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> |
98aaa913 | 272 | #include <linux/errqueue.h> |
60e2a778 | 273 | #include <linux/static_key.h> |
1da177e4 LT |
274 | |
275 | #include <net/icmp.h> | |
cf60af03 | 276 | #include <net/inet_common.h> |
1da177e4 LT |
277 | #include <net/tcp.h> |
278 | #include <net/xfrm.h> | |
279 | #include <net/ip.h> | |
9c55e01c | 280 | #include <net/sock.h> |
1da177e4 | 281 | |
7c0f6ba6 | 282 | #include <linux/uaccess.h> |
1da177e4 | 283 | #include <asm/ioctls.h> |
076bb0c8 | 284 | #include <net/busy_poll.h> |
1da177e4 | 285 | |
dd24c001 | 286 | struct percpu_counter tcp_orphan_count; |
0a5578cf ACM |
287 | EXPORT_SYMBOL_GPL(tcp_orphan_count); |
288 | ||
a4fe34bf | 289 | long sysctl_tcp_mem[3] __read_mostly; |
a4fe34bf | 290 | EXPORT_SYMBOL(sysctl_tcp_mem); |
1da177e4 | 291 | |
8d987e5c | 292 | atomic_long_t tcp_memory_allocated; /* Current allocated memory. */ |
1da177e4 | 293 | EXPORT_SYMBOL(tcp_memory_allocated); |
1748376b | 294 | |
60e2a778 UB |
295 | #if IS_ENABLED(CONFIG_SMC) |
296 | DEFINE_STATIC_KEY_FALSE(tcp_have_smc); | |
297 | EXPORT_SYMBOL(tcp_have_smc); | |
298 | #endif | |
299 | ||
1748376b ED |
300 | /* |
301 | * Current number of TCP sockets. | |
302 | */ | |
303 | struct percpu_counter tcp_sockets_allocated; | |
1da177e4 LT |
304 | EXPORT_SYMBOL(tcp_sockets_allocated); |
305 | ||
9c55e01c JA |
306 | /* |
307 | * TCP splice context | |
308 | */ | |
309 | struct tcp_splice_state { | |
310 | struct pipe_inode_info *pipe; | |
311 | size_t len; | |
312 | unsigned int flags; | |
313 | }; | |
314 | ||
1da177e4 LT |
315 | /* |
316 | * Pressure flag: try to collapse. | |
317 | * Technical note: it is used by multiple contexts non atomically. | |
3ab224be | 318 | * All the __sk_mem_schedule() is of this nature: accounting |
1da177e4 LT |
319 | * is strict, actions are advisory and have some latency. |
320 | */ | |
06044751 ED |
321 | unsigned long tcp_memory_pressure __read_mostly; |
322 | EXPORT_SYMBOL_GPL(tcp_memory_pressure); | |
1da177e4 | 323 | |
5c52ba17 | 324 | void tcp_enter_memory_pressure(struct sock *sk) |
1da177e4 | 325 | { |
06044751 ED |
326 | unsigned long val; |
327 | ||
328 | if (tcp_memory_pressure) | |
329 | return; | |
330 | val = jiffies; | |
331 | ||
332 | if (!val) | |
333 | val--; | |
334 | if (!cmpxchg(&tcp_memory_pressure, 0, val)) | |
4e673444 | 335 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES); |
1da177e4 | 336 | } |
06044751 ED |
337 | EXPORT_SYMBOL_GPL(tcp_enter_memory_pressure); |
338 | ||
339 | void tcp_leave_memory_pressure(struct sock *sk) | |
340 | { | |
341 | unsigned long val; | |
342 | ||
343 | if (!tcp_memory_pressure) | |
344 | return; | |
345 | val = xchg(&tcp_memory_pressure, 0); | |
346 | if (val) | |
347 | NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURESCHRONO, | |
348 | jiffies_to_msecs(jiffies - val)); | |
349 | } | |
350 | EXPORT_SYMBOL_GPL(tcp_leave_memory_pressure); | |
1da177e4 | 351 | |
b103cf34 JA |
352 | /* Convert seconds to retransmits based on initial and max timeout */ |
353 | static u8 secs_to_retrans(int seconds, int timeout, int rto_max) | |
354 | { | |
355 | u8 res = 0; | |
356 | ||
357 | if (seconds > 0) { | |
358 | int period = timeout; | |
359 | ||
360 | res = 1; | |
361 | while (seconds > period && res < 255) { | |
362 | res++; | |
363 | timeout <<= 1; | |
364 | if (timeout > rto_max) | |
365 | timeout = rto_max; | |
366 | period += timeout; | |
367 | } | |
368 | } | |
369 | return res; | |
370 | } | |
371 | ||
372 | /* Convert retransmits to seconds based on initial and max timeout */ | |
373 | static int retrans_to_secs(u8 retrans, int timeout, int rto_max) | |
374 | { | |
375 | int period = 0; | |
376 | ||
377 | if (retrans > 0) { | |
378 | period = timeout; | |
379 | while (--retrans) { | |
380 | timeout <<= 1; | |
381 | if (timeout > rto_max) | |
382 | timeout = rto_max; | |
383 | period += timeout; | |
384 | } | |
385 | } | |
386 | return period; | |
387 | } | |
388 | ||
0263598c WW |
389 | static u64 tcp_compute_delivery_rate(const struct tcp_sock *tp) |
390 | { | |
391 | u32 rate = READ_ONCE(tp->rate_delivered); | |
392 | u32 intv = READ_ONCE(tp->rate_interval_us); | |
393 | u64 rate64 = 0; | |
394 | ||
395 | if (rate && intv) { | |
396 | rate64 = (u64)rate * tp->mss_cache * USEC_PER_SEC; | |
397 | do_div(rate64, intv); | |
398 | } | |
399 | return rate64; | |
400 | } | |
401 | ||
900f65d3 NC |
402 | /* Address-family independent initialization for a tcp_sock. |
403 | * | |
404 | * NOTE: A lot of things set to zero explicitly by call to | |
405 | * sk_alloc() so need not be done here. | |
406 | */ | |
407 | void tcp_init_sock(struct sock *sk) | |
408 | { | |
409 | struct inet_connection_sock *icsk = inet_csk(sk); | |
410 | struct tcp_sock *tp = tcp_sk(sk); | |
411 | ||
9f5afeae | 412 | tp->out_of_order_queue = RB_ROOT; |
75c119af | 413 | sk->tcp_rtx_queue = RB_ROOT; |
900f65d3 | 414 | tcp_init_xmit_timers(sk); |
46d3ceab | 415 | INIT_LIST_HEAD(&tp->tsq_node); |
e2080072 | 416 | INIT_LIST_HEAD(&tp->tsorted_sent_queue); |
900f65d3 NC |
417 | |
418 | icsk->icsk_rto = TCP_TIMEOUT_INIT; | |
740b0f18 | 419 | tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT); |
ac9517fc | 420 | minmax_reset(&tp->rtt_min, tcp_jiffies32, ~0U); |
900f65d3 NC |
421 | |
422 | /* So many TCP implementations out there (incorrectly) count the | |
423 | * initial SYN frame in their delayed-ACK and congestion control | |
424 | * algorithms that we must have the following bandaid to talk | |
425 | * efficiently to them. -DaveM | |
426 | */ | |
427 | tp->snd_cwnd = TCP_INIT_CWND; | |
428 | ||
d7722e85 SHY |
429 | /* There's a bubble in the pipe until at least the first ACK. */ |
430 | tp->app_limited = ~0U; | |
431 | ||
900f65d3 NC |
432 | /* See draft-stevens-tcpca-spec-01 for discussion of the |
433 | * initialization of these values. | |
434 | */ | |
435 | tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; | |
436 | tp->snd_cwnd_clamp = ~0; | |
437 | tp->mss_cache = TCP_MSS_DEFAULT; | |
438 | ||
1043e25f | 439 | tp->reordering = sock_net(sk)->ipv4.sysctl_tcp_reordering; |
55d8694f | 440 | tcp_assign_congestion_control(sk); |
900f65d3 | 441 | |
ceaa1fef | 442 | tp->tsoffset = 0; |
1f255691 | 443 | tp->rack.reo_wnd_steps = 1; |
ceaa1fef | 444 | |
900f65d3 NC |
445 | sk->sk_state = TCP_CLOSE; |
446 | ||
447 | sk->sk_write_space = sk_stream_write_space; | |
448 | sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); | |
449 | ||
450 | icsk->icsk_sync_mss = tcp_sync_mss; | |
451 | ||
356d1833 ED |
452 | sk->sk_sndbuf = sock_net(sk)->ipv4.sysctl_tcp_wmem[1]; |
453 | sk->sk_rcvbuf = sock_net(sk)->ipv4.sysctl_tcp_rmem[1]; | |
900f65d3 | 454 | |
900f65d3 | 455 | sk_sockets_allocated_inc(sk); |
900f65d3 NC |
456 | } |
457 | EXPORT_SYMBOL(tcp_init_sock); | |
458 | ||
27204aaa WW |
459 | void tcp_init_transfer(struct sock *sk, int bpf_op) |
460 | { | |
461 | struct inet_connection_sock *icsk = inet_csk(sk); | |
462 | ||
463 | tcp_mtup_init(sk); | |
464 | icsk->icsk_af_ops->rebuild_header(sk); | |
465 | tcp_init_metrics(sk); | |
466 | tcp_call_bpf(sk, bpf_op); | |
467 | tcp_init_congestion_control(sk); | |
468 | tcp_init_buffer_space(sk); | |
469 | } | |
470 | ||
4e8cc228 | 471 | static void tcp_tx_timestamp(struct sock *sk, u16 tsflags) |
4ed2d765 | 472 | { |
4e8cc228 ED |
473 | struct sk_buff *skb = tcp_write_queue_tail(sk); |
474 | ||
ad02c4f5 | 475 | if (tsflags && skb) { |
f066e2b0 | 476 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
6b084928 | 477 | struct tcp_skb_cb *tcb = TCP_SKB_CB(skb); |
4ed2d765 | 478 | |
c14ac945 | 479 | sock_tx_timestamp(sk, tsflags, &shinfo->tx_flags); |
0a2cf20c SHY |
480 | if (tsflags & SOF_TIMESTAMPING_TX_ACK) |
481 | tcb->txstamp_ack = 1; | |
482 | if (tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) | |
f066e2b0 WB |
483 | shinfo->tskey = TCP_SKB_CB(skb)->seq + skb->len - 1; |
484 | } | |
4ed2d765 WB |
485 | } |
486 | ||
1da177e4 LT |
487 | /* |
488 | * Wait for a TCP event. | |
489 | * | |
490 | * Note that we don't need to lock the socket, as the upper poll layers | |
491 | * take care of normal races (between the test and the event) and we don't | |
492 | * go look at any of the socket buffers directly. | |
493 | */ | |
494 | unsigned int tcp_poll(struct file *file, struct socket *sock, poll_table *wait) | |
495 | { | |
496 | unsigned int mask; | |
497 | struct sock *sk = sock->sk; | |
cf533ea5 | 498 | const struct tcp_sock *tp = tcp_sk(sk); |
00fd38d9 | 499 | int state; |
1da177e4 | 500 | |
c3f1dbaf DM |
501 | sock_rps_record_flow(sk); |
502 | ||
aa395145 | 503 | sock_poll_wait(file, sk_sleep(sk), wait); |
00fd38d9 | 504 | |
986ffdfd | 505 | state = inet_sk_state_load(sk); |
00fd38d9 | 506 | if (state == TCP_LISTEN) |
dc40c7bc | 507 | return inet_csk_listen_poll(sk); |
1da177e4 LT |
508 | |
509 | /* Socket is not locked. We are protected from async events | |
70efce27 WN |
510 | * by poll logic and correct handling of state changes |
511 | * made by other threads is impossible in any case. | |
1da177e4 LT |
512 | */ |
513 | ||
514 | mask = 0; | |
1da177e4 LT |
515 | |
516 | /* | |
517 | * POLLHUP is certainly not done right. But poll() doesn't | |
518 | * have a notion of HUP in just one direction, and for a | |
519 | * socket the read side is more interesting. | |
520 | * | |
521 | * Some poll() documentation says that POLLHUP is incompatible | |
522 | * with the POLLOUT/POLLWR flags, so somebody should check this | |
523 | * all. But careful, it tends to be safer to return too many | |
524 | * bits than too few, and you can easily break real applications | |
525 | * if you don't tell them that something has hung up! | |
526 | * | |
527 | * Check-me. | |
528 | * | |
529 | * Check number 1. POLLHUP is _UNMASKABLE_ event (see UNIX98 and | |
530 | * our fs/select.c). It means that after we received EOF, | |
531 | * poll always returns immediately, making impossible poll() on write() | |
532 | * in state CLOSE_WAIT. One solution is evident --- to set POLLHUP | |
533 | * if and only if shutdown has been made in both directions. | |
534 | * Actually, it is interesting to look how Solaris and DUX | |
70efce27 | 535 | * solve this dilemma. I would prefer, if POLLHUP were maskable, |
1da177e4 LT |
536 | * then we could set it on SND_SHUTDOWN. BTW examples given |
537 | * in Stevens' books assume exactly this behaviour, it explains | |
70efce27 | 538 | * why POLLHUP is incompatible with POLLOUT. --ANK |
1da177e4 LT |
539 | * |
540 | * NOTE. Check for TCP_CLOSE is added. The goal is to prevent | |
541 | * blocking on fresh not-connected or disconnected socket. --ANK | |
542 | */ | |
00fd38d9 | 543 | if (sk->sk_shutdown == SHUTDOWN_MASK || state == TCP_CLOSE) |
1da177e4 LT |
544 | mask |= POLLHUP; |
545 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
f348d70a | 546 | mask |= POLLIN | POLLRDNORM | POLLRDHUP; |
1da177e4 | 547 | |
8336886f | 548 | /* Connected or passive Fast Open socket? */ |
00fd38d9 ED |
549 | if (state != TCP_SYN_SENT && |
550 | (state != TCP_SYN_RECV || tp->fastopen_rsk)) { | |
c7004482 DM |
551 | int target = sock_rcvlowat(sk, 0, INT_MAX); |
552 | ||
553 | if (tp->urg_seq == tp->copied_seq && | |
554 | !sock_flag(sk, SOCK_URGINLINE) && | |
555 | tp->urg_data) | |
b634f875 | 556 | target++; |
c7004482 | 557 | |
c7004482 | 558 | if (tp->rcv_nxt - tp->copied_seq >= target) |
1da177e4 LT |
559 | mask |= POLLIN | POLLRDNORM; |
560 | ||
561 | if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { | |
64dc6130 | 562 | if (sk_stream_is_writeable(sk)) { |
1da177e4 LT |
563 | mask |= POLLOUT | POLLWRNORM; |
564 | } else { /* send SIGIO later */ | |
9cd3e072 | 565 | sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 LT |
566 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
567 | ||
568 | /* Race breaker. If space is freed after | |
569 | * wspace test but before the flags are set, | |
3c715127 | 570 | * IO signal will be lost. Memory barrier |
571 | * pairs with the input side. | |
1da177e4 | 572 | */ |
3c715127 | 573 | smp_mb__after_atomic(); |
64dc6130 | 574 | if (sk_stream_is_writeable(sk)) |
1da177e4 LT |
575 | mask |= POLLOUT | POLLWRNORM; |
576 | } | |
d84ba638 KM |
577 | } else |
578 | mask |= POLLOUT | POLLWRNORM; | |
1da177e4 LT |
579 | |
580 | if (tp->urg_data & TCP_URG_VALID) | |
581 | mask |= POLLPRI; | |
d68be71e | 582 | } else if (state == TCP_SYN_SENT && inet_sk(sk)->defer_connect) { |
19f6d3f3 WW |
583 | /* Active TCP fastopen socket with defer_connect |
584 | * Return POLLOUT so application can call write() | |
585 | * in order for kernel to generate SYN+data | |
586 | */ | |
587 | mask |= POLLOUT | POLLWRNORM; | |
1da177e4 | 588 | } |
a4d25803 TM |
589 | /* This barrier is coupled with smp_wmb() in tcp_reset() */ |
590 | smp_rmb(); | |
4ed2d765 | 591 | if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue)) |
a4d25803 TM |
592 | mask |= POLLERR; |
593 | ||
1da177e4 LT |
594 | return mask; |
595 | } | |
4bc2f18b | 596 | EXPORT_SYMBOL(tcp_poll); |
1da177e4 LT |
597 | |
598 | int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg) | |
599 | { | |
600 | struct tcp_sock *tp = tcp_sk(sk); | |
601 | int answ; | |
0e71c55c | 602 | bool slow; |
1da177e4 LT |
603 | |
604 | switch (cmd) { | |
605 | case SIOCINQ: | |
606 | if (sk->sk_state == TCP_LISTEN) | |
607 | return -EINVAL; | |
608 | ||
0e71c55c | 609 | slow = lock_sock_fast(sk); |
473bd239 | 610 | answ = tcp_inq(sk); |
0e71c55c | 611 | unlock_sock_fast(sk, slow); |
1da177e4 LT |
612 | break; |
613 | case SIOCATMARK: | |
614 | answ = tp->urg_data && tp->urg_seq == tp->copied_seq; | |
615 | break; | |
616 | case SIOCOUTQ: | |
617 | if (sk->sk_state == TCP_LISTEN) | |
618 | return -EINVAL; | |
619 | ||
620 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) | |
621 | answ = 0; | |
622 | else | |
623 | answ = tp->write_seq - tp->snd_una; | |
624 | break; | |
2f4e1b39 MS |
625 | case SIOCOUTQNSD: |
626 | if (sk->sk_state == TCP_LISTEN) | |
627 | return -EINVAL; | |
628 | ||
629 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) | |
630 | answ = 0; | |
631 | else | |
632 | answ = tp->write_seq - tp->snd_nxt; | |
633 | break; | |
1da177e4 LT |
634 | default: |
635 | return -ENOIOCTLCMD; | |
3ff50b79 | 636 | } |
1da177e4 LT |
637 | |
638 | return put_user(answ, (int __user *)arg); | |
639 | } | |
4bc2f18b | 640 | EXPORT_SYMBOL(tcp_ioctl); |
1da177e4 | 641 | |
1da177e4 LT |
642 | static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb) |
643 | { | |
4de075e0 | 644 | TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH; |
1da177e4 LT |
645 | tp->pushed_seq = tp->write_seq; |
646 | } | |
647 | ||
a2a385d6 | 648 | static inline bool forced_push(const struct tcp_sock *tp) |
1da177e4 LT |
649 | { |
650 | return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1)); | |
651 | } | |
652 | ||
f4a775d1 | 653 | static void skb_entail(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 654 | { |
9e412ba7 | 655 | struct tcp_sock *tp = tcp_sk(sk); |
352d4800 ACM |
656 | struct tcp_skb_cb *tcb = TCP_SKB_CB(skb); |
657 | ||
658 | skb->csum = 0; | |
659 | tcb->seq = tcb->end_seq = tp->write_seq; | |
4de075e0 | 660 | tcb->tcp_flags = TCPHDR_ACK; |
352d4800 | 661 | tcb->sacked = 0; |
f4a775d1 | 662 | __skb_header_release(skb); |
fe067e8a | 663 | tcp_add_write_queue_tail(sk, skb); |
3ab224be HA |
664 | sk->sk_wmem_queued += skb->truesize; |
665 | sk_mem_charge(sk, skb->truesize); | |
89ebd197 | 666 | if (tp->nonagle & TCP_NAGLE_PUSH) |
e905a9ed | 667 | tp->nonagle &= ~TCP_NAGLE_PUSH; |
6f021c62 ED |
668 | |
669 | tcp_slow_start_after_idle_check(sk); | |
1da177e4 LT |
670 | } |
671 | ||
afeca340 | 672 | static inline void tcp_mark_urg(struct tcp_sock *tp, int flags) |
1da177e4 | 673 | { |
33f5f57e | 674 | if (flags & MSG_OOB) |
1da177e4 | 675 | tp->snd_up = tp->write_seq; |
1da177e4 LT |
676 | } |
677 | ||
f54b3111 | 678 | /* If a not yet filled skb is pushed, do not send it if |
