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); |
0a6b2a1d | 456 | sk->sk_route_forced_caps = NETIF_F_GSO; |
900f65d3 NC |
457 | } |
458 | EXPORT_SYMBOL(tcp_init_sock); | |
459 | ||
27204aaa WW |
460 | void tcp_init_transfer(struct sock *sk, int bpf_op) |
461 | { | |
462 | struct inet_connection_sock *icsk = inet_csk(sk); | |
463 | ||
464 | tcp_mtup_init(sk); | |
465 | icsk->icsk_af_ops->rebuild_header(sk); | |
466 | tcp_init_metrics(sk); | |
de525be2 | 467 | tcp_call_bpf(sk, bpf_op, 0, NULL); |
27204aaa WW |
468 | tcp_init_congestion_control(sk); |
469 | tcp_init_buffer_space(sk); | |
470 | } | |
471 | ||
4e8cc228 | 472 | static void tcp_tx_timestamp(struct sock *sk, u16 tsflags) |
4ed2d765 | 473 | { |
4e8cc228 ED |
474 | struct sk_buff *skb = tcp_write_queue_tail(sk); |
475 | ||
ad02c4f5 | 476 | if (tsflags && skb) { |
f066e2b0 | 477 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
6b084928 | 478 | struct tcp_skb_cb *tcb = TCP_SKB_CB(skb); |
4ed2d765 | 479 | |
c14ac945 | 480 | sock_tx_timestamp(sk, tsflags, &shinfo->tx_flags); |
0a2cf20c SHY |
481 | if (tsflags & SOF_TIMESTAMPING_TX_ACK) |
482 | tcb->txstamp_ack = 1; | |
483 | if (tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) | |
f066e2b0 WB |
484 | shinfo->tskey = TCP_SKB_CB(skb)->seq + skb->len - 1; |
485 | } | |
4ed2d765 WB |
486 | } |
487 | ||
8934ce2f JF |
488 | static inline bool tcp_stream_is_readable(const struct tcp_sock *tp, |
489 | int target, struct sock *sk) | |
490 | { | |
491 | return (tp->rcv_nxt - tp->copied_seq >= target) || | |
492 | (sk->sk_prot->stream_memory_read ? | |
493 | sk->sk_prot->stream_memory_read(sk) : false); | |
494 | } | |
495 | ||
1da177e4 LT |
496 | /* |
497 | * Wait for a TCP event. | |
498 | * | |
499 | * Note that we don't need to lock the socket, as the upper poll layers | |
500 | * take care of normal races (between the test and the event) and we don't | |
501 | * go look at any of the socket buffers directly. | |
502 | */ | |
ade994f4 | 503 | __poll_t tcp_poll(struct file *file, struct socket *sock, poll_table *wait) |
1da177e4 | 504 | { |
ade994f4 | 505 | __poll_t mask; |
1da177e4 | 506 | struct sock *sk = sock->sk; |
cf533ea5 | 507 | const struct tcp_sock *tp = tcp_sk(sk); |
00fd38d9 | 508 | int state; |
1da177e4 | 509 | |
aa395145 | 510 | sock_poll_wait(file, sk_sleep(sk), wait); |
00fd38d9 | 511 | |
986ffdfd | 512 | state = inet_sk_state_load(sk); |
00fd38d9 | 513 | if (state == TCP_LISTEN) |
dc40c7bc | 514 | return inet_csk_listen_poll(sk); |
1da177e4 LT |
515 | |
516 | /* Socket is not locked. We are protected from async events | |
70efce27 WN |
517 | * by poll logic and correct handling of state changes |
518 | * made by other threads is impossible in any case. | |
1da177e4 LT |
519 | */ |
520 | ||
521 | mask = 0; | |
1da177e4 LT |
522 | |
523 | /* | |
a9a08845 | 524 | * EPOLLHUP is certainly not done right. But poll() doesn't |
1da177e4 LT |
525 | * have a notion of HUP in just one direction, and for a |
526 | * socket the read side is more interesting. | |
527 | * | |
a9a08845 LT |
528 | * Some poll() documentation says that EPOLLHUP is incompatible |
529 | * with the EPOLLOUT/POLLWR flags, so somebody should check this | |
1da177e4 LT |
530 | * all. But careful, it tends to be safer to return too many |
531 | * bits than too few, and you can easily break real applications | |
532 | * if you don't tell them that something has hung up! | |
533 | * | |
534 | * Check-me. | |
535 | * | |
a9a08845 | 536 | * Check number 1. EPOLLHUP is _UNMASKABLE_ event (see UNIX98 and |
1da177e4 LT |
537 | * our fs/select.c). It means that after we received EOF, |
538 | * poll always returns immediately, making impossible poll() on write() | |
a9a08845 | 539 | * in state CLOSE_WAIT. One solution is evident --- to set EPOLLHUP |
1da177e4 LT |
540 | * if and only if shutdown has been made in both directions. |
541 | * Actually, it is interesting to look how Solaris and DUX | |
a9a08845 | 542 | * solve this dilemma. I would prefer, if EPOLLHUP were maskable, |
1da177e4 LT |
543 | * then we could set it on SND_SHUTDOWN. BTW examples given |
544 | * in Stevens' books assume exactly this behaviour, it explains | |
a9a08845 | 545 | * why EPOLLHUP is incompatible with EPOLLOUT. --ANK |
1da177e4 LT |
546 | * |
547 | * NOTE. Check for TCP_CLOSE is added. The goal is to prevent | |
548 | * blocking on fresh not-connected or disconnected socket. --ANK | |
549 | */ | |
00fd38d9 | 550 | if (sk->sk_shutdown == SHUTDOWN_MASK || state == TCP_CLOSE) |
a9a08845 | 551 | mask |= EPOLLHUP; |
1da177e4 | 552 | if (sk->sk_shutdown & RCV_SHUTDOWN) |
a9a08845 | 553 | mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; |
1da177e4 | 554 | |
8336886f | 555 | /* Connected or passive Fast Open socket? */ |
00fd38d9 ED |
556 | if (state != TCP_SYN_SENT && |
557 | (state != TCP_SYN_RECV || tp->fastopen_rsk)) { | |
c7004482 DM |
558 | int target = sock_rcvlowat(sk, 0, INT_MAX); |
559 | ||
560 | if (tp->urg_seq == tp->copied_seq && | |
561 | !sock_flag(sk, SOCK_URGINLINE) && | |
562 | tp->urg_data) | |
b634f875 | 563 | target++; |
c7004482 | 564 | |
8934ce2f | 565 | if (tcp_stream_is_readable(tp, target, sk)) |
a9a08845 | 566 | mask |= EPOLLIN | EPOLLRDNORM; |
1da177e4 LT |
567 | |
568 | if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { | |
64dc6130 | 569 | if (sk_stream_is_writeable(sk)) { |
a9a08845 | 570 | mask |= EPOLLOUT | EPOLLWRNORM; |
1da177e4 | 571 | } else { /* send SIGIO later */ |
9cd3e072 | 572 | sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 LT |
573 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
574 | ||
575 | /* Race breaker. If space is freed after | |
576 | * wspace test but before the flags are set, | |
3c715127 | 577 | * IO signal will be lost. Memory barrier |
578 | * pairs with the input side. | |
1da177e4 | 579 | */ |
3c715127 | 580 | smp_mb__after_atomic(); |
64dc6130 | 581 | if (sk_stream_is_writeable(sk)) |
a9a08845 | 582 | mask |= EPOLLOUT | EPOLLWRNORM; |
1da177e4 | 583 | } |
d84ba638 | 584 | } else |
a9a08845 | 585 | mask |= EPOLLOUT | EPOLLWRNORM; |
1da177e4 LT |
586 | |
587 | if (tp->urg_data & TCP_URG_VALID) | |
a9a08845 | 588 | mask |= EPOLLPRI; |
d68be71e | 589 | } else if (state == TCP_SYN_SENT && inet_sk(sk)->defer_connect) { |
19f6d3f3 | 590 | /* Active TCP fastopen socket with defer_connect |
a9a08845 | 591 | * Return EPOLLOUT so application can call write() |
19f6d3f3 WW |
592 | * in order for kernel to generate SYN+data |
593 | */ | |
a9a08845 | 594 | mask |= EPOLLOUT | EPOLLWRNORM; |
1da177e4 | 595 | } |
a4d25803 TM |
596 | /* This barrier is coupled with smp_wmb() in tcp_reset() */ |
597 | smp_rmb(); | |
4ed2d765 | 598 | if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue)) |
a9a08845 | 599 | mask |= EPOLLERR; |
a4d25803 | 600 | |
1da177e4 LT |
601 | return mask; |
602 | } | |
4bc2f18b | 603 | EXPORT_SYMBOL(tcp_poll); |
1da177e4 LT |
604 | |
605 | int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg) | |
606 | { | |
607 | struct tcp_sock *tp = tcp_sk(sk); | |
608 | int answ; | |
0e71c55c | 609 | bool slow; |
1da177e4 LT |
610 | |
611 | switch (cmd) { | |
612 | case SIOCINQ: | |
613 | if (sk->sk_state == TCP_LISTEN) | |
614 | return -EINVAL; | |
615 | ||
0e71c55c | 616 | slow = lock_sock_fast(sk); |
473bd239 | 617 | answ = tcp_inq(sk); |
0e71c55c | 618 | unlock_sock_fast(sk, slow); |
1da177e4 LT |
619 | break; |
620 | case SIOCATMARK: | |
621 | answ = tp->urg_data && tp->urg_seq == tp->copied_seq; | |
622 | break; | |
623 | case SIOCOUTQ: | |
624 | if (sk->sk_state == TCP_LISTEN) | |
625 | return -EINVAL; | |
626 | ||
627 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) | |
628 | answ = 0; | |
629 | else | |
630 | answ = tp->write_seq - tp->snd_una; | |
631 | break; | |
2f4e1b39 MS |
632 | case SIOCOUTQNSD: |
633 | if (sk->sk_state == TCP_LISTEN) | |
634 | return -EINVAL; | |
635 | ||
636 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) | |
637 | answ = 0; | |
638 | else | |
639 | answ = tp->write_seq - tp->snd_nxt; | |
640 | break; | |
1da177e4 LT |
641 | default: |
642 | return -ENOIOCTLCMD; | |
3ff50b79 | 643 | } |
1da177e4 LT |
644 | |
645 | return put_user(answ, (int __user *)arg); | |
646 | } | |
4bc2f18b | 647 | EXPORT_SYMBOL(tcp_ioctl); |
1da177e4 | 648 | |
1da177e4 LT |
649 | static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb) |
650 | { | |
4de075e0 | 651 | TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH; |
1da177e4 LT |
652 | tp->pushed_seq = tp->write_seq; |
653 | } | |
654 | ||
a2a385d6 | 655 | static inline bool forced_push(const struct tcp_sock *tp) |
1da177e4 LT |
656 | { |
657 | return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1)); | |
658 | } | |
659 | ||
f4a775d1 | 660 | static void skb_entail(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 661 | { |
9e412ba7 | 662 | struct tcp_sock *tp = tcp_sk(sk); |
352d4800 ACM |
663 | struct tcp_skb_cb *tcb = TCP_SKB_CB(skb); |
664 | ||
665 | skb->csum = 0; | |
666 | tcb->seq = tcb->end_seq = tp->write_seq; | |
4de075e0 | 667 | tcb->tcp_flags = TCPHDR_ACK; |
352d4800 | 668 | tcb->sacked = 0; |
f4a775d1 | 669 | __skb_header_release(skb); |
fe067e8a | 670 | tcp_add_write_queue_tail(sk, skb); |
3ab224be HA |
671 | sk->sk_wmem_queued += skb->truesize; |
672 | sk_mem_charge(sk, skb->truesize); | |
89ebd197 | 673 | if (tp->nonagle & TCP_NAGLE_PUSH) |
e905a9ed | 674 | tp->nonagle &= ~TCP_NAGLE_PUSH; |
6f021c62 ED |
675 | |
676 | tcp_slow_start_after_idle_check(sk); | |
1da177e4 LT |
677 | } |
678 | ||
afeca340 | 679 | static inline void tcp_mark_urg(struct tcp_sock *tp, int flags) |
1da177e4 | 680 | { |
33f5f57e | 681 | if (flags & MSG_OOB) |
1da177e4 | 682 | tp->snd_up = tp->write_seq; |
1da177e4 LT |
683 | } |
684 | ||
f54b3111 | 685 | /* If a not yet filled skb is pushed, do not send it if |
a181ceb5 | 686 | * we have data packets in Qdisc or NIC queues : |
f54b3111 ED |
687 | * Because TX completion will happen shortly, it gives a chance |
688 | * to coalesce future sendmsg() payload into this skb, without | |
689 | * need for a timer, and with no latency trade off. | |
690 | * As packets containing data payload have a bigger truesize | |
a181ceb5 ED |
691 | * than pure acks (dataless) packets, the last checks prevent |
692 | * autocorking if we only have an ACK in Qdisc/NIC queues, | |
693 | * or if TX completion was delayed after we processed ACK packet. | |
f54b3111 ED |
694 | */ |
695 | static bool tcp_should_autocork(struct sock *sk, struct sk_buff *skb, | |
696 | int size_goal) | |
1da177e4 | 697 | { |
f54b3111 | 698 | return skb->len < size_goal && |
790f00e1 | 699 | sock_net(sk)->ipv4.sysctl_tcp_autocorking && |
114f39fe | 700 | !tcp_rtx_queue_empty(sk) && |
14afee4b | 701 | refcount_read(&sk->sk_wmem_alloc) > skb->truesize; |
f54b3111 ED |
702 | } |
703 | ||
704 | static void tcp_push(struct sock *sk, int flags, int mss_now, | |
705 | int nonagle, int size_goal) | |
706 | { | |
707 | struct tcp_sock *tp = tcp_sk(sk); | |
708 | struct sk_buff *skb; | |
afeca340 | 709 | |
f54b3111 | 710 | skb = tcp_write_queue_tail(sk); |
75c119af ED |
711 | if (!skb) |
712 | return; | |
f54b3111 ED |
713 | if (!(flags & MSG_MORE) || forced_push(tp)) |
714 | tcp_mark_push(tp, skb); | |
715 | ||
716 | tcp_mark_urg(tp, flags); | |
717 | ||
718 | if (tcp_should_autocork(sk, skb, size_goal)) { | |
719 | ||
720 | /* avoid atomic op if TSQ_THROTTLED bit is already set */ | |
7aa5470c | 721 | if (!test_bit(TSQ_THROTTLED, &sk->sk_tsq_flags)) { |
f54b3111 | 722 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAUTOCORKING); |
7aa5470c | 723 | set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags); |
f54b3111 | 724 | } |
a181ceb5 ED |
725 | /* It is possible TX completion already happened |
726 | * before we set TSQ_THROTTLED. | |
727 | */ | |
14afee4b | 728 | if (refcount_read(&sk->sk_wmem_alloc) > skb->truesize) |
a181ceb5 | 729 | return; |
1da177e4 | 730 | } |
f54b3111 ED |
731 | |
732 | if (flags & MSG_MORE) | |
733 | nonagle = TCP_NAGLE_CORK; | |
734 | ||
735 | __tcp_push_pending_frames(sk, mss_now, nonagle); | |
1da177e4 LT |
736 | } |
737 | ||
6ff7751d AB |
738 | static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, |
739 | unsigned int offset, size_t len) | |
9c55e01c JA |
740 | { |
741 | struct tcp_splice_state *tss = rd_desc->arg.data; | |
33966dd0 | 742 | int ret; |
9c55e01c | 743 | |
a60e3cc7 | 744 | ret = skb_splice_bits(skb, skb->sk, offset, tss->pipe, |
25869262 | 745 | min(rd_desc->count, len), tss->flags); |
33966dd0 WT |
746 | if (ret > 0) |
747 | rd_desc->count -= ret; | |
748 | return ret; | |
9c55e01c JA |
749 | } |
750 | ||
751 | static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss) | |
752 | { | |
753 | /* Store TCP splice context information in read_descriptor_t. */ | |
754 | read_descriptor_t rd_desc = { | |
755 | .arg.data = tss, | |
33966dd0 | 756 | .count = tss->len, |
9c55e01c JA |
757 | }; |
758 | ||
759 | return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv); | |
760 | } | |
761 | ||
762 | /** | |
763 | * tcp_splice_read - splice data from TCP socket to a pipe | |
764 | * @sock: socket to splice from | |
765 | * @ppos: position (not valid) | |
766 | * @pipe: pipe to splice to | |
767 | * @len: number of bytes to splice | |
768 | * @flags: splice modifier flags | |
769 | * | |
770 | * Description: | |
771 | * Will read pages from given socket and fill them into a pipe. | |
772 | * | |
773 | **/ | |
774 | ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos, | |
775 | struct pipe_inode_info *pipe, size_t len, | |
776 | unsigned int flags) | |
777 | { | |
778 | struct sock *sk = sock->sk; | |
779 | struct tcp_splice_state tss = { | |
780 | .pipe = pipe, | |
781 | .len = len, | |
782 | .flags = flags, | |
783 | }; | |
784 | long timeo; | |
785 | ssize_t spliced; | |
786 | int ret; | |
787 | ||
3a047bf8 | 788 | sock_rps_record_flow(sk); |
9c55e01c JA |
789 | /* |
790 | * We can't seek on a socket input | |
791 | */ | |
792 | if (unlikely(*ppos)) | |
793 | return -ESPIPE; | |
794 | ||
795 | ret = spliced = 0; | |
796 | ||
797 | lock_sock(sk); | |
798 | ||
42324c62 | 799 | timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK); |
9c55e01c JA |
800 | while (tss.len) { |
801 | ret = __tcp_splice_read(sk, &tss); | |
802 | if (ret < 0) | |
803 | break; | |
804 | else if (!ret) { | |
805 | if (spliced) | |
806 | break; | |
9c55e01c JA |
807 | if (sock_flag(sk, SOCK_DONE)) |
808 | break; | |
809 | if (sk->sk_err) { | |
810 | ret = sock_error(sk); | |
811 | break; | |
812 | } | |
813 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
814 | break; | |
815 | if (sk->sk_state == TCP_CLOSE) { | |
816 | /* | |
817 | * This occurs when user tries to read | |
818 | * from never connected socket. | |
