfree_skb(skb); | |
833 | } else { | |
5c52ba17 | 834 | sk->sk_prot->enter_memory_pressure(sk); |
f561d0f2 PE |
835 | sk_stream_moderate_sndbuf(sk); |
836 | } | |
837 | return NULL; | |
838 | } | |
839 | ||
0c54b85f IJ |
840 | static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now, |
841 | int large_allowed) | |
842 | { | |
843 | struct tcp_sock *tp = tcp_sk(sk); | |
6c09fa09 | 844 | u32 new_size_goal, size_goal; |
605ad7f1 ED |
845 | |
846 | if (!large_allowed || !sk_can_gso(sk)) | |
847 | return mss_now; | |
848 | ||
6c09fa09 ED |
849 | /* Note : tcp_tso_autosize() will eventually split this later */ |
850 | new_size_goal = sk->sk_gso_max_size - 1 - MAX_TCP_HEADER; | |
605ad7f1 ED |
851 | new_size_goal = tcp_bound_to_half_wnd(tp, new_size_goal); |
852 | ||
853 | /* We try hard to avoid divides here */ | |
854 | size_goal = tp->gso_segs * mss_now; | |
855 | if (unlikely(new_size_goal < size_goal || | |
856 | new_size_goal >= size_goal + mss_now)) { | |
857 | tp->gso_segs = min_t(u16, new_size_goal / mss_now, | |
858 | sk->sk_gso_max_segs); | |
859 | size_goal = tp->gso_segs * mss_now; | |
0c54b85f IJ |
860 | } |
861 | ||
605ad7f1 | 862 | return max(size_goal, mss_now); |
0c54b85f IJ |
863 | } |
864 | ||
865 | static int tcp_send_mss(struct sock *sk, int *size_goal, int flags) | |
866 | { | |
867 | int mss_now; | |
868 | ||
869 | mss_now = tcp_current_mss(sk); | |
870 | *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB)); | |
871 | ||
872 | return mss_now; | |
873 | } | |
874 | ||
64022d0b ED |
875 | static ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, |
876 | size_t size, int flags) | |
1da177e4 LT |
877 | { |
878 | struct tcp_sock *tp = tcp_sk(sk); | |
c1b4a7e6 | 879 | int mss_now, size_goal; |
1da177e4 LT |
880 | int err; |
881 | ssize_t copied; | |
882 | long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); | |
883 | ||
8336886f JC |
884 | /* Wait for a connection to finish. One exception is TCP Fast Open |
885 | * (passive side) where data is allowed to be sent before a connection | |
886 | * is fully established. | |
887 | */ | |
888 | if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) && | |
889 | !tcp_passive_fastopen(sk)) { | |
686a5624 YM |
890 | err = sk_stream_wait_connect(sk, &timeo); |
891 | if (err != 0) | |
1da177e4 | 892 | goto out_err; |
8336886f | 893 | } |
1da177e4 | 894 | |
9cd3e072 | 895 | sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 | 896 | |
0c54b85f | 897 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
898 | copied = 0; |
899 | ||
900 | err = -EPIPE; | |
901 | if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) | |
0d6a775e | 902 | goto out_err; |
1da177e4 | 903 | |
64022d0b | 904 | while (size > 0) { |
fe067e8a | 905 | struct sk_buff *skb = tcp_write_queue_tail(sk); |
38ba0a65 | 906 | int copy, i; |
38ba0a65 | 907 | bool can_coalesce; |
1da177e4 | 908 | |
fe067e8a | 909 | if (!tcp_send_head(sk) || (copy = size_goal - skb->len) <= 0) { |
1da177e4 LT |
910 | new_segment: |
911 | if (!sk_stream_memory_free(sk)) | |
912 | goto wait_for_sndbuf; | |
913 | ||
eb934478 ED |
914 | skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation, |
915 | skb_queue_empty(&sk->sk_write_queue)); | |
1da177e4 LT |
916 | if (!skb) |
917 | goto wait_for_memory; | |
918 | ||
9e412ba7 | 919 | skb_entail(sk, skb); |
c1b4a7e6 | 920 | copy = size_goal; |
1da177e4 LT |
921 | } |
922 | ||
923 | if (copy > size) | |
924 | copy = size; | |
925 | ||
926 | i = skb_shinfo(skb)->nr_frags; | |
927 | can_coalesce = skb_can_coalesce(skb, i, page, offset); | |
5f74f82e | 928 | if (!can_coalesce && i >= sysctl_max_skb_frags) { |
1da177e4 LT |
929 | tcp_mark_push(tp, skb); |
930 | goto new_segment; | |
931 | } | |
3ab224be | 932 | if (!sk_wmem_schedule(sk, copy)) |
1da177e4 | 933 | goto wait_for_memory; |
e905a9ed | 934 | |
1da177e4 | 935 | if (can_coalesce) { |
9e903e08 | 936 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); |
1da177e4 LT |
937 | } else { |
938 | get_page(page); | |
939 | skb_fill_page_desc(skb, i, page, offset, copy); | |
940 | } | |
c9af6db4 | 941 | skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; |
cef401de | 942 | |
1da177e4 LT |
943 | skb->len += copy; |
944 | skb->data_len += copy; | |
945 | skb->truesize += copy; | |
946 | sk->sk_wmem_queued += copy; | |
3ab224be | 947 | sk_mem_charge(sk, copy); |
84fa7933 | 948 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 LT |
949 | tp->write_seq += copy; |
950 | TCP_SKB_CB(skb)->end_seq += copy; | |
cd7d8498 | 951 | tcp_skb_pcount_set(skb, 0); |
1da177e4 LT |
952 | |
953 | if (!copied) | |
4de075e0 | 954 | TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH; |
1da177e4 LT |
955 | |
956 | copied += copy; | |
64022d0b | 957 | offset += copy; |
686a5624 YM |
958 | size -= copy; |
959 | if (!size) { | |
4ed2d765 | 960 | tcp_tx_timestamp(sk, skb); |
1da177e4 | 961 | goto out; |
4ed2d765 | 962 | } |
1da177e4 | 963 | |
69d15067 | 964 | if (skb->len < size_goal || (flags & MSG_OOB)) |
1da177e4 LT |
965 | continue; |
966 | ||
967 | if (forced_push(tp)) { | |
968 | tcp_mark_push(tp, skb); | |
9e412ba7 | 969 | __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH); |
fe067e8a | 970 | } else if (skb == tcp_send_head(sk)) |
1da177e4 LT |
971 | tcp_push_one(sk, mss_now); |
972 | continue; | |
973 | ||
974 | wait_for_sndbuf: | |
975 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
976 | wait_for_memory: | |
f54b3111 ED |
977 | tcp_push(sk, flags & ~MSG_MORE, mss_now, |
978 | TCP_NAGLE_PUSH, size_goal); | |
1da177e4 | 979 | |
686a5624 YM |
980 | err = sk_stream_wait_memory(sk, &timeo); |
981 | if (err != 0) | |
1da177e4 LT |
982 | goto do_error; |
983 | ||
0c54b85f | 984 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
985 | } |
986 | ||
987 | out: | |
35f9c09f | 988 | if (copied && !(flags & MSG_SENDPAGE_NOTLAST)) |
f54b3111 | 989 | tcp_push(sk, flags, mss_now, tp->nonagle, size_goal); |
1da177e4 LT |
990 | return copied; |
991 | ||
992 | do_error: | |
993 | if (copied) | |
994 | goto out; | |
995 | out_err: | |
ce5ec440 JB |
996 | /* make sure we wake any epoll edge trigger waiter */ |
997 | if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN)) | |
998 | sk->sk_write_space(sk); | |
1da177e4 LT |
999 | return sk_stream_error(sk, flags, err); |
1000 | } | |
1001 | ||
7ba42910 CG |
1002 | int tcp_sendpage(struct sock *sk, struct page *page, int offset, |
1003 | size_t size, int flags) | |
1da177e4 LT |
1004 | { |
1005 | ssize_t res; | |
1da177e4 | 1006 | |
1da177e4 | 1007 | if (!(sk->sk_route_caps & NETIF_F_SG) || |
9a49850d | 1008 | !sk_check_csum_caps(sk)) |
7ba42910 CG |
1009 | return sock_no_sendpage(sk->sk_socket, page, offset, size, |
1010 | flags); | |
1da177e4 | 1011 | |
1da177e4 | 1012 | lock_sock(sk); |
64022d0b | 1013 | res = do_tcp_sendpages(sk, page, offset, size, flags); |
1da177e4 LT |
1014 | release_sock(sk); |
1015 | return res; | |
1016 | } | |
4bc2f18b | 1017 | EXPORT_SYMBOL(tcp_sendpage); |
1da177e4 | 1018 | |
690e99c4 | 1019 | static inline int select_size(const struct sock *sk, bool sg) |
1da177e4 | 1020 | { |
cf533ea5 | 1021 | const struct tcp_sock *tp = tcp_sk(sk); |
c1b4a7e6 | 1022 | int tmp = tp->mss_cache; |
1da177e4 | 1023 | |
def87cf4 | 1024 | if (sg) { |
f07d960d ED |
1025 | if (sk_can_gso(sk)) { |
1026 | /* Small frames wont use a full page: | |
1027 | * Payload will immediately follow tcp header. | |
1028 | */ | |
1029 | tmp = SKB_WITH_OVERHEAD(2048 - MAX_TCP_HEADER); | |
1030 | } else { | |
b4e26f5e DM |
1031 | int pgbreak = SKB_MAX_HEAD(MAX_TCP_HEADER); |
1032 | ||
1033 | if (tmp >= pgbreak && | |
1034 | tmp <= pgbreak + (MAX_SKB_FRAGS - 1) * PAGE_SIZE) | |
1035 | tmp = pgbreak; | |
1036 | } | |
1037 | } | |
1da177e4 | 1038 | |
1da177e4 LT |
1039 | return tmp; |
1040 | } | |
1041 | ||
cf60af03 YC |
1042 | void tcp_free_fastopen_req(struct tcp_sock *tp) |
1043 | { | |
00db4124 | 1044 | if (tp->fastopen_req) { |
cf60af03 YC |
1045 | kfree(tp->fastopen_req); |
1046 | tp->fastopen_req = NULL; | |
1047 | } | |
1048 | } | |
1049 | ||
f5ddcbbb ED |
1050 | static int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, |
1051 | int *copied, size_t size) | |
cf60af03 YC |
1052 | { |
1053 | struct tcp_sock *tp = tcp_sk(sk); | |
1054 | int err, flags; | |
1055 | ||
1056 | if (!(sysctl_tcp_fastopen & TFO_CLIENT_ENABLE)) | |
1057 | return -EOPNOTSUPP; | |
00db4124 | 1058 | if (tp->fastopen_req) |
cf60af03 YC |
1059 | return -EALREADY; /* Another Fast Open is in progress */ |
1060 | ||
1061 | tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request), | |
1062 | sk->sk_allocation); | |
51456b29 | 1063 | if (unlikely(!tp->fastopen_req)) |
cf60af03 YC |
1064 | return -ENOBUFS; |
1065 | tp->fastopen_req->data = msg; | |
f5ddcbbb | 1066 | tp->fastopen_req->size = size; |
cf60af03 YC |
1067 | |
1068 | flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0; | |
1069 | err = __inet_stream_connect(sk->sk_socket, msg->msg_name, | |
1070 | msg->msg_namelen, flags); | |
f5ddcbbb | 1071 | *copied = tp->fastopen_req->copied; |
cf60af03 YC |
1072 | tcp_free_fastopen_req(tp); |
1073 | return err; | |
1074 | } | |
1075 | ||
1b784140 | 1076 | int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size) |
1da177e4 | 1077 | { |
1da177e4 LT |
1078 | struct tcp_sock *tp = tcp_sk(sk); |
1079 | struct sk_buff *skb; | |
57be5bda AV |
1080 | int flags, err, copied = 0; |
1081 | int mss_now = 0, size_goal, copied_syn = 0; | |
690e99c4 | 1082 | bool sg; |
1da177e4 LT |
1083 | long timeo; |
1084 | ||
1085 | lock_sock(sk); | |
1da177e4 LT |
1086 | |
1087 | flags = msg->msg_flags; | |
cf60af03 | 1088 | if (flags & MSG_FASTOPEN) { |
f5ddcbbb | 1089 | err = tcp_sendmsg_fastopen(sk, msg, &copied_syn, size); |
cf60af03 YC |
1090 | if (err == -EINPROGRESS && copied_syn > 0) |
1091 | goto out; | |
1092 | else if (err) | |
1093 | goto out_err; | |
cf60af03 YC |
1094 | } |
1095 | ||
1da177e4 LT |
1096 | timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); |
1097 | ||
8336886f JC |
1098 | /* Wait for a connection to finish. One exception is TCP Fast Open |
1099 | * (passive side) where data is allowed to be sent before a connection | |
1100 | * is fully established. | |
1101 | */ | |
1102 | if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) && | |
1103 | !tcp_passive_fastopen(sk)) { | |
686a5624 YM |
1104 | err = sk_stream_wait_connect(sk, &timeo); |
1105 | if (err != 0) | |
cf60af03 | 1106 | goto do_error; |
8336886f | 1107 | } |
1da177e4 | 1108 | |
c0e88ff0 PE |
1109 | if (unlikely(tp->repair)) { |
1110 | if (tp->repair_queue == TCP_RECV_QUEUE) { | |
1111 | copied = tcp_send_rcvq(sk, msg, size); | |
5924f17a | 1112 | goto out_nopush; |
c0e88ff0 PE |
1113 | } |
1114 | ||
1115 | err = -EINVAL; | |
1116 | if (tp->repair_queue == TCP_NO_QUEUE) | |
1117 | goto out_err; | |
1118 | ||
1119 | /* 'common' sending to sendq */ | |
1120 | } | |
1121 | ||
1da177e4 | 1122 | /* This should be in poll */ |
9cd3e072 | 1123 | sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 | 1124 | |
0c54b85f | 1125 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
1126 | |
1127 | /* Ok commence sending. */ | |
1da177e4 LT |
1128 | copied = 0; |
1129 | ||
1130 | err = -EPIPE; | |
1131 | if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) | |
0d6a775e | 1132 | goto out_err; |
1da177e4 | 1133 | |
690e99c4 | 1134 | sg = !!(sk->sk_route_caps & NETIF_F_SG); |
def87cf4 | 1135 | |