a181ceb5 | 679 | * we have data packets in Qdisc or NIC queues : |
f54b3111 ED |
680 | * Because TX completion will happen shortly, it gives a chance |
681 | * to coalesce future sendmsg() payload into this skb, without | |
682 | * need for a timer, and with no latency trade off. | |
683 | * As packets containing data payload have a bigger truesize | |
a181ceb5 ED |
684 | * than pure acks (dataless) packets, the last checks prevent |
685 | * autocorking if we only have an ACK in Qdisc/NIC queues, | |
686 | * or if TX completion was delayed after we processed ACK packet. | |
f54b3111 ED |
687 | */ |
688 | static bool tcp_should_autocork(struct sock *sk, struct sk_buff *skb, | |
689 | int size_goal) | |
1da177e4 | 690 | { |
f54b3111 | 691 | return skb->len < size_goal && |
790f00e1 | 692 | sock_net(sk)->ipv4.sysctl_tcp_autocorking && |
a181ceb5 | 693 | skb != tcp_write_queue_head(sk) && |
14afee4b | 694 | refcount_read(&sk->sk_wmem_alloc) > skb->truesize; |
f54b3111 ED |
695 | } |
696 | ||
697 | static void tcp_push(struct sock *sk, int flags, int mss_now, | |
698 | int nonagle, int size_goal) | |
699 | { | |
700 | struct tcp_sock *tp = tcp_sk(sk); | |
701 | struct sk_buff *skb; | |
afeca340 | 702 | |
f54b3111 | 703 | skb = tcp_write_queue_tail(sk); |
75c119af ED |
704 | if (!skb) |
705 | return; | |
f54b3111 ED |
706 | if (!(flags & MSG_MORE) || forced_push(tp)) |
707 | tcp_mark_push(tp, skb); | |
708 | ||
709 | tcp_mark_urg(tp, flags); | |
710 | ||
711 | if (tcp_should_autocork(sk, skb, size_goal)) { | |
712 | ||
713 | /* avoid atomic op if TSQ_THROTTLED bit is already set */ | |
7aa5470c | 714 | if (!test_bit(TSQ_THROTTLED, &sk->sk_tsq_flags)) { |
f54b3111 | 715 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAUTOCORKING); |
7aa5470c | 716 | set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags); |
f54b3111 | 717 | } |
a181ceb5 ED |
718 | /* It is possible TX completion already happened |
719 | * before we set TSQ_THROTTLED. | |
720 | */ | |
14afee4b | 721 | if (refcount_read(&sk->sk_wmem_alloc) > skb->truesize) |
a181ceb5 | 722 | return; |
1da177e4 | 723 | } |
f54b3111 ED |
724 | |
725 | if (flags & MSG_MORE) | |
726 | nonagle = TCP_NAGLE_CORK; | |
727 | ||
728 | __tcp_push_pending_frames(sk, mss_now, nonagle); | |
1da177e4 LT |
729 | } |
730 | ||
6ff7751d AB |
731 | static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, |
732 | unsigned int offset, size_t len) | |
9c55e01c JA |
733 | { |
734 | struct tcp_splice_state *tss = rd_desc->arg.data; | |
33966dd0 | 735 | int ret; |
9c55e01c | 736 | |
a60e3cc7 | 737 | ret = skb_splice_bits(skb, skb->sk, offset, tss->pipe, |
25869262 | 738 | min(rd_desc->count, len), tss->flags); |
33966dd0 WT |
739 | if (ret > 0) |
740 | rd_desc->count -= ret; | |
741 | return ret; | |
9c55e01c JA |
742 | } |
743 | ||
744 | static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss) | |
745 | { | |
746 | /* Store TCP splice context information in read_descriptor_t. */ | |
747 | read_descriptor_t rd_desc = { | |
748 | .arg.data = tss, | |
33966dd0 | 749 | .count = tss->len, |
9c55e01c JA |
750 | }; |
751 | ||
752 | return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv); | |
753 | } | |
754 | ||
755 | /** | |
756 | * tcp_splice_read - splice data from TCP socket to a pipe | |
757 | * @sock: socket to splice from | |
758 | * @ppos: position (not valid) | |
759 | * @pipe: pipe to splice to | |
760 | * @len: number of bytes to splice | |
761 | * @flags: splice modifier flags | |
762 | * | |
763 | * Description: | |
764 | * Will read pages from given socket and fill them into a pipe. | |
765 | * | |
766 | **/ | |
767 | ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos, | |
768 | struct pipe_inode_info *pipe, size_t len, | |
769 | unsigned int flags) | |
770 | { | |
771 | struct sock *sk = sock->sk; | |
772 | struct tcp_splice_state tss = { | |
773 | .pipe = pipe, | |
774 | .len = len, | |
775 | .flags = flags, | |
776 | }; | |
777 | long timeo; | |
778 | ssize_t spliced; | |
779 | int ret; | |
780 | ||
3a047bf8 | 781 | sock_rps_record_flow(sk); |
9c55e01c JA |
782 | /* |
783 | * We can't seek on a socket input | |
784 | */ | |
785 | if (unlikely(*ppos)) | |
786 | return -ESPIPE; | |
787 | ||
788 | ret = spliced = 0; | |
789 | ||
790 | lock_sock(sk); | |
791 | ||
42324c62 | 792 | timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK); |
9c55e01c JA |
793 | while (tss.len) { |
794 | ret = __tcp_splice_read(sk, &tss); | |
795 | if (ret < 0) | |
796 | break; | |
797 | else if (!ret) { | |
798 | if (spliced) | |
799 | break; | |
9c55e01c JA |
800 | if (sock_flag(sk, SOCK_DONE)) |
801 | break; | |
802 | if (sk->sk_err) { | |
803 | ret = sock_error(sk); | |
804 | break; | |
805 | } | |
806 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
807 | break; | |
808 | if (sk->sk_state == TCP_CLOSE) { | |
809 | /* | |
810 | * This occurs when user tries to read | |
811 | * from never connected socket. | |
812 | */ | |
813 | if (!sock_flag(sk, SOCK_DONE)) | |
814 | ret = -ENOTCONN; | |
815 | break; | |
816 | } | |
817 | if (!timeo) { | |
818 | ret = -EAGAIN; | |
819 | break; | |
820 | } | |
ccf7abb9 ED |
821 | /* if __tcp_splice_read() got nothing while we have |
822 | * an skb in receive queue, we do not want to loop. | |
823 | * This might happen with URG data. | |
824 | */ | |
825 | if (!skb_queue_empty(&sk->sk_receive_queue)) | |
826 | break; | |
dfbafc99 | 827 | sk_wait_data(sk, &timeo, NULL); |
9c55e01c JA |
828 | if (signal_pending(current)) { |
829 | ret = sock_intr_errno(timeo); | |
830 | break; | |
831 | } | |
832 | continue; | |
833 | } | |
834 | tss.len -= ret; | |
835 | spliced += ret; | |
836 | ||
33966dd0 WT |
837 | if (!timeo) |
838 | break; | |
9c55e01c JA |
839 | release_sock(sk); |
840 | lock_sock(sk); | |
841 | ||
842 | if (sk->sk_err || sk->sk_state == TCP_CLOSE || | |
33966dd0 | 843 | (sk->sk_shutdown & RCV_SHUTDOWN) || |
9c55e01c JA |
844 | signal_pending(current)) |
845 | break; | |
846 | } | |
847 | ||
848 | release_sock(sk); | |
849 | ||
850 | if (spliced) | |
851 | return spliced; | |
852 | ||
853 | return ret; | |
854 | } | |
4bc2f18b | 855 | EXPORT_SYMBOL(tcp_splice_read); |
9c55e01c | 856 | |
eb934478 ED |
857 | struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, |
858 | bool force_schedule) | |
f561d0f2 PE |
859 | { |
860 | struct sk_buff *skb; | |
861 | ||
862 | /* The TCP header must be at least 32-bit aligned. */ | |
863 | size = ALIGN(size, 4); | |
864 | ||
8e4d980a ED |
865 | if (unlikely(tcp_under_memory_pressure(sk))) |
866 | sk_mem_reclaim_partial(sk); | |
867 | ||
f561d0f2 | 868 | skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp); |
8e4d980a | 869 | if (likely(skb)) { |
eb934478 | 870 | bool mem_scheduled; |
8e4d980a | 871 | |
eb934478 ED |
872 | if (force_schedule) { |
873 | mem_scheduled = true; | |
8e4d980a ED |
874 | sk_forced_mem_schedule(sk, skb->truesize); |
875 | } else { | |
eb934478 | 876 | mem_scheduled = sk_wmem_schedule(sk, skb->truesize); |
8e4d980a | 877 | } |
eb934478 | 878 | if (likely(mem_scheduled)) { |
a21d4572 | 879 | skb_reserve(skb, sk->sk_prot->max_header); |
f561d0f2 PE |
880 | /* |
881 | * Make sure that we have exactly size bytes | |
882 | * available to the caller, no more, no less. | |
883 | */ | |
16fad69c | 884 | skb->reserved_tailroom = skb->end - skb->tail - size; |
e2080072 | 885 | INIT_LIST_HEAD(&skb->tcp_tsorted_anchor); |
f561d0f2 PE |
886 | return skb; |
887 | } | |
888 | __kfree_skb(skb); | |
889 | } else { | |
5c52ba17 | 890 | sk->sk_prot->enter_memory_pressure(sk); |
f561d0f2 PE |
891 | sk_stream_moderate_sndbuf(sk); |
892 | } | |
893 | return NULL; | |
894 | } | |
895 | ||
0c54b85f IJ |
896 | static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now, |
897 | int large_allowed) | |
898 | { | |
899 | struct tcp_sock *tp = tcp_sk(sk); | |
6c09fa09 | 900 | u32 new_size_goal, size_goal; |
605ad7f1 ED |
901 | |
902 | if (!large_allowed || !sk_can_gso(sk)) | |
903 | return mss_now; | |
904 | ||
6c09fa09 ED |
905 | /* Note : tcp_tso_autosize() will eventually split this later */ |
906 | new_size_goal = sk->sk_gso_max_size - 1 - MAX_TCP_HEADER; | |
605ad7f1 ED |
907 | new_size_goal = tcp_bound_to_half_wnd(tp, new_size_goal); |
908 | ||
909 | /* We try hard to avoid divides here */ | |
910 | size_goal = tp->gso_segs * mss_now; | |
911 | if (unlikely(new_size_goal < size_goal || | |
912 | new_size_goal >= size_goal + mss_now)) { | |
913 | tp->gso_segs = min_t(u16, new_size_goal / mss_now, | |
914 | sk->sk_gso_max_segs); | |
915 | size_goal = tp->gso_segs * mss_now; | |
0c54b85f IJ |
916 | } |
917 | ||
605ad7f1 | 918 | return max(size_goal, mss_now); |
0c54b85f IJ |
919 | } |
920 | ||
921 | static int tcp_send_mss(struct sock *sk, int *size_goal, int flags) | |
922 | { | |
923 | int mss_now; | |
924 | ||
925 | mss_now = tcp_current_mss(sk); | |
926 | *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB)); | |
927 | ||
928 | return mss_now; | |
929 | } | |
930 | ||
e3b5616a DW |
931 | ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, |
932 | size_t size, int flags) | |
1da177e4 LT |
933 | { |
934 | struct tcp_sock *tp = tcp_sk(sk); | |
c1b4a7e6 | 935 | int mss_now, size_goal; |
1da177e4 LT |
936 | int err; |
937 | ssize_t copied; | |
938 | long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); | |
939 | ||
8336886f JC |
940 | /* Wait for a connection to finish. One exception is TCP Fast Open |
941 | * (passive side) where data is allowed to be sent before a connection | |
942 | * is fully established. | |
943 | */ | |
944 | if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) && | |
945 | !tcp_passive_fastopen(sk)) { | |
686a5624 YM |
946 | err = sk_stream_wait_connect(sk, &timeo); |
947 | if (err != 0) | |
1da177e4 | 948 | goto out_err; |
8336886f | 949 | } |
1da177e4 | 950 | |
9cd3e072 | 951 | sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 | 952 | |
0c54b85f | 953 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
954 | copied = 0; |
955 | ||
956 | err = -EPIPE; | |
957 | if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) | |
0d6a775e | 958 | goto out_err; |
1da177e4 | 959 | |
64022d0b | 960 | while (size > 0) { |
fe067e8a | 961 | struct sk_buff *skb = tcp_write_queue_tail(sk); |
38ba0a65 | 962 | int copy, i; |
38ba0a65 | 963 | bool can_coalesce; |
1da177e4 | 964 | |
75c119af | 965 | if (!skb || (copy = size_goal - skb->len) <= 0 || |
c134ecb8 | 966 | !tcp_skb_can_collapse_to(skb)) { |
1da177e4 LT |
967 | new_segment: |
968 | if (!sk_stream_memory_free(sk)) | |
969 | goto wait_for_sndbuf; | |
970 | ||
eb934478 | 971 | skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation, |
75c119af | 972 | tcp_rtx_and_write_queues_empty(sk)); |
1da177e4 LT |
973 | if (!skb) |
974 | goto wait_for_memory; | |
975 | ||
9e412ba7 | 976 | skb_entail(sk, skb); |
c1b4a7e6 | 977 | copy = size_goal; |
1da177e4 LT |
978 | } |
979 | ||
980 | if (copy > size) | |
981 | copy = size; | |
982 | ||
983 | i = skb_shinfo(skb)->nr_frags; | |
984 | can_coalesce = skb_can_coalesce(skb, i, page, offset); | |
5f74f82e | 985 | if (!can_coalesce && i >= sysctl_max_skb_frags) { |
1da177e4 LT |
986 | tcp_mark_push(tp, skb); |
987 | goto new_segment; | |
988 | } | |
3ab224be | 989 | if (!sk_wmem_schedule(sk, copy)) |
1da177e4 | 990 | goto wait_for_memory; |
e905a9ed | 991 | |
1da177e4 | 992 | if (can_coalesce) { |
9e903e08 | 993 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); |
1da177e4 LT |
994 | } else { |
995 | get_page(page); | |
996 | skb_fill_page_desc(skb, i, page, offset, copy); | |
997 | } | |
c9af6db4 | 998 | skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; |
cef401de | 999 | |
1da177e4 LT |
1000 | skb->len += copy; |
1001 | skb->data_len += copy; | |
1002 | skb->truesize += copy; | |
1003 | sk->sk_wmem_queued += copy; | |
3ab224be | 1004 | sk_mem_charge(sk, copy); |
84fa7933 | 1005 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 LT |
1006 | tp->write_seq += copy; |
1007 | TCP_SKB_CB(skb)->end_seq += copy; | |
cd7d8498 | 1008 | tcp_skb_pcount_set(skb, 0); |
1da177e4 LT |
1009 | |
1010 | if (!copied) | |
4de075e0 | 1011 | TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH; |
1da177e4 LT |
1012 | |
1013 | copied += copy; | |
64022d0b | 1014 | offset += copy; |
686a5624 | 1015 | size -= copy; |
ad02c4f5 | 1016 | if (!size) |
1da177e4 LT |
1017 | goto out; |
1018 | ||
69d15067 | 1019 | if (skb->len < size_goal || (flags & MSG_OOB)) |
1da177e4 LT |
1020 | continue; |
1021 | ||
1022 | if (forced_push(tp)) { | |
1023 | tcp_mark_push(tp, skb); | |
9e412ba7 | 1024 | __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH); |
fe067e8a | 1025 | } else if (skb == tcp_send_head(sk)) |
1da177e4 LT |
1026 | tcp_push_one(sk, mss_now); |
1027 | continue; | |
1028 | ||
1029 | wait_for_sndbuf: | |
1030 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
1031 | wait_for_memory: | |
f54b3111 ED |
1032 | tcp_push(sk, flags & ~MSG_MORE, mss_now, |
1033 | TCP_NAGLE_PUSH, size_goal); | |
1da177e4 | 1034 | |
686a5624 YM |
1035 | err = sk_stream_wait_memory(sk, &timeo); |
1036 | if (err != 0) | |
1da177e4 LT |
1037 | goto do_error; |
1038 | ||
0c54b85f | 1039 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
1040 | } |
1041 | ||
1042 | out: | |
ad02c4f5 | 1043 | if (copied) { |
4e8cc228 | 1044 | tcp_tx_timestamp(sk, sk->sk_tsflags); |
ad02c4f5 SHY |
1045 | if (!(flags & MSG_SENDPAGE_NOTLAST)) |
1046 | tcp_push(sk, flags, mss_now, tp->nonagle, size_goal); | |
1047 | } | |
1da177e4 LT |
1048 | return copied; |
1049 | ||
1050 | do_error: | |
1051 | if (copied) | |
1052 | goto out; | |
1053 | out_err: | |
ce5ec440 | 1054 | /* make sure we wake any epoll edge trigger waiter */ |
b0f71bd3 FY |
1055 | if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && |
1056 | err == -EAGAIN)) { | |
ce5ec440 | 1057 | sk->sk_write_space(sk); |
b0f71bd3 FY |
1058 | tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED); |
1059 | } | |
1da177e4 LT |
1060 | return sk_stream_error(sk, flags, err); |
1061 | } | |
e3b5616a | 1062 | EXPORT_SYMBOL_GPL(do_tcp_sendpages); |
1da177e4 | 1063 | |
306b13eb TH |
1064 | int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, |
1065 | size_t size, int flags) | |
1da177e4 | 1066 | { |
1da177e4 | 1067 | if (!(sk->sk_route_caps & NETIF_F_SG) || |
9a49850d | 1068 | !sk_check_csum_caps(sk)) |
bd9dfc54 | 1069 | return sock_no_sendpage_locked(sk, page, offset, size, flags); |
1da177e4 | 1070 | |
d7722e85 SHY |
1071 | tcp_rate_check_app_limited(sk); /* is sending application-limited? */ |
1072 | ||
306b13eb TH |
1073 | return do_tcp_sendpages(sk, page, offset, size, flags); |
1074 | } | |
774c4673 | 1075 | EXPORT_SYMBOL_GPL(tcp_sendpage_locked); |
306b13eb TH |
1076 | |
1077 | int tcp_sendpage(struct sock *sk, struct page *page, int offset, | |
1078 | size_t size, int flags) | |
1079 | { | |
1080 | int ret; | |
1081 | ||
1082 | lock_sock(sk); | |
1083 | ret = tcp_sendpage_locked(sk, page, offset, size, flags); | |
1da177e4 | 1084 | release_sock(sk); |
306b13eb TH |
1085 | |
1086 | return ret; | |
1da177e4 | 1087 | } |
4bc2f18b | 1088 | EXPORT_SYMBOL(tcp_sendpage); |
1da177e4 | 1089 | |
3613b3db ED |
1090 | /* Do not bother using a page frag for very small frames. |
1091 | * But use this heuristic only for the first skb in write queue. | |
1092 | * | |
1093 | * Having no payload in skb->head allows better SACK shifting | |
1094 | * in tcp_shift_skb_data(), reducing sack/rack overhead, because | |
1095 | * write queue has less skbs. | |
1096 | * Each skb can hold up to MAX_SKB_FRAGS * 32Kbytes, or ~0.5 MB. | |
1097 | * This also speeds up tso_fragment(), since it wont fallback | |
1098 | * to tcp_fragment(). | |
1099 | */ | |
1100 | static int linear_payload_sz(bool first_skb) | |
1101 | { | |
1102 | if (first_skb) | |
1103 | return SKB_WITH_OVERHEAD(2048 - MAX_TCP_HEADER); | |
1104 | return 0; | |
1105 | } | |
1106 | ||
1107 | static int select_size(const struct sock *sk, bool sg, bool first_skb) | |