819 | */ | |
820 | if (!sock_flag(sk, SOCK_DONE)) | |
821 | ret = -ENOTCONN; | |
822 | break; | |
823 | } | |
824 | if (!timeo) { | |
825 | ret = -EAGAIN; | |
826 | break; | |
827 | } | |
ccf7abb9 ED |
828 | /* if __tcp_splice_read() got nothing while we have |
829 | * an skb in receive queue, we do not want to loop. | |
830 | * This might happen with URG data. | |
831 | */ | |
832 | if (!skb_queue_empty(&sk->sk_receive_queue)) | |
833 | break; | |
dfbafc99 | 834 | sk_wait_data(sk, &timeo, NULL); |
9c55e01c JA |
835 | if (signal_pending(current)) { |
836 | ret = sock_intr_errno(timeo); | |
837 | break; | |
838 | } | |
839 | continue; | |
840 | } | |
841 | tss.len -= ret; | |
842 | spliced += ret; | |
843 | ||
33966dd0 WT |
844 | if (!timeo) |
845 | break; | |
9c55e01c JA |
846 | release_sock(sk); |
847 | lock_sock(sk); | |
848 | ||
849 | if (sk->sk_err || sk->sk_state == TCP_CLOSE || | |
33966dd0 | 850 | (sk->sk_shutdown & RCV_SHUTDOWN) || |
9c55e01c JA |
851 | signal_pending(current)) |
852 | break; | |
853 | } | |
854 | ||
855 | release_sock(sk); | |
856 | ||
857 | if (spliced) | |
858 | return spliced; | |
859 | ||
860 | return ret; | |
861 | } | |
4bc2f18b | 862 | EXPORT_SYMBOL(tcp_splice_read); |
9c55e01c | 863 | |
eb934478 ED |
864 | struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, |
865 | bool force_schedule) | |
f561d0f2 PE |
866 | { |
867 | struct sk_buff *skb; | |
868 | ||
869 | /* The TCP header must be at least 32-bit aligned. */ | |
870 | size = ALIGN(size, 4); | |
871 | ||
8e4d980a ED |
872 | if (unlikely(tcp_under_memory_pressure(sk))) |
873 | sk_mem_reclaim_partial(sk); | |
874 | ||
f561d0f2 | 875 | skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp); |
8e4d980a | 876 | if (likely(skb)) { |
eb934478 | 877 | bool mem_scheduled; |
8e4d980a | 878 | |
eb934478 ED |
879 | if (force_schedule) { |
880 | mem_scheduled = true; | |
8e4d980a ED |
881 | sk_forced_mem_schedule(sk, skb->truesize); |
882 | } else { | |
eb934478 | 883 | mem_scheduled = sk_wmem_schedule(sk, skb->truesize); |
8e4d980a | 884 | } |
eb934478 | 885 | if (likely(mem_scheduled)) { |
a21d4572 | 886 | skb_reserve(skb, sk->sk_prot->max_header); |
f561d0f2 PE |
887 | /* |
888 | * Make sure that we have exactly size bytes | |
889 | * available to the caller, no more, no less. | |
890 | */ | |
16fad69c | 891 | skb->reserved_tailroom = skb->end - skb->tail - size; |
e2080072 | 892 | INIT_LIST_HEAD(&skb->tcp_tsorted_anchor); |
f561d0f2 PE |
893 | return skb; |
894 | } | |
895 | __kfree_skb(skb); | |
896 | } else { | |
5c52ba17 | 897 | sk->sk_prot->enter_memory_pressure(sk); |
f561d0f2 PE |
898 | sk_stream_moderate_sndbuf(sk); |
899 | } | |
900 | return NULL; | |
901 | } | |
902 | ||
0c54b85f IJ |
903 | static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now, |
904 | int large_allowed) | |
905 | { | |
906 | struct tcp_sock *tp = tcp_sk(sk); | |
6c09fa09 | 907 | u32 new_size_goal, size_goal; |
605ad7f1 | 908 | |
74d4a8f8 | 909 | if (!large_allowed) |
605ad7f1 ED |
910 | return mss_now; |
911 | ||
6c09fa09 ED |
912 | /* Note : tcp_tso_autosize() will eventually split this later */ |
913 | new_size_goal = sk->sk_gso_max_size - 1 - MAX_TCP_HEADER; | |
605ad7f1 ED |
914 | new_size_goal = tcp_bound_to_half_wnd(tp, new_size_goal); |
915 | ||
916 | /* We try hard to avoid divides here */ | |
917 | size_goal = tp->gso_segs * mss_now; | |
918 | if (unlikely(new_size_goal < size_goal || | |
919 | new_size_goal >= size_goal + mss_now)) { | |
920 | tp->gso_segs = min_t(u16, new_size_goal / mss_now, | |
921 | sk->sk_gso_max_segs); | |
922 | size_goal = tp->gso_segs * mss_now; | |
0c54b85f IJ |
923 | } |
924 | ||
605ad7f1 | 925 | return max(size_goal, mss_now); |
0c54b85f IJ |
926 | } |
927 | ||
928 | static int tcp_send_mss(struct sock *sk, int *size_goal, int flags) | |
929 | { | |
930 | int mss_now; | |
931 | ||
932 | mss_now = tcp_current_mss(sk); | |
933 | *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB)); | |
934 | ||
935 | return mss_now; | |
936 | } | |
937 | ||
e3b5616a DW |
938 | ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, |
939 | size_t size, int flags) | |
1da177e4 LT |
940 | { |
941 | struct tcp_sock *tp = tcp_sk(sk); | |
c1b4a7e6 | 942 | int mss_now, size_goal; |
1da177e4 LT |
943 | int err; |
944 | ssize_t copied; | |
945 | long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); | |
946 | ||
8336886f JC |
947 | /* Wait for a connection to finish. One exception is TCP Fast Open |
948 | * (passive side) where data is allowed to be sent before a connection | |
949 | * is fully established. | |
950 | */ | |
951 | if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) && | |
952 | !tcp_passive_fastopen(sk)) { | |
686a5624 YM |
953 | err = sk_stream_wait_connect(sk, &timeo); |
954 | if (err != 0) | |
1da177e4 | 955 | goto out_err; |
8336886f | 956 | } |
1da177e4 | 957 | |
9cd3e072 | 958 | sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 | 959 | |
0c54b85f | 960 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
961 | copied = 0; |
962 | ||
963 | err = -EPIPE; | |
964 | if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) | |
0d6a775e | 965 | goto out_err; |
1da177e4 | 966 | |
64022d0b | 967 | while (size > 0) { |
fe067e8a | 968 | struct sk_buff *skb = tcp_write_queue_tail(sk); |
38ba0a65 | 969 | int copy, i; |
38ba0a65 | 970 | bool can_coalesce; |
1da177e4 | 971 | |
75c119af | 972 | if (!skb || (copy = size_goal - skb->len) <= 0 || |
c134ecb8 | 973 | !tcp_skb_can_collapse_to(skb)) { |
1da177e4 LT |
974 | new_segment: |
975 | if (!sk_stream_memory_free(sk)) | |
976 | goto wait_for_sndbuf; | |
977 | ||
eb934478 | 978 | skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation, |
75c119af | 979 | tcp_rtx_and_write_queues_empty(sk)); |
1da177e4 LT |
980 | if (!skb) |
981 | goto wait_for_memory; | |
982 | ||
9e412ba7 | 983 | skb_entail(sk, skb); |
c1b4a7e6 | 984 | copy = size_goal; |
1da177e4 LT |
985 | } |
986 | ||
987 | if (copy > size) | |
988 | copy = size; | |
989 | ||
990 | i = skb_shinfo(skb)->nr_frags; | |
991 | can_coalesce = skb_can_coalesce(skb, i, page, offset); | |
5f74f82e | 992 | if (!can_coalesce && i >= sysctl_max_skb_frags) { |
1da177e4 LT |
993 | tcp_mark_push(tp, skb); |
994 | goto new_segment; | |
995 | } | |
3ab224be | 996 | if (!sk_wmem_schedule(sk, copy)) |
1da177e4 | 997 | goto wait_for_memory; |
e905a9ed | 998 | |
1da177e4 | 999 | if (can_coalesce) { |
9e903e08 | 1000 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); |
1da177e4 LT |
1001 | } else { |
1002 | get_page(page); | |
1003 | skb_fill_page_desc(skb, i, page, offset, copy); | |
1004 | } | |
312fc2b4 JF |
1005 | |
1006 | if (!(flags & MSG_NO_SHARED_FRAGS)) | |
1007 | skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; | |
cef401de | 1008 | |
1da177e4 LT |
1009 | skb->len += copy; |
1010 | skb->data_len += copy; | |
1011 | skb->truesize += copy; | |
1012 | sk->sk_wmem_queued += copy; | |
3ab224be | 1013 | sk_mem_charge(sk, copy); |
84fa7933 | 1014 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 LT |
1015 | tp->write_seq += copy; |
1016 | TCP_SKB_CB(skb)->end_seq += copy; | |
cd7d8498 | 1017 | tcp_skb_pcount_set(skb, 0); |
1da177e4 LT |
1018 | |
1019 | if (!copied) | |
4de075e0 | 1020 | TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH; |
1da177e4 LT |
1021 | |
1022 | copied += copy; | |
64022d0b | 1023 | offset += copy; |
686a5624 | 1024 | size -= copy; |
ad02c4f5 | 1025 | if (!size) |
1da177e4 LT |
1026 | goto out; |
1027 | ||
69d15067 | 1028 | if (skb->len < size_goal || (flags & MSG_OOB)) |
1da177e4 LT |
1029 | continue; |
1030 | ||
1031 | if (forced_push(tp)) { | |
1032 | tcp_mark_push(tp, skb); | |
9e412ba7 | 1033 | __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH); |
fe067e8a | 1034 | } else if (skb == tcp_send_head(sk)) |
1da177e4 LT |
1035 | tcp_push_one(sk, mss_now); |
1036 | continue; | |
1037 | ||
1038 | wait_for_sndbuf: | |
1039 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
1040 | wait_for_memory: | |
f54b3111 ED |
1041 | tcp_push(sk, flags & ~MSG_MORE, mss_now, |
1042 | TCP_NAGLE_PUSH, size_goal); | |
1da177e4 | 1043 | |
686a5624 YM |
1044 | err = sk_stream_wait_memory(sk, &timeo); |
1045 | if (err != 0) | |
1da177e4 LT |
1046 | goto do_error; |
1047 | ||
0c54b85f | 1048 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
1049 | } |
1050 | ||
1051 | out: | |
ad02c4f5 | 1052 | if (copied) { |
4e8cc228 | 1053 | tcp_tx_timestamp(sk, sk->sk_tsflags); |
ad02c4f5 SHY |
1054 | if (!(flags & MSG_SENDPAGE_NOTLAST)) |
1055 | tcp_push(sk, flags, mss_now, tp->nonagle, size_goal); | |
1056 | } | |
1da177e4 LT |
1057 | return copied; |
1058 | ||
1059 | do_error: | |
1060 | if (copied) | |
1061 | goto out; | |
1062 | out_err: | |
ce5ec440 | 1063 | /* make sure we wake any epoll edge trigger waiter */ |
b0f71bd3 FY |
1064 | if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && |
1065 | err == -EAGAIN)) { | |
ce5ec440 | 1066 | sk->sk_write_space(sk); |
b0f71bd3 FY |
1067 | tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED); |
1068 | } | |
1da177e4 LT |
1069 | return sk_stream_error(sk, flags, err); |
1070 | } | |
e3b5616a | 1071 | EXPORT_SYMBOL_GPL(do_tcp_sendpages); |
1da177e4 | 1072 | |
306b13eb TH |
1073 | int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, |
1074 | size_t size, int flags) | |
1da177e4 | 1075 | { |
dead7cdb | 1076 | if (!(sk->sk_route_caps & NETIF_F_SG)) |
bd9dfc54 | 1077 | return sock_no_sendpage_locked(sk, page, offset, size, flags); |
1da177e4 | 1078 | |
d7722e85 SHY |
1079 | tcp_rate_check_app_limited(sk); /* is sending application-limited? */ |
1080 | ||
306b13eb TH |
1081 | return do_tcp_sendpages(sk, page, offset, size, flags); |
1082 | } | |
774c4673 | 1083 | EXPORT_SYMBOL_GPL(tcp_sendpage_locked); |
306b13eb TH |
1084 | |
1085 | int tcp_sendpage(struct sock *sk, struct page *page, int offset, | |
1086 | size_t size, int flags) | |
1087 | { | |
1088 | int ret; | |
1089 | ||
1090 | lock_sock(sk); | |
1091 | ret = tcp_sendpage_locked(sk, page, offset, size, flags); | |
1da177e4 | 1092 | release_sock(sk); |
306b13eb TH |
1093 | |
1094 | return ret; | |
1da177e4 | 1095 | } |
4bc2f18b | 1096 | EXPORT_SYMBOL(tcp_sendpage); |
1da177e4 | 1097 | |
3613b3db ED |
1098 | /* Do not bother using a page frag for very small frames. |
1099 | * But use this heuristic only for the first skb in write queue. | |
1100 | * | |
1101 | * Having no payload in skb->head allows better SACK shifting | |
1102 | * in tcp_shift_skb_data(), reducing sack/rack overhead, because | |
1103 | * write queue has less skbs. | |
1104 | * Each skb can hold up to MAX_SKB_FRAGS * 32Kbytes, or ~0.5 MB. | |
1105 | * This also speeds up tso_fragment(), since it wont fallback | |
1106 | * to tcp_fragment(). | |
1107 | */ | |
1108 | static int linear_payload_sz(bool first_skb) | |
1109 | { | |
1110 | if (first_skb) | |
1111 | return SKB_WITH_OVERHEAD(2048 - MAX_TCP_HEADER); | |
1112 | return 0; | |
1113 | } | |
1114 | ||
74d4a8f8 | 1115 | static int select_size(bool first_skb, bool zc) |
1da177e4 | 1116 | { |
74d4a8f8 ED |
1117 | if (zc) |
1118 | return 0; | |
1119 | return linear_payload_sz(first_skb); | |
1da177e4 LT |
1120 | } |
1121 | ||
cf60af03 YC |
1122 | void tcp_free_fastopen_req(struct tcp_sock *tp) |
1123 | { | |
00db4124 | 1124 | if (tp->fastopen_req) { |
cf60af03 YC |
1125 | kfree(tp->fastopen_req); |
1126 | tp->fastopen_req = NULL; | |
1127 | } | |
1128 | } | |
1129 | ||
f5ddcbbb ED |
1130 | static int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, |
1131 | int *copied, size_t size) | |
cf60af03 YC |
1132 | { |
1133 | struct tcp_sock *tp = tcp_sk(sk); | |
19f6d3f3 | 1134 | struct inet_sock *inet = inet_sk(sk); |
ba615f67 | 1135 | struct sockaddr *uaddr = msg->msg_name; |
cf60af03 YC |
1136 | int err, flags; |
1137 | ||
e1cfcbe8 | 1138 | if (!(sock_net(sk)->ipv4.sysctl_tcp_fastopen & TFO_CLIENT_ENABLE) || |
ba615f67 WW |
1139 | (uaddr && msg->msg_namelen >= sizeof(uaddr->sa_family) && |
1140 | uaddr->sa_family == AF_UNSPEC)) | |
cf60af03 | 1141 | return -EOPNOTSUPP; |
00db4124 | 1142 | if (tp->fastopen_req) |
cf60af03 YC |
1143 | return -EALREADY; /* Another Fast Open is in progress */ |
1144 | ||
1145 | tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request), | |
1146 | sk->sk_allocation); | |
51456b29 | 1147 | if (unlikely(!tp->fastopen_req)) |
cf60af03 YC |
1148 | return -ENOBUFS; |
1149 | tp->fastopen_req->data = msg; | |
f5ddcbbb | 1150 | tp->fastopen_req->size = size; |
cf60af03 | 1151 | |
19f6d3f3 WW |
1152 | if (inet->defer_connect) { |
1153 | err = tcp_connect(sk); | |
1154 | /* Same failure procedure as in tcp_v4/6_connect */ | |
1155 | if (err) { | |
1156 | tcp_set_state(sk, TCP_CLOSE); | |
1157 | inet->inet_dport = 0; | |
1158 | sk->sk_route_caps = 0; | |
1159 | } | |
1160 | } | |
cf60af03 | 1161 | flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0; |
ba615f67 | 1162 | err = __inet_stream_connect(sk->sk_socket, uaddr, |
3979ad7e | 1163 | msg->msg_namelen, flags, 1); |
7db92362 WW |
1164 | /* fastopen_req could already be freed in __inet_stream_connect |
1165 | * if the connection times out or gets rst | |
1166 | */ | |
1167 | if (tp->fastopen_req) { | |
1168 | *copied = tp->fastopen_req->copied; | |
1169 | tcp_free_fastopen_req(tp); | |
1170 | inet->defer_connect = 0; | |
1171 | } | |
cf60af03 YC |
1172 | return err; |
1173 | } | |
1174 | ||
306b13eb | 1175 | int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size) |
1da177e4 | 1176 | { |
1da177e4 | 1177 | struct tcp_sock *tp = tcp_sk(sk); |
f214f915 | 1178 | struct ubuf_info *uarg = NULL; |
1da177e4 | 1179 | struct sk_buff *skb; |
c14ac945 | 1180 | struct sockcm_cookie sockc; |
57be5bda AV |
1181 | int flags, err, copied = 0; |
1182 | int mss_now = 0, size_goal, copied_syn = 0; | |
d4011239 | 1183 | bool process_backlog = false; |
74d4a8f8 | 1184 | bool zc = false; |
1da177e4 LT |
1185 | long timeo; |
1186 | ||
1da177e4 | 1187 | flags = msg->msg_flags; |
f214f915 WB |
1188 | |
1189 | if (flags & MSG_ZEROCOPY && size) { | |
1190 | if (sk->sk_state != TCP_ESTABLISHED) { | |
1191 | err = -EINVAL; | |
1192 | goto out_err; | |
1193 | } | |
1194 | ||
75c119af | 1195 | skb = tcp_write_queue_tail(sk); |
f214f915 WB |
1196 | uarg = sock_zerocopy_realloc(sk, size, skb_zcopy(skb)); |
1197 | if (!uarg) { | |
1198 | err = -ENOBUFS; | |
1199 | goto out_err; | |
1200 | } | |
1201 | ||
dead7cdb | 1202 | zc = sk->sk_route_caps & NETIF_F_SG; |
02583ade | 1203 | if (!zc) |
f214f915 WB |
1204 | uarg->zerocopy = 0; |
1205 | } | |
1206 | ||
16ae6aa1 YC |
1207 | if (unlikely(flags & MSG_FASTOPEN || inet_sk(sk)->defer_connect) && |
1208 | !tp->repair) { | |
f5ddcbbb | 1209 | err = tcp_sendmsg_fastopen(sk, msg, &copied_syn, size); |
cf60af03 YC |
1210 | if (err == -EINPROGRESS && copied_syn > 0) |