01e97e65 | 1136 | while (msg_data_left(msg)) { |
57be5bda AV |
1137 | int copy = 0; |
1138 | int max = size_goal; | |
1da177e4 | 1139 | |
57be5bda AV |
1140 | skb = tcp_write_queue_tail(sk); |
1141 | if (tcp_send_head(sk)) { | |
1142 | if (skb->ip_summed == CHECKSUM_NONE) | |
1143 | max = mss_now; | |
1144 | copy = max - skb->len; | |
cf60af03 | 1145 | } |
1da177e4 | 1146 | |
57be5bda | 1147 | if (copy <= 0) { |
1da177e4 | 1148 | new_segment: |
57be5bda AV |
1149 | /* Allocate new segment. If the interface is SG, |
1150 | * allocate skb fitting to single page. | |
1151 | */ | |
1152 | if (!sk_stream_memory_free(sk)) | |
1153 | goto wait_for_sndbuf; | |
1da177e4 | 1154 | |
57be5bda AV |
1155 | skb = sk_stream_alloc_skb(sk, |
1156 | select_size(sk, sg), | |
eb934478 ED |
1157 | sk->sk_allocation, |
1158 | skb_queue_empty(&sk->sk_write_queue)); | |
57be5bda AV |
1159 | if (!skb) |
1160 | goto wait_for_memory; | |
1da177e4 | 1161 | |
57be5bda AV |
1162 | /* |
1163 | * Check whether we can use HW checksum. | |
1164 | */ | |
9a49850d | 1165 | if (sk_check_csum_caps(sk)) |
57be5bda | 1166 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 | 1167 | |
57be5bda AV |
1168 | skb_entail(sk, skb); |
1169 | copy = size_goal; | |
1170 | max = size_goal; | |
9d186cac | 1171 | |
57be5bda AV |
1172 | /* All packets are restored as if they have |
1173 | * already been sent. skb_mstamp isn't set to | |
1174 | * avoid wrong rtt estimation. | |
1175 | */ | |
1176 | if (tp->repair) | |
1177 | TCP_SKB_CB(skb)->sacked |= TCPCB_REPAIRED; | |
1178 | } | |
1da177e4 | 1179 | |
57be5bda | 1180 | /* Try to append data to the end of skb. */ |
01e97e65 AV |
1181 | if (copy > msg_data_left(msg)) |
1182 | copy = msg_data_left(msg); | |
57be5bda AV |
1183 | |
1184 | /* Where to copy to? */ | |
1185 | if (skb_availroom(skb) > 0) { | |
1186 | /* We have some space in skb head. Superb! */ | |
1187 | copy = min_t(int, copy, skb_availroom(skb)); | |
1188 | err = skb_add_data_nocache(sk, skb, &msg->msg_iter, copy); | |
1189 | if (err) | |
1190 | goto do_fault; | |
1191 | } else { | |
1192 | bool merge = true; | |
1193 | int i = skb_shinfo(skb)->nr_frags; | |
1194 | struct page_frag *pfrag = sk_page_frag(sk); | |
1195 | ||
1196 | if (!sk_page_frag_refill(sk, pfrag)) | |
1197 | goto wait_for_memory; | |
ef015786 | 1198 | |
57be5bda AV |
1199 | if (!skb_can_coalesce(skb, i, pfrag->page, |
1200 | pfrag->offset)) { | |
5f74f82e | 1201 | if (i == sysctl_max_skb_frags || !sg) { |
57be5bda AV |
1202 | tcp_mark_push(tp, skb); |
1203 | goto new_segment; | |
1da177e4 | 1204 | } |
57be5bda | 1205 | merge = false; |
1da177e4 LT |
1206 | } |
1207 | ||
57be5bda AV |
1208 | copy = min_t(int, copy, pfrag->size - pfrag->offset); |
1209 | ||
1210 | if (!sk_wmem_schedule(sk, copy)) | |
1211 | goto wait_for_memory; | |
1da177e4 | 1212 | |
57be5bda AV |
1213 | err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb, |
1214 | pfrag->page, | |
1215 | pfrag->offset, | |
1216 | copy); | |
1217 | if (err) | |
1218 | goto do_error; | |
1da177e4 | 1219 | |
57be5bda AV |
1220 | /* Update the skb. */ |
1221 | if (merge) { | |
1222 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); | |
1223 | } else { | |
1224 | skb_fill_page_desc(skb, i, pfrag->page, | |
1225 | pfrag->offset, copy); | |
1226 | get_page(pfrag->page); | |
4ed2d765 | 1227 | } |
57be5bda AV |
1228 | pfrag->offset += copy; |
1229 | } | |
1230 | ||
1231 | if (!copied) | |
1232 | TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH; | |
1233 | ||
1234 | tp->write_seq += copy; | |
1235 | TCP_SKB_CB(skb)->end_seq += copy; | |
1236 | tcp_skb_pcount_set(skb, 0); | |
1da177e4 | 1237 | |
57be5bda | 1238 | copied += copy; |
01e97e65 | 1239 | if (!msg_data_left(msg)) { |
57be5bda AV |
1240 | tcp_tx_timestamp(sk, skb); |
1241 | goto out; | |
1242 | } | |
1da177e4 | 1243 | |
57be5bda | 1244 | if (skb->len < max || (flags & MSG_OOB) || unlikely(tp->repair)) |
1da177e4 LT |
1245 | continue; |
1246 | ||
57be5bda AV |
1247 | if (forced_push(tp)) { |
1248 | tcp_mark_push(tp, skb); | |
1249 | __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH); | |
1250 | } else if (skb == tcp_send_head(sk)) | |
1251 | tcp_push_one(sk, mss_now); | |
1252 | continue; | |
1253 | ||
1da177e4 | 1254 | wait_for_sndbuf: |
57be5bda | 1255 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
1da177e4 | 1256 | wait_for_memory: |
57be5bda AV |
1257 | if (copied) |
1258 | tcp_push(sk, flags & ~MSG_MORE, mss_now, | |
1259 | TCP_NAGLE_PUSH, size_goal); | |
1da177e4 | 1260 | |
686a5624 YM |
1261 | err = sk_stream_wait_memory(sk, &timeo); |
1262 | if (err != 0) | |
57be5bda | 1263 | goto do_error; |
1da177e4 | 1264 | |
57be5bda | 1265 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
1266 | } |
1267 | ||
1268 | out: | |
ec342325 | 1269 | if (copied) |
f54b3111 | 1270 | tcp_push(sk, flags, mss_now, tp->nonagle, size_goal); |
5924f17a | 1271 | out_nopush: |
1da177e4 | 1272 | release_sock(sk); |
cf60af03 | 1273 | return copied + copied_syn; |
1da177e4 LT |
1274 | |
1275 | do_fault: | |
1276 | if (!skb->len) { | |
fe067e8a DM |
1277 | tcp_unlink_write_queue(skb, sk); |
1278 | /* It is the one place in all of TCP, except connection | |
1279 | * reset, where we can be unlinking the send_head. | |
1280 | */ | |
1281 | tcp_check_send_head(sk, skb); | |
3ab224be | 1282 | sk_wmem_free_skb(sk, skb); |
1da177e4 LT |
1283 | } |
1284 | ||
1285 | do_error: | |
cf60af03 | 1286 | if (copied + copied_syn) |
1da177e4 LT |
1287 | goto out; |
1288 | out_err: | |
1289 | err = sk_stream_error(sk, flags, err); | |
ce5ec440 JB |
1290 | /* make sure we wake any epoll edge trigger waiter */ |
1291 | if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN)) | |
1292 | sk->sk_write_space(sk); | |
1da177e4 LT |
1293 | release_sock(sk); |
1294 | return err; | |
1295 | } | |
4bc2f18b | 1296 | EXPORT_SYMBOL(tcp_sendmsg); |
1da177e4 LT |
1297 | |
1298 | /* | |
1299 | * Handle reading urgent data. BSD has very simple semantics for | |
1300 | * this, no blocking and very strange errors 8) | |
1301 | */ | |
1302 | ||
377f0a08 | 1303 | static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags) |
1da177e4 LT |
1304 | { |
1305 | struct tcp_sock *tp = tcp_sk(sk); | |
1306 | ||
1307 | /* No URG data to read. */ | |
1308 | if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data || | |
1309 | tp->urg_data == TCP_URG_READ) | |
1310 | return -EINVAL; /* Yes this is right ! */ | |
1311 | ||
1312 | if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE)) | |
1313 | return -ENOTCONN; | |
1314 | ||
1315 | if (tp->urg_data & TCP_URG_VALID) { | |
1316 | int err = 0; | |
1317 | char c = tp->urg_data; | |
1318 | ||
1319 | if (!(flags & MSG_PEEK)) | |
1320 | tp->urg_data = TCP_URG_READ; | |
1321 | ||
1322 | /* Read urgent data. */ | |
1323 | msg->msg_flags |= MSG_OOB; | |
1324 | ||
1325 | if (len > 0) { | |
1326 | if (!(flags & MSG_TRUNC)) | |
7eab8d9e | 1327 | err = memcpy_to_msg(msg, &c, 1); |
1da177e4 LT |
1328 | len = 1; |
1329 | } else | |
1330 | msg->msg_flags |= MSG_TRUNC; | |
1331 | ||
1332 | return err ? -EFAULT : len; | |
1333 | } | |
1334 | ||
1335 | if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN)) | |
1336 | return 0; | |
1337 | ||
1338 | /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and | |
1339 | * the available implementations agree in this case: | |
1340 | * this call should never block, independent of the | |
1341 | * blocking state of the socket. | |
1342 | * Mike <pall@rz.uni-karlsruhe.de> | |
1343 | */ | |
1344 | return -EAGAIN; | |
1345 | } | |
1346 | ||
c0e88ff0 PE |
1347 | static int tcp_peek_sndq(struct sock *sk, struct msghdr *msg, int len) |
1348 | { | |
1349 | struct sk_buff *skb; | |
1350 | int copied = 0, err = 0; | |
1351 | ||
1352 | /* XXX -- need to support SO_PEEK_OFF */ | |
1353 | ||
1354 | skb_queue_walk(&sk->sk_write_queue, skb) { | |
51f3d02b | 1355 | err = skb_copy_datagram_msg(skb, 0, msg, skb->len); |
c0e88ff0 PE |
1356 | if (err) |
1357 | break; | |
1358 | ||
1359 | copied += skb->len; | |
1360 | } | |
1361 | ||
1362 | return err ?: copied; | |
1363 | } | |
1364 | ||
1da177e4 LT |
1365 | /* Clean up the receive buffer for full frames taken by the user, |
1366 | * then send an ACK if necessary. COPIED is the number of bytes | |
1367 | * tcp_recvmsg has given to the user so far, it speeds up the | |
1368 | * calculation of whether or not we must ACK for the sake of | |
1369 | * a window update. | |
1370 | */ | |
3f334078 | 1371 | static void tcp_cleanup_rbuf(struct sock *sk, int copied) |
1da177e4 LT |
1372 | { |
1373 | struct tcp_sock *tp = tcp_sk(sk); | |
a2a385d6 | 1374 | bool time_to_ack = false; |
1da177e4 | 1375 | |
1da177e4 LT |
1376 | struct sk_buff *skb = skb_peek(&sk->sk_receive_queue); |
1377 | ||
d792c100 | 1378 | WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq), |
2af6fd8b | 1379 | "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n", |
d792c100 | 1380 | tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt); |
1da177e4 | 1381 | |
463c84b9 ACM |
1382 | if (inet_csk_ack_scheduled(sk)) { |
1383 | const struct inet_connection_sock *icsk = inet_csk(sk); | |
1da177e4 LT |
1384 | /* Delayed ACKs frequently hit locked sockets during bulk |
1385 | * receive. */ | |
463c84b9 | 1386 | if (icsk->icsk_ack.blocked || |
1da177e4 | 1387 | /* Once-per-two-segments ACK was not sent by tcp_input.c */ |
463c84b9 | 1388 | tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss || |
1da177e4 LT |
1389 | /* |
1390 | * If this read emptied read buffer, we send ACK, if | |
1391 | * connection is not bidirectional, user drained | |
1392 | * receive buffer and there was a small segment | |
1393 | * in queue. | |
1394 | */ | |
1ef9696c AK |
1395 | (copied > 0 && |
1396 | ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) || | |
1397 | ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) && | |
1398 | !icsk->icsk_ack.pingpong)) && | |
1399 | !atomic_read(&sk->sk_rmem_alloc))) | |
a2a385d6 | 1400 | time_to_ack = true; |
1da177e4 LT |
1401 | } |
1402 | ||
1403 | /* We send an ACK if we can now advertise a non-zero window | |
1404 | * which has been raised "significantly". | |
1405 | * | |
1406 | * Even if window raised up to infinity, do not send window open ACK | |
1407 | * in states, where we will not receive more. It is useless. | |
1408 | */ | |
1409 | if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) { | |
1410 | __u32 rcv_window_now = tcp_receive_window(tp); | |
1411 | ||
1412 | /* Optimize, __tcp_select_window() is not cheap. */ | |
1413 | if (2*rcv_window_now <= tp->window_clamp) { | |
1414 | __u32 new_window = __tcp_select_window(sk); | |
1415 | ||
1416 | /* Send ACK now, if this read freed lots of space | |
1417 | * in our buffer. Certainly, new_window is new window. | |
1418 | * We can advertise it now, if it is not less than current one. | |
1419 | * "Lots" means "at least twice" here. | |
1420 | */ | |
1421 | if (new_window && new_window >= 2 * rcv_window_now) | |
a2a385d6 | 1422 | time_to_ack = true; |
1da177e4 LT |
1423 | } |
1424 | } | |
1425 | if (time_to_ack) | |
1426 | tcp_send_ack(sk); | |
1427 | } | |
1428 | ||
1429 | static void tcp_prequeue_process(struct sock *sk) | |
1430 | { | |
1431 | struct sk_buff *skb; | |
1432 | struct tcp_sock *tp = tcp_sk(sk); | |
1433 | ||
6f67c817 | 1434 | NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPPREQUEUED); |
1da177e4 LT |
1435 | |