1da177e4 | 1108 | { |
cf533ea5 | 1109 | const struct tcp_sock *tp = tcp_sk(sk); |
c1b4a7e6 | 1110 | int tmp = tp->mss_cache; |
1da177e4 | 1111 | |
def87cf4 | 1112 | if (sg) { |
f07d960d | 1113 | if (sk_can_gso(sk)) { |
3613b3db | 1114 | tmp = linear_payload_sz(first_skb); |
f07d960d | 1115 | } else { |
b4e26f5e DM |
1116 | int pgbreak = SKB_MAX_HEAD(MAX_TCP_HEADER); |
1117 | ||
1118 | if (tmp >= pgbreak && | |
1119 | tmp <= pgbreak + (MAX_SKB_FRAGS - 1) * PAGE_SIZE) | |
1120 | tmp = pgbreak; | |
1121 | } | |
1122 | } | |
1da177e4 | 1123 | |
1da177e4 LT |
1124 | return tmp; |
1125 | } | |
1126 | ||
cf60af03 YC |
1127 | void tcp_free_fastopen_req(struct tcp_sock *tp) |
1128 | { | |
00db4124 | 1129 | if (tp->fastopen_req) { |
cf60af03 YC |
1130 | kfree(tp->fastopen_req); |
1131 | tp->fastopen_req = NULL; | |
1132 | } | |
1133 | } | |
1134 | ||
f5ddcbbb ED |
1135 | static int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, |
1136 | int *copied, size_t size) | |
cf60af03 YC |
1137 | { |
1138 | struct tcp_sock *tp = tcp_sk(sk); | |
19f6d3f3 | 1139 | struct inet_sock *inet = inet_sk(sk); |
ba615f67 | 1140 | struct sockaddr *uaddr = msg->msg_name; |
cf60af03 YC |
1141 | int err, flags; |
1142 | ||
e1cfcbe8 | 1143 | if (!(sock_net(sk)->ipv4.sysctl_tcp_fastopen & TFO_CLIENT_ENABLE) || |
ba615f67 WW |
1144 | (uaddr && msg->msg_namelen >= sizeof(uaddr->sa_family) && |
1145 | uaddr->sa_family == AF_UNSPEC)) | |
cf60af03 | 1146 | return -EOPNOTSUPP; |
00db4124 | 1147 | if (tp->fastopen_req) |
cf60af03 YC |
1148 | return -EALREADY; /* Another Fast Open is in progress */ |
1149 | ||
1150 | tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request), | |
1151 | sk->sk_allocation); | |
51456b29 | 1152 | if (unlikely(!tp->fastopen_req)) |
cf60af03 YC |
1153 | return -ENOBUFS; |
1154 | tp->fastopen_req->data = msg; | |
f5ddcbbb | 1155 | tp->fastopen_req->size = size; |
cf60af03 | 1156 | |
19f6d3f3 WW |
1157 | if (inet->defer_connect) { |
1158 | err = tcp_connect(sk); | |
1159 | /* Same failure procedure as in tcp_v4/6_connect */ | |
1160 | if (err) { | |
1161 | tcp_set_state(sk, TCP_CLOSE); | |
1162 | inet->inet_dport = 0; | |
1163 | sk->sk_route_caps = 0; | |
1164 | } | |
1165 | } | |
cf60af03 | 1166 | flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0; |
ba615f67 | 1167 | err = __inet_stream_connect(sk->sk_socket, uaddr, |
3979ad7e | 1168 | msg->msg_namelen, flags, 1); |
7db92362 WW |
1169 | /* fastopen_req could already be freed in __inet_stream_connect |
1170 | * if the connection times out or gets rst | |
1171 | */ | |
1172 | if (tp->fastopen_req) { | |
1173 | *copied = tp->fastopen_req->copied; | |
1174 | tcp_free_fastopen_req(tp); | |
1175 | inet->defer_connect = 0; | |
1176 | } | |
cf60af03 YC |
1177 | return err; |
1178 | } | |
1179 | ||
306b13eb | 1180 | int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size) |
1da177e4 | 1181 | { |
1da177e4 | 1182 | struct tcp_sock *tp = tcp_sk(sk); |
f214f915 | 1183 | struct ubuf_info *uarg = NULL; |
1da177e4 | 1184 | struct sk_buff *skb; |
c14ac945 | 1185 | struct sockcm_cookie sockc; |
57be5bda AV |
1186 | int flags, err, copied = 0; |
1187 | int mss_now = 0, size_goal, copied_syn = 0; | |
d4011239 | 1188 | bool process_backlog = false; |
02583ade | 1189 | bool sg, zc = false; |
1da177e4 LT |
1190 | long timeo; |
1191 | ||
1da177e4 | 1192 | flags = msg->msg_flags; |
f214f915 WB |
1193 | |
1194 | if (flags & MSG_ZEROCOPY && size) { | |
1195 | if (sk->sk_state != TCP_ESTABLISHED) { | |
1196 | err = -EINVAL; | |
1197 | goto out_err; | |
1198 | } | |
1199 | ||
75c119af | 1200 | skb = tcp_write_queue_tail(sk); |
f214f915 WB |
1201 | uarg = sock_zerocopy_realloc(sk, size, skb_zcopy(skb)); |
1202 | if (!uarg) { | |
1203 | err = -ENOBUFS; | |
1204 | goto out_err; | |
1205 | } | |
1206 | ||
02583ade WB |
1207 | zc = sk_check_csum_caps(sk) && sk->sk_route_caps & NETIF_F_SG; |
1208 | if (!zc) | |
f214f915 WB |
1209 | uarg->zerocopy = 0; |
1210 | } | |
1211 | ||
19f6d3f3 | 1212 | if (unlikely(flags & MSG_FASTOPEN || inet_sk(sk)->defer_connect)) { |
f5ddcbbb | 1213 | err = tcp_sendmsg_fastopen(sk, msg, &copied_syn, size); |
cf60af03 YC |
1214 | if (err == -EINPROGRESS && copied_syn > 0) |
1215 | goto out; | |
1216 | else if (err) | |
1217 | goto out_err; | |
cf60af03 YC |
1218 | } |
1219 | ||
1da177e4 LT |
1220 | timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); |
1221 | ||
d7722e85 SHY |
1222 | tcp_rate_check_app_limited(sk); /* is sending application-limited? */ |
1223 | ||
8336886f JC |
1224 | /* Wait for a connection to finish. One exception is TCP Fast Open |
1225 | * (passive side) where data is allowed to be sent before a connection | |
1226 | * is fully established. | |
1227 | */ | |
1228 | if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) && | |
1229 | !tcp_passive_fastopen(sk)) { | |
686a5624 YM |
1230 | err = sk_stream_wait_connect(sk, &timeo); |
1231 | if (err != 0) | |
cf60af03 | 1232 | goto do_error; |
8336886f | 1233 | } |
1da177e4 | 1234 | |
c0e88ff0 PE |
1235 | if (unlikely(tp->repair)) { |
1236 | if (tp->repair_queue == TCP_RECV_QUEUE) { | |
1237 | copied = tcp_send_rcvq(sk, msg, size); | |
5924f17a | 1238 | goto out_nopush; |
c0e88ff0 PE |
1239 | } |
1240 | ||
1241 | err = -EINVAL; | |
1242 | if (tp->repair_queue == TCP_NO_QUEUE) | |
1243 | goto out_err; | |
1244 | ||
1245 | /* 'common' sending to sendq */ | |
1246 | } | |
1247 | ||
c14ac945 SHY |
1248 | sockc.tsflags = sk->sk_tsflags; |
1249 | if (msg->msg_controllen) { | |
1250 | err = sock_cmsg_send(sk, msg, &sockc); | |
1251 | if (unlikely(err)) { | |
1252 | err = -EINVAL; | |
1253 | goto out_err; | |
1254 | } | |
1255 | } | |
1256 | ||
1da177e4 | 1257 | /* This should be in poll */ |
9cd3e072 | 1258 | sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 | 1259 | |
1da177e4 | 1260 | /* Ok commence sending. */ |
1da177e4 LT |
1261 | copied = 0; |
1262 | ||
d41a69f1 ED |
1263 | restart: |
1264 | mss_now = tcp_send_mss(sk, &size_goal, flags); | |
1265 | ||
1da177e4 LT |
1266 | err = -EPIPE; |
1267 | if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) | |
79d8665b | 1268 | goto do_error; |
1da177e4 | 1269 | |
690e99c4 | 1270 | sg = !!(sk->sk_route_caps & NETIF_F_SG); |
def87cf4 | 1271 | |
01e97e65 | 1272 | while (msg_data_left(msg)) { |
57be5bda AV |
1273 | int copy = 0; |
1274 | int max = size_goal; | |
1da177e4 | 1275 | |
57be5bda | 1276 | skb = tcp_write_queue_tail(sk); |
75c119af | 1277 | if (skb) { |
57be5bda AV |
1278 | if (skb->ip_summed == CHECKSUM_NONE) |
1279 | max = mss_now; | |
1280 | copy = max - skb->len; | |
cf60af03 | 1281 | } |
1da177e4 | 1282 | |
c134ecb8 | 1283 | if (copy <= 0 || !tcp_skb_can_collapse_to(skb)) { |
3613b3db ED |
1284 | bool first_skb; |
1285 | ||
1da177e4 | 1286 | new_segment: |
57be5bda AV |
1287 | /* Allocate new segment. If the interface is SG, |
1288 | * allocate skb fitting to single page. | |
1289 | */ | |
1290 | if (!sk_stream_memory_free(sk)) | |
1291 | goto wait_for_sndbuf; | |
1da177e4 | 1292 | |
d4011239 ED |
1293 | if (process_backlog && sk_flush_backlog(sk)) { |
1294 | process_backlog = false; | |
d41a69f1 | 1295 | goto restart; |
d4011239 | 1296 | } |
75c119af | 1297 | first_skb = tcp_rtx_and_write_queues_empty(sk); |
57be5bda | 1298 | skb = sk_stream_alloc_skb(sk, |
3613b3db | 1299 | select_size(sk, sg, first_skb), |
eb934478 | 1300 | sk->sk_allocation, |
3613b3db | 1301 | first_skb); |
57be5bda AV |
1302 | if (!skb) |
1303 | goto wait_for_memory; | |
1da177e4 | 1304 | |
d4011239 | 1305 | process_backlog = true; |
57be5bda AV |
1306 | /* |
1307 | * Check whether we can use HW checksum. | |
1308 | */ | |
9a49850d | 1309 | if (sk_check_csum_caps(sk)) |
57be5bda | 1310 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 | 1311 | |
57be5bda AV |
1312 | skb_entail(sk, skb); |
1313 | copy = size_goal; | |
1314 | max = size_goal; | |
9d186cac | 1315 | |
57be5bda AV |
1316 | /* All packets are restored as if they have |
1317 | * already been sent. skb_mstamp isn't set to | |
1318 | * avoid wrong rtt estimation. | |
1319 | */ | |
1320 | if (tp->repair) | |
1321 | TCP_SKB_CB(skb)->sacked |= TCPCB_REPAIRED; | |
1322 | } | |
1da177e4 | 1323 | |
57be5bda | 1324 | /* Try to append data to the end of skb. */ |
01e97e65 AV |
1325 | if (copy > msg_data_left(msg)) |
1326 | copy = msg_data_left(msg); | |
57be5bda AV |
1327 | |
1328 | /* Where to copy to? */ | |
02583ade | 1329 | if (skb_availroom(skb) > 0 && !zc) { |
57be5bda AV |
1330 | /* We have some space in skb head. Superb! */ |
1331 | copy = min_t(int, copy, skb_availroom(skb)); | |
1332 | err = skb_add_data_nocache(sk, skb, &msg->msg_iter, copy); | |
1333 | if (err) | |
1334 | goto do_fault; | |
02583ade | 1335 | } else if (!zc) { |
57be5bda AV |
1336 | bool merge = true; |
1337 | int i = skb_shinfo(skb)->nr_frags; | |
1338 | struct page_frag *pfrag = sk_page_frag(sk); | |
1339 | ||
1340 | if (!sk_page_frag_refill(sk, pfrag)) | |
1341 | goto wait_for_memory; | |
ef015786 | 1342 | |
57be5bda AV |
1343 | if (!skb_can_coalesce(skb, i, pfrag->page, |
1344 | pfrag->offset)) { | |
ac9e70b1 | 1345 | if (i >= sysctl_max_skb_frags || !sg) { |
57be5bda AV |
1346 | tcp_mark_push(tp, skb); |
1347 | goto new_segment; | |
1da177e4 | 1348 | } |
57be5bda | 1349 | merge = false; |
1da177e4 LT |
1350 | } |
1351 | ||
57be5bda AV |
1352 | copy = min_t(int, copy, pfrag->size - pfrag->offset); |
1353 | ||
1354 | if (!sk_wmem_schedule(sk, copy)) | |
1355 | goto wait_for_memory; | |
1da177e4 | 1356 | |
57be5bda AV |
1357 | err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb, |
1358 | pfrag->page, | |
1359 | pfrag->offset, | |
1360 | copy); | |
1361 | if (err) | |
1362 | goto do_error; | |
1da177e4 | 1363 | |
57be5bda AV |
1364 | /* Update the skb. */ |
1365 | if (merge) { | |
1366 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); | |
1367 | } else { | |
1368 | skb_fill_page_desc(skb, i, pfrag->page, | |
1369 | pfrag->offset, copy); | |
4e33e346 | 1370 | page_ref_inc(pfrag->page); |
4ed2d765 | 1371 | } |
57be5bda | 1372 | pfrag->offset += copy; |
f214f915 WB |
1373 | } else { |
1374 | err = skb_zerocopy_iter_stream(sk, skb, msg, copy, uarg); | |
111856c7 WB |
1375 | if (err == -EMSGSIZE || err == -EEXIST) { |
1376 | tcp_mark_push(tp, skb); | |
f214f915 | 1377 | goto new_segment; |
111856c7 | 1378 | } |
f214f915 WB |
1379 | if (err < 0) |
1380 | goto do_error; | |
1381 | copy = err; | |
57be5bda AV |
1382 | } |
1383 | ||
1384 | if (!copied) | |
1385 | TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH; | |
1386 | ||
1387 | tp->write_seq += copy; | |
1388 | TCP_SKB_CB(skb)->end_seq += copy; | |
1389 | tcp_skb_pcount_set(skb, 0); | |
1da177e4 | 1390 | |
57be5bda | 1391 | copied += copy; |
01e97e65 | 1392 | if (!msg_data_left(msg)) { |
c134ecb8 MKL |
1393 | if (unlikely(flags & MSG_EOR)) |
1394 | TCP_SKB_CB(skb)->eor = 1; | |
57be5bda AV |
1395 | goto out; |
1396 | } | |
1da177e4 | 1397 | |
57be5bda | 1398 | if (skb->len < max || (flags & MSG_OOB) || unlikely(tp->repair)) |
1da177e4 LT |
1399 | continue; |
1400 | ||
57be5bda AV |
1401 | if (forced_push(tp)) { |
1402 | tcp_mark_push(tp, skb); | |
1403 | __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH); | |
1404 | } else if (skb == tcp_send_head(sk)) | |
1405 | tcp_push_one(sk, mss_now); | |
1406 | continue; | |
1407 | ||
1da177e4 | 1408 | wait_for_sndbuf: |
57be5bda | 1409 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
1da177e4 | 1410 | wait_for_memory: |
57be5bda AV |
1411 | if (copied) |
1412 | tcp_push(sk, flags & ~MSG_MORE, mss_now, | |
1413 | TCP_NAGLE_PUSH, size_goal); | |
1da177e4 | 1414 | |
686a5624 YM |
1415 | err = sk_stream_wait_memory(sk, &timeo); |
1416 | if (err != 0) | |
57be5bda | 1417 | goto do_error; |
1da177e4 | 1418 | |
57be5bda | 1419 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
1420 | } |
1421 | ||
1422 | out: | |
ad02c4f5 | 1423 | if (copied) { |
4e8cc228 | 1424 | tcp_tx_timestamp(sk, sockc.tsflags); |
f54b3111 | 1425 | tcp_push(sk, flags, mss_now, tp->nonagle, size_goal); |
ad02c4f5 | 1426 | } |
5924f17a | 1427 | out_nopush: |
f214f915 | 1428 | sock_zerocopy_put(uarg); |
cf60af03 | 1429 | return copied + copied_syn; |
1da177e4 LT |
1430 | |
1431 | do_fault: | |
1432 | if (!skb->len) { | |
fe067e8a DM |
1433 | tcp_unlink_write_queue(skb, sk); |
1434 | /* It is the one place in all of TCP, except connection | |
1435 | * reset, where we can be unlinking the send_head. | |
1436 | */ | |
1437 | tcp_check_send_head(sk, skb); | |
3ab224be | 1438 | sk_wmem_free_skb(sk, skb); |
1da177e4 LT |
1439 | } |
1440 | ||
1441 | do_error: | |
cf60af03 | 1442 | if (copied + copied_syn) |
1da177e4 LT |
1443 | goto out; |
1444 | out_err: | |
f214f915 | 1445 | sock_zerocopy_put_abort(uarg); |
1da177e4 | 1446 | err = sk_stream_error(sk, flags, err); |
ce5ec440 | 1447 | /* make sure we wake any epoll edge trigger waiter */ |
b0f71bd3 FY |
1448 | if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && |
1449 | err == -EAGAIN)) { | |
ce5ec440 | 1450 | sk->sk_write_space(sk); |
b0f71bd3 FY |
1451 | tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED); |
1452 | } | |
1da177e4 LT |
1453 | return err; |
1454 | } | |
774c4673 | 1455 | EXPORT_SYMBOL_GPL(tcp_sendmsg_locked); |
306b13eb TH |
1456 | |
1457 | int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size) | |
1458 | { | |
1459 | int ret; | |
1460 | ||
1461 | lock_sock(sk); | |
1462 | ret = tcp_sendmsg_locked(sk, msg, size); | |
1463 | release_sock(sk); | |
1464 | ||
1465 | return ret; | |
1466 | } | |
4bc2f18b | 1467 | EXPORT_SYMBOL(tcp_sendmsg); |
1da177e4 LT |
1468 | |
1469 | /* | |
1470 | * Handle reading urgent data. BSD has very simple semantics for | |
1471 | * this, no blocking and very strange errors 8) | |
1472 | */ | |
1473 | ||
377f0a08 | 1474 | static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags) |
1da177e4 LT |
1475 | { |
1476 | struct tcp_sock *tp = tcp_sk(sk); | |
1477 | ||
1478 | /* No URG data to read. */ | |
1479 | if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data || | |
1480 | tp->urg_data == TCP_URG_READ) | |
1481 | return -EINVAL; /* Yes this is right ! */ | |
1482 | ||
1483 | if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE)) | |
1484 | return -ENOTCONN; | |
1485 | ||
1486 | if (tp->urg_data & TCP_URG_VALID) { | |
1487 | int err = 0; | |
1488 | char c = tp->urg_data; | |
1489 | ||
1490 | if (!(flags & MSG_PEEK)) | |
1491 | tp->urg_data = TCP_URG_READ; | |
1492 | ||
1493 | /* Read urgent data. */ | |
1494 | msg->msg_flags |= MSG_OOB; | |
1495 | ||
1496 | if (len > 0) { | |
1497 | if (!(flags & MSG_TRUNC)) | |
7eab8d9e | 1498 | err = memcpy_to_msg(msg, &c, 1); |
1da177e4 LT |
1499 | len = 1; |
1500 | } else | |
1501 | msg->msg_flags |= MSG_TRUNC; | |
1502 | ||
1503 | return err ? -EFAULT : len; | |
1504 | } | |
1505 | ||
1506 | if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN)) | |
1507 | return 0; | |
1508 | ||
1509 | /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and | |
1510 | * the available implementations agree in this case: | |
1511 | * this call should never block, independent of the | |
1512 | * blocking state of the socket. | |
1513 | * Mike <pall@rz.uni-karlsruhe.de> | |
1514 | */ | |
1515 | return -EAGAIN; | |
1516 | } | |
1517 | ||
c0e88ff0 PE |
1518 | static int tcp_peek_sndq(struct sock *sk, struct msghdr *msg, int len) |
1519 | { | |
1520 | struct sk_buff *skb; | |
1521 | int copied = 0, err = 0; | |
1522 | ||
1523 | /* XXX -- need to support SO_PEEK_OFF */ | |
1524 | ||
75c119af ED |
1525 | skb_rbtree_walk(skb, &sk->tcp_rtx_queue) { |
1526 | err = skb_copy_datagram_msg(skb, 0, msg, skb->len); | |
1527 | if (err) | |
1528 | return err; | |
1529 | copied += skb->len; | |
1530 | } | |
1531 | ||
c0e88ff0 | 1532 | skb_queue_walk(&sk->sk_write_queue, skb) { |
51f3d02b | 1533 | err = skb_copy_datagram_msg(skb, 0, msg, skb->len); |
c0e88ff0 PE |
1534 | if (err) |
1535 | break; | |
1536 | ||
1537 | copied += skb->len; | |