1211 | goto out; | |
1212 | else if (err) | |
1213 | goto out_err; | |
cf60af03 YC |
1214 | } |
1215 | ||
1da177e4 LT |
1216 | timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); |
1217 | ||
d7722e85 SHY |
1218 | tcp_rate_check_app_limited(sk); /* is sending application-limited? */ |
1219 | ||
8336886f JC |
1220 | /* Wait for a connection to finish. One exception is TCP Fast Open |
1221 | * (passive side) where data is allowed to be sent before a connection | |
1222 | * is fully established. | |
1223 | */ | |
1224 | if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) && | |
1225 | !tcp_passive_fastopen(sk)) { | |
686a5624 YM |
1226 | err = sk_stream_wait_connect(sk, &timeo); |
1227 | if (err != 0) | |
cf60af03 | 1228 | goto do_error; |
8336886f | 1229 | } |
1da177e4 | 1230 | |
c0e88ff0 PE |
1231 | if (unlikely(tp->repair)) { |
1232 | if (tp->repair_queue == TCP_RECV_QUEUE) { | |
1233 | copied = tcp_send_rcvq(sk, msg, size); | |
5924f17a | 1234 | goto out_nopush; |
c0e88ff0 PE |
1235 | } |
1236 | ||
1237 | err = -EINVAL; | |
1238 | if (tp->repair_queue == TCP_NO_QUEUE) | |
1239 | goto out_err; | |
1240 | ||
1241 | /* 'common' sending to sendq */ | |
1242 | } | |
1243 | ||
c14ac945 SHY |
1244 | sockc.tsflags = sk->sk_tsflags; |
1245 | if (msg->msg_controllen) { | |
1246 | err = sock_cmsg_send(sk, msg, &sockc); | |
1247 | if (unlikely(err)) { | |
1248 | err = -EINVAL; | |
1249 | goto out_err; | |
1250 | } | |
1251 | } | |
1252 | ||
1da177e4 | 1253 | /* This should be in poll */ |
9cd3e072 | 1254 | sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 | 1255 | |
1da177e4 | 1256 | /* Ok commence sending. */ |
1da177e4 LT |
1257 | copied = 0; |
1258 | ||
d41a69f1 ED |
1259 | restart: |
1260 | mss_now = tcp_send_mss(sk, &size_goal, flags); | |
1261 | ||
1da177e4 LT |
1262 | err = -EPIPE; |
1263 | if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) | |
79d8665b | 1264 | goto do_error; |
1da177e4 | 1265 | |
01e97e65 | 1266 | while (msg_data_left(msg)) { |
57be5bda | 1267 | int copy = 0; |
1da177e4 | 1268 | |
57be5bda | 1269 | skb = tcp_write_queue_tail(sk); |
65ec6097 ED |
1270 | if (skb) |
1271 | copy = size_goal - skb->len; | |
1da177e4 | 1272 | |
c134ecb8 | 1273 | if (copy <= 0 || !tcp_skb_can_collapse_to(skb)) { |
3613b3db | 1274 | bool first_skb; |
8ddab508 | 1275 | int linear; |
3613b3db | 1276 | |
1da177e4 | 1277 | new_segment: |
57be5bda AV |
1278 | /* Allocate new segment. If the interface is SG, |
1279 | * allocate skb fitting to single page. | |
1280 | */ | |
1281 | if (!sk_stream_memory_free(sk)) | |
1282 | goto wait_for_sndbuf; | |
1da177e4 | 1283 | |
d4011239 ED |
1284 | if (process_backlog && sk_flush_backlog(sk)) { |
1285 | process_backlog = false; | |
d41a69f1 | 1286 | goto restart; |
d4011239 | 1287 | } |
75c119af | 1288 | first_skb = tcp_rtx_and_write_queues_empty(sk); |
74d4a8f8 | 1289 | linear = select_size(first_skb, zc); |
8ddab508 | 1290 | skb = sk_stream_alloc_skb(sk, linear, sk->sk_allocation, |
3613b3db | 1291 | first_skb); |
57be5bda AV |
1292 | if (!skb) |
1293 | goto wait_for_memory; | |
1da177e4 | 1294 | |
d4011239 | 1295 | process_backlog = true; |
dead7cdb | 1296 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 | 1297 | |
57be5bda AV |
1298 | skb_entail(sk, skb); |
1299 | copy = size_goal; | |
9d186cac | 1300 | |
57be5bda AV |
1301 | /* All packets are restored as if they have |
1302 | * already been sent. skb_mstamp isn't set to | |
1303 | * avoid wrong rtt estimation. | |
1304 | */ | |
1305 | if (tp->repair) | |
1306 | TCP_SKB_CB(skb)->sacked |= TCPCB_REPAIRED; | |
1307 | } | |
1da177e4 | 1308 | |
57be5bda | 1309 | /* Try to append data to the end of skb. */ |
01e97e65 AV |
1310 | if (copy > msg_data_left(msg)) |
1311 | copy = msg_data_left(msg); | |
57be5bda AV |
1312 | |
1313 | /* Where to copy to? */ | |
02583ade | 1314 | if (skb_availroom(skb) > 0 && !zc) { |
57be5bda AV |
1315 | /* We have some space in skb head. Superb! */ |
1316 | copy = min_t(int, copy, skb_availroom(skb)); | |
1317 | err = skb_add_data_nocache(sk, skb, &msg->msg_iter, copy); | |
1318 | if (err) | |
1319 | goto do_fault; | |
02583ade | 1320 | } else if (!zc) { |
57be5bda AV |
1321 | bool merge = true; |
1322 | int i = skb_shinfo(skb)->nr_frags; | |
1323 | struct page_frag *pfrag = sk_page_frag(sk); | |
1324 | ||
1325 | if (!sk_page_frag_refill(sk, pfrag)) | |
1326 | goto wait_for_memory; | |
ef015786 | 1327 | |
57be5bda AV |
1328 | if (!skb_can_coalesce(skb, i, pfrag->page, |
1329 | pfrag->offset)) { | |
74d4a8f8 | 1330 | if (i >= sysctl_max_skb_frags) { |
57be5bda AV |
1331 | tcp_mark_push(tp, skb); |
1332 | goto new_segment; | |
1da177e4 | 1333 | } |
57be5bda | 1334 | merge = false; |
1da177e4 LT |
1335 | } |
1336 | ||
57be5bda AV |
1337 | copy = min_t(int, copy, pfrag->size - pfrag->offset); |
1338 | ||
1339 | if (!sk_wmem_schedule(sk, copy)) | |
1340 | goto wait_for_memory; | |
1da177e4 | 1341 | |
57be5bda AV |
1342 | err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb, |
1343 | pfrag->page, | |
1344 | pfrag->offset, | |
1345 | copy); | |
1346 | if (err) | |
1347 | goto do_error; | |
1da177e4 | 1348 | |
57be5bda AV |
1349 | /* Update the skb. */ |
1350 | if (merge) { | |
1351 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); | |
1352 | } else { | |
1353 | skb_fill_page_desc(skb, i, pfrag->page, | |
1354 | pfrag->offset, copy); | |
4e33e346 | 1355 | page_ref_inc(pfrag->page); |
4ed2d765 | 1356 | } |
57be5bda | 1357 | pfrag->offset += copy; |
f214f915 WB |
1358 | } else { |
1359 | err = skb_zerocopy_iter_stream(sk, skb, msg, copy, uarg); | |
111856c7 WB |
1360 | if (err == -EMSGSIZE || err == -EEXIST) { |
1361 | tcp_mark_push(tp, skb); | |
f214f915 | 1362 | goto new_segment; |
111856c7 | 1363 | } |
f214f915 WB |
1364 | if (err < 0) |
1365 | goto do_error; | |
1366 | copy = err; | |
57be5bda AV |
1367 | } |
1368 | ||
1369 | if (!copied) | |
1370 | TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH; | |
1371 | ||
1372 | tp->write_seq += copy; | |
1373 | TCP_SKB_CB(skb)->end_seq += copy; | |
1374 | tcp_skb_pcount_set(skb, 0); | |
1da177e4 | 1375 | |
57be5bda | 1376 | copied += copy; |
01e97e65 | 1377 | if (!msg_data_left(msg)) { |
c134ecb8 MKL |
1378 | if (unlikely(flags & MSG_EOR)) |
1379 | TCP_SKB_CB(skb)->eor = 1; | |
57be5bda AV |
1380 | goto out; |
1381 | } | |
1da177e4 | 1382 | |
65ec6097 | 1383 | if (skb->len < size_goal || (flags & MSG_OOB) || unlikely(tp->repair)) |
1da177e4 LT |
1384 | continue; |
1385 | ||
57be5bda AV |
1386 | if (forced_push(tp)) { |
1387 | tcp_mark_push(tp, skb); | |
1388 | __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH); | |
1389 | } else if (skb == tcp_send_head(sk)) | |
1390 | tcp_push_one(sk, mss_now); | |
1391 | continue; | |
1392 | ||
1da177e4 | 1393 | wait_for_sndbuf: |
57be5bda | 1394 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
1da177e4 | 1395 | wait_for_memory: |
57be5bda AV |
1396 | if (copied) |
1397 | tcp_push(sk, flags & ~MSG_MORE, mss_now, | |
1398 | TCP_NAGLE_PUSH, size_goal); | |
1da177e4 | 1399 | |
686a5624 YM |
1400 | err = sk_stream_wait_memory(sk, &timeo); |
1401 | if (err != 0) | |
57be5bda | 1402 | goto do_error; |
1da177e4 | 1403 | |
57be5bda | 1404 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
1405 | } |
1406 | ||
1407 | out: | |
ad02c4f5 | 1408 | if (copied) { |
4e8cc228 | 1409 | tcp_tx_timestamp(sk, sockc.tsflags); |
f54b3111 | 1410 | tcp_push(sk, flags, mss_now, tp->nonagle, size_goal); |
ad02c4f5 | 1411 | } |
5924f17a | 1412 | out_nopush: |
f214f915 | 1413 | sock_zerocopy_put(uarg); |
cf60af03 | 1414 | return copied + copied_syn; |
1da177e4 LT |
1415 | |
1416 | do_fault: | |
1417 | if (!skb->len) { | |
fe067e8a DM |
1418 | tcp_unlink_write_queue(skb, sk); |
1419 | /* It is the one place in all of TCP, except connection | |
1420 | * reset, where we can be unlinking the send_head. | |
1421 | */ | |
1422 | tcp_check_send_head(sk, skb); | |
3ab224be | 1423 | sk_wmem_free_skb(sk, skb); |
1da177e4 LT |
1424 | } |
1425 | ||
1426 | do_error: | |
cf60af03 | 1427 | if (copied + copied_syn) |
1da177e4 LT |
1428 | goto out; |
1429 | out_err: | |
f214f915 | 1430 | sock_zerocopy_put_abort(uarg); |
1da177e4 | 1431 | err = sk_stream_error(sk, flags, err); |
ce5ec440 | 1432 | /* make sure we wake any epoll edge trigger waiter */ |
b0f71bd3 FY |
1433 | if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && |
1434 | err == -EAGAIN)) { | |
ce5ec440 | 1435 | sk->sk_write_space(sk); |
b0f71bd3 FY |
1436 | tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED); |
1437 | } | |
1da177e4 LT |
1438 | return err; |
1439 | } | |
774c4673 | 1440 | EXPORT_SYMBOL_GPL(tcp_sendmsg_locked); |
306b13eb TH |
1441 | |
1442 | int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size) | |
1443 | { | |
1444 | int ret; | |
1445 | ||
1446 | lock_sock(sk); | |
1447 | ret = tcp_sendmsg_locked(sk, msg, size); | |
1448 | release_sock(sk); | |
1449 | ||
1450 | return ret; | |
1451 | } | |
4bc2f18b | 1452 | EXPORT_SYMBOL(tcp_sendmsg); |
1da177e4 LT |
1453 | |
1454 | /* | |
1455 | * Handle reading urgent data. BSD has very simple semantics for | |
1456 | * this, no blocking and very strange errors 8) | |
1457 | */ | |
1458 | ||
377f0a08 | 1459 | static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags) |
1da177e4 LT |
1460 | { |
1461 | struct tcp_sock *tp = tcp_sk(sk); | |
1462 | ||
1463 | /* No URG data to read. */ | |
1464 | if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data || | |
1465 | tp->urg_data == TCP_URG_READ) | |
1466 | return -EINVAL; /* Yes this is right ! */ | |
1467 | ||
1468 | if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE)) | |
1469 | return -ENOTCONN; | |
1470 | ||
1471 | if (tp->urg_data & TCP_URG_VALID) { | |
1472 | int err = 0; | |
1473 | char c = tp->urg_data; | |
1474 | ||
1475 | if (!(flags & MSG_PEEK)) | |
1476 | tp->urg_data = TCP_URG_READ; | |
1477 | ||
1478 | /* Read urgent data. */ | |
1479 | msg->msg_flags |= MSG_OOB; | |
1480 | ||
1481 | if (len > 0) { | |
1482 | if (!(flags & MSG_TRUNC)) | |
7eab8d9e | 1483 | err = memcpy_to_msg(msg, &c, 1); |
1da177e4 LT |
1484 | len = 1; |
1485 | } else | |
1486 | msg->msg_flags |= MSG_TRUNC; | |
1487 | ||
1488 | return err ? -EFAULT : len; | |
1489 | } | |
1490 | ||
1491 | if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN)) | |
1492 | return 0; | |
1493 | ||
1494 | /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and | |
1495 | * the available implementations agree in this case: | |
1496 | * this call should never block, independent of the | |
1497 | * blocking state of the socket. | |
1498 | * Mike <pall@rz.uni-karlsruhe.de> | |
1499 | */ | |
1500 | return -EAGAIN; | |
1501 | } | |
1502 | ||
c0e88ff0 PE |
1503 | static int tcp_peek_sndq(struct sock *sk, struct msghdr *msg, int len) |
1504 | { | |
1505 | struct sk_buff *skb; | |
1506 | int copied = 0, err = 0; | |
1507 | ||
1508 | /* XXX -- need to support SO_PEEK_OFF */ | |
1509 | ||
75c119af ED |
1510 | skb_rbtree_walk(skb, &sk->tcp_rtx_queue) { |
1511 | err = skb_copy_datagram_msg(skb, 0, msg, skb->len); | |
1512 | if (err) | |
1513 | return err; | |
1514 | copied += skb->len; | |
1515 | } | |
1516 | ||
c0e88ff0 | 1517 | skb_queue_walk(&sk->sk_write_queue, skb) { |
51f3d02b | 1518 | err = skb_copy_datagram_msg(skb, 0, msg, skb->len); |
c0e88ff0 PE |
1519 | if (err) |
1520 | break; | |
1521 | ||
1522 | copied += skb->len; | |
1523 | } | |
1524 | ||
1525 | return err ?: copied; | |
1526 | } | |
1527 | ||
1da177e4 LT |
1528 | /* Clean up the receive buffer for full frames taken by the user, |
1529 | * then send an ACK if necessary. COPIED is the number of bytes | |
1530 | * tcp_recvmsg has given to the user so far, it speeds up the | |
1531 | * calculation of whether or not we must ACK for the sake of | |
1532 | * a window update. | |
1533 | */ | |
3f334078 | 1534 | static void tcp_cleanup_rbuf(struct sock *sk, int copied) |
1da177e4 LT |
1535 | { |
1536 | struct tcp_sock *tp = tcp_sk(sk); | |
a2a385d6 | 1537 | bool time_to_ack = false; |
1da177e4 | 1538 | |
1da177e4 LT |
1539 | struct sk_buff *skb = skb_peek(&sk->sk_receive_queue); |
1540 | ||
d792c100 | 1541 | WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq), |
2af6fd8b | 1542 | "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n", |
d792c100 | 1543 | tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt); |
1da177e4 | 1544 | |
463c84b9 ACM |
1545 | if (inet_csk_ack_scheduled(sk)) { |
1546 | const struct inet_connection_sock *icsk = inet_csk(sk); | |
1da177e4 LT |
1547 | /* Delayed ACKs frequently hit locked sockets during bulk |
1548 | * receive. */ | |
463c84b9 | 1549 | if (icsk->icsk_ack.blocked || |
1da177e4 | 1550 | /* Once-per-two-segments ACK was not sent by tcp_input.c */ |
463c84b9 | 1551 | tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss || |
1da177e4 LT |
1552 | /* |
1553 | * If this read emptied read buffer, we send ACK, if | |
1554 | * connection is not bidirectional, user drained | |
1555 | * receive buffer and there was a small segment | |
1556 | * in queue. | |
1557 | */ | |
1ef9696c AK |
1558 | (copied > 0 && |
1559 | ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) || | |
1560 | ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) && | |
1561 | !icsk->icsk_ack.pingpong)) && | |
1562 | !atomic_read(&sk->sk_rmem_alloc))) | |
a2a385d6 | 1563 | time_to_ack = true; |
1da177e4 LT |
1564 | } |
1565 | ||
1566 | /* We send an ACK if we can now advertise a non-zero window | |
1567 | * which has been raised "significantly". | |
1568 | * | |
1569 | * Even if window raised up to infinity, do not send window open ACK | |
1570 | * in states, where we will not receive more. It is useless. | |
1571 | */ | |
1572 | if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) { | |
1573 | __u32 rcv_window_now = tcp_receive_window(tp); | |
1574 | ||
1575 | /* Optimize, __tcp_select_window() is not cheap. */ | |
1576 | if (2*rcv_window_now <= tp->window_clamp) { | |
1577 | __u32 new_window = __tcp_select_window(sk); | |
1578 | ||
1579 | /* Send ACK now, if this read freed lots of space | |
1580 | * in our buffer. Certainly, new_window is new window. | |
1581 | * We can advertise it now, if it is not less than current one. | |
1582 | * "Lots" means "at least twice" here. | |
1583 | */ | |
1584 | if (new_window && new_window >= 2 * rcv_window_now) | |
a2a385d6 | 1585 | time_to_ack = true; |
1da177e4 LT |
1586 | } |
1587 | } | |
1588 | if (time_to_ack) | |
1589 | tcp_send_ack(sk); | |
1590 | } | |
1591 | ||
f26845b4 | 1592 | static struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off) |
1da177e4 LT |
1593 | { |
1594 | struct sk_buff *skb; | |
1595 | u32 offset; | |
1596 | ||
f26845b4 | 1597 | while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) { |