1436 | /* RX process wants to run with disabled BHs, though it is not | |
1437 | * necessary */ | |
1438 | local_bh_disable(); | |
1439 | while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) | |
c57943a1 | 1440 | sk_backlog_rcv(sk, skb); |
1da177e4 LT |
1441 | local_bh_enable(); |
1442 | ||
1443 | /* Clear memory counter. */ | |
1444 | tp->ucopy.memory = 0; | |
1445 | } | |
1446 | ||
f26845b4 | 1447 | static struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off) |
1da177e4 LT |
1448 | { |
1449 | struct sk_buff *skb; | |
1450 | u32 offset; | |
1451 | ||
f26845b4 | 1452 | while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) { |
1da177e4 | 1453 | offset = seq - TCP_SKB_CB(skb)->seq; |
9d691539 ED |
1454 | if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { |
1455 | pr_err_once("%s: found a SYN, please report !\n", __func__); | |
1da177e4 | 1456 | offset--; |
9d691539 | 1457 | } |
e11ecddf | 1458 | if (offset < skb->len || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) { |
1da177e4 LT |
1459 | *off = offset; |
1460 | return skb; | |
1461 | } | |
f26845b4 ED |
1462 | /* This looks weird, but this can happen if TCP collapsing |
1463 | * splitted a fat GRO packet, while we released socket lock | |
1464 | * in skb_splice_bits() | |
1465 | */ | |
7bced397 | 1466 | sk_eat_skb(sk, skb); |
1da177e4 LT |
1467 | } |
1468 | return NULL; | |
1469 | } | |
1470 | ||
1471 | /* | |
1472 | * This routine provides an alternative to tcp_recvmsg() for routines | |
1473 | * that would like to handle copying from skbuffs directly in 'sendfile' | |
1474 | * fashion. | |
1475 | * Note: | |
1476 | * - It is assumed that the socket was locked by the caller. | |
1477 | * - The routine does not block. | |
1478 | * - At present, there is no support for reading OOB data | |
1479 | * or for 'peeking' the socket using this routine | |
1480 | * (although both would be easy to implement). | |
1481 | */ | |
1482 | int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, | |
1483 | sk_read_actor_t recv_actor) | |
1484 | { | |
1485 | struct sk_buff *skb; | |
1486 | struct tcp_sock *tp = tcp_sk(sk); | |
1487 | u32 seq = tp->copied_seq; | |
1488 | u32 offset; | |
1489 | int copied = 0; | |
1490 | ||
1491 | if (sk->sk_state == TCP_LISTEN) | |
1492 | return -ENOTCONN; | |
1493 | while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) { | |
1494 | if (offset < skb->len) { | |
374e7b59 OP |
1495 | int used; |
1496 | size_t len; | |
1da177e4 LT |
1497 | |
1498 | len = skb->len - offset; | |
1499 | /* Stop reading if we hit a patch of urgent data */ | |
1500 | if (tp->urg_data) { | |
1501 | u32 urg_offset = tp->urg_seq - seq; | |
1502 | if (urg_offset < len) | |
1503 | len = urg_offset; | |
1504 | if (!len) | |
1505 | break; | |
1506 | } | |
1507 | used = recv_actor(desc, skb, offset, len); | |
ff905b1e | 1508 | if (used <= 0) { |
ddb61a57 JA |
1509 | if (!copied) |
1510 | copied = used; | |
1511 | break; | |
1512 | } else if (used <= len) { | |
1da177e4 LT |
1513 | seq += used; |
1514 | copied += used; | |
1515 | offset += used; | |
1516 | } | |
02275a2e | 1517 | /* If recv_actor drops the lock (e.g. TCP splice |
293ad604 OP |
1518 | * receive) the skb pointer might be invalid when |
1519 | * getting here: tcp_collapse might have deleted it | |
1520 | * while aggregating skbs from the socket queue. | |
1521 | */ | |
02275a2e WT |
1522 | skb = tcp_recv_skb(sk, seq - 1, &offset); |
1523 | if (!skb) | |
1da177e4 | 1524 | break; |
02275a2e WT |
1525 | /* TCP coalescing might have appended data to the skb. |
1526 | * Try to splice more frags | |
1527 | */ | |
1528 | if (offset + 1 != skb->len) | |
1529 | continue; | |
1da177e4 | 1530 | } |
e11ecddf | 1531 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) { |
7bced397 | 1532 | sk_eat_skb(sk, skb); |
1da177e4 LT |
1533 | ++seq; |
1534 | break; | |
1535 | } | |
7bced397 | 1536 | sk_eat_skb(sk, skb); |
1da177e4 LT |
1537 | if (!desc->count) |
1538 | break; | |
baff42ab | 1539 | tp->copied_seq = seq; |
1da177e4 LT |
1540 | } |
1541 | tp->copied_seq = seq; | |
1542 | ||
1543 | tcp_rcv_space_adjust(sk); | |
1544 | ||
1545 | /* Clean up data we have read: This will do ACK frames. */ | |
f26845b4 ED |
1546 | if (copied > 0) { |
1547 | tcp_recv_skb(sk, seq, &offset); | |
0e4b4992 | 1548 | tcp_cleanup_rbuf(sk, copied); |
f26845b4 | 1549 | } |
1da177e4 LT |
1550 | return copied; |
1551 | } | |
4bc2f18b | 1552 | EXPORT_SYMBOL(tcp_read_sock); |
1da177e4 LT |
1553 | |
1554 | /* | |
1555 | * This routine copies from a sock struct into the user buffer. | |
1556 | * | |
1557 | * Technical note: in 2.3 we work on _locked_ socket, so that | |
1558 | * tricks with *seq access order and skb->users are not required. | |
1559 | * Probably, code can be easily improved even more. | |
1560 | */ | |
1561 | ||
1b784140 YX |
1562 | int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, |
1563 | int flags, int *addr_len) | |
1da177e4 LT |
1564 | { |
1565 | struct tcp_sock *tp = tcp_sk(sk); | |
1566 | int copied = 0; | |
1567 | u32 peek_seq; | |
1568 | u32 *seq; | |
1569 | unsigned long used; | |
1570 | int err; | |
1571 | int target; /* Read at least this many bytes */ | |
1572 | long timeo; | |
1573 | struct task_struct *user_recv = NULL; | |
dfbafc99 | 1574 | struct sk_buff *skb, *last; |
77527313 | 1575 | u32 urg_hole = 0; |
1da177e4 | 1576 | |
4ed2d765 | 1577 | if (unlikely(flags & MSG_ERRQUEUE)) |
f4713a3d | 1578 | return inet_recv_error(sk, msg, len, addr_len); |
4ed2d765 | 1579 | |
cbf55001 ET |
1580 | if (sk_can_busy_loop(sk) && skb_queue_empty(&sk->sk_receive_queue) && |
1581 | (sk->sk_state == TCP_ESTABLISHED)) | |
1582 | sk_busy_loop(sk, nonblock); | |
d30e383b | 1583 | |
1da177e4 LT |
1584 | lock_sock(sk); |
1585 | ||
1da177e4 LT |
1586 | err = -ENOTCONN; |
1587 | if (sk->sk_state == TCP_LISTEN) | |
1588 | goto out; | |
1589 | ||
1590 | timeo = sock_rcvtimeo(sk, nonblock); | |
1591 | ||
1592 | /* Urgent data needs to be handled specially. */ | |
1593 | if (flags & MSG_OOB) | |
1594 | goto recv_urg; | |
1595 | ||
c0e88ff0 PE |
1596 | if (unlikely(tp->repair)) { |
1597 | err = -EPERM; | |
1598 | if (!(flags & MSG_PEEK)) | |
1599 | goto out; | |
1600 | ||
1601 | if (tp->repair_queue == TCP_SEND_QUEUE) | |
1602 | goto recv_sndq; | |
1603 | ||
1604 | err = -EINVAL; | |
1605 | if (tp->repair_queue == TCP_NO_QUEUE) | |
1606 | goto out; | |
1607 | ||
1608 | /* 'common' recv queue MSG_PEEK-ing */ | |
1609 | } | |
1610 | ||
1da177e4 LT |
1611 | seq = &tp->copied_seq; |
1612 | if (flags & MSG_PEEK) { | |
1613 | peek_seq = tp->copied_seq; | |
1614 | seq = &peek_seq; | |
1615 | } | |
1616 | ||
1617 | target = sock_rcvlowat(sk, flags & MSG_WAITALL, len); | |
1618 | ||
1619 | do { | |
1da177e4 LT |
1620 | u32 offset; |
1621 | ||
1622 | /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */ | |
1623 | if (tp->urg_data && tp->urg_seq == *seq) { | |
1624 | if (copied) | |
1625 | break; | |
1626 | if (signal_pending(current)) { | |
1627 | copied = timeo ? sock_intr_errno(timeo) : -EAGAIN; | |
1628 | break; | |
1629 | } | |
1630 | } | |
1631 | ||
1632 | /* Next get a buffer. */ | |
1633 | ||
dfbafc99 | 1634 | last = skb_peek_tail(&sk->sk_receive_queue); |
91521944 | 1635 | skb_queue_walk(&sk->sk_receive_queue, skb) { |
dfbafc99 | 1636 | last = skb; |
1da177e4 LT |
1637 | /* Now that we have two receive queues this |
1638 | * shouldn't happen. | |
1639 | */ | |
d792c100 | 1640 | if (WARN(before(*seq, TCP_SKB_CB(skb)->seq), |
2af6fd8b JP |
1641 | "recvmsg bug: copied %X seq %X rcvnxt %X fl %X\n", |
1642 | *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, | |
1643 | flags)) | |
1da177e4 | 1644 | break; |
d792c100 | 1645 | |
1da177e4 | 1646 | offset = *seq - TCP_SKB_CB(skb)->seq; |
9d691539 ED |
1647 | if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { |
1648 | pr_err_once("%s: found a SYN, please report !\n", __func__); | |
1da177e4 | 1649 | offset--; |
9d691539 | 1650 | } |
1da177e4 LT |
1651 | if (offset < skb->len) |
1652 | goto found_ok_skb; | |
e11ecddf | 1653 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) |
1da177e4 | 1654 | goto found_fin_ok; |
2af6fd8b JP |
1655 | WARN(!(flags & MSG_PEEK), |
1656 | "recvmsg bug 2: copied %X seq %X rcvnxt %X fl %X\n", | |
1657 | *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags); | |
91521944 | 1658 | } |
1da177e4 LT |
1659 | |
1660 | /* Well, if we have backlog, try to process it now yet. */ | |
1661 | ||
1662 | if (copied >= target && !sk->sk_backlog.tail) | |
1663 | break; | |
1664 | ||
1665 | if (copied) { | |
1666 | if (sk->sk_err || | |
1667 | sk->sk_state == TCP_CLOSE || | |
1668 | (sk->sk_shutdown & RCV_SHUTDOWN) || | |
1669 | !timeo || | |
518a09ef | 1670 | signal_pending(current)) |
1da177e4 LT |
1671 | break; |
1672 | } else { | |
1673 | if (sock_flag(sk, SOCK_DONE)) | |
1674 | break; | |
1675 | ||
1676 | if (sk->sk_err) { | |
1677 | copied = sock_error(sk); | |
1678 | break; | |
1679 | } | |
1680 | ||
1681 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
1682 | break; | |
1683 | ||
1684 | if (sk->sk_state == TCP_CLOSE) { | |
1685 | if (!sock_flag(sk, SOCK_DONE)) { | |
1686 | /* This occurs when user tries to read | |
1687 | * from never connected socket. | |
1688 | */ | |
1689 | copied = -ENOTCONN; | |
1690 | break; | |
1691 | } | |
1692 | break; | |
1693 | } | |
1694 | ||
1695 | if (!timeo) { | |
1696 | copied = -EAGAIN; | |
1697 | break; | |
1698 | } | |
1699 | ||
1700 | if (signal_pending(current)) { | |
1701 | copied = sock_intr_errno(timeo); | |
1702 | break; | |
1703 | } | |
1704 | } | |
1705 | ||
0e4b4992 | 1706 | tcp_cleanup_rbuf(sk, copied); |
1da177e4 | 1707 | |
7df55125 | 1708 | if (!sysctl_tcp_low_latency && tp->ucopy.task == user_recv) { |
1da177e4 LT |
1709 | /* Install new reader */ |
1710 | if (!user_recv && !(flags & (MSG_TRUNC | MSG_PEEK))) { | |
1711 | user_recv = current; | |
1712 | tp->ucopy.task = user_recv; | |
f4362a2c | 1713 | tp->ucopy.msg = msg; |
1da177e4 LT |
1714 | } |
1715 | ||
1716 | tp->ucopy.len = len; | |
1717 | ||
547b792c IJ |
1718 | WARN_ON(tp->copied_seq != tp->rcv_nxt && |
1719 | !(flags & (MSG_PEEK | MSG_TRUNC))); | |
1da177e4 LT |
1720 | |
1721 | /* Ugly... If prequeue is not empty, we have to | |
1722 | * process it before releasing socket, otherwise | |
1723 | * order will be broken at second iteration. | |
1724 | * More elegant solution is required!!! | |
1725 | * | |
1726 | * Look: we have the following (pseudo)queues: | |
1727 | * | |
1728 | * 1. packets in flight | |
1729 | * 2. backlog | |
1730 | * 3. prequeue | |
1731 | * 4. receive_queue | |
1732 | * | |
1733 | * Each queue can be processed only if the next ones | |
1734 | * are empty. At this point we have empty receive_queue. | |
1735 | * But prequeue _can_ be not empty after 2nd iteration, | |
1736 | * when we jumped to start of loop because backlog | |
1737 | * processing added something to receive_queue. | |
1738 | * We cannot release_sock(), because backlog contains | |
1739 | * packets arrived _after_ prequeued ones. | |
1740 | * | |
1741 | * Shortly, algorithm is clear --- to process all | |
1742 | * the queues in order. We could make it more directly, | |
1743 | * requeueing packets from backlog to prequeue, if | |
1744 | * is not empty. It is more elegant, but eats cycles, | |
1745 | * unfortunately. | |
1746 | */ | |
b03efcfb | 1747 | if (!skb_queue_empty(&tp->ucopy.prequeue)) |
1da177e4 LT |