1538 | } | |
1539 | ||
1540 | return err ?: copied; | |
1541 | } | |
1542 | ||
1da177e4 LT |
1543 | /* Clean up the receive buffer for full frames taken by the user, |
1544 | * then send an ACK if necessary. COPIED is the number of bytes | |
1545 | * tcp_recvmsg has given to the user so far, it speeds up the | |
1546 | * calculation of whether or not we must ACK for the sake of | |
1547 | * a window update. | |
1548 | */ | |
3f334078 | 1549 | static void tcp_cleanup_rbuf(struct sock *sk, int copied) |
1da177e4 LT |
1550 | { |
1551 | struct tcp_sock *tp = tcp_sk(sk); | |
a2a385d6 | 1552 | bool time_to_ack = false; |
1da177e4 | 1553 | |
1da177e4 LT |
1554 | struct sk_buff *skb = skb_peek(&sk->sk_receive_queue); |
1555 | ||
d792c100 | 1556 | WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq), |
2af6fd8b | 1557 | "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n", |
d792c100 | 1558 | tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt); |
1da177e4 | 1559 | |
463c84b9 ACM |
1560 | if (inet_csk_ack_scheduled(sk)) { |
1561 | const struct inet_connection_sock *icsk = inet_csk(sk); | |
1da177e4 LT |
1562 | /* Delayed ACKs frequently hit locked sockets during bulk |
1563 | * receive. */ | |
463c84b9 | 1564 | if (icsk->icsk_ack.blocked || |
1da177e4 | 1565 | /* Once-per-two-segments ACK was not sent by tcp_input.c */ |
463c84b9 | 1566 | tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss || |
1da177e4 LT |
1567 | /* |
1568 | * If this read emptied read buffer, we send ACK, if | |
1569 | * connection is not bidirectional, user drained | |
1570 | * receive buffer and there was a small segment | |
1571 | * in queue. | |
1572 | */ | |
1ef9696c AK |
1573 | (copied > 0 && |
1574 | ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) || | |
1575 | ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) && | |
1576 | !icsk->icsk_ack.pingpong)) && | |
1577 | !atomic_read(&sk->sk_rmem_alloc))) | |
a2a385d6 | 1578 | time_to_ack = true; |
1da177e4 LT |
1579 | } |
1580 | ||
1581 | /* We send an ACK if we can now advertise a non-zero window | |
1582 | * which has been raised "significantly". | |
1583 | * | |
1584 | * Even if window raised up to infinity, do not send window open ACK | |
1585 | * in states, where we will not receive more. It is useless. | |
1586 | */ | |
1587 | if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) { | |
1588 | __u32 rcv_window_now = tcp_receive_window(tp); | |
1589 | ||
1590 | /* Optimize, __tcp_select_window() is not cheap. */ | |
1591 | if (2*rcv_window_now <= tp->window_clamp) { | |
1592 | __u32 new_window = __tcp_select_window(sk); | |
1593 | ||
1594 | /* Send ACK now, if this read freed lots of space | |
1595 | * in our buffer. Certainly, new_window is new window. | |
1596 | * We can advertise it now, if it is not less than current one. | |
1597 | * "Lots" means "at least twice" here. | |
1598 | */ | |
1599 | if (new_window && new_window >= 2 * rcv_window_now) | |
a2a385d6 | 1600 | time_to_ack = true; |
1da177e4 LT |
1601 | } |
1602 | } | |
1603 | if (time_to_ack) | |
1604 | tcp_send_ack(sk); | |
1605 | } | |
1606 | ||
f26845b4 | 1607 | static struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off) |
1da177e4 LT |
1608 | { |
1609 | struct sk_buff *skb; | |
1610 | u32 offset; | |
1611 | ||
f26845b4 | 1612 | while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) { |
1da177e4 | 1613 | offset = seq - TCP_SKB_CB(skb)->seq; |
9d691539 ED |
1614 | if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { |
1615 | pr_err_once("%s: found a SYN, please report !\n", __func__); | |
1da177e4 | 1616 | offset--; |
9d691539 | 1617 | } |
e11ecddf | 1618 | if (offset < skb->len || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) { |
1da177e4 LT |
1619 | *off = offset; |
1620 | return skb; | |
1621 | } | |
f26845b4 ED |
1622 | /* This looks weird, but this can happen if TCP collapsing |
1623 | * splitted a fat GRO packet, while we released socket lock | |
1624 | * in skb_splice_bits() | |
1625 | */ | |
7bced397 | 1626 | sk_eat_skb(sk, skb); |
1da177e4 LT |
1627 | } |
1628 | return NULL; | |
1629 | } | |
1630 | ||
1631 | /* | |
1632 | * This routine provides an alternative to tcp_recvmsg() for routines | |
1633 | * that would like to handle copying from skbuffs directly in 'sendfile' | |
1634 | * fashion. | |
1635 | * Note: | |
1636 | * - It is assumed that the socket was locked by the caller. | |
1637 | * - The routine does not block. | |
1638 | * - At present, there is no support for reading OOB data | |
1639 | * or for 'peeking' the socket using this routine | |
1640 | * (although both would be easy to implement). | |
1641 | */ | |
1642 | int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, | |
1643 | sk_read_actor_t recv_actor) | |
1644 | { | |
1645 | struct sk_buff *skb; | |
1646 | struct tcp_sock *tp = tcp_sk(sk); | |
1647 | u32 seq = tp->copied_seq; | |
1648 | u32 offset; | |
1649 | int copied = 0; | |
1650 | ||
1651 | if (sk->sk_state == TCP_LISTEN) | |
1652 | return -ENOTCONN; | |
1653 | while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) { | |
1654 | if (offset < skb->len) { | |
374e7b59 OP |
1655 | int used; |
1656 | size_t len; | |
1da177e4 LT |
1657 | |
1658 | len = skb->len - offset; | |
1659 | /* Stop reading if we hit a patch of urgent data */ | |
1660 | if (tp->urg_data) { | |
1661 | u32 urg_offset = tp->urg_seq - seq; | |
1662 | if (urg_offset < len) | |
1663 | len = urg_offset; | |
1664 | if (!len) | |
1665 | break; | |
1666 | } | |
1667 | used = recv_actor(desc, skb, offset, len); | |
ff905b1e | 1668 | if (used <= 0) { |
ddb61a57 JA |
1669 | if (!copied) |
1670 | copied = used; | |
1671 | break; | |
1672 | } else if (used <= len) { | |
1da177e4 LT |
1673 | seq += used; |
1674 | copied += used; | |
1675 | offset += used; | |
1676 | } | |
02275a2e | 1677 | /* If recv_actor drops the lock (e.g. TCP splice |
293ad604 OP |
1678 | * receive) the skb pointer might be invalid when |
1679 | * getting here: tcp_collapse might have deleted it | |
1680 | * while aggregating skbs from the socket queue. | |
1681 | */ | |
02275a2e WT |
1682 | skb = tcp_recv_skb(sk, seq - 1, &offset); |
1683 | if (!skb) | |
1da177e4 | 1684 | break; |
02275a2e WT |
1685 | /* TCP coalescing might have appended data to the skb. |
1686 | * Try to splice more frags | |
1687 | */ | |
1688 | if (offset + 1 != skb->len) | |
1689 | continue; | |
1da177e4 | 1690 | } |
e11ecddf | 1691 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) { |
7bced397 | 1692 | sk_eat_skb(sk, skb); |
1da177e4 LT |
1693 | ++seq; |
1694 | break; | |
1695 | } | |
7bced397 | 1696 | sk_eat_skb(sk, skb); |
1da177e4 LT |
1697 | if (!desc->count) |
1698 | break; | |
baff42ab | 1699 | tp->copied_seq = seq; |
1da177e4 LT |
1700 | } |
1701 | tp->copied_seq = seq; | |
1702 | ||
1703 | tcp_rcv_space_adjust(sk); | |
1704 | ||
1705 | /* Clean up data we have read: This will do ACK frames. */ | |
f26845b4 ED |
1706 | if (copied > 0) { |
1707 | tcp_recv_skb(sk, seq, &offset); | |
0e4b4992 | 1708 | tcp_cleanup_rbuf(sk, copied); |
f26845b4 | 1709 | } |
1da177e4 LT |
1710 | return copied; |
1711 | } | |
4bc2f18b | 1712 | EXPORT_SYMBOL(tcp_read_sock); |
1da177e4 | 1713 | |
32035585 TH |
1714 | int tcp_peek_len(struct socket *sock) |
1715 | { | |
1716 | return tcp_inq(sock->sk); | |
1717 | } | |
1718 | EXPORT_SYMBOL(tcp_peek_len); | |
1719 | ||
98aaa913 MM |
1720 | static void tcp_update_recv_tstamps(struct sk_buff *skb, |
1721 | struct scm_timestamping *tss) | |
1722 | { | |
1723 | if (skb->tstamp) | |
1724 | tss->ts[0] = ktime_to_timespec(skb->tstamp); | |
1725 | else | |
1726 | tss->ts[0] = (struct timespec) {0}; | |
1727 | ||
1728 | if (skb_hwtstamps(skb)->hwtstamp) | |
1729 | tss->ts[2] = ktime_to_timespec(skb_hwtstamps(skb)->hwtstamp); | |
1730 | else | |
1731 | tss->ts[2] = (struct timespec) {0}; | |
1732 | } | |
1733 | ||
1734 | /* Similar to __sock_recv_timestamp, but does not require an skb */ | |
1735 | void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, | |
1736 | struct scm_timestamping *tss) | |
1737 | { | |
1738 | struct timeval tv; | |
1739 | bool has_timestamping = false; | |
1740 | ||
1741 | if (tss->ts[0].tv_sec || tss->ts[0].tv_nsec) { | |
1742 | if (sock_flag(sk, SOCK_RCVTSTAMP)) { | |
1743 | if (sock_flag(sk, SOCK_RCVTSTAMPNS)) { | |
1744 | put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPNS, | |
1745 | sizeof(tss->ts[0]), &tss->ts[0]); | |
1746 | } else { | |
1747 | tv.tv_sec = tss->ts[0].tv_sec; | |
1748 | tv.tv_usec = tss->ts[0].tv_nsec / 1000; | |
1749 | ||
1750 | put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP, | |
1751 | sizeof(tv), &tv); | |
1752 | } | |
1753 | } | |
1754 | ||
1755 | if (sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) | |
1756 | has_timestamping = true; | |
1757 | else | |
1758 | tss->ts[0] = (struct timespec) {0}; | |
1759 | } | |
1760 | ||
1761 | if (tss->ts[2].tv_sec || tss->ts[2].tv_nsec) { | |
1762 | if (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) | |
1763 | has_timestamping = true; | |
1764 | else | |
1765 | tss->ts[2] = (struct timespec) {0}; | |
1766 | } | |
1767 | ||
1768 | if (has_timestamping) { | |
1769 | tss->ts[1] = (struct timespec) {0}; | |
1770 | put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING, | |
1771 | sizeof(*tss), tss); | |
1772 | } | |
1773 | } | |
1774 | ||
1da177e4 LT |
1775 | /* |
1776 | * This routine copies from a sock struct into the user buffer. | |
1777 | * | |
1778 | * Technical note: in 2.3 we work on _locked_ socket, so that | |
1779 | * tricks with *seq access order and skb->users are not required. | |
1780 | * Probably, code can be easily improved even more. | |
1781 | */ | |
1782 | ||
1b784140 YX |
1783 | int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, |
1784 | int flags, int *addr_len) | |
1da177e4 LT |
1785 | { |
1786 | struct tcp_sock *tp = tcp_sk(sk); | |
1787 | int copied = 0; | |
1788 | u32 peek_seq; | |
1789 | u32 *seq; | |
1790 | unsigned long used; | |
1791 | int err; | |
1792 | int target; /* Read at least this many bytes */ | |
1793 | long timeo; | |
dfbafc99 | 1794 | struct sk_buff *skb, *last; |
77527313 | 1795 | u32 urg_hole = 0; |
98aaa913 MM |
1796 | struct scm_timestamping tss; |
1797 | bool has_tss = false; | |
1da177e4 | 1798 | |
4ed2d765 | 1799 | if (unlikely(flags & MSG_ERRQUEUE)) |
f4713a3d | 1800 | return inet_recv_error(sk, msg, len, addr_len); |
4ed2d765 | 1801 | |
cbf55001 ET |
1802 | if (sk_can_busy_loop(sk) && skb_queue_empty(&sk->sk_receive_queue) && |
1803 | (sk->sk_state == TCP_ESTABLISHED)) | |
1804 | sk_busy_loop(sk, nonblock); | |
d30e383b | 1805 | |
1da177e4 LT |
1806 | lock_sock(sk); |
1807 | ||
1da177e4 LT |
1808 | err = -ENOTCONN; |
1809 | if (sk->sk_state == TCP_LISTEN) | |
1810 | goto out; | |
1811 | ||
1812 | timeo = sock_rcvtimeo(sk, nonblock); | |
1813 | ||
1814 | /* Urgent data needs to be handled specially. */ | |
1815 | if (flags & MSG_OOB) | |
1816 | goto recv_urg; | |
1817 | ||
c0e88ff0 PE |
1818 | if (unlikely(tp->repair)) { |
1819 | err = -EPERM; | |
1820 | if (!(flags & MSG_PEEK)) | |
1821 | goto out; | |
1822 | ||
1823 | if (tp->repair_queue == TCP_SEND_QUEUE) | |
1824 | goto recv_sndq; | |
1825 | ||
1826 | err = -EINVAL; | |
1827 | if (tp->repair_queue == TCP_NO_QUEUE) | |
1828 | goto out; | |
1829 | ||
1830 | /* 'common' recv queue MSG_PEEK-ing */ | |
1831 | } | |
1832 | ||
1da177e4 LT |
1833 | seq = &tp->copied_seq; |
1834 | if (flags & MSG_PEEK) { | |
1835 | peek_seq = tp->copied_seq; | |
1836 | seq = &peek_seq; | |
1837 | } | |
1838 | ||
1839 | target = sock_rcvlowat(sk, flags & MSG_WAITALL, len); | |
1840 | ||
1841 | do { | |
1da177e4 LT |
1842 | u32 offset; |
1843 | ||
1844 | /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */ | |
1845 | if (tp->urg_data && tp->urg_seq == *seq) { | |
1846 | if (copied) | |
1847 | break; | |
1848 | if (signal_pending(current)) { | |
1849 | copied = timeo ? sock_intr_errno(timeo) : -EAGAIN; | |
1850 | break; | |
1851 | } | |
1852 | } | |
1853 | ||
1854 | /* Next get a buffer. */ | |
1855 | ||
dfbafc99 | 1856 | last = skb_peek_tail(&sk->sk_receive_queue); |
91521944 | 1857 | skb_queue_walk(&sk->sk_receive_queue, skb) { |
dfbafc99 | 1858 | last = skb; |
1da177e4 LT |
1859 | /* Now that we have two receive queues this |
1860 | * shouldn't happen. | |
1861 | */ | |
d792c100 | 1862 | if (WARN(before(*seq, TCP_SKB_CB(skb)->seq), |
2af6fd8b JP |
1863 | "recvmsg bug: copied %X seq %X rcvnxt %X fl %X\n", |
1864 | *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, | |
1865 | flags)) | |
1da177e4 | 1866 | break; |
d792c100 | 1867 | |
1da177e4 | 1868 | offset = *seq - TCP_SKB_CB(skb)->seq; |
9d691539 ED |
1869 | if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { |
1870 | pr_err_once("%s: found a SYN, please report !\n", __func__); | |
1da177e4 | 1871 | offset--; |
9d691539 | 1872 | } |
1da177e4 LT |
1873 | if (offset < skb->len) |
1874 | goto found_ok_skb; | |
e11ecddf | 1875 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) |
1da177e4 | 1876 | goto found_fin_ok; |
2af6fd8b JP |
1877 | WARN(!(flags & MSG_PEEK), |
1878 | "recvmsg bug 2: copied %X seq %X rcvnxt %X fl %X\n", | |
1879 | *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags); | |
91521944 | 1880 | } |
1da177e4 LT |
1881 | |
1882 | /* Well, if we have backlog, try to process it now yet. */ | |
1883 | ||
1884 | if (copied >= target && !sk->sk_backlog.tail) | |
1885 | break; | |
1886 | ||
1887 | if (copied) { | |
1888 | if (sk->sk_err || | |
1889 | sk->sk_state == TCP_CLOSE || | |
1890 | (sk->sk_shutdown & RCV_SHUTDOWN) || | |
1891 | !timeo || | |
518a09ef | 1892 | signal_pending(current)) |
1da177e4 LT |
1893 | break; |
1894 | } else { | |
1895 | if (sock_flag(sk, SOCK_DONE)) | |
1896 | break; | |
1897 | ||
1898 | if (sk->sk_err) { | |
1899 | copied = sock_error(sk); | |
1900 | break; | |
1901 | } | |
1902 | ||
1903 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
1904 | break; | |
1905 | ||
1906 | if (sk->sk_state == TCP_CLOSE) { | |
1907 | if (!sock_flag(sk, SOCK_DONE)) { | |
1908 | /* This occurs when user tries to read | |
1909 | * from never connected socket. | |
1910 | */ | |
1911 | copied = -ENOTCONN; | |
1912 | break; | |
1913 | } | |
1914 | break; | |
1915 | } | |
1916 | ||
1917 | if (!timeo) { | |
1918 | copied = -EAGAIN; | |
1919 | break; | |
1920 | } | |
1921 | ||
1922 | if (signal_pending(current)) { | |
1923 | copied = sock_intr_errno(timeo); | |
1924 | break; | |
1925 | } | |
1926 | } | |
1927 | ||
0e4b4992 | 1928 | tcp_cleanup_rbuf(sk, copied); |
1da177e4 | 1929 | |
1da177e4 LT |
1930 | if (copied >= target) { |
1931 | /* Do not sleep, just process backlog. */ | |
1932 | release_sock(sk); | |
1933 | lock_sock(sk); | |
dfbafc99 SD |
1934 | } else { |
1935 | sk_wait_data(sk, &timeo, last); | |
1936 | } | |
1da177e4 | 1937 | |
77527313 IJ |
1938 | if ((flags & MSG_PEEK) && |
1939 | (peek_seq - copied - urg_hole != tp->copied_seq)) { | |
e87cc472 JP |
1940 | net_dbg_ratelimited("TCP(%s:%d): Application bug, race in MSG_PEEK\n", |
1941 | current->comm, | |
1942 | task_pid_nr(current)); | |
1da177e4 LT |
1943 | peek_seq = tp->copied_seq; |
1944 | } | |
1945 | continue; | |
1946 | ||
1947 | found_ok_skb: | |
1948 | /* Ok so how much can we use? */ | |
1949 | used = skb->len - offset; | |
1950 | if (len < used) | |
1951 | used = len; | |
1952 | ||
1953 | /* Do we have urgent data here? */ | |
1954 | if (tp->urg_data) { | |
1955 | u32 urg_offset = tp->urg_seq - *seq; | |
1956 | if (urg_offset < used) { | |
1957 | if (!urg_offset) { | |
1958 | if (!sock_flag(sk, SOCK_URGINLINE)) { | |
1959 | ++*seq; | |
77527313 | 1960 | urg_hole++; |
1da177e4 LT |
1961 | offset++; |
1962 | used--; | |
1963 | if (!used) | |
1964 | goto skip_copy; | |
1965 | } | |
1966 | } else | |
1967 | used = urg_offset; | |
1968 | } | |
1969 | } | |
1970 | ||
1971 | if (!(flags & MSG_TRUNC)) { | |
51f3d02b | 1972 | err = skb_copy_datagram_msg(skb, offset, msg, used); |