1da177e4 | 1598 | offset = seq - TCP_SKB_CB(skb)->seq; |
9d691539 ED |
1599 | if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { |
1600 | pr_err_once("%s: found a SYN, please report !\n", __func__); | |
1da177e4 | 1601 | offset--; |
9d691539 | 1602 | } |
e11ecddf | 1603 | if (offset < skb->len || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) { |
1da177e4 LT |
1604 | *off = offset; |
1605 | return skb; | |
1606 | } | |
f26845b4 ED |
1607 | /* This looks weird, but this can happen if TCP collapsing |
1608 | * splitted a fat GRO packet, while we released socket lock | |
1609 | * in skb_splice_bits() | |
1610 | */ | |
7bced397 | 1611 | sk_eat_skb(sk, skb); |
1da177e4 LT |
1612 | } |
1613 | return NULL; | |
1614 | } | |
1615 | ||
1616 | /* | |
1617 | * This routine provides an alternative to tcp_recvmsg() for routines | |
1618 | * that would like to handle copying from skbuffs directly in 'sendfile' | |
1619 | * fashion. | |
1620 | * Note: | |
1621 | * - It is assumed that the socket was locked by the caller. | |
1622 | * - The routine does not block. | |
1623 | * - At present, there is no support for reading OOB data | |
1624 | * or for 'peeking' the socket using this routine | |
1625 | * (although both would be easy to implement). | |
1626 | */ | |
1627 | int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, | |
1628 | sk_read_actor_t recv_actor) | |
1629 | { | |
1630 | struct sk_buff *skb; | |
1631 | struct tcp_sock *tp = tcp_sk(sk); | |
1632 | u32 seq = tp->copied_seq; | |
1633 | u32 offset; | |
1634 | int copied = 0; | |
1635 | ||
1636 | if (sk->sk_state == TCP_LISTEN) | |
1637 | return -ENOTCONN; | |
1638 | while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) { | |
1639 | if (offset < skb->len) { | |
374e7b59 OP |
1640 | int used; |
1641 | size_t len; | |
1da177e4 LT |
1642 | |
1643 | len = skb->len - offset; | |
1644 | /* Stop reading if we hit a patch of urgent data */ | |
1645 | if (tp->urg_data) { | |
1646 | u32 urg_offset = tp->urg_seq - seq; | |
1647 | if (urg_offset < len) | |
1648 | len = urg_offset; | |
1649 | if (!len) | |
1650 | break; | |
1651 | } | |
1652 | used = recv_actor(desc, skb, offset, len); | |
ff905b1e | 1653 | if (used <= 0) { |
ddb61a57 JA |
1654 | if (!copied) |
1655 | copied = used; | |
1656 | break; | |
1657 | } else if (used <= len) { | |
1da177e4 LT |
1658 | seq += used; |
1659 | copied += used; | |
1660 | offset += used; | |
1661 | } | |
02275a2e | 1662 | /* If recv_actor drops the lock (e.g. TCP splice |
293ad604 OP |
1663 | * receive) the skb pointer might be invalid when |
1664 | * getting here: tcp_collapse might have deleted it | |
1665 | * while aggregating skbs from the socket queue. | |
1666 | */ | |
02275a2e WT |
1667 | skb = tcp_recv_skb(sk, seq - 1, &offset); |
1668 | if (!skb) | |
1da177e4 | 1669 | break; |
02275a2e WT |
1670 | /* TCP coalescing might have appended data to the skb. |
1671 | * Try to splice more frags | |
1672 | */ | |
1673 | if (offset + 1 != skb->len) | |
1674 | continue; | |
1da177e4 | 1675 | } |
e11ecddf | 1676 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) { |
7bced397 | 1677 | sk_eat_skb(sk, skb); |
1da177e4 LT |
1678 | ++seq; |
1679 | break; | |
1680 | } | |
7bced397 | 1681 | sk_eat_skb(sk, skb); |
1da177e4 LT |
1682 | if (!desc->count) |
1683 | break; | |
baff42ab | 1684 | tp->copied_seq = seq; |
1da177e4 LT |
1685 | } |
1686 | tp->copied_seq = seq; | |
1687 | ||
1688 | tcp_rcv_space_adjust(sk); | |
1689 | ||
1690 | /* Clean up data we have read: This will do ACK frames. */ | |
f26845b4 ED |
1691 | if (copied > 0) { |
1692 | tcp_recv_skb(sk, seq, &offset); | |
0e4b4992 | 1693 | tcp_cleanup_rbuf(sk, copied); |
f26845b4 | 1694 | } |
1da177e4 LT |
1695 | return copied; |
1696 | } | |
4bc2f18b | 1697 | EXPORT_SYMBOL(tcp_read_sock); |
1da177e4 | 1698 | |
32035585 TH |
1699 | int tcp_peek_len(struct socket *sock) |
1700 | { | |
1701 | return tcp_inq(sock->sk); | |
1702 | } | |
1703 | EXPORT_SYMBOL(tcp_peek_len); | |
1704 | ||
98aaa913 MM |
1705 | static void tcp_update_recv_tstamps(struct sk_buff *skb, |
1706 | struct scm_timestamping *tss) | |
1707 | { | |
1708 | if (skb->tstamp) | |
1709 | tss->ts[0] = ktime_to_timespec(skb->tstamp); | |
1710 | else | |
1711 | tss->ts[0] = (struct timespec) {0}; | |
1712 | ||
1713 | if (skb_hwtstamps(skb)->hwtstamp) | |
1714 | tss->ts[2] = ktime_to_timespec(skb_hwtstamps(skb)->hwtstamp); | |
1715 | else | |
1716 | tss->ts[2] = (struct timespec) {0}; | |
1717 | } | |
1718 | ||
1719 | /* Similar to __sock_recv_timestamp, but does not require an skb */ | |
809a79e9 WY |
1720 | static void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, |
1721 | struct scm_timestamping *tss) | |
98aaa913 MM |
1722 | { |
1723 | struct timeval tv; | |
1724 | bool has_timestamping = false; | |
1725 | ||
1726 | if (tss->ts[0].tv_sec || tss->ts[0].tv_nsec) { | |
1727 | if (sock_flag(sk, SOCK_RCVTSTAMP)) { | |
1728 | if (sock_flag(sk, SOCK_RCVTSTAMPNS)) { | |
1729 | put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPNS, | |
1730 | sizeof(tss->ts[0]), &tss->ts[0]); | |
1731 | } else { | |
1732 | tv.tv_sec = tss->ts[0].tv_sec; | |
1733 | tv.tv_usec = tss->ts[0].tv_nsec / 1000; | |
1734 | ||
1735 | put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP, | |
1736 | sizeof(tv), &tv); | |
1737 | } | |
1738 | } | |
1739 | ||
1740 | if (sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) | |
1741 | has_timestamping = true; | |
1742 | else | |
1743 | tss->ts[0] = (struct timespec) {0}; | |
1744 | } | |
1745 | ||
1746 | if (tss->ts[2].tv_sec || tss->ts[2].tv_nsec) { | |
1747 | if (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) | |
1748 | has_timestamping = true; | |
1749 | else | |
1750 | tss->ts[2] = (struct timespec) {0}; | |
1751 | } | |
1752 | ||
1753 | if (has_timestamping) { | |
1754 | tss->ts[1] = (struct timespec) {0}; | |
1755 | put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING, | |
1756 | sizeof(*tss), tss); | |
1757 | } | |
1758 | } | |
1759 | ||
1da177e4 LT |
1760 | /* |
1761 | * This routine copies from a sock struct into the user buffer. | |
1762 | * | |
1763 | * Technical note: in 2.3 we work on _locked_ socket, so that | |
1764 | * tricks with *seq access order and skb->users are not required. | |
1765 | * Probably, code can be easily improved even more. | |
1766 | */ | |
1767 | ||
1b784140 YX |
1768 | int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, |
1769 | int flags, int *addr_len) | |
1da177e4 LT |
1770 | { |
1771 | struct tcp_sock *tp = tcp_sk(sk); | |
1772 | int copied = 0; | |
1773 | u32 peek_seq; | |
1774 | u32 *seq; | |
1775 | unsigned long used; | |
1776 | int err; | |
1777 | int target; /* Read at least this many bytes */ | |
1778 | long timeo; | |
dfbafc99 | 1779 | struct sk_buff *skb, *last; |
77527313 | 1780 | u32 urg_hole = 0; |
98aaa913 MM |
1781 | struct scm_timestamping tss; |
1782 | bool has_tss = false; | |
1da177e4 | 1783 | |
4ed2d765 | 1784 | if (unlikely(flags & MSG_ERRQUEUE)) |
f4713a3d | 1785 | return inet_recv_error(sk, msg, len, addr_len); |
4ed2d765 | 1786 | |
cbf55001 ET |
1787 | if (sk_can_busy_loop(sk) && skb_queue_empty(&sk->sk_receive_queue) && |
1788 | (sk->sk_state == TCP_ESTABLISHED)) | |
1789 | sk_busy_loop(sk, nonblock); | |
d30e383b | 1790 | |
1da177e4 LT |
1791 | lock_sock(sk); |
1792 | ||
1da177e4 LT |
1793 | err = -ENOTCONN; |
1794 | if (sk->sk_state == TCP_LISTEN) | |
1795 | goto out; | |
1796 | ||
1797 | timeo = sock_rcvtimeo(sk, nonblock); | |
1798 | ||
1799 | /* Urgent data needs to be handled specially. */ | |
1800 | if (flags & MSG_OOB) | |
1801 | goto recv_urg; | |
1802 | ||
c0e88ff0 PE |
1803 | if (unlikely(tp->repair)) { |
1804 | err = -EPERM; | |
1805 | if (!(flags & MSG_PEEK)) | |
1806 | goto out; | |
1807 | ||
1808 | if (tp->repair_queue == TCP_SEND_QUEUE) | |
1809 | goto recv_sndq; | |
1810 | ||
1811 | err = -EINVAL; | |
1812 | if (tp->repair_queue == TCP_NO_QUEUE) | |
1813 | goto out; | |
1814 | ||
1815 | /* 'common' recv queue MSG_PEEK-ing */ | |
1816 | } | |
1817 | ||
1da177e4 LT |
1818 | seq = &tp->copied_seq; |
1819 | if (flags & MSG_PEEK) { | |
1820 | peek_seq = tp->copied_seq; | |
1821 | seq = &peek_seq; | |
1822 | } | |
1823 | ||
1824 | target = sock_rcvlowat(sk, flags & MSG_WAITALL, len); | |
1825 | ||
1826 | do { | |
1da177e4 LT |
1827 | u32 offset; |
1828 | ||
1829 | /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */ | |
1830 | if (tp->urg_data && tp->urg_seq == *seq) { | |
1831 | if (copied) | |
1832 | break; | |
1833 | if (signal_pending(current)) { | |
1834 | copied = timeo ? sock_intr_errno(timeo) : -EAGAIN; | |
1835 | break; | |
1836 | } | |
1837 | } | |
1838 | ||
1839 | /* Next get a buffer. */ | |
1840 | ||
dfbafc99 | 1841 | last = skb_peek_tail(&sk->sk_receive_queue); |
91521944 | 1842 | skb_queue_walk(&sk->sk_receive_queue, skb) { |
dfbafc99 | 1843 | last = skb; |
1da177e4 LT |
1844 | /* Now that we have two receive queues this |
1845 | * shouldn't happen. | |
1846 | */ | |
d792c100 | 1847 | if (WARN(before(*seq, TCP_SKB_CB(skb)->seq), |
2af6fd8b JP |
1848 | "recvmsg bug: copied %X seq %X rcvnxt %X fl %X\n", |
1849 | *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, | |
1850 | flags)) | |
1da177e4 | 1851 | break; |
d792c100 | 1852 | |
1da177e4 | 1853 | offset = *seq - TCP_SKB_CB(skb)->seq; |
9d691539 ED |
1854 | if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { |
1855 | pr_err_once("%s: found a SYN, please report !\n", __func__); | |
1da177e4 | 1856 | offset--; |
9d691539 | 1857 | } |
1da177e4 LT |
1858 | if (offset < skb->len) |
1859 | goto found_ok_skb; | |
e11ecddf | 1860 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) |
1da177e4 | 1861 | goto found_fin_ok; |
2af6fd8b JP |
1862 | WARN(!(flags & MSG_PEEK), |
1863 | "recvmsg bug 2: copied %X seq %X rcvnxt %X fl %X\n", | |
1864 | *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags); | |
91521944 | 1865 | } |
1da177e4 LT |
1866 | |
1867 | /* Well, if we have backlog, try to process it now yet. */ | |
1868 | ||
1869 | if (copied >= target && !sk->sk_backlog.tail) | |
1870 | break; | |
1871 | ||
1872 | if (copied) { | |
1873 | if (sk->sk_err || | |
1874 | sk->sk_state == TCP_CLOSE || | |
1875 | (sk->sk_shutdown & RCV_SHUTDOWN) || | |
1876 | !timeo || | |
518a09ef | 1877 | signal_pending(current)) |
1da177e4 LT |
1878 | break; |
1879 | } else { | |
1880 | if (sock_flag(sk, SOCK_DONE)) | |
1881 | break; | |
1882 | ||
1883 | if (sk->sk_err) { | |
1884 | copied = sock_error(sk); | |
1885 | break; | |
1886 | } | |
1887 | ||
1888 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
1889 | break; | |
1890 | ||
1891 | if (sk->sk_state == TCP_CLOSE) { | |
1892 | if (!sock_flag(sk, SOCK_DONE)) { | |
1893 | /* This occurs when user tries to read | |
1894 | * from never connected socket. | |
1895 | */ | |
1896 | copied = -ENOTCONN; | |
1897 | break; | |
1898 | } | |
1899 | break; | |
1900 | } | |
1901 | ||
1902 | if (!timeo) { | |
1903 | copied = -EAGAIN; | |
1904 | break; | |
1905 | } | |
1906 | ||
1907 | if (signal_pending(current)) { | |
1908 | copied = sock_intr_errno(timeo); | |
1909 | break; | |
1910 | } | |
1911 | } | |
1912 | ||
0e4b4992 | 1913 | tcp_cleanup_rbuf(sk, copied); |
1da177e4 | 1914 | |
1da177e4 LT |
1915 | if (copied >= target) { |
1916 | /* Do not sleep, just process backlog. */ | |
1917 | release_sock(sk); | |
1918 | lock_sock(sk); | |
dfbafc99 SD |
1919 | } else { |
1920 | sk_wait_data(sk, &timeo, last); | |
1921 | } | |
1da177e4 | 1922 | |
77527313 IJ |
1923 | if ((flags & MSG_PEEK) && |
1924 | (peek_seq - copied - urg_hole != tp->copied_seq)) { | |
e87cc472 JP |
1925 | net_dbg_ratelimited("TCP(%s:%d): Application bug, race in MSG_PEEK\n", |
1926 | current->comm, | |
1927 | task_pid_nr(current)); | |
1da177e4 LT |
1928 | peek_seq = tp->copied_seq; |
1929 | } | |
1930 | continue; | |
1931 | ||
1932 | found_ok_skb: | |
1933 | /* Ok so how much can we use? */ | |
1934 | used = skb->len - offset; | |
1935 | if (len < used) | |
1936 | used = len; | |
1937 | ||
1938 | /* Do we have urgent data here? */ | |
1939 | if (tp->urg_data) { | |
1940 | u32 urg_offset = tp->urg_seq - *seq; | |
1941 | if (urg_offset < used) { | |
1942 | if (!urg_offset) { | |
1943 | if (!sock_flag(sk, SOCK_URGINLINE)) { | |
1944 | ++*seq; | |
77527313 | 1945 | urg_hole++; |
1da177e4 LT |
1946 | offset++; |
1947 | used--; | |
1948 | if (!used) | |
1949 | goto skip_copy; | |
1950 | } | |
1951 | } else | |
1952 | used = urg_offset; | |
1953 | } | |
1954 | } | |
1955 | ||
1956 | if (!(flags & MSG_TRUNC)) { | |
51f3d02b | 1957 | err = skb_copy_datagram_msg(skb, offset, msg, used); |
7bced397 DW |
1958 | if (err) { |
1959 | /* Exception. Bailout! */ | |
1960 | if (!copied) | |
1961 | copied = -EFAULT; | |
1962 | break; | |
1da177e4 LT |
1963 | } |
1964 | } | |
1965 | ||
1966 | *seq += used; | |
1967 | copied += used; | |
1968 | len -= used; | |
1969 | ||
1970 | tcp_rcv_space_adjust(sk); | |
1971 | ||
1972 | skip_copy: | |
31770e34 | 1973 | if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) { |
1da177e4 | 1974 | tp->urg_data = 0; |
31770e34 FW |
1975 | tcp_fast_path_check(sk); |
1976 | } | |
1da177e4 LT |
1977 | if (used + offset < skb->len) |
1978 | continue; | |
1979 | ||
98aaa913 MM |
1980 | if (TCP_SKB_CB(skb)->has_rxtstamp) { |
1981 | tcp_update_recv_tstamps(skb, &tss); | |
1982 | has_tss = true; | |
1983 | } | |
e11ecddf | 1984 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) |
1da177e4 | 1985 | goto found_fin_ok; |
7bced397 DW |
1986 | if (!(flags & MSG_PEEK)) |
1987 | sk_eat_skb(sk, skb); | |
1da177e4 LT |
1988 | continue; |
1989 | ||
1990 | found_fin_ok: | |
1991 | /* Process the FIN. */ | |
1992 | ++*seq; | |
7bced397 DW |
1993 | if (!(flags & MSG_PEEK)) |
1994 | sk_eat_skb(sk, skb); | |
1da177e4 LT |
1995 | break; |
1996 | } while (len > 0); | |
1997 | ||
1da177e4 LT |
1998 | /* According to UNIX98, msg_name/msg_namelen are ignored |
1999 | * on connected socket. I was just happy when found this 8) --ANK | |
2000 | */ | |
2001 | ||
98aaa913 MM |
2002 | if (has_tss) |
2003 | tcp_recv_timestamp(msg, sk, &tss); | |
2004 | ||
1da177e4 | 2005 | /* Clean up data we have read: This will do ACK frames. */ |
0e4b4992 | 2006 | tcp_cleanup_rbuf(sk, copied); |
1da177e4 | 2007 | |
1da177e4 LT |
2008 | release_sock(sk); |
2009 | return copied; | |
2010 | ||
2011 | out: | |
1da177e4 LT |
2012 | release_sock(sk); |
2013 | return err; | |
2014 | ||
2015 | recv_urg: | |
377f0a08 | 2016 | err = tcp_recv_urg(sk, msg, len, flags); |
1da177e4 | 2017 | goto out; |
c0e88ff0 PE |
2018 | |
2019 | recv_sndq: | |
2020 | err = tcp_peek_sndq(sk, msg, len); | |
2021 | goto out; | |
1da177e4 | 2022 | } |
4bc2f18b | 2023 | EXPORT_SYMBOL(tcp_recvmsg); |
1da177e4 | 2024 | |
490d5046 IJ |