1748 | goto do_prequeue; |
1749 | ||
1750 | /* __ Set realtime policy in scheduler __ */ | |
1751 | } | |
1752 | ||
1753 | if (copied >= target) { | |
1754 | /* Do not sleep, just process backlog. */ | |
1755 | release_sock(sk); | |
1756 | lock_sock(sk); | |
dfbafc99 SD |
1757 | } else { |
1758 | sk_wait_data(sk, &timeo, last); | |
1759 | } | |
1da177e4 LT |
1760 | |
1761 | if (user_recv) { | |
1762 | int chunk; | |
1763 | ||
1764 | /* __ Restore normal policy in scheduler __ */ | |
1765 | ||
686a5624 YM |
1766 | chunk = len - tp->ucopy.len; |
1767 | if (chunk != 0) { | |
ed88098e | 1768 | NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMBACKLOG, chunk); |
1da177e4 LT |
1769 | len -= chunk; |
1770 | copied += chunk; | |
1771 | } | |
1772 | ||
1773 | if (tp->rcv_nxt == tp->copied_seq && | |
b03efcfb | 1774 | !skb_queue_empty(&tp->ucopy.prequeue)) { |
1da177e4 LT |
1775 | do_prequeue: |
1776 | tcp_prequeue_process(sk); | |
1777 | ||
686a5624 YM |
1778 | chunk = len - tp->ucopy.len; |
1779 | if (chunk != 0) { | |
ed88098e | 1780 | NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk); |
1da177e4 LT |
1781 | len -= chunk; |
1782 | copied += chunk; | |
1783 | } | |
1784 | } | |
1785 | } | |
77527313 IJ |
1786 | if ((flags & MSG_PEEK) && |
1787 | (peek_seq - copied - urg_hole != tp->copied_seq)) { | |
e87cc472 JP |
1788 | net_dbg_ratelimited("TCP(%s:%d): Application bug, race in MSG_PEEK\n", |
1789 | current->comm, | |
1790 | task_pid_nr(current)); | |
1da177e4 LT |
1791 | peek_seq = tp->copied_seq; |
1792 | } | |
1793 | continue; | |
1794 | ||
1795 | found_ok_skb: | |
1796 | /* Ok so how much can we use? */ | |
1797 | used = skb->len - offset; | |
1798 | if (len < used) | |
1799 | used = len; | |
1800 | ||
1801 | /* Do we have urgent data here? */ | |
1802 | if (tp->urg_data) { | |
1803 | u32 urg_offset = tp->urg_seq - *seq; | |
1804 | if (urg_offset < used) { | |
1805 | if (!urg_offset) { | |
1806 | if (!sock_flag(sk, SOCK_URGINLINE)) { | |
1807 | ++*seq; | |
77527313 | 1808 | urg_hole++; |
1da177e4 LT |
1809 | offset++; |
1810 | used--; | |
1811 | if (!used) | |
1812 | goto skip_copy; | |
1813 | } | |
1814 | } else | |
1815 | used = urg_offset; | |
1816 | } | |
1817 | } | |
1818 | ||
1819 | if (!(flags & MSG_TRUNC)) { | |
51f3d02b | 1820 | err = skb_copy_datagram_msg(skb, offset, msg, used); |
7bced397 DW |
1821 | if (err) { |
1822 | /* Exception. Bailout! */ | |
1823 | if (!copied) | |
1824 | copied = -EFAULT; | |
1825 | break; | |
1da177e4 LT |
1826 | } |
1827 | } | |
1828 | ||
1829 | *seq += used; | |
1830 | copied += used; | |
1831 | len -= used; | |
1832 | ||
1833 | tcp_rcv_space_adjust(sk); | |
1834 | ||
1835 | skip_copy: | |
1836 | if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) { | |
1837 | tp->urg_data = 0; | |
9e412ba7 | 1838 | tcp_fast_path_check(sk); |
1da177e4 LT |
1839 | } |
1840 | if (used + offset < skb->len) | |
1841 | continue; | |
1842 | ||
e11ecddf | 1843 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) |
1da177e4 | 1844 | goto found_fin_ok; |
7bced397 DW |
1845 | if (!(flags & MSG_PEEK)) |
1846 | sk_eat_skb(sk, skb); | |
1da177e4 LT |
1847 | continue; |
1848 | ||
1849 | found_fin_ok: | |
1850 | /* Process the FIN. */ | |
1851 | ++*seq; | |
7bced397 DW |
1852 | if (!(flags & MSG_PEEK)) |
1853 | sk_eat_skb(sk, skb); | |
1da177e4 LT |
1854 | break; |
1855 | } while (len > 0); | |
1856 | ||
1857 | if (user_recv) { | |
b03efcfb | 1858 | if (!skb_queue_empty(&tp->ucopy.prequeue)) { |
1da177e4 LT |
1859 | int chunk; |
1860 | ||
1861 | tp->ucopy.len = copied > 0 ? len : 0; | |
1862 | ||
1863 | tcp_prequeue_process(sk); | |
1864 | ||
1865 | if (copied > 0 && (chunk = len - tp->ucopy.len) != 0) { | |
ed88098e | 1866 | NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk); |
1da177e4 LT |
1867 | len -= chunk; |
1868 | copied += chunk; | |
1869 | } | |
1870 | } | |
1871 | ||
1872 | tp->ucopy.task = NULL; | |
1873 | tp->ucopy.len = 0; | |
1874 | } | |
1875 | ||
1876 | /* According to UNIX98, msg_name/msg_namelen are ignored | |
1877 | * on connected socket. I was just happy when found this 8) --ANK | |
1878 | */ | |
1879 | ||
1880 | /* Clean up data we have read: This will do ACK frames. */ | |
0e4b4992 | 1881 | tcp_cleanup_rbuf(sk, copied); |
1da177e4 | 1882 | |
1da177e4 LT |
1883 | release_sock(sk); |
1884 | return copied; | |
1885 | ||
1886 | out: | |
1da177e4 LT |
1887 | release_sock(sk); |
1888 | return err; | |
1889 | ||
1890 | recv_urg: | |
377f0a08 | 1891 | err = tcp_recv_urg(sk, msg, len, flags); |
1da177e4 | 1892 | goto out; |
c0e88ff0 PE |
1893 | |
1894 | recv_sndq: | |
1895 | err = tcp_peek_sndq(sk, msg, len); | |
1896 | goto out; | |
1da177e4 | 1897 | } |
4bc2f18b | 1898 | EXPORT_SYMBOL(tcp_recvmsg); |
1da177e4 | 1899 | |
490d5046 IJ |
1900 | void tcp_set_state(struct sock *sk, int state) |
1901 | { | |
1902 | int oldstate = sk->sk_state; | |
1903 | ||
1904 | switch (state) { | |
1905 | case TCP_ESTABLISHED: | |
1906 | if (oldstate != TCP_ESTABLISHED) | |
81cc8a75 | 1907 | TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB); |
490d5046 IJ |
1908 | break; |
1909 | ||
1910 | case TCP_CLOSE: | |
1911 | if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED) | |
81cc8a75 | 1912 | TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS); |
490d5046 IJ |
1913 | |
1914 | sk->sk_prot->unhash(sk); | |
1915 | if (inet_csk(sk)->icsk_bind_hash && | |
1916 | !(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) | |
ab1e0a13 | 1917 | inet_put_port(sk); |
490d5046 IJ |
1918 | /* fall through */ |
1919 | default: | |
5a5f3a8d | 1920 | if (oldstate == TCP_ESTABLISHED) |
74688e48 | 1921 | TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB); |
490d5046 IJ |
1922 | } |
1923 | ||
1924 | /* Change state AFTER socket is unhashed to avoid closed | |
1925 | * socket sitting in hash tables. | |
1926 | */ | |
00fd38d9 | 1927 | sk_state_store(sk, state); |
490d5046 IJ |
1928 | |
1929 | #ifdef STATE_TRACE | |
5a5f3a8d | 1930 | SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n", sk, statename[oldstate], statename[state]); |
490d5046 IJ |
1931 | #endif |
1932 | } | |
1933 | EXPORT_SYMBOL_GPL(tcp_set_state); | |
1934 | ||
1da177e4 LT |
1935 | /* |
1936 | * State processing on a close. This implements the state shift for | |
1937 | * sending our FIN frame. Note that we only send a FIN for some | |
1938 | * states. A shutdown() may have already sent the FIN, or we may be | |
1939 | * closed. | |
1940 | */ | |
1941 | ||
9b5b5cff | 1942 | static const unsigned char new_state[16] = { |
1da177e4 | 1943 | /* current state: new state: action: */ |
0980c1e3 ED |
1944 | [0 /* (Invalid) */] = TCP_CLOSE, |
1945 | [TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN, | |
1946 | [TCP_SYN_SENT] = TCP_CLOSE, | |
1947 | [TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN, | |
1948 | [TCP_FIN_WAIT1] = TCP_FIN_WAIT1, | |
1949 | [TCP_FIN_WAIT2] = TCP_FIN_WAIT2, | |
1950 | [TCP_TIME_WAIT] = TCP_CLOSE, | |
1951 | [TCP_CLOSE] = TCP_CLOSE, | |
1952 | [TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN, | |
1953 | [TCP_LAST_ACK] = TCP_LAST_ACK, | |
1954 | [TCP_LISTEN] = TCP_CLOSE, | |
1955 | [TCP_CLOSING] = TCP_CLOSING, | |
1956 | [TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */ | |
1da177e4 LT |
1957 | }; |
1958 | ||
1959 | static int tcp_close_state(struct sock *sk) | |
1960 | { | |
1961 | int next = (int)new_state[sk->sk_state]; | |
1962 | int ns = next & TCP_STATE_MASK; | |
1963 | ||
1964 | tcp_set_state(sk, ns); | |
1965 | ||
1966 | return next & TCP_ACTION_FIN; | |
1967 | } | |
1968 | ||
1969 | /* | |
1970 | * Shutdown the sending side of a connection. Much like close except | |
1f29b058 | 1971 | * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD). |
1da177e4 LT |
1972 | */ |
1973 | ||
1974 | void tcp_shutdown(struct sock *sk, int how) | |
1975 | { | |
1976 | /* We need to grab some memory, and put together a FIN, | |
1977 | * and then put it into the queue to be sent. | |
1978 | * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92. | |
1979 | */ | |
1980 | if (!(how & SEND_SHUTDOWN)) | |
1981 | return; | |
1982 | ||
1983 | /* If we've already sent a FIN, or it's a closed state, skip this. */ | |
1984 | if ((1 << sk->sk_state) & | |
1985 | (TCPF_ESTABLISHED | TCPF_SYN_SENT | | |
1986 | TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) { | |
1987 | /* Clear out any half completed packets. FIN if needed. */ | |
1988 | if (tcp_close_state(sk)) | |
1989 | tcp_send_fin(sk); | |
1990 | } | |
1991 | } | |
4bc2f18b | 1992 | EXPORT_SYMBOL(tcp_shutdown); |
1da177e4 | 1993 | |
efcdbf24 AS |
1994 | bool tcp_check_oom(struct sock *sk, int shift) |
1995 | { | |
1996 | bool too_many_orphans, out_of_socket_memory; | |
1997 | ||
1998 | too_many_orphans = tcp_too_many_orphans(sk, shift); | |
1999 | out_of_socket_memory = tcp_out_of_memory(sk); | |
2000 | ||
e87cc472 JP |
2001 | if (too_many_orphans) |
2002 | net_info_ratelimited("too many orphaned sockets\n"); | |
2003 | if (out_of_socket_memory) | |
2004 | net_info_ratelimited("out of memory -- consider tuning tcp_mem\n"); | |
efcdbf24 AS |
2005 | return too_many_orphans || out_of_socket_memory; |
2006 | } | |
2007 | ||
1da177e4 LT |
2008 | void tcp_close(struct sock *sk, long timeout) |
2009 | { | |
2010 | struct sk_buff *skb; | |
2011 | int data_was_unread = 0; | |
75c2d907 | 2012 | int state; |
1da177e4 LT |
2013 | |
2014 | lock_sock(sk); | |
2015 | sk->sk_shutdown = SHUTDOWN_MASK; | |
2016 | ||
2017 | if (sk->sk_state == TCP_LISTEN) { | |
2018 | tcp_set_state(sk, TCP_CLOSE); | |
2019 | ||
2020 | /* Special case. */ | |
0a5578cf | 2021 | inet_csk_listen_stop(sk); |
1da177e4 LT |
2022 | |
2023 | goto adjudge_to_death; | |
2024 | } | |
2025 | ||
2026 | /* We need to flush the recv. buffs. We do this only on the | |
2027 | * descriptor close, not protocol-sourced closes, because the | |
2028 | * reader process may not have drained the data yet! | |
2029 | */ | |
2030 | while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) { | |
e11ecddf ED |
2031 | u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq; |
2032 | ||
2033 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) | |
2034 | len--; | |
1da177e4 LT |
2035 | data_was_unread += len; |
2036 | __kfree_skb(skb); | |
2037 | } | |
2038 | ||
3ab224be | 2039 | sk_mem_reclaim(sk); |
1da177e4 | 2040 | |
565b7b2d KK |
2041 | /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */ |
2042 | if (sk->sk_state == TCP_CLOSE) | |
2043 | goto adjudge_to_death; | |
2044 | ||
65bb723c GR |
2045 | /* As outlined in RFC 2525, section 2.17, we send a RST here because |
2046 | * data was lost. To witness the awful effects of the old behavior of | |
2047 | * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk | |
2048 | * GET in an FTP client, suspend the process, wait for the client to | |
2049 | * advertise a zero window, then kill -9 the FTP client, wheee... | |
2050 | * Note: timeout is always zero in such a case. | |
1da177e4 | 2051 | */ |
ee995283 PE |
2052 | if (unlikely(tcp_sk(sk)->repair)) { |
2053 | sk->sk_prot->disconnect(sk, 0); | |
2054 | } else if (data_was_unread) { | |
1da177e4 | 2055 | /* Unread data was tossed, zap the connection. */ |
6f67c817 | 2056 | NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE); |
1da177e4 | 2057 | tcp_set_state(sk, TCP_CLOSE); |
aa133076 | 2058 | tcp_send_active_reset(sk, sk->sk_allocation); |
1da177e4 LT |