7bced397 DW |
1973 | if (err) { |
1974 | /* Exception. Bailout! */ | |
1975 | if (!copied) | |
1976 | copied = -EFAULT; | |
1977 | break; | |
1da177e4 LT |
1978 | } |
1979 | } | |
1980 | ||
1981 | *seq += used; | |
1982 | copied += used; | |
1983 | len -= used; | |
1984 | ||
1985 | tcp_rcv_space_adjust(sk); | |
1986 | ||
1987 | skip_copy: | |
31770e34 | 1988 | if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) { |
1da177e4 | 1989 | tp->urg_data = 0; |
31770e34 FW |
1990 | tcp_fast_path_check(sk); |
1991 | } | |
1da177e4 LT |
1992 | if (used + offset < skb->len) |
1993 | continue; | |
1994 | ||
98aaa913 MM |
1995 | if (TCP_SKB_CB(skb)->has_rxtstamp) { |
1996 | tcp_update_recv_tstamps(skb, &tss); | |
1997 | has_tss = true; | |
1998 | } | |
e11ecddf | 1999 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) |
1da177e4 | 2000 | goto found_fin_ok; |
7bced397 DW |
2001 | if (!(flags & MSG_PEEK)) |
2002 | sk_eat_skb(sk, skb); | |
1da177e4 LT |
2003 | continue; |
2004 | ||
2005 | found_fin_ok: | |
2006 | /* Process the FIN. */ | |
2007 | ++*seq; | |
7bced397 DW |
2008 | if (!(flags & MSG_PEEK)) |
2009 | sk_eat_skb(sk, skb); | |
1da177e4 LT |
2010 | break; |
2011 | } while (len > 0); | |
2012 | ||
1da177e4 LT |
2013 | /* According to UNIX98, msg_name/msg_namelen are ignored |
2014 | * on connected socket. I was just happy when found this 8) --ANK | |
2015 | */ | |
2016 | ||
98aaa913 MM |
2017 | if (has_tss) |
2018 | tcp_recv_timestamp(msg, sk, &tss); | |
2019 | ||
1da177e4 | 2020 | /* Clean up data we have read: This will do ACK frames. */ |
0e4b4992 | 2021 | tcp_cleanup_rbuf(sk, copied); |
1da177e4 | 2022 | |
1da177e4 LT |
2023 | release_sock(sk); |
2024 | return copied; | |
2025 | ||
2026 | out: | |
1da177e4 LT |
2027 | release_sock(sk); |
2028 | return err; | |
2029 | ||
2030 | recv_urg: | |
377f0a08 | 2031 | err = tcp_recv_urg(sk, msg, len, flags); |
1da177e4 | 2032 | goto out; |
c0e88ff0 PE |
2033 | |
2034 | recv_sndq: | |
2035 | err = tcp_peek_sndq(sk, msg, len); | |
2036 | goto out; | |
1da177e4 | 2037 | } |
4bc2f18b | 2038 | EXPORT_SYMBOL(tcp_recvmsg); |
1da177e4 | 2039 | |
490d5046 IJ |
2040 | void tcp_set_state(struct sock *sk, int state) |
2041 | { | |
2042 | int oldstate = sk->sk_state; | |
2043 | ||
2044 | switch (state) { | |
2045 | case TCP_ESTABLISHED: | |
2046 | if (oldstate != TCP_ESTABLISHED) | |
81cc8a75 | 2047 | TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB); |
490d5046 IJ |
2048 | break; |
2049 | ||
2050 | case TCP_CLOSE: | |
2051 | if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED) | |
81cc8a75 | 2052 | TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS); |
490d5046 IJ |
2053 | |
2054 | sk->sk_prot->unhash(sk); | |
2055 | if (inet_csk(sk)->icsk_bind_hash && | |
2056 | !(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) | |
ab1e0a13 | 2057 | inet_put_port(sk); |
490d5046 IJ |
2058 | /* fall through */ |
2059 | default: | |
5a5f3a8d | 2060 | if (oldstate == TCP_ESTABLISHED) |
74688e48 | 2061 | TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB); |
490d5046 IJ |
2062 | } |
2063 | ||
2064 | /* Change state AFTER socket is unhashed to avoid closed | |
2065 | * socket sitting in hash tables. | |
2066 | */ | |
563e0bb0 | 2067 | inet_sk_state_store(sk, state); |
490d5046 IJ |
2068 | |
2069 | #ifdef STATE_TRACE | |
5a5f3a8d | 2070 | SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n", sk, statename[oldstate], statename[state]); |
490d5046 IJ |
2071 | #endif |
2072 | } | |
2073 | EXPORT_SYMBOL_GPL(tcp_set_state); | |
2074 | ||
1da177e4 LT |
2075 | /* |
2076 | * State processing on a close. This implements the state shift for | |
2077 | * sending our FIN frame. Note that we only send a FIN for some | |
2078 | * states. A shutdown() may have already sent the FIN, or we may be | |
2079 | * closed. | |
2080 | */ | |
2081 | ||
9b5b5cff | 2082 | static const unsigned char new_state[16] = { |
1da177e4 | 2083 | /* current state: new state: action: */ |
0980c1e3 ED |
2084 | [0 /* (Invalid) */] = TCP_CLOSE, |
2085 | [TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN, | |
2086 | [TCP_SYN_SENT] = TCP_CLOSE, | |
2087 | [TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN, | |
2088 | [TCP_FIN_WAIT1] = TCP_FIN_WAIT1, | |
2089 | [TCP_FIN_WAIT2] = TCP_FIN_WAIT2, | |
2090 | [TCP_TIME_WAIT] = TCP_CLOSE, | |
2091 | [TCP_CLOSE] = TCP_CLOSE, | |
2092 | [TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN, | |
2093 | [TCP_LAST_ACK] = TCP_LAST_ACK, | |
2094 | [TCP_LISTEN] = TCP_CLOSE, | |
2095 | [TCP_CLOSING] = TCP_CLOSING, | |
2096 | [TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */ | |
1da177e4 LT |
2097 | }; |
2098 | ||
2099 | static int tcp_close_state(struct sock *sk) | |
2100 | { | |
2101 | int next = (int)new_state[sk->sk_state]; | |
2102 | int ns = next & TCP_STATE_MASK; | |
2103 | ||
2104 | tcp_set_state(sk, ns); | |
2105 | ||
2106 | return next & TCP_ACTION_FIN; | |
2107 | } | |
2108 | ||
2109 | /* | |
2110 | * Shutdown the sending side of a connection. Much like close except | |
1f29b058 | 2111 | * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD). |
1da177e4 LT |
2112 | */ |
2113 | ||
2114 | void tcp_shutdown(struct sock *sk, int how) | |
2115 | { | |
2116 | /* We need to grab some memory, and put together a FIN, | |
2117 | * and then put it into the queue to be sent. | |
2118 | * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92. | |
2119 | */ | |
2120 | if (!(how & SEND_SHUTDOWN)) | |
2121 | return; | |
2122 | ||
2123 | /* If we've already sent a FIN, or it's a closed state, skip this. */ | |
2124 | if ((1 << sk->sk_state) & | |
2125 | (TCPF_ESTABLISHED | TCPF_SYN_SENT | | |
2126 | TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) { | |
2127 | /* Clear out any half completed packets. FIN if needed. */ | |
2128 | if (tcp_close_state(sk)) | |
2129 | tcp_send_fin(sk); | |
2130 | } | |
2131 | } | |
4bc2f18b | 2132 | EXPORT_SYMBOL(tcp_shutdown); |
1da177e4 | 2133 | |
efcdbf24 AS |
2134 | bool tcp_check_oom(struct sock *sk, int shift) |
2135 | { | |
2136 | bool too_many_orphans, out_of_socket_memory; | |
2137 | ||
2138 | too_many_orphans = tcp_too_many_orphans(sk, shift); | |
2139 | out_of_socket_memory = tcp_out_of_memory(sk); | |
2140 | ||
e87cc472 JP |
2141 | if (too_many_orphans) |
2142 | net_info_ratelimited("too many orphaned sockets\n"); | |
2143 | if (out_of_socket_memory) | |
2144 | net_info_ratelimited("out of memory -- consider tuning tcp_mem\n"); | |
efcdbf24 AS |
2145 | return too_many_orphans || out_of_socket_memory; |
2146 | } | |
2147 | ||
1da177e4 LT |
2148 | void tcp_close(struct sock *sk, long timeout) |
2149 | { | |
2150 | struct sk_buff *skb; | |
2151 | int data_was_unread = 0; | |
75c2d907 | 2152 | int state; |
1da177e4 LT |
2153 | |
2154 | lock_sock(sk); | |
2155 | sk->sk_shutdown = SHUTDOWN_MASK; | |
2156 | ||
2157 | if (sk->sk_state == TCP_LISTEN) { | |
2158 | tcp_set_state(sk, TCP_CLOSE); | |
2159 | ||
2160 | /* Special case. */ | |
0a5578cf | 2161 | inet_csk_listen_stop(sk); |
1da177e4 LT |
2162 | |
2163 | goto adjudge_to_death; | |
2164 | } | |
2165 | ||
2166 | /* We need to flush the recv. buffs. We do this only on the | |
2167 | * descriptor close, not protocol-sourced closes, because the | |
2168 | * reader process may not have drained the data yet! | |
2169 | */ | |
2170 | while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) { | |
e11ecddf ED |
2171 | u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq; |
2172 | ||
2173 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) | |
2174 | len--; | |
1da177e4 LT |
2175 | data_was_unread += len; |
2176 | __kfree_skb(skb); | |
2177 | } | |
2178 | ||
3ab224be | 2179 | sk_mem_reclaim(sk); |
1da177e4 | 2180 | |
565b7b2d KK |
2181 | /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */ |
2182 | if (sk->sk_state == TCP_CLOSE) | |
2183 | goto adjudge_to_death; | |
2184 | ||
65bb723c GR |
2185 | /* As outlined in RFC 2525, section 2.17, we send a RST here because |
2186 | * data was lost. To witness the awful effects of the old behavior of | |
2187 | * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk | |
2188 | * GET in an FTP client, suspend the process, wait for the client to | |
2189 | * advertise a zero window, then kill -9 the FTP client, wheee... | |
2190 | * Note: timeout is always zero in such a case. | |
1da177e4 | 2191 | */ |
ee995283 PE |
2192 | if (unlikely(tcp_sk(sk)->repair)) { |
2193 | sk->sk_prot->disconnect(sk, 0); | |
2194 | } else if (data_was_unread) { | |
1da177e4 | 2195 | /* Unread data was tossed, zap the connection. */ |
6aef70a8 | 2196 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE); |
1da177e4 | 2197 | tcp_set_state(sk, TCP_CLOSE); |
aa133076 | 2198 | tcp_send_active_reset(sk, sk->sk_allocation); |
1da177e4 LT |
2199 | } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) { |
2200 | /* Check zero linger _after_ checking for unread data. */ | |
2201 | sk->sk_prot->disconnect(sk, 0); | |
6aef70a8 | 2202 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONDATA); |
1da177e4 LT |
2203 | } else if (tcp_close_state(sk)) { |
2204 | /* We FIN if the application ate all the data before | |
2205 | * zapping the connection. | |
2206 | */ | |
2207 | ||
2208 | /* RED-PEN. Formally speaking, we have broken TCP state | |
2209 | * machine. State transitions: | |
2210 | * | |
2211 | * TCP_ESTABLISHED -> TCP_FIN_WAIT1 | |
2212 | * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible) | |
2213 | * TCP_CLOSE_WAIT -> TCP_LAST_ACK | |
2214 | * | |
2215 | * are legal only when FIN has been sent (i.e. in window), | |
2216 | * rather than queued out of window. Purists blame. | |
2217 | * | |
2218 | * F.e. "RFC state" is ESTABLISHED, | |
2219 | * if Linux state is FIN-WAIT-1, but FIN is still not sent. | |
2220 | * | |
2221 | * The visible declinations are that sometimes | |
2222 | * we enter time-wait state, when it is not required really | |
2223 | * (harmless), do not send active resets, when they are | |
2224 | * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when | |
2225 | * they look as CLOSING or LAST_ACK for Linux) | |
2226 | * Probably, I missed some more holelets. | |
2227 | * --ANK | |
8336886f JC |
2228 | * XXX (TFO) - To start off we don't support SYN+ACK+FIN |
2229 | * in a single packet! (May consider it later but will | |
2230 | * probably need API support or TCP_CORK SYN-ACK until | |
2231 | * data is written and socket is closed.) | |
1da177e4 LT |
2232 | */ |
2233 | tcp_send_fin(sk); | |
2234 | } | |
2235 | ||
2236 | sk_stream_wait_close(sk, timeout); | |
2237 | ||
2238 | adjudge_to_death: | |
75c2d907 HX |
2239 | state = sk->sk_state; |
2240 | sock_hold(sk); | |
2241 | sock_orphan(sk); | |
75c2d907 | 2242 | |
1da177e4 LT |
2243 | /* It is the last release_sock in its life. It will remove backlog. */ |
2244 | release_sock(sk); | |
2245 | ||
2246 | ||
2247 | /* Now socket is owned by kernel and we acquire BH lock | |
a777f715 | 2248 | * to finish close. No need to check for user refs. |
1da177e4 LT |
2249 | */ |
2250 | local_bh_disable(); | |
2251 | bh_lock_sock(sk); | |
547b792c | 2252 | WARN_ON(sock_owned_by_user(sk)); |
1da177e4 | 2253 | |
eb4dea58 HX |
2254 | percpu_counter_inc(sk->sk_prot->orphan_count); |
2255 | ||
75c2d907 HX |
2256 | /* Have we already been destroyed by a softirq or backlog? */ |
2257 | if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE) | |
2258 | goto out; | |
1da177e4 LT |
2259 | |
2260 | /* This is a (useful) BSD violating of the RFC. There is a | |
2261 | * problem with TCP as specified in that the other end could | |
2262 | * keep a socket open forever with no application left this end. | |
b10bd54c | 2263 | * We use a 1 minute timeout (about the same as BSD) then kill |
1da177e4 LT |
2264 | * our end. If they send after that then tough - BUT: long enough |
2265 | * that we won't make the old 4*rto = almost no time - whoops | |
2266 | * reset mistake. | |
2267 | * | |
2268 | * Nope, it was not mistake. It is really desired behaviour | |
2269 | * f.e. on http servers, when such sockets are useless, but | |
2270 | * consume significant resources. Let's do it with special | |
2271 | * linger2 option. --ANK | |
2272 | */ | |
2273 | ||
2274 | if (sk->sk_state == TCP_FIN_WAIT2) { | |
2275 | struct tcp_sock *tp = tcp_sk(sk); | |
2276 | if (tp->linger2 < 0) { | |
2277 | tcp_set_state(sk, TCP_CLOSE); | |
2278 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
02a1d6e7 | 2279 | __NET_INC_STATS(sock_net(sk), |
de0744af | 2280 | LINUX_MIB_TCPABORTONLINGER); |
1da177e4 | 2281 | } else { |
463c84b9 | 2282 | const int tmo = tcp_fin_time(sk); |
1da177e4 LT |
2283 | |
2284 | if (tmo > TCP_TIMEWAIT_LEN) { | |
52499afe DM |
2285 | inet_csk_reset_keepalive_timer(sk, |
2286 | tmo - TCP_TIMEWAIT_LEN); | |
1da177e4 | 2287 | } else { |
1da177e4 LT |
2288 | tcp_time_wait(sk, TCP_FIN_WAIT2, tmo); |
2289 | goto out; | |
2290 | } | |
2291 | } | |
2292 | } | |
2293 | if (sk->sk_state != TCP_CLOSE) { | |
3ab224be | 2294 | sk_mem_reclaim(sk); |
efcdbf24 | 2295 | if (tcp_check_oom(sk, 0)) { |
1da177e4 LT |
2296 | tcp_set_state(sk, TCP_CLOSE); |
2297 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
02a1d6e7 | 2298 | __NET_INC_STATS(sock_net(sk), |
de0744af | 2299 | LINUX_MIB_TCPABORTONMEMORY); |
1da177e4 LT |
2300 | } |
2301 | } | |
1da177e4 | 2302 | |
8336886f JC |
2303 | if (sk->sk_state == TCP_CLOSE) { |
2304 | struct request_sock *req = tcp_sk(sk)->fastopen_rsk; | |
2305 | /* We could get here with a non-NULL req if the socket is | |
2306 | * aborted (e.g., closed with unread data) before 3WHS | |
2307 | * finishes. | |
2308 | */ | |
00db4124 | 2309 | if (req) |
8336886f | 2310 | reqsk_fastopen_remove(sk, req, false); |
0a5578cf | 2311 | inet_csk_destroy_sock(sk); |
8336886f | 2312 | } |
1da177e4 LT |
2313 | /* Otherwise, socket is reprieved until protocol close. */ |
2314 | ||
2315 | out: | |
2316 | bh_unlock_sock(sk); | |
2317 | local_bh_enable(); | |
2318 | sock_put(sk); | |
2319 | } | |
4bc2f18b | 2320 | EXPORT_SYMBOL(tcp_close); |
1da177e4 LT |
2321 | |
2322 | /* These states need RST on ABORT according to RFC793 */ | |
2323 | ||
a2a385d6 | 2324 | static inline bool tcp_need_reset(int state) |
1da177e4 LT |
2325 | { |
2326 | return (1 << state) & | |
2327 | (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 | | |
2328 | TCPF_FIN_WAIT2 | TCPF_SYN_RECV); | |
2329 | } | |
2330 | ||
75c119af ED |
2331 | static void tcp_rtx_queue_purge(struct sock *sk) |
2332 | { | |
2333 | struct rb_node *p = rb_first(&sk->tcp_rtx_queue); | |
2334 | ||
2335 | while (p) { | |
2336 | struct sk_buff *skb = rb_to_skb(p); | |
2337 | ||
2338 | p = rb_next(p); | |
2339 | /* Since we are deleting whole queue, no need to | |
2340 | * list_del(&skb->tcp_tsorted_anchor) | |
2341 | */ | |
2342 | tcp_rtx_queue_unlink(skb, sk); | |
2343 | sk_wmem_free_skb(sk, skb); | |
2344 | } | |
2345 | } | |
2346 | ||
ac3f09ba ED |
2347 | void tcp_write_queue_purge(struct sock *sk) |
2348 | { | |
2349 | struct sk_buff *skb; | |
2350 | ||
2351 | tcp_chrono_stop(sk, TCP_CHRONO_BUSY); | |
2352 | while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) { | |
2353 | tcp_skb_tsorted_anchor_cleanup(skb); | |
2354 | sk_wmem_free_skb(sk, skb); | |
2355 | } | |
75c119af | 2356 | tcp_rtx_queue_purge(sk); |
ac3f09ba ED |
2357 | INIT_LIST_HEAD(&tcp_sk(sk)->tsorted_sent_queue); |
2358 | sk_mem_reclaim(sk); | |
2359 | tcp_clear_all_retrans_hints(tcp_sk(sk)); | |
2360 | } | |
2361 | ||
1da177e4 LT |
2362 | int tcp_disconnect(struct sock *sk, int flags) |
2363 | { | |
2364 | struct inet_sock *inet = inet_sk(sk); | |
463c84b9 | 2365 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
2366 | struct tcp_sock *tp = tcp_sk(sk); |
2367 | int err = 0; | |
2368 | int old_state = sk->sk_state; | |
2369 | ||
2370 | if (old_state != TCP_CLOSE) | |
2371 | tcp_set_state(sk, TCP_CLOSE); | |
2372 | ||
2373 | /* ABORT function of RFC793 */ | |