2025 | void tcp_set_state(struct sock *sk, int state) |
2026 | { | |
2027 | int oldstate = sk->sk_state; | |
2028 | ||
d4487491 LB |
2029 | /* We defined a new enum for TCP states that are exported in BPF |
2030 | * so as not force the internal TCP states to be frozen. The | |
2031 | * following checks will detect if an internal state value ever | |
2032 | * differs from the BPF value. If this ever happens, then we will | |
2033 | * need to remap the internal value to the BPF value before calling | |
2034 | * tcp_call_bpf_2arg. | |
2035 | */ | |
2036 | BUILD_BUG_ON((int)BPF_TCP_ESTABLISHED != (int)TCP_ESTABLISHED); | |
2037 | BUILD_BUG_ON((int)BPF_TCP_SYN_SENT != (int)TCP_SYN_SENT); | |
2038 | BUILD_BUG_ON((int)BPF_TCP_SYN_RECV != (int)TCP_SYN_RECV); | |
2039 | BUILD_BUG_ON((int)BPF_TCP_FIN_WAIT1 != (int)TCP_FIN_WAIT1); | |
2040 | BUILD_BUG_ON((int)BPF_TCP_FIN_WAIT2 != (int)TCP_FIN_WAIT2); | |
2041 | BUILD_BUG_ON((int)BPF_TCP_TIME_WAIT != (int)TCP_TIME_WAIT); | |
2042 | BUILD_BUG_ON((int)BPF_TCP_CLOSE != (int)TCP_CLOSE); | |
2043 | BUILD_BUG_ON((int)BPF_TCP_CLOSE_WAIT != (int)TCP_CLOSE_WAIT); | |
2044 | BUILD_BUG_ON((int)BPF_TCP_LAST_ACK != (int)TCP_LAST_ACK); | |
2045 | BUILD_BUG_ON((int)BPF_TCP_LISTEN != (int)TCP_LISTEN); | |
2046 | BUILD_BUG_ON((int)BPF_TCP_CLOSING != (int)TCP_CLOSING); | |
2047 | BUILD_BUG_ON((int)BPF_TCP_NEW_SYN_RECV != (int)TCP_NEW_SYN_RECV); | |
2048 | BUILD_BUG_ON((int)BPF_TCP_MAX_STATES != (int)TCP_MAX_STATES); | |
2049 | ||
2050 | if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_STATE_CB_FLAG)) | |
2051 | tcp_call_bpf_2arg(sk, BPF_SOCK_OPS_STATE_CB, oldstate, state); | |
e8fce239 | 2052 | |
490d5046 IJ |
2053 | switch (state) { |
2054 | case TCP_ESTABLISHED: | |
2055 | if (oldstate != TCP_ESTABLISHED) | |
81cc8a75 | 2056 | TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB); |
490d5046 IJ |
2057 | break; |
2058 | ||
2059 | case TCP_CLOSE: | |
2060 | if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED) | |
81cc8a75 | 2061 | TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS); |
490d5046 IJ |
2062 | |
2063 | sk->sk_prot->unhash(sk); | |
2064 | if (inet_csk(sk)->icsk_bind_hash && | |
2065 | !(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) | |
ab1e0a13 | 2066 | inet_put_port(sk); |
490d5046 IJ |
2067 | /* fall through */ |
2068 | default: | |
5a5f3a8d | 2069 | if (oldstate == TCP_ESTABLISHED) |
74688e48 | 2070 | TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB); |
490d5046 IJ |
2071 | } |
2072 | ||
2073 | /* Change state AFTER socket is unhashed to avoid closed | |
2074 | * socket sitting in hash tables. | |
2075 | */ | |
563e0bb0 | 2076 | inet_sk_state_store(sk, state); |
490d5046 IJ |
2077 | |
2078 | #ifdef STATE_TRACE | |
5a5f3a8d | 2079 | SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n", sk, statename[oldstate], statename[state]); |
490d5046 IJ |
2080 | #endif |
2081 | } | |
2082 | EXPORT_SYMBOL_GPL(tcp_set_state); | |
2083 | ||
1da177e4 LT |
2084 | /* |
2085 | * State processing on a close. This implements the state shift for | |
2086 | * sending our FIN frame. Note that we only send a FIN for some | |
2087 | * states. A shutdown() may have already sent the FIN, or we may be | |
2088 | * closed. | |
2089 | */ | |
2090 | ||
9b5b5cff | 2091 | static const unsigned char new_state[16] = { |
1da177e4 | 2092 | /* current state: new state: action: */ |
0980c1e3 ED |
2093 | [0 /* (Invalid) */] = TCP_CLOSE, |
2094 | [TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN, | |
2095 | [TCP_SYN_SENT] = TCP_CLOSE, | |
2096 | [TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN, | |
2097 | [TCP_FIN_WAIT1] = TCP_FIN_WAIT1, | |
2098 | [TCP_FIN_WAIT2] = TCP_FIN_WAIT2, | |
2099 | [TCP_TIME_WAIT] = TCP_CLOSE, | |
2100 | [TCP_CLOSE] = TCP_CLOSE, | |
2101 | [TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN, | |
2102 | [TCP_LAST_ACK] = TCP_LAST_ACK, | |
2103 | [TCP_LISTEN] = TCP_CLOSE, | |
2104 | [TCP_CLOSING] = TCP_CLOSING, | |
2105 | [TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */ | |
1da177e4 LT |
2106 | }; |
2107 | ||
2108 | static int tcp_close_state(struct sock *sk) | |
2109 | { | |
2110 | int next = (int)new_state[sk->sk_state]; | |
2111 | int ns = next & TCP_STATE_MASK; | |
2112 | ||
2113 | tcp_set_state(sk, ns); | |
2114 | ||
2115 | return next & TCP_ACTION_FIN; | |
2116 | } | |
2117 | ||
2118 | /* | |
2119 | * Shutdown the sending side of a connection. Much like close except | |
1f29b058 | 2120 | * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD). |
1da177e4 LT |
2121 | */ |
2122 | ||
2123 | void tcp_shutdown(struct sock *sk, int how) | |
2124 | { | |
2125 | /* We need to grab some memory, and put together a FIN, | |
2126 | * and then put it into the queue to be sent. | |
2127 | * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92. | |
2128 | */ | |
2129 | if (!(how & SEND_SHUTDOWN)) | |
2130 | return; | |
2131 | ||
2132 | /* If we've already sent a FIN, or it's a closed state, skip this. */ | |
2133 | if ((1 << sk->sk_state) & | |
2134 | (TCPF_ESTABLISHED | TCPF_SYN_SENT | | |
2135 | TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) { | |
2136 | /* Clear out any half completed packets. FIN if needed. */ | |
2137 | if (tcp_close_state(sk)) | |
2138 | tcp_send_fin(sk); | |
2139 | } | |
2140 | } | |
4bc2f18b | 2141 | EXPORT_SYMBOL(tcp_shutdown); |
1da177e4 | 2142 | |
efcdbf24 AS |
2143 | bool tcp_check_oom(struct sock *sk, int shift) |
2144 | { | |
2145 | bool too_many_orphans, out_of_socket_memory; | |
2146 | ||
2147 | too_many_orphans = tcp_too_many_orphans(sk, shift); | |
2148 | out_of_socket_memory = tcp_out_of_memory(sk); | |
2149 | ||
e87cc472 JP |
2150 | if (too_many_orphans) |
2151 | net_info_ratelimited("too many orphaned sockets\n"); | |
2152 | if (out_of_socket_memory) | |
2153 | net_info_ratelimited("out of memory -- consider tuning tcp_mem\n"); | |
efcdbf24 AS |
2154 | return too_many_orphans || out_of_socket_memory; |
2155 | } | |
2156 | ||
1da177e4 LT |
2157 | void tcp_close(struct sock *sk, long timeout) |
2158 | { | |
2159 | struct sk_buff *skb; | |
2160 | int data_was_unread = 0; | |
75c2d907 | 2161 | int state; |
1da177e4 LT |
2162 | |
2163 | lock_sock(sk); | |
2164 | sk->sk_shutdown = SHUTDOWN_MASK; | |
2165 | ||
2166 | if (sk->sk_state == TCP_LISTEN) { | |
2167 | tcp_set_state(sk, TCP_CLOSE); | |
2168 | ||
2169 | /* Special case. */ | |
0a5578cf | 2170 | inet_csk_listen_stop(sk); |
1da177e4 LT |
2171 | |
2172 | goto adjudge_to_death; | |
2173 | } | |
2174 | ||
2175 | /* We need to flush the recv. buffs. We do this only on the | |
2176 | * descriptor close, not protocol-sourced closes, because the | |
2177 | * reader process may not have drained the data yet! | |
2178 | */ | |
2179 | while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) { | |
e11ecddf ED |
2180 | u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq; |
2181 | ||
2182 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) | |
2183 | len--; | |
1da177e4 LT |
2184 | data_was_unread += len; |
2185 | __kfree_skb(skb); | |
2186 | } | |
2187 | ||
3ab224be | 2188 | sk_mem_reclaim(sk); |
1da177e4 | 2189 | |
565b7b2d KK |
2190 | /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */ |
2191 | if (sk->sk_state == TCP_CLOSE) | |
2192 | goto adjudge_to_death; | |
2193 | ||
65bb723c GR |
2194 | /* As outlined in RFC 2525, section 2.17, we send a RST here because |
2195 | * data was lost. To witness the awful effects of the old behavior of | |
2196 | * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk | |
2197 | * GET in an FTP client, suspend the process, wait for the client to | |
2198 | * advertise a zero window, then kill -9 the FTP client, wheee... | |
2199 | * Note: timeout is always zero in such a case. | |
1da177e4 | 2200 | */ |
ee995283 PE |
2201 | if (unlikely(tcp_sk(sk)->repair)) { |
2202 | sk->sk_prot->disconnect(sk, 0); | |
2203 | } else if (data_was_unread) { | |
1da177e4 | 2204 | /* Unread data was tossed, zap the connection. */ |
6aef70a8 | 2205 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE); |
1da177e4 | 2206 | tcp_set_state(sk, TCP_CLOSE); |
aa133076 | 2207 | tcp_send_active_reset(sk, sk->sk_allocation); |
1da177e4 LT |
2208 | } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) { |
2209 | /* Check zero linger _after_ checking for unread data. */ | |
2210 | sk->sk_prot->disconnect(sk, 0); | |
6aef70a8 | 2211 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONDATA); |
1da177e4 LT |
2212 | } else if (tcp_close_state(sk)) { |
2213 | /* We FIN if the application ate all the data before | |
2214 | * zapping the connection. | |
2215 | */ | |
2216 | ||
2217 | /* RED-PEN. Formally speaking, we have broken TCP state | |
2218 | * machine. State transitions: | |
2219 | * | |
2220 | * TCP_ESTABLISHED -> TCP_FIN_WAIT1 | |
2221 | * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible) | |
2222 | * TCP_CLOSE_WAIT -> TCP_LAST_ACK | |
2223 | * | |
2224 | * are legal only when FIN has been sent (i.e. in window), | |
2225 | * rather than queued out of window. Purists blame. | |
2226 | * | |
2227 | * F.e. "RFC state" is ESTABLISHED, | |
2228 | * if Linux state is FIN-WAIT-1, but FIN is still not sent. | |
2229 | * | |
2230 | * The visible declinations are that sometimes | |
2231 | * we enter time-wait state, when it is not required really | |
2232 | * (harmless), do not send active resets, when they are | |
2233 | * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when | |
2234 | * they look as CLOSING or LAST_ACK for Linux) | |
2235 | * Probably, I missed some more holelets. | |
2236 | * --ANK | |
8336886f JC |
2237 | * XXX (TFO) - To start off we don't support SYN+ACK+FIN |
2238 | * in a single packet! (May consider it later but will | |
2239 | * probably need API support or TCP_CORK SYN-ACK until | |
2240 | * data is written and socket is closed.) | |
1da177e4 LT |
2241 | */ |
2242 | tcp_send_fin(sk); | |
2243 | } | |
2244 | ||
2245 | sk_stream_wait_close(sk, timeout); | |
2246 | ||
2247 | adjudge_to_death: | |
75c2d907 HX |
2248 | state = sk->sk_state; |
2249 | sock_hold(sk); | |
2250 | sock_orphan(sk); | |
75c2d907 | 2251 | |
1da177e4 LT |
2252 | /* It is the last release_sock in its life. It will remove backlog. */ |
2253 | release_sock(sk); | |
2254 | ||
2255 | ||
2256 | /* Now socket is owned by kernel and we acquire BH lock | |
a777f715 | 2257 | * to finish close. No need to check for user refs. |
1da177e4 LT |
2258 | */ |
2259 | local_bh_disable(); | |
2260 | bh_lock_sock(sk); | |
547b792c | 2261 | WARN_ON(sock_owned_by_user(sk)); |
1da177e4 | 2262 | |
eb4dea58 HX |
2263 | percpu_counter_inc(sk->sk_prot->orphan_count); |
2264 | ||
75c2d907 HX |
2265 | /* Have we already been destroyed by a softirq or backlog? */ |
2266 | if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE) | |
2267 | goto out; | |
1da177e4 LT |
2268 | |
2269 | /* This is a (useful) BSD violating of the RFC. There is a | |
2270 | * problem with TCP as specified in that the other end could | |
2271 | * keep a socket open forever with no application left this end. | |
b10bd54c | 2272 | * We use a 1 minute timeout (about the same as BSD) then kill |
1da177e4 LT |
2273 | * our end. If they send after that then tough - BUT: long enough |
2274 | * that we won't make the old 4*rto = almost no time - whoops | |
2275 | * reset mistake. | |
2276 | * | |
2277 | * Nope, it was not mistake. It is really desired behaviour | |
2278 | * f.e. on http servers, when such sockets are useless, but | |
2279 | * consume significant resources. Let's do it with special | |
2280 | * linger2 option. --ANK | |
2281 | */ | |
2282 | ||
2283 | if (sk->sk_state == TCP_FIN_WAIT2) { | |
2284 | struct tcp_sock *tp = tcp_sk(sk); | |
2285 | if (tp->linger2 < 0) { | |
2286 | tcp_set_state(sk, TCP_CLOSE); | |
2287 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
02a1d6e7 | 2288 | __NET_INC_STATS(sock_net(sk), |
de0744af | 2289 | LINUX_MIB_TCPABORTONLINGER); |
1da177e4 | 2290 | } else { |
463c84b9 | 2291 | const int tmo = tcp_fin_time(sk); |
1da177e4 LT |
2292 | |
2293 | if (tmo > TCP_TIMEWAIT_LEN) { | |
52499afe DM |
2294 | inet_csk_reset_keepalive_timer(sk, |
2295 | tmo - TCP_TIMEWAIT_LEN); | |
1da177e4 | 2296 | } else { |
1da177e4 LT |
2297 | tcp_time_wait(sk, TCP_FIN_WAIT2, tmo); |
2298 | goto out; | |
2299 | } | |
2300 | } | |
2301 | } | |
2302 | if (sk->sk_state != TCP_CLOSE) { | |
3ab224be | 2303 | sk_mem_reclaim(sk); |
efcdbf24 | 2304 | if (tcp_check_oom(sk, 0)) { |
1da177e4 LT |
2305 | tcp_set_state(sk, TCP_CLOSE); |
2306 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
02a1d6e7 | 2307 | __NET_INC_STATS(sock_net(sk), |
de0744af | 2308 | LINUX_MIB_TCPABORTONMEMORY); |
4ee806d5 DS |
2309 | } else if (!check_net(sock_net(sk))) { |
2310 | /* Not possible to send reset; just close */ | |
2311 | tcp_set_state(sk, TCP_CLOSE); | |
1da177e4 LT |
2312 | } |
2313 | } | |
1da177e4 | 2314 | |
8336886f JC |
2315 | if (sk->sk_state == TCP_CLOSE) { |
2316 | struct request_sock *req = tcp_sk(sk)->fastopen_rsk; | |
2317 | /* We could get here with a non-NULL req if the socket is | |
2318 | * aborted (e.g., closed with unread data) before 3WHS | |
2319 | * finishes. | |
2320 | */ | |
00db4124 | 2321 | if (req) |
8336886f | 2322 | reqsk_fastopen_remove(sk, req, false); |
0a5578cf | 2323 | inet_csk_destroy_sock(sk); |
8336886f | 2324 | } |
1da177e4 LT |
2325 | /* Otherwise, socket is reprieved until protocol close. */ |
2326 | ||
2327 | out: | |
2328 | bh_unlock_sock(sk); | |
2329 | local_bh_enable(); | |
2330 | sock_put(sk); | |
2331 | } | |
4bc2f18b | 2332 | EXPORT_SYMBOL(tcp_close); |
1da177e4 LT |
2333 | |
2334 | /* These states need RST on ABORT according to RFC793 */ | |
2335 | ||
a2a385d6 | 2336 | static inline bool tcp_need_reset(int state) |
1da177e4 LT |
2337 | { |
2338 | return (1 << state) & | |
2339 | (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 | | |
2340 | TCPF_FIN_WAIT2 | TCPF_SYN_RECV); | |
2341 | } | |
2342 | ||
75c119af ED |
2343 | static void tcp_rtx_queue_purge(struct sock *sk) |
2344 | { | |
2345 | struct rb_node *p = rb_first(&sk->tcp_rtx_queue); | |
2346 | ||
2347 | while (p) { | |
2348 | struct sk_buff *skb = rb_to_skb(p); | |
2349 | ||
2350 | p = rb_next(p); | |
2351 | /* Since we are deleting whole queue, no need to | |
2352 | * list_del(&skb->tcp_tsorted_anchor) | |
2353 | */ | |
2354 | tcp_rtx_queue_unlink(skb, sk); | |
2355 | sk_wmem_free_skb(sk, skb); | |
2356 | } | |
2357 | } | |
2358 | ||
ac3f09ba ED |
2359 | void tcp_write_queue_purge(struct sock *sk) |
2360 | { | |
2361 | struct sk_buff *skb; | |
2362 | ||
2363 | tcp_chrono_stop(sk, TCP_CHRONO_BUSY); | |
2364 | while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) { | |
2365 | tcp_skb_tsorted_anchor_cleanup(skb); | |
2366 | sk_wmem_free_skb(sk, skb); | |
2367 | } | |