2059 | } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) { |
2060 | /* Check zero linger _after_ checking for unread data. */ | |
2061 | sk->sk_prot->disconnect(sk, 0); | |
6f67c817 | 2062 | NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONDATA); |
1da177e4 LT |
2063 | } else if (tcp_close_state(sk)) { |
2064 | /* We FIN if the application ate all the data before | |
2065 | * zapping the connection. | |
2066 | */ | |
2067 | ||
2068 | /* RED-PEN. Formally speaking, we have broken TCP state | |
2069 | * machine. State transitions: | |
2070 | * | |
2071 | * TCP_ESTABLISHED -> TCP_FIN_WAIT1 | |
2072 | * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible) | |
2073 | * TCP_CLOSE_WAIT -> TCP_LAST_ACK | |
2074 | * | |
2075 | * are legal only when FIN has been sent (i.e. in window), | |
2076 | * rather than queued out of window. Purists blame. | |
2077 | * | |
2078 | * F.e. "RFC state" is ESTABLISHED, | |
2079 | * if Linux state is FIN-WAIT-1, but FIN is still not sent. | |
2080 | * | |
2081 | * The visible declinations are that sometimes | |
2082 | * we enter time-wait state, when it is not required really | |
2083 | * (harmless), do not send active resets, when they are | |
2084 | * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when | |
2085 | * they look as CLOSING or LAST_ACK for Linux) | |
2086 | * Probably, I missed some more holelets. | |
2087 | * --ANK | |
8336886f JC |
2088 | * XXX (TFO) - To start off we don't support SYN+ACK+FIN |
2089 | * in a single packet! (May consider it later but will | |
2090 | * probably need API support or TCP_CORK SYN-ACK until | |
2091 | * data is written and socket is closed.) | |
1da177e4 LT |
2092 | */ |
2093 | tcp_send_fin(sk); | |
2094 | } | |
2095 | ||
2096 | sk_stream_wait_close(sk, timeout); | |
2097 | ||
2098 | adjudge_to_death: | |
75c2d907 HX |
2099 | state = sk->sk_state; |
2100 | sock_hold(sk); | |
2101 | sock_orphan(sk); | |
75c2d907 | 2102 | |
1da177e4 LT |
2103 | /* It is the last release_sock in its life. It will remove backlog. */ |
2104 | release_sock(sk); | |
2105 | ||
2106 | ||
2107 | /* Now socket is owned by kernel and we acquire BH lock | |
2108 | to finish close. No need to check for user refs. | |
2109 | */ | |
2110 | local_bh_disable(); | |
2111 | bh_lock_sock(sk); | |
547b792c | 2112 | WARN_ON(sock_owned_by_user(sk)); |
1da177e4 | 2113 | |
eb4dea58 HX |
2114 | percpu_counter_inc(sk->sk_prot->orphan_count); |
2115 | ||
75c2d907 HX |
2116 | /* Have we already been destroyed by a softirq or backlog? */ |
2117 | if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE) | |
2118 | goto out; | |
1da177e4 LT |
2119 | |
2120 | /* This is a (useful) BSD violating of the RFC. There is a | |
2121 | * problem with TCP as specified in that the other end could | |
2122 | * keep a socket open forever with no application left this end. | |
b10bd54c | 2123 | * We use a 1 minute timeout (about the same as BSD) then kill |
1da177e4 LT |
2124 | * our end. If they send after that then tough - BUT: long enough |
2125 | * that we won't make the old 4*rto = almost no time - whoops | |
2126 | * reset mistake. | |
2127 | * | |
2128 | * Nope, it was not mistake. It is really desired behaviour | |
2129 | * f.e. on http servers, when such sockets are useless, but | |
2130 | * consume significant resources. Let's do it with special | |
2131 | * linger2 option. --ANK | |
2132 | */ | |
2133 | ||
2134 | if (sk->sk_state == TCP_FIN_WAIT2) { | |
2135 | struct tcp_sock *tp = tcp_sk(sk); | |
2136 | if (tp->linger2 < 0) { | |
2137 | tcp_set_state(sk, TCP_CLOSE); | |
2138 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
de0744af PE |
2139 | NET_INC_STATS_BH(sock_net(sk), |
2140 | LINUX_MIB_TCPABORTONLINGER); | |
1da177e4 | 2141 | } else { |
463c84b9 | 2142 | const int tmo = tcp_fin_time(sk); |
1da177e4 LT |
2143 | |
2144 | if (tmo > TCP_TIMEWAIT_LEN) { | |
52499afe DM |
2145 | inet_csk_reset_keepalive_timer(sk, |
2146 | tmo - TCP_TIMEWAIT_LEN); | |
1da177e4 | 2147 | } else { |
1da177e4 LT |
2148 | tcp_time_wait(sk, TCP_FIN_WAIT2, tmo); |
2149 | goto out; | |
2150 | } | |
2151 | } | |
2152 | } | |
2153 | if (sk->sk_state != TCP_CLOSE) { | |
3ab224be | 2154 | sk_mem_reclaim(sk); |
efcdbf24 | 2155 | if (tcp_check_oom(sk, 0)) { |
1da177e4 LT |
2156 | tcp_set_state(sk, TCP_CLOSE); |
2157 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
de0744af PE |
2158 | NET_INC_STATS_BH(sock_net(sk), |
2159 | LINUX_MIB_TCPABORTONMEMORY); | |
1da177e4 LT |
2160 | } |
2161 | } | |
1da177e4 | 2162 | |
8336886f JC |
2163 | if (sk->sk_state == TCP_CLOSE) { |
2164 | struct request_sock *req = tcp_sk(sk)->fastopen_rsk; | |
2165 | /* We could get here with a non-NULL req if the socket is | |
2166 | * aborted (e.g., closed with unread data) before 3WHS | |
2167 | * finishes. | |
2168 | */ | |
00db4124 | 2169 | if (req) |
8336886f | 2170 | reqsk_fastopen_remove(sk, req, false); |
0a5578cf | 2171 | inet_csk_destroy_sock(sk); |
8336886f | 2172 | } |
1da177e4 LT |
2173 | /* Otherwise, socket is reprieved until protocol close. */ |
2174 | ||
2175 | out: | |
2176 | bh_unlock_sock(sk); | |
2177 | local_bh_enable(); | |
2178 | sock_put(sk); | |
2179 | } | |
4bc2f18b | 2180 | EXPORT_SYMBOL(tcp_close); |
1da177e4 LT |
2181 | |
2182 | /* These states need RST on ABORT according to RFC793 */ | |
2183 | ||
a2a385d6 | 2184 | static inline bool tcp_need_reset(int state) |
1da177e4 LT |
2185 | { |
2186 | return (1 << state) & | |
2187 | (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 | | |
2188 | TCPF_FIN_WAIT2 | TCPF_SYN_RECV); | |
2189 | } | |
2190 | ||
2191 | int tcp_disconnect(struct sock *sk, int flags) | |
2192 | { | |
2193 | struct inet_sock *inet = inet_sk(sk); | |
463c84b9 | 2194 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
2195 | struct tcp_sock *tp = tcp_sk(sk); |
2196 | int err = 0; | |
2197 | int old_state = sk->sk_state; | |
2198 | ||
2199 | if (old_state != TCP_CLOSE) | |
2200 | tcp_set_state(sk, TCP_CLOSE); | |
2201 | ||
2202 | /* ABORT function of RFC793 */ | |
2203 | if (old_state == TCP_LISTEN) { | |
0a5578cf | 2204 | inet_csk_listen_stop(sk); |
ee995283 PE |
2205 | } else if (unlikely(tp->repair)) { |
2206 | sk->sk_err = ECONNABORTED; | |
1da177e4 LT |
2207 | } else if (tcp_need_reset(old_state) || |
2208 | (tp->snd_nxt != tp->write_seq && | |
2209 | (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) { | |
caa20d9a | 2210 | /* The last check adjusts for discrepancy of Linux wrt. RFC |
1da177e4 LT |
2211 | * states |
2212 | */ | |
2213 | tcp_send_active_reset(sk, gfp_any()); | |
2214 | sk->sk_err = ECONNRESET; | |
2215 | } else if (old_state == TCP_SYN_SENT) | |
2216 | sk->sk_err = ECONNRESET; | |
2217 | ||
2218 | tcp_clear_xmit_timers(sk); | |
2219 | __skb_queue_purge(&sk->sk_receive_queue); | |
fe067e8a | 2220 | tcp_write_queue_purge(sk); |
1da177e4 LT |
2221 | __skb_queue_purge(&tp->out_of_order_queue); |
2222 | ||
c720c7e8 | 2223 | inet->inet_dport = 0; |
1da177e4 LT |
2224 | |
2225 | if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) | |
2226 | inet_reset_saddr(sk); | |
2227 | ||
2228 | sk->sk_shutdown = 0; | |
2229 | sock_reset_flag(sk, SOCK_DONE); | |
740b0f18 | 2230 | tp->srtt_us = 0; |
686a5624 YM |
2231 | tp->write_seq += tp->max_window + 2; |
2232 | if (tp->write_seq == 0) | |
1da177e4 | 2233 | tp->write_seq = 1; |
463c84b9 | 2234 | icsk->icsk_backoff = 0; |
1da177e4 | 2235 | tp->snd_cwnd = 2; |
6687e988 | 2236 | icsk->icsk_probes_out = 0; |
1da177e4 | 2237 | tp->packets_out = 0; |
0b6a05c1 | 2238 | tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; |
1da177e4 | 2239 | tp->snd_cwnd_cnt = 0; |
1fdf475a | 2240 | tp->window_clamp = 0; |
6687e988 | 2241 | tcp_set_ca_state(sk, TCP_CA_Open); |
1da177e4 | 2242 | tcp_clear_retrans(tp); |
463c84b9 | 2243 | inet_csk_delack_init(sk); |
fe067e8a | 2244 | tcp_init_send_head(sk); |
b40b4f79 | 2245 | memset(&tp->rx_opt, 0, sizeof(tp->rx_opt)); |
1da177e4 LT |
2246 | __sk_dst_reset(sk); |
2247 | ||
c720c7e8 | 2248 | WARN_ON(inet->inet_num && !icsk->icsk_bind_hash); |
1da177e4 LT |
2249 | |
2250 | sk->sk_error_report(sk); | |
2251 | return err; | |
2252 | } | |
4bc2f18b | 2253 | EXPORT_SYMBOL(tcp_disconnect); |
1da177e4 | 2254 | |
a2a385d6 | 2255 | static inline bool tcp_can_repair_sock(const struct sock *sk) |
ee995283 | 2256 | { |
52e804c6 | 2257 | return ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN) && |
ee995283 PE |
2258 | ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_ESTABLISHED)); |
2259 | } | |
2260 | ||
de248a75 PE |
2261 | static int tcp_repair_options_est(struct tcp_sock *tp, |
2262 | struct tcp_repair_opt __user *optbuf, unsigned int len) | |
b139ba4e | 2263 | { |
de248a75 | 2264 | struct tcp_repair_opt opt; |
b139ba4e | 2265 | |
de248a75 PE |
2266 | while (len >= sizeof(opt)) { |
2267 | if (copy_from_user(&opt, optbuf, sizeof(opt))) | |
b139ba4e PE |
2268 | return -EFAULT; |
2269 | ||
2270 | optbuf++; | |
de248a75 | 2271 | len -= sizeof(opt); |
b139ba4e | 2272 | |
de248a75 PE |
2273 | switch (opt.opt_code) { |
2274 | case TCPOPT_MSS: | |
2275 | tp->rx_opt.mss_clamp = opt.opt_val; | |
b139ba4e | 2276 | break; |
de248a75 | 2277 | case TCPOPT_WINDOW: |
bc26ccd8 AV |
2278 | { |
2279 | u16 snd_wscale = opt.opt_val & 0xFFFF; | |
2280 | u16 rcv_wscale = opt.opt_val >> 16; | |
2281 | ||
2282 | if (snd_wscale > 14 || rcv_wscale > 14) | |
2283 | return -EFBIG; | |
b139ba4e | 2284 | |
bc26ccd8 AV |
2285 | tp->rx_opt.snd_wscale = snd_wscale; |
2286 | tp->rx_opt.rcv_wscale = rcv_wscale; | |
2287 | tp->rx_opt.wscale_ok = 1; | |
2288 | } | |
b139ba4e | 2289 | break; |
b139ba4e | 2290 | case TCPOPT_SACK_PERM: |
de248a75 PE |
2291 | if (opt.opt_val != 0) |
2292 | return -EINVAL; | |
2293 | ||
b139ba4e PE |
2294 | tp->rx_opt.sack_ok |= TCP_SACK_SEEN; |
2295 | if (sysctl_tcp_fack) | |
2296 | tcp_enable_fack(tp); | |
2297 | break; | |
2298 | case TCPOPT_TIMESTAMP: | |
de248a75 PE |
2299 | if (opt.opt_val != 0) |
2300 | return -EINVAL; | |
2301 | ||
b139ba4e PE |
2302 | tp->rx_opt.tstamp_ok = 1; |
2303 | break; | |
2304 | } | |
2305 | } | |
2306 | ||
2307 | return 0; | |
2308 | } | |
2309 | ||
1da177e4 LT |
2310 | /* |
2311 | * Socket option code for TCP. | |
2312 | */ | |
3fdadf7d | 2313 | static int do_tcp_setsockopt(struct sock *sk, int level, |
b7058842 | 2314 | int optname, char __user *optval, unsigned int optlen) |
1da177e4 LT |
2315 | { |
2316 | struct tcp_sock *tp = tcp_sk(sk); | |
463c84b9 | 2317 | struct inet_connection_sock *icsk = inet_csk(sk); |
1e579caa | 2318 | struct net *net = sock_net(sk); |
1da177e4 LT |
2319 | int val; |
2320 | int err = 0; | |
2321 | ||
e56fb50f WAS |
2322 | /* These are data/string values, all the others are ints */ |
2323 | switch (optname) { | |
2324 | case TCP_CONGESTION: { | |
5f8ef48d SH |
2325 | char name[TCP_CA_NAME_MAX]; |
2326 | ||
2327 | if (optlen < 1) | |
2328 | return -EINVAL; | |
2329 | ||
2330 | val = strncpy_from_user(name, optval, | |
4fdb78d3 | 2331 | min_t(long, TCP_CA_NAME_MAX-1, optlen)); |
5f8ef48d SH |
2332 | if (val < 0) |
2333 | return -EFAULT; | |
2334 | name[val] = 0; | |
2335 | ||
2336 | lock_sock(sk); | |
6687e988 | 2337 | err = tcp_set_congestion_control(sk, name); |
5f8ef48d SH |
2338 | release_sock(sk); |
2339 | return err; | |
2340 | } | |
e56fb50f WAS |
2341 | default: |
2342 | /* fallthru */ | |
2343 | break; | |
ccbd6a5a | 2344 | } |
5f8ef48d | 2345 | |
1da177e4 LT |
2346 | if (optlen < sizeof(int)) |