2374 | if (old_state == TCP_LISTEN) { | |
0a5578cf | 2375 | inet_csk_listen_stop(sk); |
ee995283 PE |
2376 | } else if (unlikely(tp->repair)) { |
2377 | sk->sk_err = ECONNABORTED; | |
1da177e4 LT |
2378 | } else if (tcp_need_reset(old_state) || |
2379 | (tp->snd_nxt != tp->write_seq && | |
2380 | (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) { | |
caa20d9a | 2381 | /* The last check adjusts for discrepancy of Linux wrt. RFC |
1da177e4 LT |
2382 | * states |
2383 | */ | |
2384 | tcp_send_active_reset(sk, gfp_any()); | |
2385 | sk->sk_err = ECONNRESET; | |
2386 | } else if (old_state == TCP_SYN_SENT) | |
2387 | sk->sk_err = ECONNRESET; | |
2388 | ||
2389 | tcp_clear_xmit_timers(sk); | |
2390 | __skb_queue_purge(&sk->sk_receive_queue); | |
fe067e8a | 2391 | tcp_write_queue_purge(sk); |
cf1ef3f0 | 2392 | tcp_fastopen_active_disable_ofo_check(sk); |
9f5afeae | 2393 | skb_rbtree_purge(&tp->out_of_order_queue); |
1da177e4 | 2394 | |
c720c7e8 | 2395 | inet->inet_dport = 0; |
1da177e4 LT |
2396 | |
2397 | if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) | |
2398 | inet_reset_saddr(sk); | |
2399 | ||
2400 | sk->sk_shutdown = 0; | |
2401 | sock_reset_flag(sk, SOCK_DONE); | |
740b0f18 | 2402 | tp->srtt_us = 0; |
686a5624 YM |
2403 | tp->write_seq += tp->max_window + 2; |
2404 | if (tp->write_seq == 0) | |
1da177e4 | 2405 | tp->write_seq = 1; |
463c84b9 | 2406 | icsk->icsk_backoff = 0; |
1da177e4 | 2407 | tp->snd_cwnd = 2; |
6687e988 | 2408 | icsk->icsk_probes_out = 0; |
1da177e4 | 2409 | tp->packets_out = 0; |
0b6a05c1 | 2410 | tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; |
1da177e4 | 2411 | tp->snd_cwnd_cnt = 0; |
1fdf475a | 2412 | tp->window_clamp = 0; |
6687e988 | 2413 | tcp_set_ca_state(sk, TCP_CA_Open); |
d4761754 | 2414 | tp->is_sack_reneg = 0; |
1da177e4 | 2415 | tcp_clear_retrans(tp); |
463c84b9 | 2416 | inet_csk_delack_init(sk); |
499350a5 WW |
2417 | /* Initialize rcv_mss to TCP_MIN_MSS to avoid division by 0 |
2418 | * issue in __tcp_select_window() | |
2419 | */ | |
2420 | icsk->icsk_ack.rcv_mss = TCP_MIN_MSS; | |
b40b4f79 | 2421 | memset(&tp->rx_opt, 0, sizeof(tp->rx_opt)); |
1da177e4 | 2422 | __sk_dst_reset(sk); |
d747a7a5 WC |
2423 | dst_release(sk->sk_rx_dst); |
2424 | sk->sk_rx_dst = NULL; | |
17c3060b | 2425 | tcp_saved_syn_free(tp); |
1da177e4 | 2426 | |
7db92362 WW |
2427 | /* Clean up fastopen related fields */ |
2428 | tcp_free_fastopen_req(tp); | |
2429 | inet->defer_connect = 0; | |
2430 | ||
c720c7e8 | 2431 | WARN_ON(inet->inet_num && !icsk->icsk_bind_hash); |
1da177e4 LT |
2432 | |
2433 | sk->sk_error_report(sk); | |
2434 | return err; | |
2435 | } | |
4bc2f18b | 2436 | EXPORT_SYMBOL(tcp_disconnect); |
1da177e4 | 2437 | |
a2a385d6 | 2438 | static inline bool tcp_can_repair_sock(const struct sock *sk) |
ee995283 | 2439 | { |
52e804c6 | 2440 | return ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN) && |
319b0534 | 2441 | (sk->sk_state != TCP_LISTEN); |
ee995283 PE |
2442 | } |
2443 | ||
b1ed4c4f AV |
2444 | static int tcp_repair_set_window(struct tcp_sock *tp, char __user *optbuf, int len) |
2445 | { | |
2446 | struct tcp_repair_window opt; | |
2447 | ||
2448 | if (!tp->repair) | |
2449 | return -EPERM; | |
2450 | ||
2451 | if (len != sizeof(opt)) | |
2452 | return -EINVAL; | |
2453 | ||
2454 | if (copy_from_user(&opt, optbuf, sizeof(opt))) | |
2455 | return -EFAULT; | |
2456 | ||
2457 | if (opt.max_window < opt.snd_wnd) | |
2458 | return -EINVAL; | |
2459 | ||
2460 | if (after(opt.snd_wl1, tp->rcv_nxt + opt.rcv_wnd)) | |
2461 | return -EINVAL; | |
2462 | ||
2463 | if (after(opt.rcv_wup, tp->rcv_nxt)) | |
2464 | return -EINVAL; | |
2465 | ||
2466 | tp->snd_wl1 = opt.snd_wl1; | |
2467 | tp->snd_wnd = opt.snd_wnd; | |
2468 | tp->max_window = opt.max_window; | |
2469 | ||
2470 | tp->rcv_wnd = opt.rcv_wnd; | |
2471 | tp->rcv_wup = opt.rcv_wup; | |
2472 | ||
2473 | return 0; | |
2474 | } | |
2475 | ||
15e56515 | 2476 | static int tcp_repair_options_est(struct sock *sk, |
de248a75 | 2477 | struct tcp_repair_opt __user *optbuf, unsigned int len) |
b139ba4e | 2478 | { |
15e56515 | 2479 | struct tcp_sock *tp = tcp_sk(sk); |
de248a75 | 2480 | struct tcp_repair_opt opt; |
b139ba4e | 2481 | |
de248a75 PE |
2482 | while (len >= sizeof(opt)) { |
2483 | if (copy_from_user(&opt, optbuf, sizeof(opt))) | |
b139ba4e PE |
2484 | return -EFAULT; |
2485 | ||
2486 | optbuf++; | |
de248a75 | 2487 | len -= sizeof(opt); |
b139ba4e | 2488 | |
de248a75 PE |
2489 | switch (opt.opt_code) { |
2490 | case TCPOPT_MSS: | |
2491 | tp->rx_opt.mss_clamp = opt.opt_val; | |
15e56515 | 2492 | tcp_mtup_init(sk); |
b139ba4e | 2493 | break; |
de248a75 | 2494 | case TCPOPT_WINDOW: |
bc26ccd8 AV |
2495 | { |
2496 | u16 snd_wscale = opt.opt_val & 0xFFFF; | |
2497 | u16 rcv_wscale = opt.opt_val >> 16; | |
2498 | ||
589c49cb | 2499 | if (snd_wscale > TCP_MAX_WSCALE || rcv_wscale > TCP_MAX_WSCALE) |
bc26ccd8 | 2500 | return -EFBIG; |
b139ba4e | 2501 | |
bc26ccd8 AV |
2502 | tp->rx_opt.snd_wscale = snd_wscale; |
2503 | tp->rx_opt.rcv_wscale = rcv_wscale; | |
2504 | tp->rx_opt.wscale_ok = 1; | |
2505 | } | |
b139ba4e | 2506 | break; |
b139ba4e | 2507 | case TCPOPT_SACK_PERM: |
de248a75 PE |
2508 | if (opt.opt_val != 0) |
2509 | return -EINVAL; | |
2510 | ||
b139ba4e | 2511 | tp->rx_opt.sack_ok |= TCP_SACK_SEEN; |
b139ba4e PE |
2512 | break; |
2513 | case TCPOPT_TIMESTAMP: | |
de248a75 PE |
2514 | if (opt.opt_val != 0) |
2515 | return -EINVAL; | |
2516 | ||
b139ba4e PE |
2517 | tp->rx_opt.tstamp_ok = 1; |
2518 | break; | |
2519 | } | |
2520 | } | |
2521 | ||
2522 | return 0; | |
2523 | } | |
2524 | ||
1da177e4 LT |
2525 | /* |
2526 | * Socket option code for TCP. | |
2527 | */ | |
3fdadf7d | 2528 | static int do_tcp_setsockopt(struct sock *sk, int level, |
b7058842 | 2529 | int optname, char __user *optval, unsigned int optlen) |
1da177e4 LT |
2530 | { |
2531 | struct tcp_sock *tp = tcp_sk(sk); | |
463c84b9 | 2532 | struct inet_connection_sock *icsk = inet_csk(sk); |
1e579caa | 2533 | struct net *net = sock_net(sk); |
1da177e4 LT |
2534 | int val; |
2535 | int err = 0; | |
2536 | ||
e56fb50f WAS |
2537 | /* These are data/string values, all the others are ints */ |
2538 | switch (optname) { | |
2539 | case TCP_CONGESTION: { | |
5f8ef48d SH |
2540 | char name[TCP_CA_NAME_MAX]; |
2541 | ||
2542 | if (optlen < 1) | |
2543 | return -EINVAL; | |
2544 | ||
2545 | val = strncpy_from_user(name, optval, | |
4fdb78d3 | 2546 | min_t(long, TCP_CA_NAME_MAX-1, optlen)); |
5f8ef48d SH |
2547 | if (val < 0) |
2548 | return -EFAULT; | |
2549 | name[val] = 0; | |
2550 | ||
2551 | lock_sock(sk); | |
ebfa00c5 | 2552 | err = tcp_set_congestion_control(sk, name, true, true); |
5f8ef48d SH |
2553 | release_sock(sk); |
2554 | return err; | |
2555 | } | |
734942cc DW |
2556 | case TCP_ULP: { |
2557 | char name[TCP_ULP_NAME_MAX]; | |
2558 | ||
2559 | if (optlen < 1) | |
2560 | return -EINVAL; | |
2561 | ||
2562 | val = strncpy_from_user(name, optval, | |
2563 | min_t(long, TCP_ULP_NAME_MAX - 1, | |
2564 | optlen)); | |
2565 | if (val < 0) | |
2566 | return -EFAULT; | |
2567 | name[val] = 0; | |
2568 | ||
2569 | lock_sock(sk); | |
2570 | err = tcp_set_ulp(sk, name); | |
2571 | release_sock(sk); | |
2572 | return err; | |
2573 | } | |
1fba70e5 YC |
2574 | case TCP_FASTOPEN_KEY: { |
2575 | __u8 key[TCP_FASTOPEN_KEY_LENGTH]; | |
2576 | ||
2577 | if (optlen != sizeof(key)) | |
2578 | return -EINVAL; | |
2579 | ||
2580 | if (copy_from_user(key, optval, optlen)) | |
2581 | return -EFAULT; | |
2582 | ||
2583 | return tcp_fastopen_reset_cipher(net, sk, key, sizeof(key)); | |
2584 | } | |
e56fb50f WAS |
2585 | default: |
2586 | /* fallthru */ | |
2587 | break; | |
ccbd6a5a | 2588 | } |
5f8ef48d | 2589 | |
1da177e4 LT |
2590 | if (optlen < sizeof(int)) |
2591 | return -EINVAL; | |
2592 | ||
2593 | if (get_user(val, (int __user *)optval)) | |
2594 | return -EFAULT; | |
2595 | ||
2596 | lock_sock(sk); | |
2597 | ||
2598 | switch (optname) { | |
2599 | case TCP_MAXSEG: | |
2600 | /* Values greater than interface MTU won't take effect. However | |
2601 | * at the point when this call is done we typically don't yet | |
a777f715 RC |
2602 | * know which interface is going to be used |
2603 | */ | |
cfc62d87 | 2604 | if (val && (val < TCP_MIN_MSS || val > MAX_TCP_WINDOW)) { |
1da177e4 LT |
2605 | err = -EINVAL; |
2606 | break; | |
2607 | } | |
2608 | tp->rx_opt.user_mss = val; | |
2609 | break; | |
2610 | ||
2611 | case TCP_NODELAY: | |
2612 | if (val) { | |
2613 | /* TCP_NODELAY is weaker than TCP_CORK, so that | |
2614 | * this option on corked socket is remembered, but | |
2615 | * it is not activated until cork is cleared. | |
2616 | * | |
2617 | * However, when TCP_NODELAY is set we make | |
2618 | * an explicit push, which overrides even TCP_CORK | |
2619 | * for currently queued segments. | |
2620 | */ | |
2621 | tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH; | |
9e412ba7 | 2622 | tcp_push_pending_frames(sk); |
1da177e4 LT |
2623 | } else { |
2624 | tp->nonagle &= ~TCP_NAGLE_OFF; | |
2625 | } | |
2626 | break; | |
2627 | ||
36e31b0a AP |
2628 | case TCP_THIN_LINEAR_TIMEOUTS: |
2629 | if (val < 0 || val > 1) | |
2630 | err = -EINVAL; | |
2631 | else | |
2632 | tp->thin_lto = val; | |
2633 | break; | |
2634 | ||
7e380175 AP |
2635 | case TCP_THIN_DUPACK: |
2636 | if (val < 0 || val > 1) | |
2637 | err = -EINVAL; | |
7e380175 AP |
2638 | break; |
2639 | ||
ee995283 PE |
2640 | case TCP_REPAIR: |
2641 | if (!tcp_can_repair_sock(sk)) | |
2642 | err = -EPERM; | |
2643 | else if (val == 1) { | |
2644 | tp->repair = 1; | |
2645 | sk->sk_reuse = SK_FORCE_REUSE; | |
2646 | tp->repair_queue = TCP_NO_QUEUE; | |
2647 | } else if (val == 0) { | |
2648 | tp->repair = 0; | |
2649 | sk->sk_reuse = SK_NO_REUSE; | |
2650 | tcp_send_window_probe(sk); | |
2651 | } else | |
2652 | err = -EINVAL; | |
2653 | ||
2654 | break; | |
2655 | ||
2656 | case TCP_REPAIR_QUEUE: | |
2657 | if (!tp->repair) | |
2658 | err = -EPERM; | |
2659 | else if (val < TCP_QUEUES_NR) | |
2660 | tp->repair_queue = val; | |
2661 | else | |
2662 | err = -EINVAL; | |
2663 | break; | |
2664 | ||
2665 | case TCP_QUEUE_SEQ: | |
2666 | if (sk->sk_state != TCP_CLOSE) | |
2667 | err = -EPERM; | |
2668 | else if (tp->repair_queue == TCP_SEND_QUEUE) | |
2669 | tp->write_seq = val; | |
2670 | else if (tp->repair_queue == TCP_RECV_QUEUE) | |
2671 | tp->rcv_nxt = val; | |
2672 | else | |
2673 | err = -EINVAL; | |
2674 | break; | |
2675 | ||
b139ba4e PE |
2676 | case TCP_REPAIR_OPTIONS: |
2677 | if (!tp->repair) | |
2678 | err = -EINVAL; | |
2679 | else if (sk->sk_state == TCP_ESTABLISHED) | |
15e56515 | 2680 | err = tcp_repair_options_est(sk, |
de248a75 PE |
2681 | (struct tcp_repair_opt __user *)optval, |
2682 | optlen); | |
b139ba4e PE |
2683 | else |
2684 | err = -EPERM; | |
2685 | break; | |
2686 | ||
1da177e4 LT |
2687 | case TCP_CORK: |
2688 | /* When set indicates to always queue non-full frames. | |
2689 | * Later the user clears this option and we transmit | |
2690 | * any pending partial frames in the queue. This is | |
2691 | * meant to be used alongside sendfile() to get properly | |
2692 | * filled frames when the user (for example) must write | |
2693 | * out headers with a write() call first and then use | |
2694 | * sendfile to send out the data parts. | |
2695 | * | |
2696 | * TCP_CORK can be set together with TCP_NODELAY and it is | |
2697 | * stronger than TCP_NODELAY. | |
2698 | */ | |
2699 | if (val) { | |
2700 | tp->nonagle |= TCP_NAGLE_CORK; | |
2701 | } else { | |
2702 | tp->nonagle &= ~TCP_NAGLE_CORK; | |
2703 | if (tp->nonagle&TCP_NAGLE_OFF) | |
2704 | tp->nonagle |= TCP_NAGLE_PUSH; | |
9e412ba7 | 2705 | tcp_push_pending_frames(sk); |
1da177e4 LT |
2706 | } |
2707 | break; | |
2708 | ||
2709 | case TCP_KEEPIDLE: | |
2710 | if (val < 1 || val > MAX_TCP_KEEPIDLE) | |
2711 | err = -EINVAL; | |
2712 | else { | |
2713 | tp->keepalive_time = val * HZ; | |
2714 | if (sock_flag(sk, SOCK_KEEPOPEN) && | |
2715 | !((1 << sk->sk_state) & | |
2716 | (TCPF_CLOSE | TCPF_LISTEN))) { | |
6c37e5de | 2717 | u32 elapsed = keepalive_time_elapsed(tp); |
1da177e4 LT |
2718 | if (tp->keepalive_time > elapsed) |
2719 | elapsed = tp->keepalive_time - elapsed; | |
2720 | else | |
2721 | elapsed = 0; | |
463c84b9 | 2722 | inet_csk_reset_keepalive_timer(sk, elapsed); |
1da177e4 LT |
2723 | } |
2724 | } | |
2725 | break; | |
2726 | case TCP_KEEPINTVL: | |
2727 | if (val < 1 || val > MAX_TCP_KEEPINTVL) | |
2728 | err = -EINVAL; | |
2729 | else | |
2730 | tp->keepalive_intvl = val * HZ; | |
2731 | break; | |
2732 | case TCP_KEEPCNT: | |
2733 | if (val < 1 || val > MAX_TCP_KEEPCNT) | |
2734 | err = -EINVAL; | |
2735 | else | |
2736 | tp->keepalive_probes = val; | |
2737 | break; | |
2738 | case TCP_SYNCNT: | |
2739 | if (val < 1 || val > MAX_TCP_SYNCNT) | |
2740 | err = -EINVAL; | |
2741 | else | |
463c84b9 | 2742 | icsk->icsk_syn_retries = val; |
1da177e4 LT |
2743 | break; |
2744 | ||
cd8ae852 ED |
2745 | case TCP_SAVE_SYN: |
2746 | if (val < 0 || val > 1) | |
2747 | err = -EINVAL; | |
2748 | else | |
2749 | tp->save_syn = val; | |
2750 | break; | |
2751 | ||
1da177e4 LT |
2752 | case TCP_LINGER2: |
2753 | if (val < 0) | |
2754 | tp->linger2 = -1; | |
1e579caa | 2755 | else if (val > net->ipv4.sysctl_tcp_fin_timeout / HZ) |
1da177e4 LT |
2756 | tp->linger2 = 0; |
2757 | else | |
2758 | tp->linger2 = val * HZ; | |
2759 | break; | |
2760 | ||
2761 | case TCP_DEFER_ACCEPT: | |
b103cf34 JA |
2762 | /* Translate value in seconds to number of retransmits */ |
2763 | icsk->icsk_accept_queue.rskq_defer_accept = | |
2764 | secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ, | |
2765 | TCP_RTO_MAX / HZ); | |
1da177e4 LT |
2766 | break; |
2767 | ||
2768 | case TCP_WINDOW_CLAMP: | |
2769 | if (!val) { | |
2770 | if (sk->sk_state != TCP_CLOSE) { | |
2771 | err = -EINVAL; | |
2772 | break; | |
2773 | } | |
2774 | tp->window_clamp = 0; | |
2775 | } else | |
2776 | tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ? | |
2777 | SOCK_MIN_RCVBUF / 2 : val; | |
2778 | break; | |
2779 | ||
2780 | case TCP_QUICKACK: | |
2781 | if (!val) { | |
463c84b9 | 2782 | icsk->icsk_ack.pingpong = 1; |
1da177e4 | 2783 | } else { |
463c84b9 | 2784 | icsk->icsk_ack.pingpong = 0; |
1da177e4 LT |
2785 | if ((1 << sk->sk_state) & |
2786 | (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) && | |
463c84b9 ACM |
2787 | inet_csk_ack_scheduled(sk)) { |
2788 | icsk->icsk_ack.pending |= ICSK_ACK_PUSHED; | |
0e4b4992 | 2789 | tcp_cleanup_rbuf(sk, 1); |
1da177e4 | 2790 | if (!(val & 1)) |
463c84b9 | 2791 | icsk->icsk_ack.pingpong = 1; |
1da177e4 LT |
2792 | } |
2793 | } | |
2794 | break; | |
2795 | ||
cfb6eeb4 YH |
2796 | #ifdef CONFIG_TCP_MD5SIG |
2797 | case TCP_MD5SIG: | |
8917a777 | 2798 | case TCP_MD5SIG_EXT: |
cfb6eeb4 | 2799 | /* Read the IP->Key mappings from userspace */ |
8917a777 | 2800 | err = tp->af_specific->md5_parse(sk, optname, optval, optlen); |
cfb6eeb4 YH |
2801 | break; |
2802 | #endif | |
dca43c75 | 2803 | case TCP_USER_TIMEOUT: |
b248230c | 2804 | /* Cap the max time in ms TCP will retry or probe the window |
dca43c75 JC |
2805 | * before giving up and aborting (ETIMEDOUT) a connection. |
2806 | */ | |
42493570 HL |
2807 | if (val < 0) |
2808 | err = -EINVAL; | |
2809 | else | |
2810 | icsk->icsk_user_timeout = msecs_to_jiffies(val); | |
dca43c75 | 2811 | break; |
8336886f JC |
2812 | |
2813 | case TCP_FASTOPEN: | |
2814 | if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE | | |
dfea2aa6 | 2815 | TCPF_LISTEN))) { |
43713848 | 2816 | tcp_fastopen_init_key_once(net); |
dfea2aa6 | 2817 | |
0536fcc0 | 2818 | fastopen_queue_tune(sk, val); |
dfea2aa6 | 2819 | } else { |
8336886f | 2820 | err = -EINVAL; |
dfea2aa6 | 2821 | } |
8336886f | 2822 | break; |
19f6d3f3 WW |
2823 | case TCP_FASTOPEN_CONNECT: |
2824 | if (val > 1 || val < 0) { | |
2825 | err = -EINVAL; | |
e1cfcbe8 | 2826 | } else if (net->ipv4.sysctl_tcp_fastopen & TFO_CLIENT_ENABLE) { |
19f6d3f3 WW |