75c119af | 2368 | tcp_rtx_queue_purge(sk); |
ac3f09ba ED |
2369 | INIT_LIST_HEAD(&tcp_sk(sk)->tsorted_sent_queue); |
2370 | sk_mem_reclaim(sk); | |
2371 | tcp_clear_all_retrans_hints(tcp_sk(sk)); | |
bffd168c | 2372 | tcp_sk(sk)->packets_out = 0; |
ac3f09ba ED |
2373 | } |
2374 | ||
1da177e4 LT |
2375 | int tcp_disconnect(struct sock *sk, int flags) |
2376 | { | |
2377 | struct inet_sock *inet = inet_sk(sk); | |
463c84b9 | 2378 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
2379 | struct tcp_sock *tp = tcp_sk(sk); |
2380 | int err = 0; | |
2381 | int old_state = sk->sk_state; | |
2382 | ||
2383 | if (old_state != TCP_CLOSE) | |
2384 | tcp_set_state(sk, TCP_CLOSE); | |
2385 | ||
2386 | /* ABORT function of RFC793 */ | |
2387 | if (old_state == TCP_LISTEN) { | |
0a5578cf | 2388 | inet_csk_listen_stop(sk); |
ee995283 PE |
2389 | } else if (unlikely(tp->repair)) { |
2390 | sk->sk_err = ECONNABORTED; | |
1da177e4 LT |
2391 | } else if (tcp_need_reset(old_state) || |
2392 | (tp->snd_nxt != tp->write_seq && | |
2393 | (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) { | |
caa20d9a | 2394 | /* The last check adjusts for discrepancy of Linux wrt. RFC |
1da177e4 LT |
2395 | * states |
2396 | */ | |
2397 | tcp_send_active_reset(sk, gfp_any()); | |
2398 | sk->sk_err = ECONNRESET; | |
2399 | } else if (old_state == TCP_SYN_SENT) | |
2400 | sk->sk_err = ECONNRESET; | |
2401 | ||
2402 | tcp_clear_xmit_timers(sk); | |
2403 | __skb_queue_purge(&sk->sk_receive_queue); | |
fe067e8a | 2404 | tcp_write_queue_purge(sk); |
cf1ef3f0 | 2405 | tcp_fastopen_active_disable_ofo_check(sk); |
9f5afeae | 2406 | skb_rbtree_purge(&tp->out_of_order_queue); |
1da177e4 | 2407 | |
c720c7e8 | 2408 | inet->inet_dport = 0; |
1da177e4 LT |
2409 | |
2410 | if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) | |
2411 | inet_reset_saddr(sk); | |
2412 | ||
2413 | sk->sk_shutdown = 0; | |
2414 | sock_reset_flag(sk, SOCK_DONE); | |
740b0f18 | 2415 | tp->srtt_us = 0; |
686a5624 YM |
2416 | tp->write_seq += tp->max_window + 2; |
2417 | if (tp->write_seq == 0) | |
1da177e4 | 2418 | tp->write_seq = 1; |
463c84b9 | 2419 | icsk->icsk_backoff = 0; |
1da177e4 | 2420 | tp->snd_cwnd = 2; |
6687e988 | 2421 | icsk->icsk_probes_out = 0; |
0b6a05c1 | 2422 | tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; |
1da177e4 | 2423 | tp->snd_cwnd_cnt = 0; |
1fdf475a | 2424 | tp->window_clamp = 0; |
6687e988 | 2425 | tcp_set_ca_state(sk, TCP_CA_Open); |
d4761754 | 2426 | tp->is_sack_reneg = 0; |
1da177e4 | 2427 | tcp_clear_retrans(tp); |
463c84b9 | 2428 | inet_csk_delack_init(sk); |
499350a5 WW |
2429 | /* Initialize rcv_mss to TCP_MIN_MSS to avoid division by 0 |
2430 | * issue in __tcp_select_window() | |
2431 | */ | |
2432 | icsk->icsk_ack.rcv_mss = TCP_MIN_MSS; | |
b40b4f79 | 2433 | memset(&tp->rx_opt, 0, sizeof(tp->rx_opt)); |
1da177e4 | 2434 | __sk_dst_reset(sk); |
d747a7a5 WC |
2435 | dst_release(sk->sk_rx_dst); |
2436 | sk->sk_rx_dst = NULL; | |
17c3060b | 2437 | tcp_saved_syn_free(tp); |
1da177e4 | 2438 | |
7db92362 WW |
2439 | /* Clean up fastopen related fields */ |
2440 | tcp_free_fastopen_req(tp); | |
2441 | inet->defer_connect = 0; | |
2442 | ||
c720c7e8 | 2443 | WARN_ON(inet->inet_num && !icsk->icsk_bind_hash); |
1da177e4 | 2444 | |
9b42d55a LR |
2445 | if (sk->sk_frag.page) { |
2446 | put_page(sk->sk_frag.page); | |
2447 | sk->sk_frag.page = NULL; | |
2448 | sk->sk_frag.offset = 0; | |
2449 | } | |
2450 | ||
1da177e4 LT |
2451 | sk->sk_error_report(sk); |
2452 | return err; | |
2453 | } | |
4bc2f18b | 2454 | EXPORT_SYMBOL(tcp_disconnect); |
1da177e4 | 2455 | |
a2a385d6 | 2456 | static inline bool tcp_can_repair_sock(const struct sock *sk) |
ee995283 | 2457 | { |
52e804c6 | 2458 | return ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN) && |
319b0534 | 2459 | (sk->sk_state != TCP_LISTEN); |
ee995283 PE |
2460 | } |
2461 | ||
b1ed4c4f AV |
2462 | static int tcp_repair_set_window(struct tcp_sock *tp, char __user *optbuf, int len) |
2463 | { | |
2464 | struct tcp_repair_window opt; | |
2465 | ||
2466 | if (!tp->repair) | |
2467 | return -EPERM; | |
2468 | ||
2469 | if (len != sizeof(opt)) | |
2470 | return -EINVAL; | |
2471 | ||
2472 | if (copy_from_user(&opt, optbuf, sizeof(opt))) | |
2473 | return -EFAULT; | |
2474 | ||
2475 | if (opt.max_window < opt.snd_wnd) | |
2476 | return -EINVAL; | |
2477 | ||
2478 | if (after(opt.snd_wl1, tp->rcv_nxt + opt.rcv_wnd)) | |
2479 | return -EINVAL; | |
2480 | ||
2481 | if (after(opt.rcv_wup, tp->rcv_nxt)) | |
2482 | return -EINVAL; | |
2483 | ||
2484 | tp->snd_wl1 = opt.snd_wl1; | |
2485 | tp->snd_wnd = opt.snd_wnd; | |
2486 | tp->max_window = opt.max_window; | |
2487 | ||
2488 | tp->rcv_wnd = opt.rcv_wnd; | |
2489 | tp->rcv_wup = opt.rcv_wup; | |
2490 | ||
2491 | return 0; | |
2492 | } | |
2493 | ||
15e56515 | 2494 | static int tcp_repair_options_est(struct sock *sk, |
de248a75 | 2495 | struct tcp_repair_opt __user *optbuf, unsigned int len) |
b139ba4e | 2496 | { |
15e56515 | 2497 | struct tcp_sock *tp = tcp_sk(sk); |
de248a75 | 2498 | struct tcp_repair_opt opt; |
b139ba4e | 2499 | |
de248a75 PE |
2500 | while (len >= sizeof(opt)) { |
2501 | if (copy_from_user(&opt, optbuf, sizeof(opt))) | |
b139ba4e PE |
2502 | return -EFAULT; |
2503 | ||
2504 | optbuf++; | |
de248a75 | 2505 | len -= sizeof(opt); |
b139ba4e | 2506 | |
de248a75 PE |
2507 | switch (opt.opt_code) { |
2508 | case TCPOPT_MSS: | |
2509 | tp->rx_opt.mss_clamp = opt.opt_val; | |
15e56515 | 2510 | tcp_mtup_init(sk); |
b139ba4e | 2511 | break; |
de248a75 | 2512 | case TCPOPT_WINDOW: |
bc26ccd8 AV |
2513 | { |
2514 | u16 snd_wscale = opt.opt_val & 0xFFFF; | |
2515 | u16 rcv_wscale = opt.opt_val >> 16; | |
2516 | ||
589c49cb | 2517 | if (snd_wscale > TCP_MAX_WSCALE || rcv_wscale > TCP_MAX_WSCALE) |
bc26ccd8 | 2518 | return -EFBIG; |
b139ba4e | 2519 | |
bc26ccd8 AV |
2520 | tp->rx_opt.snd_wscale = snd_wscale; |
2521 | tp->rx_opt.rcv_wscale = rcv_wscale; | |
2522 | tp->rx_opt.wscale_ok = 1; | |
2523 | } | |
b139ba4e | 2524 | break; |
b139ba4e | 2525 | case TCPOPT_SACK_PERM: |
de248a75 PE |
2526 | if (opt.opt_val != 0) |
2527 | return -EINVAL; | |
2528 | ||
b139ba4e | 2529 | tp->rx_opt.sack_ok |= TCP_SACK_SEEN; |
b139ba4e PE |
2530 | break; |
2531 | case TCPOPT_TIMESTAMP: | |
de248a75 PE |
2532 | if (opt.opt_val != 0) |
2533 | return -EINVAL; | |
2534 | ||
b139ba4e PE |
2535 | tp->rx_opt.tstamp_ok = 1; |
2536 | break; | |
2537 | } | |
2538 | } | |
2539 | ||
2540 | return 0; | |
2541 | } | |
2542 | ||
1da177e4 LT |
2543 | /* |
2544 | * Socket option code for TCP. | |
2545 | */ | |
3fdadf7d | 2546 | static int do_tcp_setsockopt(struct sock *sk, int level, |
b7058842 | 2547 | int optname, char __user *optval, unsigned int optlen) |
1da177e4 LT |
2548 | { |
2549 | struct tcp_sock *tp = tcp_sk(sk); | |
463c84b9 | 2550 | struct inet_connection_sock *icsk = inet_csk(sk); |
1e579caa | 2551 | struct net *net = sock_net(sk); |
1da177e4 LT |
2552 | int val; |
2553 | int err = 0; | |
2554 | ||
e56fb50f WAS |
2555 | /* These are data/string values, all the others are ints */ |
2556 | switch (optname) { | |
2557 | case TCP_CONGESTION: { | |
5f8ef48d SH |
2558 | char name[TCP_CA_NAME_MAX]; |
2559 | ||
2560 | if (optlen < 1) | |
2561 | return -EINVAL; | |
2562 | ||
2563 | val = strncpy_from_user(name, optval, | |
4fdb78d3 | 2564 | min_t(long, TCP_CA_NAME_MAX-1, optlen)); |
5f8ef48d SH |
2565 | if (val < 0) |
2566 | return -EFAULT; | |
2567 | name[val] = 0; | |
2568 | ||
2569 | lock_sock(sk); | |
ebfa00c5 | 2570 | err = tcp_set_congestion_control(sk, name, true, true); |
5f8ef48d SH |
2571 | release_sock(sk); |
2572 | return err; | |
2573 | } | |
734942cc DW |
2574 | case TCP_ULP: { |
2575 | char name[TCP_ULP_NAME_MAX]; | |
2576 | ||
2577 | if (optlen < 1) | |
2578 | return -EINVAL; | |
2579 | ||
2580 | val = strncpy_from_user(name, optval, | |
2581 | min_t(long, TCP_ULP_NAME_MAX - 1, | |
2582 | optlen)); | |
2583 | if (val < 0) | |
2584 | return -EFAULT; | |
2585 | name[val] = 0; | |
2586 | ||
2587 | lock_sock(sk); | |
2588 | err = tcp_set_ulp(sk, name); | |
2589 | release_sock(sk); | |
2590 | return err; | |
2591 | } | |
1fba70e5 YC |
2592 | case TCP_FASTOPEN_KEY: { |
2593 | __u8 key[TCP_FASTOPEN_KEY_LENGTH]; | |
2594 | ||
2595 | if (optlen != sizeof(key)) | |
2596 | return -EINVAL; | |
2597 | ||
2598 | if (copy_from_user(key, optval, optlen)) | |
2599 | return -EFAULT; | |
2600 | ||
2601 | return tcp_fastopen_reset_cipher(net, sk, key, sizeof(key)); | |
2602 | } | |
e56fb50f WAS |
2603 | default: |
2604 | /* fallthru */ | |
2605 | break; | |
ccbd6a5a | 2606 | } |
5f8ef48d | 2607 | |
1da177e4 LT |
2608 | if (optlen < sizeof(int)) |
2609 | return -EINVAL; | |
2610 | ||
2611 | if (get_user(val, (int __user *)optval)) | |
2612 | return -EFAULT; | |
2613 | ||
2614 | lock_sock(sk); | |
2615 | ||
2616 | switch (optname) { | |
2617 | case TCP_MAXSEG: | |
2618 | /* Values greater than interface MTU won't take effect. However | |
2619 | * at the point when this call is done we typically don't yet | |
a777f715 RC |
2620 | * know which interface is going to be used |
2621 | */ | |
cfc62d87 | 2622 | if (val && (val < TCP_MIN_MSS || val > MAX_TCP_WINDOW)) { |
1da177e4 LT |
2623 | err = -EINVAL; |
2624 | break; | |
2625 | } | |
2626 | tp->rx_opt.user_mss = val; | |
2627 | break; | |
2628 | ||
2629 | case TCP_NODELAY: | |
2630 | if (val) { | |
2631 | /* TCP_NODELAY is weaker than TCP_CORK, so that | |
2632 | * this option on corked socket is remembered, but | |
2633 | * it is not activated until cork is cleared. | |
2634 | * | |
2635 | * However, when TCP_NODELAY is set we make | |
2636 | * an explicit push, which overrides even TCP_CORK | |
2637 | * for currently queued segments. | |
2638 | */ | |
2639 | tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH; | |
9e412ba7 | 2640 | tcp_push_pending_frames(sk); |
1da177e4 LT |
2641 | } else { |
2642 | tp->nonagle &= ~TCP_NAGLE_OFF; | |
2643 | } | |
2644 | break; | |
2645 | ||
36e31b0a AP |
2646 | case TCP_THIN_LINEAR_TIMEOUTS: |
2647 | if (val < 0 || val > 1) | |
2648 | err = -EINVAL; | |
2649 | else | |
2650 | tp->thin_lto = val; | |
2651 | break; | |
2652 | ||
7e380175 AP |
2653 | case TCP_THIN_DUPACK: |
2654 | if (val < 0 || val > 1) | |
2655 | err = -EINVAL; | |
7e380175 AP |
2656 | break; |
2657 | ||
ee995283 PE |
2658 | case TCP_REPAIR: |
2659 | if (!tcp_can_repair_sock(sk)) | |
2660 | err = -EPERM; | |
2661 | else if (val == 1) { | |
2662 | tp->repair = 1; | |
2663 | sk->sk_reuse = SK_FORCE_REUSE; | |
2664 | tp->repair_queue = TCP_NO_QUEUE; | |
2665 | } else if (val == 0) { | |
2666 | tp->repair = 0; | |
2667 | sk->sk_reuse = SK_NO_REUSE; | |
2668 | tcp_send_window_probe(sk); | |
2669 | } else | |
2670 | err = -EINVAL; | |
2671 | ||
2672 | break; | |
2673 | ||
2674 | case TCP_REPAIR_QUEUE: | |
2675 | if (!tp->repair) | |
2676 | err = -EPERM; | |
bf2acc94 | 2677 | else if ((unsigned int)val < TCP_QUEUES_NR) |
ee995283 PE |
2678 | tp->repair_queue = val; |
2679 | else | |
2680 | err = -EINVAL; | |
2681 | break; | |
2682 | ||
2683 | case TCP_QUEUE_SEQ: | |
2684 | if (sk->sk_state != TCP_CLOSE) | |
2685 | err = -EPERM; | |
2686 | else if (tp->repair_queue == TCP_SEND_QUEUE) | |
2687 | tp->write_seq = val; | |
2688 | else if (tp->repair_queue == TCP_RECV_QUEUE) | |
2689 | tp->rcv_nxt = val; | |
2690 | else | |
2691 | err = -EINVAL; | |
2692 | break; | |
2693 | ||
b139ba4e PE |
2694 | case TCP_REPAIR_OPTIONS: |
2695 | if (!tp->repair) | |
2696 | err = -EINVAL; | |
2697 | else if (sk->sk_state == TCP_ESTABLISHED) | |
15e56515 | 2698 | err = tcp_repair_options_est(sk, |
de248a75 PE |
2699 | (struct tcp_repair_opt __user *)optval, |
2700 | optlen); | |
b139ba4e PE |
2701 | else |
2702 | err = -EPERM; | |
2703 | break; | |
2704 | ||
1da177e4 LT |
2705 | case TCP_CORK: |
2706 | /* When set indicates to always queue non-full frames. | |
2707 | * Later the user clears this option and we transmit | |
2708 | * any pending partial frames in the queue. This is | |
2709 | * meant to be used alongside sendfile() to get properly | |
2710 | * filled frames when the user (for example) must write | |
2711 | * out headers with a write() call first and then use | |
2712 | * sendfile to send out the data parts. | |
2713 | * | |
2714 | * TCP_CORK can be set together with TCP_NODELAY and it is | |
2715 | * stronger than TCP_NODELAY. | |
2716 | */ | |
2717 | if (val) { | |
2718 | tp->nonagle |= TCP_NAGLE_CORK; | |
2719 | } else { | |
2720 | tp->nonagle &= ~TCP_NAGLE_CORK; | |
2721 | if (tp->nonagle&TCP_NAGLE_OFF) | |
2722 | tp->nonagle |= TCP_NAGLE_PUSH; | |
9e412ba7 | 2723 | tcp_push_pending_frames(sk); |
1da177e4 LT |
2724 | } |
2725 | break; | |
2726 | ||
2727 | case TCP_KEEPIDLE: | |
2728 | if (val < 1 || val > MAX_TCP_KEEPIDLE) | |
2729 | err = -EINVAL; | |
2730 | else { | |
2731 | tp->keepalive_time = val * HZ; | |
2732 | if (sock_flag(sk, SOCK_KEEPOPEN) && | |
2733 | !((1 << sk->sk_state) & | |
2734 | (TCPF_CLOSE | TCPF_LISTEN))) { | |
6c37e5de | 2735 | u32 elapsed = keepalive_time_elapsed(tp); |
1da177e4 LT |
2736 | if (tp->keepalive_time > elapsed) |
2737 | elapsed = tp->keepalive_time - elapsed; | |
2738 | else | |
2739 | elapsed = 0; | |
463c84b9 | 2740 | inet_csk_reset_keepalive_timer(sk, elapsed); |
1da177e4 LT |
2741 | } |
2742 | } | |
2743 | break; | |
2744 | case TCP_KEEPINTVL: | |
2745 | if (val < 1 || val > MAX_TCP_KEEPINTVL) | |
2746 | err = -EINVAL; | |
2747 | else | |
2748 | tp->keepalive_intvl = val * HZ; | |
2749 | break; | |
2750 | case TCP_KEEPCNT: | |
2751 | if (val < 1 || val > MAX_TCP_KEEPCNT) | |
2752 | err = -EINVAL; | |
2753 | else | |
2754 | tp->keepalive_probes = val; | |
2755 | break; | |
2756 | case TCP_SYNCNT: | |
2757 | if (val < 1 || val > MAX_TCP_SYNCNT) | |
2758 | err = -EINVAL; | |
2759 | else | |
463c84b9 | 2760 | icsk->icsk_syn_retries = val; |
1da177e4 LT |
2761 | break; |
2762 | ||
cd8ae852 ED |
2763 | case TCP_SAVE_SYN: |
2764 | if (val < 0 || val > 1) | |
2765 | err = -EINVAL; | |
2766 | else | |
2767 | tp->save_syn = val; | |
2768 | break; | |
2769 | ||
1da177e4 LT |
2770 | case TCP_LINGER2: |
2771 | if (val < 0) | |
2772 | tp->linger2 = -1; | |
1e579caa | 2773 | else if (val > net->ipv4.sysctl_tcp_fin_timeout / HZ) |
1da177e4 LT |
2774 | tp->linger2 = 0; |
2775 | else | |
2776 | tp->linger2 = val * HZ; | |
2777 | break; | |
2778 | ||
2779 | case TCP_DEFER_ACCEPT: | |
b103cf34 JA |
2780 | /* Translate value in seconds to number of retransmits */ |
2781 | icsk->icsk_accept_queue.rskq_defer_accept = | |
2782 | secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ, | |
2783 | TCP_RTO_MAX / HZ); | |
1da177e4 LT |
2784 | break; |
2785 | ||
2786 | case TCP_WINDOW_CLAMP: | |
2787 | if (!val) { | |
2788 | if (sk->sk_state != TCP_CLOSE) { | |
2789 | err = -EINVAL; | |