2347 | return -EINVAL; | |
2348 | ||
2349 | if (get_user(val, (int __user *)optval)) | |
2350 | return -EFAULT; | |
2351 | ||
2352 | lock_sock(sk); | |
2353 | ||
2354 | switch (optname) { | |
2355 | case TCP_MAXSEG: | |
2356 | /* Values greater than interface MTU won't take effect. However | |
2357 | * at the point when this call is done we typically don't yet | |
2358 | * know which interface is going to be used */ | |
c39508d6 | 2359 | if (val < TCP_MIN_MSS || val > MAX_TCP_WINDOW) { |
1da177e4 LT |
2360 | err = -EINVAL; |
2361 | break; | |
2362 | } | |
2363 | tp->rx_opt.user_mss = val; | |
2364 | break; | |
2365 | ||
2366 | case TCP_NODELAY: | |
2367 | if (val) { | |
2368 | /* TCP_NODELAY is weaker than TCP_CORK, so that | |
2369 | * this option on corked socket is remembered, but | |
2370 | * it is not activated until cork is cleared. | |
2371 | * | |
2372 | * However, when TCP_NODELAY is set we make | |
2373 | * an explicit push, which overrides even TCP_CORK | |
2374 | * for currently queued segments. | |
2375 | */ | |
2376 | tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH; | |
9e412ba7 | 2377 | tcp_push_pending_frames(sk); |
1da177e4 LT |
2378 | } else { |
2379 | tp->nonagle &= ~TCP_NAGLE_OFF; | |
2380 | } | |
2381 | break; | |
2382 | ||
36e31b0a AP |
2383 | case TCP_THIN_LINEAR_TIMEOUTS: |
2384 | if (val < 0 || val > 1) | |
2385 | err = -EINVAL; | |
2386 | else | |
2387 | tp->thin_lto = val; | |
2388 | break; | |
2389 | ||
7e380175 AP |
2390 | case TCP_THIN_DUPACK: |
2391 | if (val < 0 || val > 1) | |
2392 | err = -EINVAL; | |
e2e5c4c0 | 2393 | else { |
7e380175 | 2394 | tp->thin_dupack = val; |
eed530b6 YC |
2395 | if (tp->thin_dupack) |
2396 | tcp_disable_early_retrans(tp); | |
e2e5c4c0 | 2397 | } |
7e380175 AP |
2398 | break; |
2399 | ||
ee995283 PE |
2400 | case TCP_REPAIR: |
2401 | if (!tcp_can_repair_sock(sk)) | |
2402 | err = -EPERM; | |
2403 | else if (val == 1) { | |
2404 | tp->repair = 1; | |
2405 | sk->sk_reuse = SK_FORCE_REUSE; | |
2406 | tp->repair_queue = TCP_NO_QUEUE; | |
2407 | } else if (val == 0) { | |
2408 | tp->repair = 0; | |
2409 | sk->sk_reuse = SK_NO_REUSE; | |
2410 | tcp_send_window_probe(sk); | |
2411 | } else | |
2412 | err = -EINVAL; | |
2413 | ||
2414 | break; | |
2415 | ||
2416 | case TCP_REPAIR_QUEUE: | |
2417 | if (!tp->repair) | |
2418 | err = -EPERM; | |
2419 | else if (val < TCP_QUEUES_NR) | |
2420 | tp->repair_queue = val; | |
2421 | else | |
2422 | err = -EINVAL; | |
2423 | break; | |
2424 | ||
2425 | case TCP_QUEUE_SEQ: | |
2426 | if (sk->sk_state != TCP_CLOSE) | |
2427 | err = -EPERM; | |
2428 | else if (tp->repair_queue == TCP_SEND_QUEUE) | |
2429 | tp->write_seq = val; | |
2430 | else if (tp->repair_queue == TCP_RECV_QUEUE) | |
2431 | tp->rcv_nxt = val; | |
2432 | else | |
2433 | err = -EINVAL; | |
2434 | break; | |
2435 | ||
b139ba4e PE |
2436 | case TCP_REPAIR_OPTIONS: |
2437 | if (!tp->repair) | |
2438 | err = -EINVAL; | |
2439 | else if (sk->sk_state == TCP_ESTABLISHED) | |
de248a75 PE |
2440 | err = tcp_repair_options_est(tp, |
2441 | (struct tcp_repair_opt __user *)optval, | |
2442 | optlen); | |
b139ba4e PE |
2443 | else |
2444 | err = -EPERM; | |
2445 | break; | |
2446 | ||
1da177e4 LT |
2447 | case TCP_CORK: |
2448 | /* When set indicates to always queue non-full frames. | |
2449 | * Later the user clears this option and we transmit | |
2450 | * any pending partial frames in the queue. This is | |
2451 | * meant to be used alongside sendfile() to get properly | |
2452 | * filled frames when the user (for example) must write | |
2453 | * out headers with a write() call first and then use | |
2454 | * sendfile to send out the data parts. | |
2455 | * | |
2456 | * TCP_CORK can be set together with TCP_NODELAY and it is | |
2457 | * stronger than TCP_NODELAY. | |
2458 | */ | |
2459 | if (val) { | |
2460 | tp->nonagle |= TCP_NAGLE_CORK; | |
2461 | } else { | |
2462 | tp->nonagle &= ~TCP_NAGLE_CORK; | |
2463 | if (tp->nonagle&TCP_NAGLE_OFF) | |
2464 | tp->nonagle |= TCP_NAGLE_PUSH; | |
9e412ba7 | 2465 | tcp_push_pending_frames(sk); |
1da177e4 LT |
2466 | } |
2467 | break; | |
2468 | ||
2469 | case TCP_KEEPIDLE: | |
2470 | if (val < 1 || val > MAX_TCP_KEEPIDLE) | |
2471 | err = -EINVAL; | |
2472 | else { | |
2473 | tp->keepalive_time = val * HZ; | |
2474 | if (sock_flag(sk, SOCK_KEEPOPEN) && | |
2475 | !((1 << sk->sk_state) & | |
2476 | (TCPF_CLOSE | TCPF_LISTEN))) { | |
6c37e5de | 2477 | u32 elapsed = keepalive_time_elapsed(tp); |
1da177e4 LT |
2478 | if (tp->keepalive_time > elapsed) |
2479 | elapsed = tp->keepalive_time - elapsed; | |
2480 | else | |
2481 | elapsed = 0; | |
463c84b9 | 2482 | inet_csk_reset_keepalive_timer(sk, elapsed); |
1da177e4 LT |
2483 | } |
2484 | } | |
2485 | break; | |
2486 | case TCP_KEEPINTVL: | |
2487 | if (val < 1 || val > MAX_TCP_KEEPINTVL) | |
2488 | err = -EINVAL; | |
2489 | else | |
2490 | tp->keepalive_intvl = val * HZ; | |
2491 | break; | |
2492 | case TCP_KEEPCNT: | |
2493 | if (val < 1 || val > MAX_TCP_KEEPCNT) | |
2494 | err = -EINVAL; | |
2495 | else | |
2496 | tp->keepalive_probes = val; | |
2497 | break; | |
2498 | case TCP_SYNCNT: | |
2499 | if (val < 1 || val > MAX_TCP_SYNCNT) | |
2500 | err = -EINVAL; | |
2501 | else | |
463c84b9 | 2502 | icsk->icsk_syn_retries = val; |
1da177e4 LT |
2503 | break; |
2504 | ||
cd8ae852 ED |
2505 | case TCP_SAVE_SYN: |
2506 | if (val < 0 || val > 1) | |
2507 | err = -EINVAL; | |
2508 | else | |
2509 | tp->save_syn = val; | |
2510 | break; | |
2511 | ||
1da177e4 LT |
2512 | case TCP_LINGER2: |
2513 | if (val < 0) | |
2514 | tp->linger2 = -1; | |
1e579caa | 2515 | else if (val > net->ipv4.sysctl_tcp_fin_timeout / HZ) |
1da177e4 LT |
2516 | tp->linger2 = 0; |
2517 | else | |
2518 | tp->linger2 = val * HZ; | |
2519 | break; | |
2520 | ||
2521 | case TCP_DEFER_ACCEPT: | |
b103cf34 JA |
2522 | /* Translate value in seconds to number of retransmits */ |
2523 | icsk->icsk_accept_queue.rskq_defer_accept = | |
2524 | secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ, | |
2525 | TCP_RTO_MAX / HZ); | |
1da177e4 LT |
2526 | break; |
2527 | ||
2528 | case TCP_WINDOW_CLAMP: | |
2529 | if (!val) { | |
2530 | if (sk->sk_state != TCP_CLOSE) { | |
2531 | err = -EINVAL; | |
2532 | break; | |
2533 | } | |
2534 | tp->window_clamp = 0; | |
2535 | } else | |
2536 | tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ? | |
2537 | SOCK_MIN_RCVBUF / 2 : val; | |
2538 | break; | |
2539 | ||
2540 | case TCP_QUICKACK: | |
2541 | if (!val) { | |
463c84b9 | 2542 | icsk->icsk_ack.pingpong = 1; |
1da177e4 | 2543 | } else { |
463c84b9 | 2544 | icsk->icsk_ack.pingpong = 0; |
1da177e4 LT |
2545 | if ((1 << sk->sk_state) & |
2546 | (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) && | |
463c84b9 ACM |
2547 | inet_csk_ack_scheduled(sk)) { |
2548 | icsk->icsk_ack.pending |= ICSK_ACK_PUSHED; | |
0e4b4992 | 2549 | tcp_cleanup_rbuf(sk, 1); |
1da177e4 | 2550 | if (!(val & 1)) |
463c84b9 | 2551 | icsk->icsk_ack.pingpong = 1; |
1da177e4 LT |
2552 | } |
2553 | } | |
2554 | break; | |
2555 | ||
cfb6eeb4 YH |
2556 | #ifdef CONFIG_TCP_MD5SIG |
2557 | case TCP_MD5SIG: | |
2558 | /* Read the IP->Key mappings from userspace */ | |
2559 | err = tp->af_specific->md5_parse(sk, optval, optlen); | |
2560 | break; | |
2561 | #endif | |
dca43c75 | 2562 | case TCP_USER_TIMEOUT: |
b248230c | 2563 | /* Cap the max time in ms TCP will retry or probe the window |
dca43c75 JC |
2564 | * before giving up and aborting (ETIMEDOUT) a connection. |
2565 | */ | |
42493570 HL |
2566 | if (val < 0) |
2567 | err = -EINVAL; | |
2568 | else | |
2569 | icsk->icsk_user_timeout = msecs_to_jiffies(val); | |
dca43c75 | 2570 | break; |
8336886f JC |
2571 | |
2572 | case TCP_FASTOPEN: | |
2573 | if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE | | |
dfea2aa6 CP |
2574 | TCPF_LISTEN))) { |
2575 | tcp_fastopen_init_key_once(true); | |
2576 | ||
0536fcc0 | 2577 | fastopen_queue_tune(sk, val); |
dfea2aa6 | 2578 | } else { |
8336886f | 2579 | err = -EINVAL; |
dfea2aa6 | 2580 | } |
8336886f | 2581 | break; |
93be6ce0 AV |
2582 | case TCP_TIMESTAMP: |
2583 | if (!tp->repair) | |
2584 | err = -EPERM; | |
2585 | else | |
2586 | tp->tsoffset = val - tcp_time_stamp; | |
2587 | break; | |
c9bee3b7 ED |
2588 | case TCP_NOTSENT_LOWAT: |
2589 | tp->notsent_lowat = val; | |
2590 | sk->sk_write_space(sk); | |
2591 | break; | |
1da177e4 LT |
2592 | default: |
2593 | err = -ENOPROTOOPT; | |
2594 | break; | |
3ff50b79 SH |
2595 | } |
2596 | ||
1da177e4 LT |
2597 | release_sock(sk); |
2598 | return err; | |
2599 | } | |
2600 | ||
3fdadf7d | 2601 | int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval, |
b7058842 | 2602 | unsigned int optlen) |
3fdadf7d | 2603 | { |
cf533ea5 | 2604 | const struct inet_connection_sock *icsk = inet_csk(sk); |
3fdadf7d DM |
2605 | |
2606 | if (level != SOL_TCP) | |
2607 | return icsk->icsk_af_ops->setsockopt(sk, level, optname, | |
2608 | optval, optlen); | |
2609 | return do_tcp_setsockopt(sk, level, optname, optval, optlen); | |
2610 | } | |
4bc2f18b | 2611 | EXPORT_SYMBOL(tcp_setsockopt); |
3fdadf7d DM |
2612 | |
2613 | #ifdef CONFIG_COMPAT | |
543d9cfe | 2614 | int compat_tcp_setsockopt(struct sock *sk, int level, int optname, |
b7058842 | 2615 | char __user *optval, unsigned int optlen) |
3fdadf7d | 2616 | { |
dec73ff0 ACM |
2617 | if (level != SOL_TCP) |
2618 | return inet_csk_compat_setsockopt(sk, level, optname, | |
2619 | optval, optlen); | |
3fdadf7d DM |
2620 | return do_tcp_setsockopt(sk, level, optname, optval, optlen); |
2621 | } | |
543d9cfe | 2622 | EXPORT_SYMBOL(compat_tcp_setsockopt); |
3fdadf7d DM |
2623 | #endif |
2624 | ||
1da177e4 | 2625 | /* Return information about state of tcp endpoint in API format. */ |
0df48c26 | 2626 | void tcp_get_info(struct sock *sk, struct tcp_info *info) |
1da177e4 | 2627 | { |
35ac838a | 2628 | const struct tcp_sock *tp = tcp_sk(sk); /* iff sk_type == SOCK_STREAM */ |
463c84b9 | 2629 | const struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 2630 | u32 now = tcp_time_stamp; |
d654976c | 2631 | unsigned int start; |
cd9b2660 | 2632 | int notsent_bytes; |
ff5d7497 | 2633 | u64 rate64; |
fad9dfef | 2634 | u32 rate; |
1da177e4 LT |
2635 | |
2636 | memset(info, 0, sizeof(*info)); | |
35ac838a CG |
2637 | if (sk->sk_type != SOCK_STREAM) |
2638 | return; | |
1da177e4 | 2639 | |
00fd38d9 ED |
2640 | info->tcpi_state = sk_state_load(sk); |
2641 | ||
6687e988 | 2642 | info->tcpi_ca_state = icsk->icsk_ca_state; |
463c84b9 | 2643 | info->tcpi_retransmits = icsk->icsk_retransmits; |
6687e988 | 2644 | info->tcpi_probes = icsk->icsk_probes_out; |
463c84b9 | 2645 | info->tcpi_backoff = icsk->icsk_backoff; |
1da177e4 LT |
2646 | |
2647 | if (tp->rx_opt.tstamp_ok) | |
2648 | info->tcpi_options |= TCPI_OPT_TIMESTAMPS; | |
e60402d0 | 2649 | if (tcp_is_sack(tp)) |
1da177e4 LT |
2650 | info->tcpi_options |= TCPI_OPT_SACK; |
2651 | if (tp->rx_opt.wscale_ok) { | |
2652 | info->tcpi_options |= TCPI_OPT_WSCALE; | |
2653 | info->tcpi_snd_wscale = tp->rx_opt.snd_wscale; | |
2654 | info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale; | |
e905a9ed | 2655 | } |
1da177e4 | 2656 | |
b5c5693b | 2657 | if (tp->ecn_flags & TCP_ECN_OK) |
1da177e4 | 2658 | info->tcpi_options |= TCPI_OPT_ECN; |
b5c5693b ED |
2659 | if (tp->ecn_flags & TCP_ECN_SEEN) |
2660 | info->tcpi_options |= TCPI_OPT_ECN_SEEN; | |
6f73601e YC |
2661 | if (tp->syn_data_acked) |