2827 | if (sk->sk_state == TCP_CLOSE) |
2828 | tp->fastopen_connect = val; | |
2829 | else | |
2830 | err = -EINVAL; | |
2831 | } else { | |
2832 | err = -EOPNOTSUPP; | |
2833 | } | |
2834 | break; | |
71c02379 CP |
2835 | case TCP_FASTOPEN_NO_COOKIE: |
2836 | if (val > 1 || val < 0) | |
2837 | err = -EINVAL; | |
2838 | else if (!((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) | |
2839 | err = -EINVAL; | |
2840 | else | |
2841 | tp->fastopen_no_cookie = val; | |
2842 | break; | |
93be6ce0 AV |
2843 | case TCP_TIMESTAMP: |
2844 | if (!tp->repair) | |
2845 | err = -EPERM; | |
2846 | else | |
9a568de4 | 2847 | tp->tsoffset = val - tcp_time_stamp_raw(); |
93be6ce0 | 2848 | break; |
b1ed4c4f AV |
2849 | case TCP_REPAIR_WINDOW: |
2850 | err = tcp_repair_set_window(tp, optval, optlen); | |
2851 | break; | |
c9bee3b7 ED |
2852 | case TCP_NOTSENT_LOWAT: |
2853 | tp->notsent_lowat = val; | |
2854 | sk->sk_write_space(sk); | |
2855 | break; | |
1da177e4 LT |
2856 | default: |
2857 | err = -ENOPROTOOPT; | |
2858 | break; | |
3ff50b79 SH |
2859 | } |
2860 | ||
1da177e4 LT |
2861 | release_sock(sk); |
2862 | return err; | |
2863 | } | |
2864 | ||
3fdadf7d | 2865 | int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval, |
b7058842 | 2866 | unsigned int optlen) |
3fdadf7d | 2867 | { |
cf533ea5 | 2868 | const struct inet_connection_sock *icsk = inet_csk(sk); |
3fdadf7d DM |
2869 | |
2870 | if (level != SOL_TCP) | |
2871 | return icsk->icsk_af_ops->setsockopt(sk, level, optname, | |
2872 | optval, optlen); | |
2873 | return do_tcp_setsockopt(sk, level, optname, optval, optlen); | |
2874 | } | |
4bc2f18b | 2875 | EXPORT_SYMBOL(tcp_setsockopt); |
3fdadf7d DM |
2876 | |
2877 | #ifdef CONFIG_COMPAT | |
543d9cfe | 2878 | int compat_tcp_setsockopt(struct sock *sk, int level, int optname, |
b7058842 | 2879 | char __user *optval, unsigned int optlen) |
3fdadf7d | 2880 | { |
dec73ff0 ACM |
2881 | if (level != SOL_TCP) |
2882 | return inet_csk_compat_setsockopt(sk, level, optname, | |
2883 | optval, optlen); | |
3fdadf7d DM |
2884 | return do_tcp_setsockopt(sk, level, optname, optval, optlen); |
2885 | } | |
543d9cfe | 2886 | EXPORT_SYMBOL(compat_tcp_setsockopt); |
3fdadf7d DM |
2887 | #endif |
2888 | ||
efd90174 FY |
2889 | static void tcp_get_info_chrono_stats(const struct tcp_sock *tp, |
2890 | struct tcp_info *info) | |
2891 | { | |
2892 | u64 stats[__TCP_CHRONO_MAX], total = 0; | |
2893 | enum tcp_chrono i; | |
2894 | ||
2895 | for (i = TCP_CHRONO_BUSY; i < __TCP_CHRONO_MAX; ++i) { | |
2896 | stats[i] = tp->chrono_stat[i - 1]; | |
2897 | if (i == tp->chrono_type) | |
628174cc | 2898 | stats[i] += tcp_jiffies32 - tp->chrono_start; |
efd90174 FY |
2899 | stats[i] *= USEC_PER_SEC / HZ; |
2900 | total += stats[i]; | |
2901 | } | |
2902 | ||
2903 | info->tcpi_busy_time = total; | |
2904 | info->tcpi_rwnd_limited = stats[TCP_CHRONO_RWND_LIMITED]; | |
2905 | info->tcpi_sndbuf_limited = stats[TCP_CHRONO_SNDBUF_LIMITED]; | |
2906 | } | |
2907 | ||
1da177e4 | 2908 | /* Return information about state of tcp endpoint in API format. */ |
0df48c26 | 2909 | void tcp_get_info(struct sock *sk, struct tcp_info *info) |
1da177e4 | 2910 | { |
35ac838a | 2911 | const struct tcp_sock *tp = tcp_sk(sk); /* iff sk_type == SOCK_STREAM */ |
463c84b9 | 2912 | const struct inet_connection_sock *icsk = inet_csk(sk); |
0263598c | 2913 | u32 now; |
ff5d7497 | 2914 | u64 rate64; |
67db3e4b | 2915 | bool slow; |
fad9dfef | 2916 | u32 rate; |
1da177e4 LT |
2917 | |
2918 | memset(info, 0, sizeof(*info)); | |
35ac838a CG |
2919 | if (sk->sk_type != SOCK_STREAM) |
2920 | return; | |
1da177e4 | 2921 | |
986ffdfd | 2922 | info->tcpi_state = inet_sk_state_load(sk); |
00fd38d9 | 2923 | |
ccbf3bfa ED |
2924 | /* Report meaningful fields for all TCP states, including listeners */ |
2925 | rate = READ_ONCE(sk->sk_pacing_rate); | |
2926 | rate64 = rate != ~0U ? rate : ~0ULL; | |
f522a5fc | 2927 | info->tcpi_pacing_rate = rate64; |
ccbf3bfa ED |
2928 | |
2929 | rate = READ_ONCE(sk->sk_max_pacing_rate); | |
2930 | rate64 = rate != ~0U ? rate : ~0ULL; | |
f522a5fc | 2931 | info->tcpi_max_pacing_rate = rate64; |
ccbf3bfa ED |
2932 | |
2933 | info->tcpi_reordering = tp->reordering; | |
2934 | info->tcpi_snd_cwnd = tp->snd_cwnd; | |
2935 | ||
2936 | if (info->tcpi_state == TCP_LISTEN) { | |
2937 | /* listeners aliased fields : | |
2938 | * tcpi_unacked -> Number of children ready for accept() | |
2939 | * tcpi_sacked -> max backlog | |
2940 | */ | |
2941 | info->tcpi_unacked = sk->sk_ack_backlog; | |
2942 | info->tcpi_sacked = sk->sk_max_ack_backlog; | |
2943 | return; | |
2944 | } | |
b369e7fd ED |
2945 | |
2946 | slow = lock_sock_fast(sk); | |
2947 | ||
6687e988 | 2948 | info->tcpi_ca_state = icsk->icsk_ca_state; |
463c84b9 | 2949 | info->tcpi_retransmits = icsk->icsk_retransmits; |
6687e988 | 2950 | info->tcpi_probes = icsk->icsk_probes_out; |
463c84b9 | 2951 | info->tcpi_backoff = icsk->icsk_backoff; |
1da177e4 LT |
2952 | |
2953 | if (tp->rx_opt.tstamp_ok) | |
2954 | info->tcpi_options |= TCPI_OPT_TIMESTAMPS; | |
e60402d0 | 2955 | if (tcp_is_sack(tp)) |
1da177e4 LT |
2956 | info->tcpi_options |= TCPI_OPT_SACK; |
2957 | if (tp->rx_opt.wscale_ok) { | |
2958 | info->tcpi_options |= TCPI_OPT_WSCALE; | |
2959 | info->tcpi_snd_wscale = tp->rx_opt.snd_wscale; | |
2960 | info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale; | |
e905a9ed | 2961 | } |
1da177e4 | 2962 | |
b5c5693b | 2963 | if (tp->ecn_flags & TCP_ECN_OK) |
1da177e4 | 2964 | info->tcpi_options |= TCPI_OPT_ECN; |
b5c5693b ED |
2965 | if (tp->ecn_flags & TCP_ECN_SEEN) |
2966 | info->tcpi_options |= TCPI_OPT_ECN_SEEN; | |
6f73601e YC |
2967 | if (tp->syn_data_acked) |
2968 | info->tcpi_options |= TCPI_OPT_SYN_DATA; | |
1da177e4 | 2969 | |
463c84b9 ACM |
2970 | info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto); |
2971 | info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato); | |
c1b4a7e6 | 2972 | info->tcpi_snd_mss = tp->mss_cache; |
463c84b9 | 2973 | info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss; |
1da177e4 | 2974 | |
ccbf3bfa ED |
2975 | info->tcpi_unacked = tp->packets_out; |
2976 | info->tcpi_sacked = tp->sacked_out; | |
2977 | ||
1da177e4 LT |
2978 | info->tcpi_lost = tp->lost_out; |
2979 | info->tcpi_retrans = tp->retrans_out; | |
1da177e4 | 2980 | |
d635fbe2 | 2981 | now = tcp_jiffies32; |
1da177e4 | 2982 | info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime); |
463c84b9 | 2983 | info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime); |
1da177e4 LT |
2984 | info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp); |
2985 | ||
d83d8461 | 2986 | info->tcpi_pmtu = icsk->icsk_pmtu_cookie; |
1da177e4 | 2987 | info->tcpi_rcv_ssthresh = tp->rcv_ssthresh; |
740b0f18 ED |
2988 | info->tcpi_rtt = tp->srtt_us >> 3; |
2989 | info->tcpi_rttvar = tp->mdev_us >> 2; | |
1da177e4 | 2990 | info->tcpi_snd_ssthresh = tp->snd_ssthresh; |
1da177e4 | 2991 | info->tcpi_advmss = tp->advmss; |
1da177e4 | 2992 | |
645f4c6f | 2993 | info->tcpi_rcv_rtt = tp->rcv_rtt_est.rtt_us >> 3; |
1da177e4 LT |
2994 | info->tcpi_rcv_space = tp->rcvq_space.space; |
2995 | ||
2996 | info->tcpi_total_retrans = tp->total_retrans; | |
977cb0ec | 2997 | |
f522a5fc ED |
2998 | info->tcpi_bytes_acked = tp->bytes_acked; |
2999 | info->tcpi_bytes_received = tp->bytes_received; | |
67db3e4b | 3000 | info->tcpi_notsent_bytes = max_t(int, 0, tp->write_seq - tp->snd_nxt); |
efd90174 | 3001 | tcp_get_info_chrono_stats(tp, info); |
67db3e4b | 3002 | |
2efd055c MRL |
3003 | info->tcpi_segs_out = tp->segs_out; |
3004 | info->tcpi_segs_in = tp->segs_in; | |
cd9b2660 | 3005 | |
cd9b2660 | 3006 | info->tcpi_min_rtt = tcp_min_rtt(tp); |
a44d6eac MKL |
3007 | info->tcpi_data_segs_in = tp->data_segs_in; |
3008 | info->tcpi_data_segs_out = tp->data_segs_out; | |
eb8329e0 YC |
3009 | |
3010 | info->tcpi_delivery_rate_app_limited = tp->rate_app_limited ? 1 : 0; | |
0263598c WW |
3011 | rate64 = tcp_compute_delivery_rate(tp); |
3012 | if (rate64) | |
f522a5fc | 3013 | info->tcpi_delivery_rate = rate64; |
b369e7fd | 3014 | unlock_sock_fast(sk, slow); |
1da177e4 | 3015 | } |
1da177e4 LT |
3016 | EXPORT_SYMBOL_GPL(tcp_get_info); |
3017 | ||
1c885808 FY |
3018 | struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk) |
3019 | { | |
3020 | const struct tcp_sock *tp = tcp_sk(sk); | |
3021 | struct sk_buff *stats; | |
3022 | struct tcp_info info; | |
bb7c19f9 WW |
3023 | u64 rate64; |
3024 | u32 rate; | |
1c885808 | 3025 | |
bb7c19f9 WW |
3026 | stats = alloc_skb(7 * nla_total_size_64bit(sizeof(u64)) + |
3027 | 3 * nla_total_size(sizeof(u32)) + | |
3028 | 2 * nla_total_size(sizeof(u8)), GFP_ATOMIC); | |
1c885808 FY |
3029 | if (!stats) |
3030 | return NULL; | |
3031 | ||
3032 | tcp_get_info_chrono_stats(tp, &info); | |
3033 | nla_put_u64_64bit(stats, TCP_NLA_BUSY, | |
3034 | info.tcpi_busy_time, TCP_NLA_PAD); | |
3035 | nla_put_u64_64bit(stats, TCP_NLA_RWND_LIMITED, | |
3036 | info.tcpi_rwnd_limited, TCP_NLA_PAD); | |
3037 | nla_put_u64_64bit(stats, TCP_NLA_SNDBUF_LIMITED, | |
3038 | info.tcpi_sndbuf_limited, TCP_NLA_PAD); | |
7e98102f YC |
3039 | nla_put_u64_64bit(stats, TCP_NLA_DATA_SEGS_OUT, |
3040 | tp->data_segs_out, TCP_NLA_PAD); | |
3041 | nla_put_u64_64bit(stats, TCP_NLA_TOTAL_RETRANS, | |
3042 | tp->total_retrans, TCP_NLA_PAD); | |
bb7c19f9 WW |
3043 | |
3044 | rate = READ_ONCE(sk->sk_pacing_rate); | |
3045 | rate64 = rate != ~0U ? rate : ~0ULL; | |
3046 | nla_put_u64_64bit(stats, TCP_NLA_PACING_RATE, rate64, TCP_NLA_PAD); | |
3047 | ||
3048 | rate64 = tcp_compute_delivery_rate(tp); | |
3049 | nla_put_u64_64bit(stats, TCP_NLA_DELIVERY_RATE, rate64, TCP_NLA_PAD); | |
3050 | ||
3051 | nla_put_u32(stats, TCP_NLA_SND_CWND, tp->snd_cwnd); | |
3052 | nla_put_u32(stats, TCP_NLA_REORDERING, tp->reordering); | |
3053 | nla_put_u32(stats, TCP_NLA_MIN_RTT, tcp_min_rtt(tp)); | |
3054 | ||
3055 | nla_put_u8(stats, TCP_NLA_RECUR_RETRANS, inet_csk(sk)->icsk_retransmits); | |
3056 | nla_put_u8(stats, TCP_NLA_DELIVERY_RATE_APP_LMT, !!tp->rate_app_limited); | |
1c885808 FY |
3057 | return stats; |
3058 | } | |
3059 | ||
3fdadf7d DM |
3060 | static int do_tcp_getsockopt(struct sock *sk, int level, |
3061 | int optname, char __user *optval, int __user *optlen) | |
1da177e4 | 3062 | { |
295f7324 | 3063 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 3064 | struct tcp_sock *tp = tcp_sk(sk); |
6fa25166 | 3065 | struct net *net = sock_net(sk); |
1da177e4 LT |
3066 | int val, len; |
3067 | ||
1da177e4 LT |
3068 | if (get_user(len, optlen)) |
3069 | return -EFAULT; | |
3070 | ||
3071 | len = min_t(unsigned int, len, sizeof(int)); | |
3072 | ||
3073 | if (len < 0) | |
3074 | return -EINVAL; | |
3075 | ||
3076 | switch (optname) { | |
3077 | case TCP_MAXSEG: | |
c1b4a7e6 | 3078 | val = tp->mss_cache; |
1da177e4 LT |
3079 | if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) |
3080 | val = tp->rx_opt.user_mss; | |
5e6a3ce6 PE |
3081 | if (tp->repair) |
3082 | val = tp->rx_opt.mss_clamp; | |
1da177e4 LT |
3083 | break; |
3084 | case TCP_NODELAY: | |
3085 | val = !!(tp->nonagle&TCP_NAGLE_OFF); | |
3086 | break; | |
3087 | case TCP_CORK: | |
3088 | val = !!(tp->nonagle&TCP_NAGLE_CORK); | |
3089 | break; | |
3090 | case TCP_KEEPIDLE: | |
df19a626 | 3091 | val = keepalive_time_when(tp) / HZ; |
1da177e4 LT |
3092 | break; |
3093 | case TCP_KEEPINTVL: | |
df19a626 | 3094 | val = keepalive_intvl_when(tp) / HZ; |
1da177e4 LT |
3095 | break; |
3096 | case TCP_KEEPCNT: | |
df19a626 | 3097 | val = keepalive_probes(tp); |
1da177e4 LT |
3098 | break; |
3099 | case TCP_SYNCNT: | |
6fa25166 | 3100 | val = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries; |
1da177e4 LT |
3101 | break; |
3102 | case TCP_LINGER2: | |
3103 | val = tp->linger2; | |
3104 | if (val >= 0) | |
1e579caa | 3105 | val = (val ? : net->ipv4.sysctl_tcp_fin_timeout) / HZ; |
1da177e4 LT |
3106 | break; |
3107 | case TCP_DEFER_ACCEPT: | |
b103cf34 JA |
3108 | val = retrans_to_secs(icsk->icsk_accept_queue.rskq_defer_accept, |
3109 | TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ); | |
1da177e4 LT |
3110 | break; |
3111 | case TCP_WINDOW_CLAMP: | |
3112 | val = tp->window_clamp; | |
3113 | break; | |
3114 | case TCP_INFO: { | |
3115 | struct tcp_info info; | |
3116 | ||
3117 | if (get_user(len, optlen)) | |
3118 | return -EFAULT; | |
3119 | ||
3120 | tcp_get_info(sk, &info); | |
3121 | ||
3122 | len = min_t(unsigned int, len, sizeof(info)); | |
3123 | if (put_user(len, optlen)) | |
3124 | return -EFAULT; | |
3125 | if (copy_to_user(optval, &info, len)) | |
3126 | return -EFAULT; | |
3127 | return 0; | |
3128 | } | |
6e9250f5 ED |
3129 | case TCP_CC_INFO: { |
3130 | const struct tcp_congestion_ops *ca_ops; | |
3131 | union tcp_cc_info info; | |
3132 | size_t sz = 0; | |
3133 | int attr; | |
3134 | ||
3135 | if (get_user(len, optlen)) | |
3136 | return -EFAULT; | |
3137 | ||
3138 | ca_ops = icsk->icsk_ca_ops; | |
3139 | if (ca_ops && ca_ops->get_info) | |
3140 | sz = ca_ops->get_info(sk, ~0U, &attr, &info); | |
3141 | ||
3142 | len = min_t(unsigned int, len, sz); | |
3143 | if (put_user(len, optlen)) | |
3144 | return -EFAULT; | |
3145 | if (copy_to_user(optval, &info, len)) | |
3146 | return -EFAULT; | |
3147 | return 0; | |
3148 | } | |
1da177e4 | 3149 | case TCP_QUICKACK: |
295f7324 | 3150 | val = !icsk->icsk_ack.pingpong; |
1da177e4 | 3151 | break; |
5f8ef48d SH |
3152 | |
3153 | case TCP_CONGESTION: | |
3154 | if (get_user(len, optlen)) | |
3155 | return -EFAULT; | |
3156 | len = min_t(unsigned int, len, TCP_CA_NAME_MAX); | |
3157 | if (put_user(len, optlen)) | |
3158 | return -EFAULT; | |
6687e988 | 3159 | if (copy_to_user(optval, icsk->icsk_ca_ops->name, len)) |
5f8ef48d SH |
3160 | return -EFAULT; |
3161 | return 0; | |
e56fb50f | 3162 | |
734942cc DW |
3163 | case TCP_ULP: |
3164 | if (get_user(len, optlen)) | |
3165 | return -EFAULT; | |
3166 | len = min_t(unsigned int, len, TCP_ULP_NAME_MAX); | |
d97af30f DW |
3167 | if (!icsk->icsk_ulp_ops) { |
3168 | if (put_user(0, optlen)) | |
3169 | return -EFAULT; | |
3170 | return 0; | |
3171 | } | |
734942cc DW |
3172 | if (put_user(len, optlen)) |
3173 | return -EFAULT; | |
3174 | if (copy_to_user(optval, icsk->icsk_ulp_ops->name, len)) | |
3175 | return -EFAULT; | |
3176 | return 0; | |
3177 | ||
1fba70e5 YC |
3178 | case TCP_FASTOPEN_KEY: { |
3179 | __u8 key[TCP_FASTOPEN_KEY_LENGTH]; | |
3180 | struct tcp_fastopen_context *ctx; | |
3181 | ||
3182 | if (get_user(len, optlen)) | |
3183 | return -EFAULT; | |
3184 | ||
3185 | rcu_read_lock(); | |
3186 | ctx = rcu_dereference(icsk->icsk_accept_queue.fastopenq.ctx); | |
3187 | if (ctx) | |
3188 | memcpy(key, ctx->key, sizeof(key)); | |
3189 | else | |
3190 | len = 0; | |
3191 | rcu_read_unlock(); | |
3192 | ||
3193 | len = min_t(unsigned int, len, sizeof(key)); | |
3194 | if (put_user(len, optlen)) | |
3195 | return -EFAULT; | |
3196 | if (copy_to_user(optval, key, len)) | |
3197 | return -EFAULT; | |
3198 | return 0; | |
3199 | } | |
3c0fef0b JH |
3200 | case TCP_THIN_LINEAR_TIMEOUTS: |
3201 | val = tp->thin_lto; | |
3202 | break; | |
4a7f6009 | 3203 | |
3c0fef0b | 3204 | case TCP_THIN_DUPACK: |
4a7f6009 | 3205 | val = 0; |
3c0fef0b | 3206 | break; |
dca43c75 | 3207 | |
ee995283 PE |
3208 | case TCP_REPAIR: |
3209 | val = tp->repair; | |
3210 | break; | |
3211 | ||
3212 | case TCP_REPAIR_QUEUE: | |
3213 | if (tp->repair) | |
3214 | val = tp->repair_queue; | |
3215 | else | |
3216 | return -EINVAL; | |
3217 | break; | |
3218 | ||
b1ed4c4f AV |
3219 | case TCP_REPAIR_WINDOW: { |
3220 | struct tcp_repair_window opt; | |
3221 | ||
3222 | if (get_user(len, optlen)) | |
3223 | return -EFAULT; | |
3224 | ||
3225 | if (len != sizeof(opt)) | |
3226 | return -EINVAL; | |
3227 | ||
3228 | if (!tp->repair) | |
3229 | return -EPERM; | |
3230 | ||
3231 | opt.snd_wl1 = tp->snd_wl1; | |
3232 | opt.snd_wnd = tp->snd_wnd; | |
3233 | opt.max_window = tp->max_window; | |
3234 | opt.rcv_wnd = tp->rcv_wnd; | |