2790 | break; | |
2791 | } | |
2792 | tp->window_clamp = 0; | |
2793 | } else | |
2794 | tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ? | |
2795 | SOCK_MIN_RCVBUF / 2 : val; | |
2796 | break; | |
2797 | ||
2798 | case TCP_QUICKACK: | |
2799 | if (!val) { | |
463c84b9 | 2800 | icsk->icsk_ack.pingpong = 1; |
1da177e4 | 2801 | } else { |
463c84b9 | 2802 | icsk->icsk_ack.pingpong = 0; |
1da177e4 LT |
2803 | if ((1 << sk->sk_state) & |
2804 | (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) && | |
463c84b9 ACM |
2805 | inet_csk_ack_scheduled(sk)) { |
2806 | icsk->icsk_ack.pending |= ICSK_ACK_PUSHED; | |
0e4b4992 | 2807 | tcp_cleanup_rbuf(sk, 1); |
1da177e4 | 2808 | if (!(val & 1)) |
463c84b9 | 2809 | icsk->icsk_ack.pingpong = 1; |
1da177e4 LT |
2810 | } |
2811 | } | |
2812 | break; | |
2813 | ||
cfb6eeb4 YH |
2814 | #ifdef CONFIG_TCP_MD5SIG |
2815 | case TCP_MD5SIG: | |
8917a777 | 2816 | case TCP_MD5SIG_EXT: |
72123032 ED |
2817 | if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) |
2818 | err = tp->af_specific->md5_parse(sk, optname, optval, optlen); | |
2819 | else | |
2820 | err = -EINVAL; | |
cfb6eeb4 YH |
2821 | break; |
2822 | #endif | |
dca43c75 | 2823 | case TCP_USER_TIMEOUT: |
b248230c | 2824 | /* Cap the max time in ms TCP will retry or probe the window |
dca43c75 JC |
2825 | * before giving up and aborting (ETIMEDOUT) a connection. |
2826 | */ | |
42493570 HL |
2827 | if (val < 0) |
2828 | err = -EINVAL; | |
2829 | else | |
2830 | icsk->icsk_user_timeout = msecs_to_jiffies(val); | |
dca43c75 | 2831 | break; |
8336886f JC |
2832 | |
2833 | case TCP_FASTOPEN: | |
2834 | if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE | | |
dfea2aa6 | 2835 | TCPF_LISTEN))) { |
43713848 | 2836 | tcp_fastopen_init_key_once(net); |
dfea2aa6 | 2837 | |
0536fcc0 | 2838 | fastopen_queue_tune(sk, val); |
dfea2aa6 | 2839 | } else { |
8336886f | 2840 | err = -EINVAL; |
dfea2aa6 | 2841 | } |
8336886f | 2842 | break; |
19f6d3f3 WW |
2843 | case TCP_FASTOPEN_CONNECT: |
2844 | if (val > 1 || val < 0) { | |
2845 | err = -EINVAL; | |
e1cfcbe8 | 2846 | } else if (net->ipv4.sysctl_tcp_fastopen & TFO_CLIENT_ENABLE) { |
19f6d3f3 WW |
2847 | if (sk->sk_state == TCP_CLOSE) |
2848 | tp->fastopen_connect = val; | |
2849 | else | |
2850 | err = -EINVAL; | |
2851 | } else { | |
2852 | err = -EOPNOTSUPP; | |
2853 | } | |
2854 | break; | |
71c02379 CP |
2855 | case TCP_FASTOPEN_NO_COOKIE: |
2856 | if (val > 1 || val < 0) | |
2857 | err = -EINVAL; | |
2858 | else if (!((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) | |
2859 | err = -EINVAL; | |
2860 | else | |
2861 | tp->fastopen_no_cookie = val; | |
2862 | break; | |
93be6ce0 AV |
2863 | case TCP_TIMESTAMP: |
2864 | if (!tp->repair) | |
2865 | err = -EPERM; | |
2866 | else | |
9a568de4 | 2867 | tp->tsoffset = val - tcp_time_stamp_raw(); |
93be6ce0 | 2868 | break; |
b1ed4c4f AV |
2869 | case TCP_REPAIR_WINDOW: |
2870 | err = tcp_repair_set_window(tp, optval, optlen); | |
2871 | break; | |
c9bee3b7 ED |
2872 | case TCP_NOTSENT_LOWAT: |
2873 | tp->notsent_lowat = val; | |
2874 | sk->sk_write_space(sk); | |
2875 | break; | |
1da177e4 LT |
2876 | default: |
2877 | err = -ENOPROTOOPT; | |
2878 | break; | |
3ff50b79 SH |
2879 | } |
2880 | ||
1da177e4 LT |
2881 | release_sock(sk); |
2882 | return err; | |
2883 | } | |
2884 | ||
3fdadf7d | 2885 | int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval, |
b7058842 | 2886 | unsigned int optlen) |
3fdadf7d | 2887 | { |
cf533ea5 | 2888 | const struct inet_connection_sock *icsk = inet_csk(sk); |
3fdadf7d DM |
2889 | |
2890 | if (level != SOL_TCP) | |
2891 | return icsk->icsk_af_ops->setsockopt(sk, level, optname, | |
2892 | optval, optlen); | |
2893 | return do_tcp_setsockopt(sk, level, optname, optval, optlen); | |
2894 | } | |
4bc2f18b | 2895 | EXPORT_SYMBOL(tcp_setsockopt); |
3fdadf7d DM |
2896 | |
2897 | #ifdef CONFIG_COMPAT | |
543d9cfe | 2898 | int compat_tcp_setsockopt(struct sock *sk, int level, int optname, |
b7058842 | 2899 | char __user *optval, unsigned int optlen) |
3fdadf7d | 2900 | { |
dec73ff0 ACM |
2901 | if (level != SOL_TCP) |
2902 | return inet_csk_compat_setsockopt(sk, level, optname, | |
2903 | optval, optlen); | |
3fdadf7d DM |
2904 | return do_tcp_setsockopt(sk, level, optname, optval, optlen); |
2905 | } | |
543d9cfe | 2906 | EXPORT_SYMBOL(compat_tcp_setsockopt); |
3fdadf7d DM |
2907 | #endif |
2908 | ||
efd90174 FY |
2909 | static void tcp_get_info_chrono_stats(const struct tcp_sock *tp, |
2910 | struct tcp_info *info) | |
2911 | { | |
2912 | u64 stats[__TCP_CHRONO_MAX], total = 0; | |
2913 | enum tcp_chrono i; | |
2914 | ||
2915 | for (i = TCP_CHRONO_BUSY; i < __TCP_CHRONO_MAX; ++i) { | |
2916 | stats[i] = tp->chrono_stat[i - 1]; | |
2917 | if (i == tp->chrono_type) | |
628174cc | 2918 | stats[i] += tcp_jiffies32 - tp->chrono_start; |
efd90174 FY |
2919 | stats[i] *= USEC_PER_SEC / HZ; |
2920 | total += stats[i]; | |
2921 | } | |
2922 | ||
2923 | info->tcpi_busy_time = total; | |
2924 | info->tcpi_rwnd_limited = stats[TCP_CHRONO_RWND_LIMITED]; | |
2925 | info->tcpi_sndbuf_limited = stats[TCP_CHRONO_SNDBUF_LIMITED]; | |
2926 | } | |
2927 | ||
1da177e4 | 2928 | /* Return information about state of tcp endpoint in API format. */ |
0df48c26 | 2929 | void tcp_get_info(struct sock *sk, struct tcp_info *info) |
1da177e4 | 2930 | { |
35ac838a | 2931 | const struct tcp_sock *tp = tcp_sk(sk); /* iff sk_type == SOCK_STREAM */ |
463c84b9 | 2932 | const struct inet_connection_sock *icsk = inet_csk(sk); |
0263598c | 2933 | u32 now; |
ff5d7497 | 2934 | u64 rate64; |
67db3e4b | 2935 | bool slow; |
fad9dfef | 2936 | u32 rate; |
1da177e4 LT |
2937 | |
2938 | memset(info, 0, sizeof(*info)); | |
35ac838a CG |
2939 | if (sk->sk_type != SOCK_STREAM) |
2940 | return; | |
1da177e4 | 2941 | |
986ffdfd | 2942 | info->tcpi_state = inet_sk_state_load(sk); |
00fd38d9 | 2943 | |
ccbf3bfa ED |
2944 | /* Report meaningful fields for all TCP states, including listeners */ |
2945 | rate = READ_ONCE(sk->sk_pacing_rate); | |
2946 | rate64 = rate != ~0U ? rate : ~0ULL; | |
f522a5fc | 2947 | info->tcpi_pacing_rate = rate64; |
ccbf3bfa ED |
2948 | |
2949 | rate = READ_ONCE(sk->sk_max_pacing_rate); | |
2950 | rate64 = rate != ~0U ? rate : ~0ULL; | |
f522a5fc | 2951 | info->tcpi_max_pacing_rate = rate64; |
ccbf3bfa ED |
2952 | |
2953 | info->tcpi_reordering = tp->reordering; | |
2954 | info->tcpi_snd_cwnd = tp->snd_cwnd; | |
2955 | ||
2956 | if (info->tcpi_state == TCP_LISTEN) { | |
2957 | /* listeners aliased fields : | |
2958 | * tcpi_unacked -> Number of children ready for accept() | |
2959 | * tcpi_sacked -> max backlog | |
2960 | */ | |
2961 | info->tcpi_unacked = sk->sk_ack_backlog; | |
2962 | info->tcpi_sacked = sk->sk_max_ack_backlog; | |
2963 | return; | |
2964 | } | |
b369e7fd ED |
2965 | |
2966 | slow = lock_sock_fast(sk); | |
2967 | ||
6687e988 | 2968 | info->tcpi_ca_state = icsk->icsk_ca_state; |
463c84b9 | 2969 | info->tcpi_retransmits = icsk->icsk_retransmits; |
6687e988 | 2970 | info->tcpi_probes = icsk->icsk_probes_out; |
463c84b9 | 2971 | info->tcpi_backoff = icsk->icsk_backoff; |
1da177e4 LT |
2972 | |
2973 | if (tp->rx_opt.tstamp_ok) | |
2974 | info->tcpi_options |= TCPI_OPT_TIMESTAMPS; | |
e60402d0 | 2975 | if (tcp_is_sack(tp)) |
1da177e4 LT |
2976 | info->tcpi_options |= TCPI_OPT_SACK; |
2977 | if (tp->rx_opt.wscale_ok) { | |
2978 | info->tcpi_options |= TCPI_OPT_WSCALE; | |
2979 | info->tcpi_snd_wscale = tp->rx_opt.snd_wscale; | |
2980 | info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale; | |
e905a9ed | 2981 | } |
1da177e4 | 2982 | |
b5c5693b | 2983 | if (tp->ecn_flags & TCP_ECN_OK) |
1da177e4 | 2984 | info->tcpi_options |= TCPI_OPT_ECN; |
b5c5693b ED |
2985 | if (tp->ecn_flags & TCP_ECN_SEEN) |
2986 | info->tcpi_options |= TCPI_OPT_ECN_SEEN; | |
6f73601e YC |
2987 | if (tp->syn_data_acked) |
2988 | info->tcpi_options |= TCPI_OPT_SYN_DATA; | |
1da177e4 | 2989 | |
463c84b9 ACM |
2990 | info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto); |
2991 | info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato); | |
c1b4a7e6 | 2992 | info->tcpi_snd_mss = tp->mss_cache; |
463c84b9 | 2993 | info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss; |
1da177e4 | 2994 | |
ccbf3bfa ED |
2995 | info->tcpi_unacked = tp->packets_out; |
2996 | info->tcpi_sacked = tp->sacked_out; | |
2997 | ||
1da177e4 LT |
2998 | info->tcpi_lost = tp->lost_out; |
2999 | info->tcpi_retrans = tp->retrans_out; | |
1da177e4 | 3000 | |
d635fbe2 | 3001 | now = tcp_jiffies32; |
1da177e4 | 3002 | info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime); |
463c84b9 | 3003 | info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime); |
1da177e4 LT |
3004 | info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp); |
3005 | ||
d83d8461 | 3006 | info->tcpi_pmtu = icsk->icsk_pmtu_cookie; |
1da177e4 | 3007 | info->tcpi_rcv_ssthresh = tp->rcv_ssthresh; |
740b0f18 ED |
3008 | info->tcpi_rtt = tp->srtt_us >> 3; |
3009 | info->tcpi_rttvar = tp->mdev_us >> 2; | |
1da177e4 | 3010 | info->tcpi_snd_ssthresh = tp->snd_ssthresh; |
1da177e4 | 3011 | info->tcpi_advmss = tp->advmss; |
1da177e4 | 3012 | |
645f4c6f | 3013 | info->tcpi_rcv_rtt = tp->rcv_rtt_est.rtt_us >> 3; |
1da177e4 LT |
3014 | info->tcpi_rcv_space = tp->rcvq_space.space; |
3015 | ||
3016 | info->tcpi_total_retrans = tp->total_retrans; | |
977cb0ec | 3017 | |
f522a5fc ED |
3018 | info->tcpi_bytes_acked = tp->bytes_acked; |
3019 | info->tcpi_bytes_received = tp->bytes_received; | |
67db3e4b | 3020 | info->tcpi_notsent_bytes = max_t(int, 0, tp->write_seq - tp->snd_nxt); |
efd90174 | 3021 | tcp_get_info_chrono_stats(tp, info); |
67db3e4b | 3022 | |
2efd055c MRL |
3023 | info->tcpi_segs_out = tp->segs_out; |
3024 | info->tcpi_segs_in = tp->segs_in; | |
cd9b2660 | 3025 | |
cd9b2660 | 3026 | info->tcpi_min_rtt = tcp_min_rtt(tp); |
a44d6eac MKL |
3027 | info->tcpi_data_segs_in = tp->data_segs_in; |
3028 | info->tcpi_data_segs_out = tp->data_segs_out; | |
eb8329e0 YC |
3029 | |
3030 | info->tcpi_delivery_rate_app_limited = tp->rate_app_limited ? 1 : 0; | |
0263598c WW |
3031 | rate64 = tcp_compute_delivery_rate(tp); |
3032 | if (rate64) | |
f522a5fc | 3033 | info->tcpi_delivery_rate = rate64; |
b369e7fd | 3034 | unlock_sock_fast(sk, slow); |
1da177e4 | 3035 | } |
1da177e4 LT |
3036 | EXPORT_SYMBOL_GPL(tcp_get_info); |
3037 | ||
1c885808 FY |
3038 | struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk) |
3039 | { | |
3040 | const struct tcp_sock *tp = tcp_sk(sk); | |
3041 | struct sk_buff *stats; | |
3042 | struct tcp_info info; | |
bb7c19f9 WW |
3043 | u64 rate64; |
3044 | u32 rate; | |
1c885808 | 3045 | |
bb7c19f9 | 3046 | stats = alloc_skb(7 * nla_total_size_64bit(sizeof(u64)) + |
7156d194 | 3047 | 5 * nla_total_size(sizeof(u32)) + |
be631892 | 3048 | 3 * nla_total_size(sizeof(u8)), GFP_ATOMIC); |
1c885808 FY |
3049 | if (!stats) |
3050 | return NULL; | |
3051 | ||
3052 | tcp_get_info_chrono_stats(tp, &info); | |
3053 | nla_put_u64_64bit(stats, TCP_NLA_BUSY, | |
3054 | info.tcpi_busy_time, TCP_NLA_PAD); | |
3055 | nla_put_u64_64bit(stats, TCP_NLA_RWND_LIMITED, | |
3056 | info.tcpi_rwnd_limited, TCP_NLA_PAD); | |
3057 | nla_put_u64_64bit(stats, TCP_NLA_SNDBUF_LIMITED, | |
3058 | info.tcpi_sndbuf_limited, TCP_NLA_PAD); | |
7e98102f YC |
3059 | nla_put_u64_64bit(stats, TCP_NLA_DATA_SEGS_OUT, |
3060 | tp->data_segs_out, TCP_NLA_PAD); | |
3061 | nla_put_u64_64bit(stats, TCP_NLA_TOTAL_RETRANS, | |
3062 | tp->total_retrans, TCP_NLA_PAD); | |
bb7c19f9 WW |
3063 | |
3064 | rate = READ_ONCE(sk->sk_pacing_rate); | |
3065 | rate64 = rate != ~0U ? rate : ~0ULL; | |
3066 | nla_put_u64_64bit(stats, TCP_NLA_PACING_RATE, rate64, TCP_NLA_PAD); | |
3067 | ||
3068 | rate64 = tcp_compute_delivery_rate(tp); | |
3069 | nla_put_u64_64bit(stats, TCP_NLA_DELIVERY_RATE, rate64, TCP_NLA_PAD); | |
3070 | ||
3071 | nla_put_u32(stats, TCP_NLA_SND_CWND, tp->snd_cwnd); | |
3072 | nla_put_u32(stats, TCP_NLA_REORDERING, tp->reordering); | |
3073 | nla_put_u32(stats, TCP_NLA_MIN_RTT, tcp_min_rtt(tp)); | |
3074 | ||
3075 | nla_put_u8(stats, TCP_NLA_RECUR_RETRANS, inet_csk(sk)->icsk_retransmits); | |
3076 | nla_put_u8(stats, TCP_NLA_DELIVERY_RATE_APP_LMT, !!tp->rate_app_limited); | |
7156d194 | 3077 | nla_put_u32(stats, TCP_NLA_SND_SSTHRESH, tp->snd_ssthresh); |
87ecc95d PJ |
3078 | |
3079 | nla_put_u32(stats, TCP_NLA_SNDQ_SIZE, tp->write_seq - tp->snd_una); | |
be631892 | 3080 | nla_put_u8(stats, TCP_NLA_CA_STATE, inet_csk(sk)->icsk_ca_state); |
1c885808 FY |
3081 | return stats; |
3082 | } | |
3083 | ||
3fdadf7d DM |
3084 | static int do_tcp_getsockopt(struct sock *sk, int level, |
3085 | int optname, char __user *optval, int __user *optlen) | |
1da177e4 | 3086 | { |
295f7324 | 3087 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 3088 | struct tcp_sock *tp = tcp_sk(sk); |
6fa25166 | 3089 | struct net *net = sock_net(sk); |
1da177e4 LT |
3090 | int val, len; |
3091 | ||
1da177e4 LT |
3092 | if (get_user(len, optlen)) |
3093 | return -EFAULT; | |
3094 | ||
3095 | len = min_t(unsigned int, len, sizeof(int)); | |
3096 | ||
3097 | if (len < 0) | |
3098 | return -EINVAL; | |
3099 | ||
3100 | switch (optname) { | |
3101 | case TCP_MAXSEG: | |
c1b4a7e6 | 3102 | val = tp->mss_cache; |
1da177e4 LT |
3103 | if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) |
3104 | val = tp->rx_opt.user_mss; | |
5e6a3ce6 PE |
3105 | if (tp->repair) |
3106 | val = tp->rx_opt.mss_clamp; | |
1da177e4 LT |
3107 | break; |
3108 | case TCP_NODELAY: | |
3109 | val = !!(tp->nonagle&TCP_NAGLE_OFF); | |
3110 | break; | |
3111 | case TCP_CORK: | |
3112 | val = !!(tp->nonagle&TCP_NAGLE_CORK); | |
3113 | break; | |
3114 | case TCP_KEEPIDLE: | |
df19a626 | 3115 | val = keepalive_time_when(tp) / HZ; |
1da177e4 LT |
3116 | break; |
3117 | case TCP_KEEPINTVL: | |
df19a626 | 3118 | val = keepalive_intvl_when(tp) / HZ; |
1da177e4 LT |
3119 | break; |
3120 | case TCP_KEEPCNT: | |
df19a626 | 3121 | val = keepalive_probes(tp); |
1da177e4 LT |
3122 | break; |
3123 | case TCP_SYNCNT: | |
6fa25166 | 3124 | val = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries; |
1da177e4 LT |
3125 | break; |
3126 | case TCP_LINGER2: | |
3127 | val = tp->linger2; | |
3128 | if (val >= 0) | |
1e579caa | 3129 | val = (val ? : net->ipv4.sysctl_tcp_fin_timeout) / HZ; |
1da177e4 LT |
3130 | break; |
3131 | case TCP_DEFER_ACCEPT: | |
b103cf34 JA |
3132 | val = retrans_to_secs(icsk->icsk_accept_queue.rskq_defer_accept, |
3133 | TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ); | |
1da177e4 LT |
3134 | break; |
3135 | case TCP_WINDOW_CLAMP: | |
3136 | val = tp->window_clamp; | |
3137 | break; | |
3138 | case TCP_INFO: { | |
3139 | struct tcp_info info; | |
3140 | ||
3141 | if (get_user(len, optlen)) | |
3142 | return -EFAULT; | |
3143 | ||