2662 | info->tcpi_options |= TCPI_OPT_SYN_DATA; | |
1da177e4 | 2663 | |
463c84b9 ACM |
2664 | info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto); |
2665 | info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato); | |
c1b4a7e6 | 2666 | info->tcpi_snd_mss = tp->mss_cache; |
463c84b9 | 2667 | info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss; |
1da177e4 | 2668 | |
00fd38d9 | 2669 | if (info->tcpi_state == TCP_LISTEN) { |
5ee3afba RJ |
2670 | info->tcpi_unacked = sk->sk_ack_backlog; |
2671 | info->tcpi_sacked = sk->sk_max_ack_backlog; | |
2672 | } else { | |
2673 | info->tcpi_unacked = tp->packets_out; | |
2674 | info->tcpi_sacked = tp->sacked_out; | |
2675 | } | |
1da177e4 LT |
2676 | info->tcpi_lost = tp->lost_out; |
2677 | info->tcpi_retrans = tp->retrans_out; | |
2678 | info->tcpi_fackets = tp->fackets_out; | |
2679 | ||
2680 | info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime); | |
463c84b9 | 2681 | info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime); |
1da177e4 LT |
2682 | info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp); |
2683 | ||
d83d8461 | 2684 | info->tcpi_pmtu = icsk->icsk_pmtu_cookie; |
1da177e4 | 2685 | info->tcpi_rcv_ssthresh = tp->rcv_ssthresh; |
740b0f18 ED |
2686 | info->tcpi_rtt = tp->srtt_us >> 3; |
2687 | info->tcpi_rttvar = tp->mdev_us >> 2; | |
1da177e4 LT |
2688 | info->tcpi_snd_ssthresh = tp->snd_ssthresh; |
2689 | info->tcpi_snd_cwnd = tp->snd_cwnd; | |
2690 | info->tcpi_advmss = tp->advmss; | |
2691 | info->tcpi_reordering = tp->reordering; | |
2692 | ||
2693 | info->tcpi_rcv_rtt = jiffies_to_usecs(tp->rcv_rtt_est.rtt)>>3; | |
2694 | info->tcpi_rcv_space = tp->rcvq_space.space; | |
2695 | ||
2696 | info->tcpi_total_retrans = tp->total_retrans; | |
977cb0ec | 2697 | |
fad9dfef | 2698 | rate = READ_ONCE(sk->sk_pacing_rate); |
ff5d7497 ED |
2699 | rate64 = rate != ~0U ? rate : ~0ULL; |
2700 | put_unaligned(rate64, &info->tcpi_pacing_rate); | |
fad9dfef ED |
2701 | |
2702 | rate = READ_ONCE(sk->sk_max_pacing_rate); | |
ff5d7497 ED |
2703 | rate64 = rate != ~0U ? rate : ~0ULL; |
2704 | put_unaligned(rate64, &info->tcpi_max_pacing_rate); | |
0df48c26 | 2705 | |
d654976c ED |
2706 | do { |
2707 | start = u64_stats_fetch_begin_irq(&tp->syncp); | |
ff5d7497 ED |
2708 | put_unaligned(tp->bytes_acked, &info->tcpi_bytes_acked); |
2709 | put_unaligned(tp->bytes_received, &info->tcpi_bytes_received); | |
d654976c | 2710 | } while (u64_stats_fetch_retry_irq(&tp->syncp, start)); |
2efd055c MRL |
2711 | info->tcpi_segs_out = tp->segs_out; |
2712 | info->tcpi_segs_in = tp->segs_in; | |
cd9b2660 ED |
2713 | |
2714 | notsent_bytes = READ_ONCE(tp->write_seq) - READ_ONCE(tp->snd_nxt); | |
2715 | info->tcpi_notsent_bytes = max(0, notsent_bytes); | |
2716 | ||
2717 | info->tcpi_min_rtt = tcp_min_rtt(tp); | |
a44d6eac MKL |
2718 | info->tcpi_data_segs_in = tp->data_segs_in; |
2719 | info->tcpi_data_segs_out = tp->data_segs_out; | |
1da177e4 | 2720 | } |
1da177e4 LT |
2721 | EXPORT_SYMBOL_GPL(tcp_get_info); |
2722 | ||
3fdadf7d DM |
2723 | static int do_tcp_getsockopt(struct sock *sk, int level, |
2724 | int optname, char __user *optval, int __user *optlen) | |
1da177e4 | 2725 | { |
295f7324 | 2726 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 2727 | struct tcp_sock *tp = tcp_sk(sk); |
6fa25166 | 2728 | struct net *net = sock_net(sk); |
1da177e4 LT |
2729 | int val, len; |
2730 | ||
1da177e4 LT |
2731 | if (get_user(len, optlen)) |
2732 | return -EFAULT; | |
2733 | ||
2734 | len = min_t(unsigned int, len, sizeof(int)); | |
2735 | ||
2736 | if (len < 0) | |
2737 | return -EINVAL; | |
2738 | ||
2739 | switch (optname) { | |
2740 | case TCP_MAXSEG: | |
c1b4a7e6 | 2741 | val = tp->mss_cache; |
1da177e4 LT |
2742 | if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) |
2743 | val = tp->rx_opt.user_mss; | |
5e6a3ce6 PE |
2744 | if (tp->repair) |
2745 | val = tp->rx_opt.mss_clamp; | |
1da177e4 LT |
2746 | break; |
2747 | case TCP_NODELAY: | |
2748 | val = !!(tp->nonagle&TCP_NAGLE_OFF); | |
2749 | break; | |
2750 | case TCP_CORK: | |
2751 | val = !!(tp->nonagle&TCP_NAGLE_CORK); | |
2752 | break; | |
2753 | case TCP_KEEPIDLE: | |
df19a626 | 2754 | val = keepalive_time_when(tp) / HZ; |
1da177e4 LT |
2755 | break; |
2756 | case TCP_KEEPINTVL: | |
df19a626 | 2757 | val = keepalive_intvl_when(tp) / HZ; |
1da177e4 LT |
2758 | break; |
2759 | case TCP_KEEPCNT: | |
df19a626 | 2760 | val = keepalive_probes(tp); |
1da177e4 LT |
2761 | break; |
2762 | case TCP_SYNCNT: | |
6fa25166 | 2763 | val = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries; |
1da177e4 LT |
2764 | break; |
2765 | case TCP_LINGER2: | |
2766 | val = tp->linger2; | |
2767 | if (val >= 0) | |
1e579caa | 2768 | val = (val ? : net->ipv4.sysctl_tcp_fin_timeout) / HZ; |
1da177e4 LT |
2769 | break; |
2770 | case TCP_DEFER_ACCEPT: | |
b103cf34 JA |
2771 | val = retrans_to_secs(icsk->icsk_accept_queue.rskq_defer_accept, |
2772 | TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ); | |
1da177e4 LT |
2773 | break; |
2774 | case TCP_WINDOW_CLAMP: | |
2775 | val = tp->window_clamp; | |
2776 | break; | |
2777 | case TCP_INFO: { | |
2778 | struct tcp_info info; | |
2779 | ||
2780 | if (get_user(len, optlen)) | |
2781 | return -EFAULT; | |
2782 | ||
2783 | tcp_get_info(sk, &info); | |
2784 | ||
2785 | len = min_t(unsigned int, len, sizeof(info)); | |
2786 | if (put_user(len, optlen)) | |
2787 | return -EFAULT; | |
2788 | if (copy_to_user(optval, &info, len)) | |
2789 | return -EFAULT; | |
2790 | return 0; | |
2791 | } | |
6e9250f5 ED |
2792 | case TCP_CC_INFO: { |
2793 | const struct tcp_congestion_ops *ca_ops; | |
2794 | union tcp_cc_info info; | |
2795 | size_t sz = 0; | |
2796 | int attr; | |
2797 | ||
2798 | if (get_user(len, optlen)) | |
2799 | return -EFAULT; | |
2800 | ||
2801 | ca_ops = icsk->icsk_ca_ops; | |
2802 | if (ca_ops && ca_ops->get_info) | |
2803 | sz = ca_ops->get_info(sk, ~0U, &attr, &info); | |
2804 | ||
2805 | len = min_t(unsigned int, len, sz); | |
2806 | if (put_user(len, optlen)) | |
2807 | return -EFAULT; | |
2808 | if (copy_to_user(optval, &info, len)) | |
2809 | return -EFAULT; | |
2810 | return 0; | |
2811 | } | |
1da177e4 | 2812 | case TCP_QUICKACK: |
295f7324 | 2813 | val = !icsk->icsk_ack.pingpong; |
1da177e4 | 2814 | break; |
5f8ef48d SH |
2815 | |
2816 | case TCP_CONGESTION: | |
2817 | if (get_user(len, optlen)) | |
2818 | return -EFAULT; | |
2819 | len = min_t(unsigned int, len, TCP_CA_NAME_MAX); | |
2820 | if (put_user(len, optlen)) | |
2821 | return -EFAULT; | |
6687e988 | 2822 | if (copy_to_user(optval, icsk->icsk_ca_ops->name, len)) |
5f8ef48d SH |
2823 | return -EFAULT; |
2824 | return 0; | |
e56fb50f | 2825 | |
3c0fef0b JH |
2826 | case TCP_THIN_LINEAR_TIMEOUTS: |
2827 | val = tp->thin_lto; | |
2828 | break; | |
2829 | case TCP_THIN_DUPACK: | |
2830 | val = tp->thin_dupack; | |
2831 | break; | |
dca43c75 | 2832 | |
ee995283 PE |
2833 | case TCP_REPAIR: |
2834 | val = tp->repair; | |
2835 | break; | |
2836 | ||
2837 | case TCP_REPAIR_QUEUE: | |
2838 | if (tp->repair) | |
2839 | val = tp->repair_queue; | |
2840 | else | |
2841 | return -EINVAL; | |
2842 | break; | |
2843 | ||
2844 | case TCP_QUEUE_SEQ: | |
2845 | if (tp->repair_queue == TCP_SEND_QUEUE) | |
2846 | val = tp->write_seq; | |
2847 | else if (tp->repair_queue == TCP_RECV_QUEUE) | |
2848 | val = tp->rcv_nxt; | |
2849 | else | |
2850 | return -EINVAL; | |
2851 | break; | |
2852 | ||
dca43c75 JC |
2853 | case TCP_USER_TIMEOUT: |
2854 | val = jiffies_to_msecs(icsk->icsk_user_timeout); | |
2855 | break; | |
1536e285 KN |
2856 | |
2857 | case TCP_FASTOPEN: | |
0536fcc0 | 2858 | val = icsk->icsk_accept_queue.fastopenq.max_qlen; |
1536e285 KN |
2859 | break; |
2860 | ||
93be6ce0 AV |
2861 | case TCP_TIMESTAMP: |
2862 | val = tcp_time_stamp + tp->tsoffset; | |
2863 | break; | |
c9bee3b7 ED |
2864 | case TCP_NOTSENT_LOWAT: |
2865 | val = tp->notsent_lowat; | |
2866 | break; | |
cd8ae852 ED |
2867 | case TCP_SAVE_SYN: |
2868 | val = tp->save_syn; | |
2869 | break; | |
2870 | case TCP_SAVED_SYN: { | |
2871 | if (get_user(len, optlen)) | |
2872 | return -EFAULT; | |
2873 | ||
2874 | lock_sock(sk); | |
2875 | if (tp->saved_syn) { | |
aea0929e EM |
2876 | if (len < tp->saved_syn[0]) { |
2877 | if (put_user(tp->saved_syn[0], optlen)) { | |
2878 | release_sock(sk); | |
2879 | return -EFAULT; | |
2880 | } | |
2881 | release_sock(sk); | |
2882 | return -EINVAL; | |
2883 | } | |
2884 | len = tp->saved_syn[0]; | |
cd8ae852 ED |
2885 | if (put_user(len, optlen)) { |
2886 | release_sock(sk); | |
2887 | return -EFAULT; | |
2888 | } | |
2889 | if (copy_to_user(optval, tp->saved_syn + 1, len)) { | |
2890 | release_sock(sk); | |
2891 | return -EFAULT; | |
2892 | } | |
2893 | tcp_saved_syn_free(tp); | |
2894 | release_sock(sk); | |
2895 | } else { | |
2896 | release_sock(sk); | |
2897 | len = 0; | |
2898 | if (put_user(len, optlen)) | |
2899 | return -EFAULT; | |
2900 | } | |
2901 | return 0; | |
2902 | } | |
1da177e4 LT |
2903 | default: |
2904 | return -ENOPROTOOPT; | |
3ff50b79 | 2905 | } |
1da177e4 LT |
2906 | |
2907 | if (put_user(len, optlen)) | |
2908 | return -EFAULT; | |
2909 | if (copy_to_user(optval, &val, len)) | |
2910 | return -EFAULT; | |
2911 | return 0; | |
2912 | } | |
2913 | ||
3fdadf7d DM |
2914 | int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval, |
2915 | int __user *optlen) | |
2916 | { | |
2917 | struct inet_connection_sock *icsk = inet_csk(sk); | |
2918 | ||
2919 | if (level != SOL_TCP) | |
2920 | return icsk->icsk_af_ops->getsockopt(sk, level, optname, | |
2921 | optval, optlen); | |
2922 | return do_tcp_getsockopt(sk, level, optname, optval, optlen); | |
2923 | } | |
4bc2f18b | 2924 | EXPORT_SYMBOL(tcp_getsockopt); |
3fdadf7d DM |
2925 | |
2926 | #ifdef CONFIG_COMPAT | |
543d9cfe ACM |
2927 | int compat_tcp_getsockopt(struct sock *sk, int level, int optname, |
2928 | char __user *optval, int __user *optlen) | |
3fdadf7d | 2929 | { |
dec73ff0 ACM |
2930 | if (level != SOL_TCP) |
2931 | return inet_csk_compat_getsockopt(sk, level, optname, | |
2932 | optval, optlen); | |
3fdadf7d DM |
2933 | return do_tcp_getsockopt(sk, level, optname, optval, optlen); |
2934 | } | |
543d9cfe | 2935 | EXPORT_SYMBOL(compat_tcp_getsockopt); |
3fdadf7d | 2936 | #endif |
1da177e4 | 2937 | |
cfb6eeb4 | 2938 | #ifdef CONFIG_TCP_MD5SIG |
349ce993 | 2939 | static DEFINE_PER_CPU(struct tcp_md5sig_pool, tcp_md5sig_pool); |
71cea17e | 2940 | static DEFINE_MUTEX(tcp_md5sig_mutex); |
349ce993 | 2941 | static bool tcp_md5sig_pool_populated = false; |
cfb6eeb4 | 2942 | |
71cea17e | 2943 | static void __tcp_alloc_md5sig_pool(void) |
cfb6eeb4 | 2944 | { |
cf80e0e4 | 2945 | struct crypto_ahash *hash; |
cfb6eeb4 | 2946 | int cpu; |
cfb6eeb4 | 2947 | |
cf80e0e4 | 2948 | hash = crypto_alloc_ahash("md5", 0, CRYPTO_ALG_ASYNC); |
70477371 | 2949 | if (IS_ERR(hash)) |
cf80e0e4 HX |
2950 | return; |
2951 | ||
cfb6eeb4 | 2952 | for_each_possible_cpu(cpu) { |
cf80e0e4 | 2953 | struct ahash_request *req; |
cfb6eeb4 | 2954 | |
cf80e0e4 HX |
2955 | if (per_cpu(tcp_md5sig_pool, cpu).md5_req) |
2956 | continue; | |
2957 | ||
2958 | req = ahash_request_alloc(hash, GFP_KERNEL); | |
2959 | if (!req) | |
2960 | return; | |
2961 | ||
2962 | ahash_request_set_callback(req, 0, NULL, NULL); | |
2963 | ||
2964 | per_cpu(tcp_md5sig_pool, cpu).md5_req = req; | |
cfb6eeb4 | 2965 | } |
349ce993 ED |
2966 | /* before setting tcp_md5sig_pool_populated, we must commit all writes |