3235 | opt.rcv_wup = tp->rcv_wup; | |
3236 | ||
3237 | if (copy_to_user(optval, &opt, len)) | |
3238 | return -EFAULT; | |
3239 | return 0; | |
3240 | } | |
ee995283 PE |
3241 | case TCP_QUEUE_SEQ: |
3242 | if (tp->repair_queue == TCP_SEND_QUEUE) | |
3243 | val = tp->write_seq; | |
3244 | else if (tp->repair_queue == TCP_RECV_QUEUE) | |
3245 | val = tp->rcv_nxt; | |
3246 | else | |
3247 | return -EINVAL; | |
3248 | break; | |
3249 | ||
dca43c75 JC |
3250 | case TCP_USER_TIMEOUT: |
3251 | val = jiffies_to_msecs(icsk->icsk_user_timeout); | |
3252 | break; | |
1536e285 KN |
3253 | |
3254 | case TCP_FASTOPEN: | |
0536fcc0 | 3255 | val = icsk->icsk_accept_queue.fastopenq.max_qlen; |
1536e285 KN |
3256 | break; |
3257 | ||
19f6d3f3 WW |
3258 | case TCP_FASTOPEN_CONNECT: |
3259 | val = tp->fastopen_connect; | |
3260 | break; | |
3261 | ||
71c02379 CP |
3262 | case TCP_FASTOPEN_NO_COOKIE: |
3263 | val = tp->fastopen_no_cookie; | |
3264 | break; | |
3265 | ||
93be6ce0 | 3266 | case TCP_TIMESTAMP: |
9a568de4 | 3267 | val = tcp_time_stamp_raw() + tp->tsoffset; |
93be6ce0 | 3268 | break; |
c9bee3b7 ED |
3269 | case TCP_NOTSENT_LOWAT: |
3270 | val = tp->notsent_lowat; | |
3271 | break; | |
cd8ae852 ED |
3272 | case TCP_SAVE_SYN: |
3273 | val = tp->save_syn; | |
3274 | break; | |
3275 | case TCP_SAVED_SYN: { | |
3276 | if (get_user(len, optlen)) | |
3277 | return -EFAULT; | |
3278 | ||
3279 | lock_sock(sk); | |
3280 | if (tp->saved_syn) { | |
aea0929e EM |
3281 | if (len < tp->saved_syn[0]) { |
3282 | if (put_user(tp->saved_syn[0], optlen)) { | |
3283 | release_sock(sk); | |
3284 | return -EFAULT; | |
3285 | } | |
3286 | release_sock(sk); | |
3287 | return -EINVAL; | |
3288 | } | |
3289 | len = tp->saved_syn[0]; | |
cd8ae852 ED |
3290 | if (put_user(len, optlen)) { |
3291 | release_sock(sk); | |
3292 | return -EFAULT; | |
3293 | } | |
3294 | if (copy_to_user(optval, tp->saved_syn + 1, len)) { | |
3295 | release_sock(sk); | |
3296 | return -EFAULT; | |
3297 | } | |
3298 | tcp_saved_syn_free(tp); | |
3299 | release_sock(sk); | |
3300 | } else { | |
3301 | release_sock(sk); | |
3302 | len = 0; | |
3303 | if (put_user(len, optlen)) | |
3304 | return -EFAULT; | |
3305 | } | |
3306 | return 0; | |
3307 | } | |
1da177e4 LT |
3308 | default: |
3309 | return -ENOPROTOOPT; | |
3ff50b79 | 3310 | } |
1da177e4 LT |
3311 | |
3312 | if (put_user(len, optlen)) | |
3313 | return -EFAULT; | |
3314 | if (copy_to_user(optval, &val, len)) | |
3315 | return -EFAULT; | |
3316 | return 0; | |
3317 | } | |
3318 | ||
3fdadf7d DM |
3319 | int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval, |
3320 | int __user *optlen) | |
3321 | { | |
3322 | struct inet_connection_sock *icsk = inet_csk(sk); | |
3323 | ||
3324 | if (level != SOL_TCP) | |
3325 | return icsk->icsk_af_ops->getsockopt(sk, level, optname, | |
3326 | optval, optlen); | |
3327 | return do_tcp_getsockopt(sk, level, optname, optval, optlen); | |
3328 | } | |
4bc2f18b | 3329 | EXPORT_SYMBOL(tcp_getsockopt); |
3fdadf7d DM |
3330 | |
3331 | #ifdef CONFIG_COMPAT | |
543d9cfe ACM |
3332 | int compat_tcp_getsockopt(struct sock *sk, int level, int optname, |
3333 | char __user *optval, int __user *optlen) | |
3fdadf7d | 3334 | { |
dec73ff0 ACM |
3335 | if (level != SOL_TCP) |
3336 | return inet_csk_compat_getsockopt(sk, level, optname, | |
3337 | optval, optlen); | |
3fdadf7d DM |
3338 | return do_tcp_getsockopt(sk, level, optname, optval, optlen); |
3339 | } | |
543d9cfe | 3340 | EXPORT_SYMBOL(compat_tcp_getsockopt); |
3fdadf7d | 3341 | #endif |
1da177e4 | 3342 | |
cfb6eeb4 | 3343 | #ifdef CONFIG_TCP_MD5SIG |
349ce993 | 3344 | static DEFINE_PER_CPU(struct tcp_md5sig_pool, tcp_md5sig_pool); |
71cea17e | 3345 | static DEFINE_MUTEX(tcp_md5sig_mutex); |
349ce993 | 3346 | static bool tcp_md5sig_pool_populated = false; |
cfb6eeb4 | 3347 | |
71cea17e | 3348 | static void __tcp_alloc_md5sig_pool(void) |
cfb6eeb4 | 3349 | { |
cf80e0e4 | 3350 | struct crypto_ahash *hash; |
cfb6eeb4 | 3351 | int cpu; |
cfb6eeb4 | 3352 | |
cf80e0e4 | 3353 | hash = crypto_alloc_ahash("md5", 0, CRYPTO_ALG_ASYNC); |
70477371 | 3354 | if (IS_ERR(hash)) |
cf80e0e4 HX |
3355 | return; |
3356 | ||
cfb6eeb4 | 3357 | for_each_possible_cpu(cpu) { |
19689e38 | 3358 | void *scratch = per_cpu(tcp_md5sig_pool, cpu).scratch; |
cf80e0e4 | 3359 | struct ahash_request *req; |
cfb6eeb4 | 3360 | |
19689e38 ED |
3361 | if (!scratch) { |
3362 | scratch = kmalloc_node(sizeof(union tcp_md5sum_block) + | |
3363 | sizeof(struct tcphdr), | |
3364 | GFP_KERNEL, | |
3365 | cpu_to_node(cpu)); | |
3366 | if (!scratch) | |
3367 | return; | |
3368 | per_cpu(tcp_md5sig_pool, cpu).scratch = scratch; | |
3369 | } | |
cf80e0e4 HX |
3370 | if (per_cpu(tcp_md5sig_pool, cpu).md5_req) |
3371 | continue; | |
3372 | ||
3373 | req = ahash_request_alloc(hash, GFP_KERNEL); | |
3374 | if (!req) | |
3375 | return; | |
3376 | ||
3377 | ahash_request_set_callback(req, 0, NULL, NULL); | |
3378 | ||
3379 | per_cpu(tcp_md5sig_pool, cpu).md5_req = req; | |
cfb6eeb4 | 3380 | } |
349ce993 ED |
3381 | /* before setting tcp_md5sig_pool_populated, we must commit all writes |
3382 | * to memory. See smp_rmb() in tcp_get_md5sig_pool() | |
71cea17e ED |
3383 | */ |
3384 | smp_wmb(); | |
349ce993 | 3385 | tcp_md5sig_pool_populated = true; |
cfb6eeb4 YH |
3386 | } |
3387 | ||
71cea17e | 3388 | bool tcp_alloc_md5sig_pool(void) |
cfb6eeb4 | 3389 | { |
349ce993 | 3390 | if (unlikely(!tcp_md5sig_pool_populated)) { |
71cea17e ED |
3391 | mutex_lock(&tcp_md5sig_mutex); |
3392 | ||
349ce993 | 3393 | if (!tcp_md5sig_pool_populated) |
71cea17e ED |
3394 | __tcp_alloc_md5sig_pool(); |
3395 | ||
3396 | mutex_unlock(&tcp_md5sig_mutex); | |
cfb6eeb4 | 3397 | } |
349ce993 | 3398 | return tcp_md5sig_pool_populated; |
cfb6eeb4 | 3399 | } |
cfb6eeb4 YH |
3400 | EXPORT_SYMBOL(tcp_alloc_md5sig_pool); |
3401 | ||
35790c04 ED |
3402 | |
3403 | /** | |
3404 | * tcp_get_md5sig_pool - get md5sig_pool for this user | |
3405 | * | |
3406 | * We use percpu structure, so if we succeed, we exit with preemption | |
3407 | * and BH disabled, to make sure another thread or softirq handling | |
3408 | * wont try to get same context. | |
3409 | */ | |
3410 | struct tcp_md5sig_pool *tcp_get_md5sig_pool(void) | |
cfb6eeb4 | 3411 | { |
35790c04 | 3412 | local_bh_disable(); |
cfb6eeb4 | 3413 | |
349ce993 ED |
3414 | if (tcp_md5sig_pool_populated) { |
3415 | /* coupled with smp_wmb() in __tcp_alloc_md5sig_pool() */ | |
3416 | smp_rmb(); | |
3417 | return this_cpu_ptr(&tcp_md5sig_pool); | |
3418 | } | |
35790c04 ED |
3419 | local_bh_enable(); |
3420 | return NULL; | |
3421 | } | |
3422 | EXPORT_SYMBOL(tcp_get_md5sig_pool); | |
cfb6eeb4 | 3423 | |
49a72dfb | 3424 | int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp, |
cf533ea5 | 3425 | const struct sk_buff *skb, unsigned int header_len) |
49a72dfb AL |
3426 | { |
3427 | struct scatterlist sg; | |
3428 | const struct tcphdr *tp = tcp_hdr(skb); | |
cf80e0e4 | 3429 | struct ahash_request *req = hp->md5_req; |
95c96174 ED |
3430 | unsigned int i; |
3431 | const unsigned int head_data_len = skb_headlen(skb) > header_len ? | |
3432 | skb_headlen(skb) - header_len : 0; | |
49a72dfb | 3433 | const struct skb_shared_info *shi = skb_shinfo(skb); |
d7fd1b57 | 3434 | struct sk_buff *frag_iter; |
49a72dfb AL |
3435 | |
3436 | sg_init_table(&sg, 1); | |
3437 | ||
3438 | sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len); | |
cf80e0e4 HX |
3439 | ahash_request_set_crypt(req, &sg, NULL, head_data_len); |
3440 | if (crypto_ahash_update(req)) | |
49a72dfb AL |
3441 | return 1; |
3442 | ||
3443 | for (i = 0; i < shi->nr_frags; ++i) { | |
3444 | const struct skb_frag_struct *f = &shi->frags[i]; | |
54d27fcb ED |
3445 | unsigned int offset = f->page_offset; |
3446 | struct page *page = skb_frag_page(f) + (offset >> PAGE_SHIFT); | |
3447 | ||
3448 | sg_set_page(&sg, page, skb_frag_size(f), | |
3449 | offset_in_page(offset)); | |
cf80e0e4 HX |
3450 | ahash_request_set_crypt(req, &sg, NULL, skb_frag_size(f)); |
3451 | if (crypto_ahash_update(req)) | |
49a72dfb AL |
3452 | return 1; |
3453 | } | |
3454 | ||
d7fd1b57 ED |
3455 | skb_walk_frags(skb, frag_iter) |
3456 | if (tcp_md5_hash_skb_data(hp, frag_iter, 0)) | |
3457 | return 1; | |
3458 | ||
49a72dfb AL |
3459 | return 0; |
3460 | } | |
49a72dfb AL |
3461 | EXPORT_SYMBOL(tcp_md5_hash_skb_data); |
3462 | ||
cf533ea5 | 3463 | int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, const struct tcp_md5sig_key *key) |
49a72dfb AL |
3464 | { |
3465 | struct scatterlist sg; | |
3466 | ||
3467 | sg_init_one(&sg, key->key, key->keylen); | |
cf80e0e4 HX |
3468 | ahash_request_set_crypt(hp->md5_req, &sg, NULL, key->keylen); |
3469 | return crypto_ahash_update(hp->md5_req); | |
49a72dfb | 3470 | } |
49a72dfb AL |
3471 | EXPORT_SYMBOL(tcp_md5_hash_key); |
3472 | ||
cfb6eeb4 YH |
3473 | #endif |
3474 | ||
4ac02bab AK |
3475 | void tcp_done(struct sock *sk) |
3476 | { | |
8336886f JC |
3477 | struct request_sock *req = tcp_sk(sk)->fastopen_rsk; |
3478 | ||
5a5f3a8d | 3479 | if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV) |
c10d9310 | 3480 | TCP_INC_STATS(sock_net(sk), TCP_MIB_ATTEMPTFAILS); |
4ac02bab AK |
3481 | |
3482 | tcp_set_state(sk, TCP_CLOSE); | |
3483 | tcp_clear_xmit_timers(sk); | |
00db4124 | 3484 | if (req) |
8336886f | 3485 | reqsk_fastopen_remove(sk, req, false); |
4ac02bab AK |
3486 | |
3487 | sk->sk_shutdown = SHUTDOWN_MASK; | |
3488 | ||
3489 | if (!sock_flag(sk, SOCK_DEAD)) | |
3490 | sk->sk_state_change(sk); | |
3491 | else | |
3492 | inet_csk_destroy_sock(sk); | |
3493 | } | |
3494 | EXPORT_SYMBOL_GPL(tcp_done); | |
3495 | ||
c1e64e29 LC |
3496 | int tcp_abort(struct sock *sk, int err) |
3497 | { | |
3498 | if (!sk_fullsock(sk)) { | |
07f6f4a3 ED |
3499 | if (sk->sk_state == TCP_NEW_SYN_RECV) { |
3500 | struct request_sock *req = inet_reqsk(sk); | |
3501 | ||
3502 | local_bh_disable(); | |
3503 | inet_csk_reqsk_queue_drop_and_put(req->rsk_listener, | |
3504 | req); | |
3505 | local_bh_enable(); | |
3506 | return 0; | |
3507 | } | |
c1e64e29 LC |
3508 | return -EOPNOTSUPP; |
3509 | } | |
3510 | ||
3511 | /* Don't race with userspace socket closes such as tcp_close. */ | |
3512 | lock_sock(sk); | |
3513 | ||
2010b93e LC |
3514 | if (sk->sk_state == TCP_LISTEN) { |
3515 | tcp_set_state(sk, TCP_CLOSE); | |
3516 | inet_csk_listen_stop(sk); | |
3517 | } | |
3518 | ||
c1e64e29 LC |
3519 | /* Don't race with BH socket closes such as inet_csk_listen_stop. */ |
3520 | local_bh_disable(); | |
3521 | bh_lock_sock(sk); | |
3522 | ||
3523 | if (!sock_flag(sk, SOCK_DEAD)) { | |
3524 | sk->sk_err = err; | |
3525 | /* This barrier is coupled with smp_rmb() in tcp_poll() */ | |
3526 | smp_wmb(); | |
3527 | sk->sk_error_report(sk); | |
3528 | if (tcp_need_reset(sk->sk_state)) | |
3529 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
3530 | tcp_done(sk); | |
3531 | } | |
3532 | ||
3533 | bh_unlock_sock(sk); | |
3534 | local_bh_enable(); | |
3535 | release_sock(sk); | |
c1e64e29 LC |
3536 | return 0; |
3537 | } | |
3538 | EXPORT_SYMBOL_GPL(tcp_abort); | |
3539 | ||
5f8ef48d | 3540 | extern struct tcp_congestion_ops tcp_reno; |
1da177e4 LT |
3541 | |
3542 | static __initdata unsigned long thash_entries; | |
3543 | static int __init set_thash_entries(char *str) | |
3544 | { | |
413c27d8 EZ |
3545 | ssize_t ret; |
3546 | ||
1da177e4 LT |
3547 | if (!str) |
3548 | return 0; | |
413c27d8 EZ |
3549 | |
3550 | ret = kstrtoul(str, 0, &thash_entries); | |
3551 | if (ret) | |
3552 | return 0; | |
3553 | ||
1da177e4 LT |
3554 | return 1; |
3555 | } | |
3556 | __setup("thash_entries=", set_thash_entries); | |
3557 | ||
47d7a88c | 3558 | static void __init tcp_init_mem(void) |
4acb4190 | 3559 | { |
b66e91cc ED |
3560 | unsigned long limit = nr_free_buffer_pages() / 16; |
3561 | ||
4acb4190 | 3562 | limit = max(limit, 128UL); |
b66e91cc ED |
3563 | sysctl_tcp_mem[0] = limit / 4 * 3; /* 4.68 % */ |
3564 | sysctl_tcp_mem[1] = limit; /* 6.25 % */ | |
3565 | sysctl_tcp_mem[2] = sysctl_tcp_mem[0] * 2; /* 9.37 % */ | |
4acb4190 GC |
3566 | } |
3567 | ||
1da177e4 LT |
3568 | void __init tcp_init(void) |
3569 | { | |
b49960a0 | 3570 | int max_rshare, max_wshare, cnt; |
b2d3ea4a | 3571 | unsigned long limit; |
074b8517 | 3572 | unsigned int i; |
1da177e4 | 3573 | |
b2d3ea4a ED |
3574 | BUILD_BUG_ON(sizeof(struct tcp_skb_cb) > |
3575 | FIELD_SIZEOF(struct sk_buff, cb)); | |
1da177e4 | 3576 | |
908c7f19 TH |
3577 | percpu_counter_init(&tcp_sockets_allocated, 0, GFP_KERNEL); |
3578 | percpu_counter_init(&tcp_orphan_count, 0, GFP_KERNEL); | |
1946e672 | 3579 | inet_hashinfo_init(&tcp_hashinfo); |
27da6d37 MKL |
3580 | inet_hashinfo2_init(&tcp_hashinfo, "tcp_listen_portaddr_hash", |
3581 | thash_entries, 21, /* one slot per 2 MB*/ | |
3582 | 0, 64 * 1024); | |
6e04e021 ACM |
3583 | tcp_hashinfo.bind_bucket_cachep = |
3584 | kmem_cache_create("tcp_bind_bucket", | |
3585 | sizeof(struct inet_bind_bucket), 0, | |
20c2df83 | 3586 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3587 | |
1da177e4 LT |
3588 | /* Size and allocate the main established and bind bucket |
3589 | * hash tables. | |
3590 | * | |
3591 | * The methodology is similar to that of the buffer cache. | |
3592 | */ | |
6e04e021 | 3593 | tcp_hashinfo.ehash = |
1da177e4 | 3594 | alloc_large_system_hash("TCP established", |
0f7ff927 | 3595 | sizeof(struct inet_ehash_bucket), |
1da177e4 | 3596 | thash_entries, |
fd90b29d | 3597 | 17, /* one slot per 128 KB of memory */ |
9e950efa | 3598 | 0, |
1da177e4 | 3599 | NULL, |
f373b53b | 3600 | &tcp_hashinfo.ehash_mask, |
31fe62b9 | 3601 | 0, |
0ccfe618 | 3602 | thash_entries ? 0 : 512 * 1024); |
05dbc7b5 | 3603 | for (i = 0; i <= tcp_hashinfo.ehash_mask; i++) |
3ab5aee7 | 3604 | INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i); |
05dbc7b5 | 3605 | |
230140cf ED |
3606 | if (inet_ehash_locks_alloc(&tcp_hashinfo)) |
3607 | panic("TCP: failed to alloc ehash_locks"); | |
6e04e021 | 3608 | tcp_hashinfo.bhash = |
1da177e4 | 3609 | alloc_large_system_hash("TCP bind", |
0f7ff927 | 3610 | sizeof(struct inet_bind_hashbucket), |
f373b53b | 3611 | tcp_hashinfo.ehash_mask + 1, |
fd90b29d | 3612 | 17, /* one slot per 128 KB of memory */ |
9e950efa | 3613 | 0, |
6e04e021 | 3614 | &tcp_hashinfo.bhash_size, |
1da177e4 | 3615 | NULL, |
31fe62b9 | 3616 | 0, |
1da177e4 | 3617 | 64 * 1024); |
074b8517 | 3618 | tcp_hashinfo.bhash_size = 1U << tcp_hashinfo.bhash_size; |
6e04e021 ACM |
3619 | for (i = 0; i < tcp_hashinfo.bhash_size; i++) { |
3620 | spin_lock_init(&tcp_hashinfo.bhash[i].lock); | |
3621 | INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain); | |
1da177e4 LT |
3622 | } |
3623 | ||
c5ed63d6 ED |
3624 | |
3625 | cnt = tcp_hashinfo.ehash_mask + 1; | |
c5ed63d6 | 3626 | sysctl_tcp_max_orphans = cnt / 2; |
1da177e4 | 3627 | |
a4fe34bf | 3628 | tcp_init_mem(); |
c43b874d | 3629 | /* Set per-socket limits to no more than 1/128 the pressure threshold */ |
5fb84b14 | 3630 | limit = nr_free_buffer_pages() << (PAGE_SHIFT - 7); |
b49960a0 ED |
3631 | max_wshare = min(4UL*1024*1024, limit); |
3632 | max_rshare = min(6UL*1024*1024, limit); | |
7b4f4b5e | 3633 | |
356d1833 ED |
3634 | init_net.ipv4.sysctl_tcp_wmem[0] = SK_MEM_QUANTUM; |
3635 | init_net.ipv4.sysctl_tcp_wmem[1] = 16*1024; | |
3636 | init_net.ipv4.sysctl_tcp_wmem[2] = max(64*1024, max_wshare); | |
7b4f4b5e | 3637 | |
356d1833 ED |
3638 | init_net.ipv4.sysctl_tcp_rmem[0] = SK_MEM_QUANTUM; |
3639 | init_net.ipv4.sysctl_tcp_rmem[1] = 87380; | |
3640 | init_net.ipv4.sysctl_tcp_rmem[2] = max(87380, max_rshare); | |
1da177e4 | 3641 | |
afd46503 | 3642 | pr_info("Hash tables configured (established %u bind %u)\n", |
058bd4d2 | 3643 | tcp_hashinfo.ehash_mask + 1, tcp_hashinfo.bhash_size); |
317a76f9 | 3644 | |
1946e672 | 3645 | tcp_v4_init(); |
51c5d0c4 | 3646 | tcp_metrics_init(); |
55d8694f | 3647 | BUG_ON(tcp_register_congestion_control(&tcp_reno) != 0); |
46d3ceab | 3648 | tcp_tasklet_init(); |
1da177e4 | 3649 | } |