2967 | * to memory. See smp_rmb() in tcp_get_md5sig_pool() | |
71cea17e ED |
2968 | */ |
2969 | smp_wmb(); | |
349ce993 | 2970 | tcp_md5sig_pool_populated = true; |
cfb6eeb4 YH |
2971 | } |
2972 | ||
71cea17e | 2973 | bool tcp_alloc_md5sig_pool(void) |
cfb6eeb4 | 2974 | { |
349ce993 | 2975 | if (unlikely(!tcp_md5sig_pool_populated)) { |
71cea17e ED |
2976 | mutex_lock(&tcp_md5sig_mutex); |
2977 | ||
349ce993 | 2978 | if (!tcp_md5sig_pool_populated) |
71cea17e ED |
2979 | __tcp_alloc_md5sig_pool(); |
2980 | ||
2981 | mutex_unlock(&tcp_md5sig_mutex); | |
cfb6eeb4 | 2982 | } |
349ce993 | 2983 | return tcp_md5sig_pool_populated; |
cfb6eeb4 | 2984 | } |
cfb6eeb4 YH |
2985 | EXPORT_SYMBOL(tcp_alloc_md5sig_pool); |
2986 | ||
35790c04 ED |
2987 | |
2988 | /** | |
2989 | * tcp_get_md5sig_pool - get md5sig_pool for this user | |
2990 | * | |
2991 | * We use percpu structure, so if we succeed, we exit with preemption | |
2992 | * and BH disabled, to make sure another thread or softirq handling | |
2993 | * wont try to get same context. | |
2994 | */ | |
2995 | struct tcp_md5sig_pool *tcp_get_md5sig_pool(void) | |
cfb6eeb4 | 2996 | { |
35790c04 | 2997 | local_bh_disable(); |
cfb6eeb4 | 2998 | |
349ce993 ED |
2999 | if (tcp_md5sig_pool_populated) { |
3000 | /* coupled with smp_wmb() in __tcp_alloc_md5sig_pool() */ | |
3001 | smp_rmb(); | |
3002 | return this_cpu_ptr(&tcp_md5sig_pool); | |
3003 | } | |
35790c04 ED |
3004 | local_bh_enable(); |
3005 | return NULL; | |
3006 | } | |
3007 | EXPORT_SYMBOL(tcp_get_md5sig_pool); | |
cfb6eeb4 | 3008 | |
49a72dfb | 3009 | int tcp_md5_hash_header(struct tcp_md5sig_pool *hp, |
ca35a0ef | 3010 | const struct tcphdr *th) |
49a72dfb AL |
3011 | { |
3012 | struct scatterlist sg; | |
ca35a0ef | 3013 | struct tcphdr hdr; |
49a72dfb | 3014 | |
ca35a0ef ED |
3015 | /* We are not allowed to change tcphdr, make a local copy */ |
3016 | memcpy(&hdr, th, sizeof(hdr)); | |
3017 | hdr.check = 0; | |
3018 | ||
49a72dfb | 3019 | /* options aren't included in the hash */ |
ca35a0ef | 3020 | sg_init_one(&sg, &hdr, sizeof(hdr)); |
cf80e0e4 HX |
3021 | ahash_request_set_crypt(hp->md5_req, &sg, NULL, sizeof(hdr)); |
3022 | return crypto_ahash_update(hp->md5_req); | |
49a72dfb | 3023 | } |
49a72dfb AL |
3024 | EXPORT_SYMBOL(tcp_md5_hash_header); |
3025 | ||
3026 | int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp, | |
cf533ea5 | 3027 | const struct sk_buff *skb, unsigned int header_len) |
49a72dfb AL |
3028 | { |
3029 | struct scatterlist sg; | |
3030 | const struct tcphdr *tp = tcp_hdr(skb); | |
cf80e0e4 | 3031 | struct ahash_request *req = hp->md5_req; |
95c96174 ED |
3032 | unsigned int i; |
3033 | const unsigned int head_data_len = skb_headlen(skb) > header_len ? | |
3034 | skb_headlen(skb) - header_len : 0; | |
49a72dfb | 3035 | const struct skb_shared_info *shi = skb_shinfo(skb); |
d7fd1b57 | 3036 | struct sk_buff *frag_iter; |
49a72dfb AL |
3037 | |
3038 | sg_init_table(&sg, 1); | |
3039 | ||
3040 | sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len); | |
cf80e0e4 HX |
3041 | ahash_request_set_crypt(req, &sg, NULL, head_data_len); |
3042 | if (crypto_ahash_update(req)) | |
49a72dfb AL |
3043 | return 1; |
3044 | ||
3045 | for (i = 0; i < shi->nr_frags; ++i) { | |
3046 | const struct skb_frag_struct *f = &shi->frags[i]; | |
54d27fcb ED |
3047 | unsigned int offset = f->page_offset; |
3048 | struct page *page = skb_frag_page(f) + (offset >> PAGE_SHIFT); | |
3049 | ||
3050 | sg_set_page(&sg, page, skb_frag_size(f), | |
3051 | offset_in_page(offset)); | |
cf80e0e4 HX |
3052 | ahash_request_set_crypt(req, &sg, NULL, skb_frag_size(f)); |
3053 | if (crypto_ahash_update(req)) | |
49a72dfb AL |
3054 | return 1; |
3055 | } | |
3056 | ||
d7fd1b57 ED |
3057 | skb_walk_frags(skb, frag_iter) |
3058 | if (tcp_md5_hash_skb_data(hp, frag_iter, 0)) | |
3059 | return 1; | |
3060 | ||
49a72dfb AL |
3061 | return 0; |
3062 | } | |
49a72dfb AL |
3063 | EXPORT_SYMBOL(tcp_md5_hash_skb_data); |
3064 | ||
cf533ea5 | 3065 | int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, const struct tcp_md5sig_key *key) |
49a72dfb AL |
3066 | { |
3067 | struct scatterlist sg; | |
3068 | ||
3069 | sg_init_one(&sg, key->key, key->keylen); | |
cf80e0e4 HX |
3070 | ahash_request_set_crypt(hp->md5_req, &sg, NULL, key->keylen); |
3071 | return crypto_ahash_update(hp->md5_req); | |
49a72dfb | 3072 | } |
49a72dfb AL |
3073 | EXPORT_SYMBOL(tcp_md5_hash_key); |
3074 | ||
cfb6eeb4 YH |
3075 | #endif |
3076 | ||
4ac02bab AK |
3077 | void tcp_done(struct sock *sk) |
3078 | { | |
8336886f JC |
3079 | struct request_sock *req = tcp_sk(sk)->fastopen_rsk; |
3080 | ||
5a5f3a8d | 3081 | if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV) |
63231bdd | 3082 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS); |
4ac02bab AK |
3083 | |
3084 | tcp_set_state(sk, TCP_CLOSE); | |
3085 | tcp_clear_xmit_timers(sk); | |
00db4124 | 3086 | if (req) |
8336886f | 3087 | reqsk_fastopen_remove(sk, req, false); |
4ac02bab AK |
3088 | |
3089 | sk->sk_shutdown = SHUTDOWN_MASK; | |
3090 | ||
3091 | if (!sock_flag(sk, SOCK_DEAD)) | |
3092 | sk->sk_state_change(sk); | |
3093 | else | |
3094 | inet_csk_destroy_sock(sk); | |
3095 | } | |
3096 | EXPORT_SYMBOL_GPL(tcp_done); | |
3097 | ||
c1e64e29 LC |
3098 | int tcp_abort(struct sock *sk, int err) |
3099 | { | |
3100 | if (!sk_fullsock(sk)) { | |
07f6f4a3 ED |
3101 | if (sk->sk_state == TCP_NEW_SYN_RECV) { |
3102 | struct request_sock *req = inet_reqsk(sk); | |
3103 | ||
3104 | local_bh_disable(); | |
3105 | inet_csk_reqsk_queue_drop_and_put(req->rsk_listener, | |
3106 | req); | |
3107 | local_bh_enable(); | |
3108 | return 0; | |
3109 | } | |
c1e64e29 LC |
3110 | sock_gen_put(sk); |
3111 | return -EOPNOTSUPP; | |
3112 | } | |
3113 | ||
3114 | /* Don't race with userspace socket closes such as tcp_close. */ | |
3115 | lock_sock(sk); | |
3116 | ||
2010b93e LC |
3117 | if (sk->sk_state == TCP_LISTEN) { |
3118 | tcp_set_state(sk, TCP_CLOSE); | |
3119 | inet_csk_listen_stop(sk); | |
3120 | } | |
3121 | ||
c1e64e29 LC |
3122 | /* Don't race with BH socket closes such as inet_csk_listen_stop. */ |
3123 | local_bh_disable(); | |
3124 | bh_lock_sock(sk); | |
3125 | ||
3126 | if (!sock_flag(sk, SOCK_DEAD)) { | |
3127 | sk->sk_err = err; | |
3128 | /* This barrier is coupled with smp_rmb() in tcp_poll() */ | |
3129 | smp_wmb(); | |
3130 | sk->sk_error_report(sk); | |
3131 | if (tcp_need_reset(sk->sk_state)) | |
3132 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
3133 | tcp_done(sk); | |
3134 | } | |
3135 | ||
3136 | bh_unlock_sock(sk); | |
3137 | local_bh_enable(); | |
3138 | release_sock(sk); | |
3139 | sock_put(sk); | |
3140 | return 0; | |
3141 | } | |
3142 | EXPORT_SYMBOL_GPL(tcp_abort); | |
3143 | ||
5f8ef48d | 3144 | extern struct tcp_congestion_ops tcp_reno; |
1da177e4 LT |
3145 | |
3146 | static __initdata unsigned long thash_entries; | |
3147 | static int __init set_thash_entries(char *str) | |
3148 | { | |
413c27d8 EZ |
3149 | ssize_t ret; |
3150 | ||
1da177e4 LT |
3151 | if (!str) |
3152 | return 0; | |
413c27d8 EZ |
3153 | |
3154 | ret = kstrtoul(str, 0, &thash_entries); | |
3155 | if (ret) | |
3156 | return 0; | |
3157 | ||
1da177e4 LT |
3158 | return 1; |
3159 | } | |
3160 | __setup("thash_entries=", set_thash_entries); | |
3161 | ||
47d7a88c | 3162 | static void __init tcp_init_mem(void) |
4acb4190 | 3163 | { |
b66e91cc ED |
3164 | unsigned long limit = nr_free_buffer_pages() / 16; |
3165 | ||
4acb4190 | 3166 | limit = max(limit, 128UL); |
b66e91cc ED |
3167 | sysctl_tcp_mem[0] = limit / 4 * 3; /* 4.68 % */ |
3168 | sysctl_tcp_mem[1] = limit; /* 6.25 % */ | |
3169 | sysctl_tcp_mem[2] = sysctl_tcp_mem[0] * 2; /* 9.37 % */ | |
4acb4190 GC |
3170 | } |
3171 | ||
1da177e4 LT |
3172 | void __init tcp_init(void) |
3173 | { | |
f03d78db | 3174 | unsigned long limit; |
b49960a0 | 3175 | int max_rshare, max_wshare, cnt; |
074b8517 | 3176 | unsigned int i; |
1da177e4 | 3177 | |
b4772ef8 | 3178 | sock_skb_cb_check_size(sizeof(struct tcp_skb_cb)); |
1da177e4 | 3179 | |
908c7f19 TH |
3180 | percpu_counter_init(&tcp_sockets_allocated, 0, GFP_KERNEL); |
3181 | percpu_counter_init(&tcp_orphan_count, 0, GFP_KERNEL); | |
6e04e021 ACM |
3182 | tcp_hashinfo.bind_bucket_cachep = |
3183 | kmem_cache_create("tcp_bind_bucket", | |
3184 | sizeof(struct inet_bind_bucket), 0, | |
20c2df83 | 3185 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3186 | |
1da177e4 LT |
3187 | /* Size and allocate the main established and bind bucket |
3188 | * hash tables. | |
3189 | * | |
3190 | * The methodology is similar to that of the buffer cache. | |
3191 | */ | |
6e04e021 | 3192 | tcp_hashinfo.ehash = |
1da177e4 | 3193 | alloc_large_system_hash("TCP established", |
0f7ff927 | 3194 | sizeof(struct inet_ehash_bucket), |
1da177e4 | 3195 | thash_entries, |
fd90b29d | 3196 | 17, /* one slot per 128 KB of memory */ |
9e950efa | 3197 | 0, |
1da177e4 | 3198 | NULL, |
f373b53b | 3199 | &tcp_hashinfo.ehash_mask, |
31fe62b9 | 3200 | 0, |
0ccfe618 | 3201 | thash_entries ? 0 : 512 * 1024); |
05dbc7b5 | 3202 | for (i = 0; i <= tcp_hashinfo.ehash_mask; i++) |
3ab5aee7 | 3203 | INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i); |
05dbc7b5 | 3204 | |
230140cf ED |
3205 | if (inet_ehash_locks_alloc(&tcp_hashinfo)) |
3206 | panic("TCP: failed to alloc ehash_locks"); | |
6e04e021 | 3207 | tcp_hashinfo.bhash = |
1da177e4 | 3208 | alloc_large_system_hash("TCP bind", |
0f7ff927 | 3209 | sizeof(struct inet_bind_hashbucket), |
f373b53b | 3210 | tcp_hashinfo.ehash_mask + 1, |
fd90b29d | 3211 | 17, /* one slot per 128 KB of memory */ |
9e950efa | 3212 | 0, |
6e04e021 | 3213 | &tcp_hashinfo.bhash_size, |
1da177e4 | 3214 | NULL, |
31fe62b9 | 3215 | 0, |
1da177e4 | 3216 | 64 * 1024); |
074b8517 | 3217 | tcp_hashinfo.bhash_size = 1U << tcp_hashinfo.bhash_size; |
6e04e021 ACM |
3218 | for (i = 0; i < tcp_hashinfo.bhash_size; i++) { |
3219 | spin_lock_init(&tcp_hashinfo.bhash[i].lock); | |
3220 | INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain); | |
1da177e4 LT |
3221 | } |
3222 | ||
c5ed63d6 ED |
3223 | |
3224 | cnt = tcp_hashinfo.ehash_mask + 1; | |
3225 | ||
3226 | tcp_death_row.sysctl_max_tw_buckets = cnt / 2; | |
3227 | sysctl_tcp_max_orphans = cnt / 2; | |
3228 | sysctl_max_syn_backlog = max(128, cnt / 256); | |
1da177e4 | 3229 | |
a4fe34bf | 3230 | tcp_init_mem(); |
c43b874d | 3231 | /* Set per-socket limits to no more than 1/128 the pressure threshold */ |
5fb84b14 | 3232 | limit = nr_free_buffer_pages() << (PAGE_SHIFT - 7); |
b49960a0 ED |
3233 | max_wshare = min(4UL*1024*1024, limit); |
3234 | max_rshare = min(6UL*1024*1024, limit); | |
7b4f4b5e | 3235 | |
3ab224be | 3236 | sysctl_tcp_wmem[0] = SK_MEM_QUANTUM; |
7b4f4b5e | 3237 | sysctl_tcp_wmem[1] = 16*1024; |
b49960a0 | 3238 | sysctl_tcp_wmem[2] = max(64*1024, max_wshare); |
7b4f4b5e | 3239 | |
3ab224be | 3240 | sysctl_tcp_rmem[0] = SK_MEM_QUANTUM; |
7b4f4b5e | 3241 | sysctl_tcp_rmem[1] = 87380; |
b49960a0 | 3242 | sysctl_tcp_rmem[2] = max(87380, max_rshare); |
1da177e4 | 3243 | |
afd46503 | 3244 | pr_info("Hash tables configured (established %u bind %u)\n", |
058bd4d2 | 3245 | tcp_hashinfo.ehash_mask + 1, tcp_hashinfo.bhash_size); |
317a76f9 | 3246 | |
51c5d0c4 | 3247 | tcp_metrics_init(); |
55d8694f | 3248 | BUG_ON(tcp_register_congestion_control(&tcp_reno) != 0); |
46d3ceab | 3249 | tcp_tasklet_init(); |
1da177e4 | 3250 | } |