3144 | tcp_get_info(sk, &info); | |
3145 | ||
3146 | len = min_t(unsigned int, len, sizeof(info)); | |
3147 | if (put_user(len, optlen)) | |
3148 | return -EFAULT; | |
3149 | if (copy_to_user(optval, &info, len)) | |
3150 | return -EFAULT; | |
3151 | return 0; | |
3152 | } | |
6e9250f5 ED |
3153 | case TCP_CC_INFO: { |
3154 | const struct tcp_congestion_ops *ca_ops; | |
3155 | union tcp_cc_info info; | |
3156 | size_t sz = 0; | |
3157 | int attr; | |
3158 | ||
3159 | if (get_user(len, optlen)) | |
3160 | return -EFAULT; | |
3161 | ||
3162 | ca_ops = icsk->icsk_ca_ops; | |
3163 | if (ca_ops && ca_ops->get_info) | |
3164 | sz = ca_ops->get_info(sk, ~0U, &attr, &info); | |
3165 | ||
3166 | len = min_t(unsigned int, len, sz); | |
3167 | if (put_user(len, optlen)) | |
3168 | return -EFAULT; | |
3169 | if (copy_to_user(optval, &info, len)) | |
3170 | return -EFAULT; | |
3171 | return 0; | |
3172 | } | |
1da177e4 | 3173 | case TCP_QUICKACK: |
295f7324 | 3174 | val = !icsk->icsk_ack.pingpong; |
1da177e4 | 3175 | break; |
5f8ef48d SH |
3176 | |
3177 | case TCP_CONGESTION: | |
3178 | if (get_user(len, optlen)) | |
3179 | return -EFAULT; | |
3180 | len = min_t(unsigned int, len, TCP_CA_NAME_MAX); | |
3181 | if (put_user(len, optlen)) | |
3182 | return -EFAULT; | |
6687e988 | 3183 | if (copy_to_user(optval, icsk->icsk_ca_ops->name, len)) |
5f8ef48d SH |
3184 | return -EFAULT; |
3185 | return 0; | |
e56fb50f | 3186 | |
734942cc DW |
3187 | case TCP_ULP: |
3188 | if (get_user(len, optlen)) | |
3189 | return -EFAULT; | |
3190 | len = min_t(unsigned int, len, TCP_ULP_NAME_MAX); | |
d97af30f DW |
3191 | if (!icsk->icsk_ulp_ops) { |
3192 | if (put_user(0, optlen)) | |
3193 | return -EFAULT; | |
3194 | return 0; | |
3195 | } | |
734942cc DW |
3196 | if (put_user(len, optlen)) |
3197 | return -EFAULT; | |
3198 | if (copy_to_user(optval, icsk->icsk_ulp_ops->name, len)) | |
3199 | return -EFAULT; | |
3200 | return 0; | |
3201 | ||
1fba70e5 YC |
3202 | case TCP_FASTOPEN_KEY: { |
3203 | __u8 key[TCP_FASTOPEN_KEY_LENGTH]; | |
3204 | struct tcp_fastopen_context *ctx; | |
3205 | ||
3206 | if (get_user(len, optlen)) | |
3207 | return -EFAULT; | |
3208 | ||
3209 | rcu_read_lock(); | |
3210 | ctx = rcu_dereference(icsk->icsk_accept_queue.fastopenq.ctx); | |
3211 | if (ctx) | |
3212 | memcpy(key, ctx->key, sizeof(key)); | |
3213 | else | |
3214 | len = 0; | |
3215 | rcu_read_unlock(); | |
3216 | ||
3217 | len = min_t(unsigned int, len, sizeof(key)); | |
3218 | if (put_user(len, optlen)) | |
3219 | return -EFAULT; | |
3220 | if (copy_to_user(optval, key, len)) | |
3221 | return -EFAULT; | |
3222 | return 0; | |
3223 | } | |
3c0fef0b JH |
3224 | case TCP_THIN_LINEAR_TIMEOUTS: |
3225 | val = tp->thin_lto; | |
3226 | break; | |
4a7f6009 | 3227 | |
3c0fef0b | 3228 | case TCP_THIN_DUPACK: |
4a7f6009 | 3229 | val = 0; |
3c0fef0b | 3230 | break; |
dca43c75 | 3231 | |
ee995283 PE |
3232 | case TCP_REPAIR: |
3233 | val = tp->repair; | |
3234 | break; | |
3235 | ||
3236 | case TCP_REPAIR_QUEUE: | |
3237 | if (tp->repair) | |
3238 | val = tp->repair_queue; | |
3239 | else | |
3240 | return -EINVAL; | |
3241 | break; | |
3242 | ||
b1ed4c4f AV |
3243 | case TCP_REPAIR_WINDOW: { |
3244 | struct tcp_repair_window opt; | |
3245 | ||
3246 | if (get_user(len, optlen)) | |
3247 | return -EFAULT; | |
3248 | ||
3249 | if (len != sizeof(opt)) | |
3250 | return -EINVAL; | |
3251 | ||
3252 | if (!tp->repair) | |
3253 | return -EPERM; | |
3254 | ||
3255 | opt.snd_wl1 = tp->snd_wl1; | |
3256 | opt.snd_wnd = tp->snd_wnd; | |
3257 | opt.max_window = tp->max_window; | |
3258 | opt.rcv_wnd = tp->rcv_wnd; | |
3259 | opt.rcv_wup = tp->rcv_wup; | |
3260 | ||
3261 | if (copy_to_user(optval, &opt, len)) | |
3262 | return -EFAULT; | |
3263 | return 0; | |
3264 | } | |
ee995283 PE |
3265 | case TCP_QUEUE_SEQ: |
3266 | if (tp->repair_queue == TCP_SEND_QUEUE) | |
3267 | val = tp->write_seq; | |
3268 | else if (tp->repair_queue == TCP_RECV_QUEUE) | |
3269 | val = tp->rcv_nxt; | |
3270 | else | |
3271 | return -EINVAL; | |
3272 | break; | |
3273 | ||
dca43c75 JC |
3274 | case TCP_USER_TIMEOUT: |
3275 | val = jiffies_to_msecs(icsk->icsk_user_timeout); | |
3276 | break; | |
1536e285 KN |
3277 | |
3278 | case TCP_FASTOPEN: | |
0536fcc0 | 3279 | val = icsk->icsk_accept_queue.fastopenq.max_qlen; |
1536e285 KN |
3280 | break; |
3281 | ||
19f6d3f3 WW |
3282 | case TCP_FASTOPEN_CONNECT: |
3283 | val = tp->fastopen_connect; | |
3284 | break; | |
3285 | ||
71c02379 CP |
3286 | case TCP_FASTOPEN_NO_COOKIE: |
3287 | val = tp->fastopen_no_cookie; | |
3288 | break; | |
3289 | ||
93be6ce0 | 3290 | case TCP_TIMESTAMP: |
9a568de4 | 3291 | val = tcp_time_stamp_raw() + tp->tsoffset; |
93be6ce0 | 3292 | break; |
c9bee3b7 ED |
3293 | case TCP_NOTSENT_LOWAT: |
3294 | val = tp->notsent_lowat; | |
3295 | break; | |
cd8ae852 ED |
3296 | case TCP_SAVE_SYN: |
3297 | val = tp->save_syn; | |
3298 | break; | |
3299 | case TCP_SAVED_SYN: { | |
3300 | if (get_user(len, optlen)) | |
3301 | return -EFAULT; | |
3302 | ||
3303 | lock_sock(sk); | |
3304 | if (tp->saved_syn) { | |
aea0929e EM |
3305 | if (len < tp->saved_syn[0]) { |
3306 | if (put_user(tp->saved_syn[0], optlen)) { | |
3307 | release_sock(sk); | |
3308 | return -EFAULT; | |
3309 | } | |
3310 | release_sock(sk); | |
3311 | return -EINVAL; | |
3312 | } | |
3313 | len = tp->saved_syn[0]; | |
cd8ae852 ED |
3314 | if (put_user(len, optlen)) { |
3315 | release_sock(sk); | |
3316 | return -EFAULT; | |
3317 | } | |
3318 | if (copy_to_user(optval, tp->saved_syn + 1, len)) { | |
3319 | release_sock(sk); | |
3320 | return -EFAULT; | |
3321 | } | |
3322 | tcp_saved_syn_free(tp); | |
3323 | release_sock(sk); | |
3324 | } else { | |
3325 | release_sock(sk); | |
3326 | len = 0; | |
3327 | if (put_user(len, optlen)) | |
3328 | return -EFAULT; | |
3329 | } | |
3330 | return 0; | |
3331 | } | |
1da177e4 LT |
3332 | default: |
3333 | return -ENOPROTOOPT; | |
3ff50b79 | 3334 | } |
1da177e4 LT |
3335 | |
3336 | if (put_user(len, optlen)) | |
3337 | return -EFAULT; | |
3338 | if (copy_to_user(optval, &val, len)) | |
3339 | return -EFAULT; | |
3340 | return 0; | |
3341 | } | |
3342 | ||
3fdadf7d DM |
3343 | int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval, |
3344 | int __user *optlen) | |
3345 | { | |
3346 | struct inet_connection_sock *icsk = inet_csk(sk); | |
3347 | ||
3348 | if (level != SOL_TCP) | |
3349 | return icsk->icsk_af_ops->getsockopt(sk, level, optname, | |
3350 | optval, optlen); | |
3351 | return do_tcp_getsockopt(sk, level, optname, optval, optlen); | |
3352 | } | |
4bc2f18b | 3353 | EXPORT_SYMBOL(tcp_getsockopt); |
3fdadf7d DM |
3354 | |
3355 | #ifdef CONFIG_COMPAT | |
543d9cfe ACM |
3356 | int compat_tcp_getsockopt(struct sock *sk, int level, int optname, |
3357 | char __user *optval, int __user *optlen) | |
3fdadf7d | 3358 | { |
dec73ff0 ACM |
3359 | if (level != SOL_TCP) |
3360 | return inet_csk_compat_getsockopt(sk, level, optname, | |
3361 | optval, optlen); | |
3fdadf7d DM |
3362 | return do_tcp_getsockopt(sk, level, optname, optval, optlen); |
3363 | } | |
543d9cfe | 3364 | EXPORT_SYMBOL(compat_tcp_getsockopt); |
3fdadf7d | 3365 | #endif |
1da177e4 | 3366 | |
cfb6eeb4 | 3367 | #ifdef CONFIG_TCP_MD5SIG |
349ce993 | 3368 | static DEFINE_PER_CPU(struct tcp_md5sig_pool, tcp_md5sig_pool); |
71cea17e | 3369 | static DEFINE_MUTEX(tcp_md5sig_mutex); |
349ce993 | 3370 | static bool tcp_md5sig_pool_populated = false; |
cfb6eeb4 | 3371 | |
71cea17e | 3372 | static void __tcp_alloc_md5sig_pool(void) |
cfb6eeb4 | 3373 | { |
cf80e0e4 | 3374 | struct crypto_ahash *hash; |
cfb6eeb4 | 3375 | int cpu; |
cfb6eeb4 | 3376 | |
cf80e0e4 | 3377 | hash = crypto_alloc_ahash("md5", 0, CRYPTO_ALG_ASYNC); |
70477371 | 3378 | if (IS_ERR(hash)) |
cf80e0e4 HX |
3379 | return; |
3380 | ||
cfb6eeb4 | 3381 | for_each_possible_cpu(cpu) { |
19689e38 | 3382 | void *scratch = per_cpu(tcp_md5sig_pool, cpu).scratch; |
cf80e0e4 | 3383 | struct ahash_request *req; |
cfb6eeb4 | 3384 | |
19689e38 ED |
3385 | if (!scratch) { |
3386 | scratch = kmalloc_node(sizeof(union tcp_md5sum_block) + | |
3387 | sizeof(struct tcphdr), | |
3388 | GFP_KERNEL, | |
3389 | cpu_to_node(cpu)); | |
3390 | if (!scratch) | |
3391 | return; | |
3392 | per_cpu(tcp_md5sig_pool, cpu).scratch = scratch; | |
3393 | } | |
cf80e0e4 HX |
3394 | if (per_cpu(tcp_md5sig_pool, cpu).md5_req) |
3395 | continue; | |
3396 | ||
3397 | req = ahash_request_alloc(hash, GFP_KERNEL); | |
3398 | if (!req) | |
3399 | return; | |
3400 | ||
3401 | ahash_request_set_callback(req, 0, NULL, NULL); | |
3402 | ||
3403 | per_cpu(tcp_md5sig_pool, cpu).md5_req = req; | |
cfb6eeb4 | 3404 | } |
349ce993 ED |
3405 | /* before setting tcp_md5sig_pool_populated, we must commit all writes |
3406 | * to memory. See smp_rmb() in tcp_get_md5sig_pool() | |
71cea17e ED |
3407 | */ |
3408 | smp_wmb(); | |
349ce993 | 3409 | tcp_md5sig_pool_populated = true; |
cfb6eeb4 YH |
3410 | } |
3411 | ||
71cea17e | 3412 | bool tcp_alloc_md5sig_pool(void) |
cfb6eeb4 | 3413 | { |
349ce993 | 3414 | if (unlikely(!tcp_md5sig_pool_populated)) { |
71cea17e ED |
3415 | mutex_lock(&tcp_md5sig_mutex); |
3416 | ||
349ce993 | 3417 | if (!tcp_md5sig_pool_populated) |
71cea17e ED |
3418 | __tcp_alloc_md5sig_pool(); |
3419 | ||
3420 | mutex_unlock(&tcp_md5sig_mutex); | |
cfb6eeb4 | 3421 | } |
349ce993 | 3422 | return tcp_md5sig_pool_populated; |
cfb6eeb4 | 3423 | } |
cfb6eeb4 YH |
3424 | EXPORT_SYMBOL(tcp_alloc_md5sig_pool); |
3425 | ||
35790c04 ED |
3426 | |
3427 | /** | |
3428 | * tcp_get_md5sig_pool - get md5sig_pool for this user | |
3429 | * | |
3430 | * We use percpu structure, so if we succeed, we exit with preemption | |
3431 | * and BH disabled, to make sure another thread or softirq handling | |
3432 | * wont try to get same context. | |
3433 | */ | |
3434 | struct tcp_md5sig_pool *tcp_get_md5sig_pool(void) | |
cfb6eeb4 | 3435 | { |
35790c04 | 3436 | local_bh_disable(); |
cfb6eeb4 | 3437 | |
349ce993 ED |
3438 | if (tcp_md5sig_pool_populated) { |
3439 | /* coupled with smp_wmb() in __tcp_alloc_md5sig_pool() */ | |
3440 | smp_rmb(); | |
3441 | return this_cpu_ptr(&tcp_md5sig_pool); | |
3442 | } | |
35790c04 ED |
3443 | local_bh_enable(); |
3444 | return NULL; | |
3445 | } | |
3446 | EXPORT_SYMBOL(tcp_get_md5sig_pool); | |
cfb6eeb4 | 3447 | |
49a72dfb | 3448 | int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp, |
cf533ea5 | 3449 | const struct sk_buff *skb, unsigned int header_len) |
49a72dfb AL |
3450 | { |
3451 | struct scatterlist sg; | |
3452 | const struct tcphdr *tp = tcp_hdr(skb); | |
cf80e0e4 | 3453 | struct ahash_request *req = hp->md5_req; |
95c96174 ED |
3454 | unsigned int i; |
3455 | const unsigned int head_data_len = skb_headlen(skb) > header_len ? | |
3456 | skb_headlen(skb) - header_len : 0; | |
49a72dfb | 3457 | const struct skb_shared_info *shi = skb_shinfo(skb); |
d7fd1b57 | 3458 | struct sk_buff *frag_iter; |
49a72dfb AL |
3459 | |
3460 | sg_init_table(&sg, 1); | |
3461 | ||
3462 | sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len); | |
cf80e0e4 HX |
3463 | ahash_request_set_crypt(req, &sg, NULL, head_data_len); |
3464 | if (crypto_ahash_update(req)) | |
49a72dfb AL |
3465 | return 1; |
3466 | ||
3467 | for (i = 0; i < shi->nr_frags; ++i) { | |
3468 | const struct skb_frag_struct *f = &shi->frags[i]; | |
54d27fcb ED |
3469 | unsigned int offset = f->page_offset; |
3470 | struct page *page = skb_frag_page(f) + (offset >> PAGE_SHIFT); | |
3471 | ||
3472 | sg_set_page(&sg, page, skb_frag_size(f), | |
3473 | offset_in_page(offset)); | |
cf80e0e4 HX |
3474 | ahash_request_set_crypt(req, &sg, NULL, skb_frag_size(f)); |
3475 | if (crypto_ahash_update(req)) | |
49a72dfb AL |
3476 | return 1; |
3477 | } | |
3478 | ||
d7fd1b57 ED |
3479 | skb_walk_frags(skb, frag_iter) |
3480 | if (tcp_md5_hash_skb_data(hp, frag_iter, 0)) | |
3481 | return 1; | |
3482 | ||
49a72dfb AL |
3483 | return 0; |
3484 | } | |
49a72dfb AL |
3485 | EXPORT_SYMBOL(tcp_md5_hash_skb_data); |
3486 | ||
cf533ea5 | 3487 | int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, const struct tcp_md5sig_key *key) |
49a72dfb AL |
3488 | { |
3489 | struct scatterlist sg; | |
3490 | ||
3491 | sg_init_one(&sg, key->key, key->keylen); | |
cf80e0e4 HX |
3492 | ahash_request_set_crypt(hp->md5_req, &sg, NULL, key->keylen); |
3493 | return crypto_ahash_update(hp->md5_req); | |
49a72dfb | 3494 | } |
49a72dfb AL |
3495 | EXPORT_SYMBOL(tcp_md5_hash_key); |
3496 | ||
cfb6eeb4 YH |
3497 | #endif |
3498 | ||
4ac02bab AK |
3499 | void tcp_done(struct sock *sk) |
3500 | { | |
8336886f JC |
3501 | struct request_sock *req = tcp_sk(sk)->fastopen_rsk; |
3502 | ||
5a5f3a8d | 3503 | if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV) |
c10d9310 | 3504 | TCP_INC_STATS(sock_net(sk), TCP_MIB_ATTEMPTFAILS); |
4ac02bab AK |
3505 | |
3506 | tcp_set_state(sk, TCP_CLOSE); | |
3507 | tcp_clear_xmit_timers(sk); | |
00db4124 | 3508 | if (req) |
8336886f | 3509 | reqsk_fastopen_remove(sk, req, false); |
4ac02bab AK |
3510 | |
3511 | sk->sk_shutdown = SHUTDOWN_MASK; | |
3512 | ||
3513 | if (!sock_flag(sk, SOCK_DEAD)) | |
3514 | sk->sk_state_change(sk); | |
3515 | else | |
3516 | inet_csk_destroy_sock(sk); | |
3517 | } | |
3518 | EXPORT_SYMBOL_GPL(tcp_done); | |
3519 | ||
c1e64e29 LC |
3520 | int tcp_abort(struct sock *sk, int err) |
3521 | { | |
3522 | if (!sk_fullsock(sk)) { | |
07f6f4a3 ED |
3523 | if (sk->sk_state == TCP_NEW_SYN_RECV) { |
3524 | struct request_sock *req = inet_reqsk(sk); | |
3525 | ||
3526 | local_bh_disable(); | |
3527 | inet_csk_reqsk_queue_drop_and_put(req->rsk_listener, | |
3528 | req); | |
3529 | local_bh_enable(); | |
3530 | return 0; | |
3531 | } | |
c1e64e29 LC |
3532 | return -EOPNOTSUPP; |
3533 | } | |
3534 | ||
3535 | /* Don't race with userspace socket closes such as tcp_close. */ | |
3536 | lock_sock(sk); | |
3537 | ||
2010b93e LC |
3538 | if (sk->sk_state == TCP_LISTEN) { |
3539 | tcp_set_state(sk, TCP_CLOSE); | |
3540 | inet_csk_listen_stop(sk); | |
3541 | } | |
3542 | ||
c1e64e29 LC |
3543 | /* Don't race with BH socket closes such as inet_csk_listen_stop. */ |
3544 | local_bh_disable(); | |
3545 | bh_lock_sock(sk); | |
3546 | ||
3547 | if (!sock_flag(sk, SOCK_DEAD)) { | |
3548 | sk->sk_err = err; | |
3549 | /* This barrier is coupled with smp_rmb() in tcp_poll() */ | |
3550 | smp_wmb(); | |
3551 | sk->sk_error_report(sk); | |
3552 | if (tcp_need_reset(sk->sk_state)) | |
3553 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
3554 | tcp_done(sk); | |
3555 | } | |
3556 | ||
3557 | bh_unlock_sock(sk); | |
3558 | local_bh_enable(); | |
e05836ac | 3559 | tcp_write_queue_purge(sk); |
c1e64e29 | 3560 | release_sock(sk); |
c1e64e29 LC |
3561 | return 0; |
3562 | } | |
3563 | EXPORT_SYMBOL_GPL(tcp_abort); | |
3564 | ||
5f8ef48d | 3565 | extern struct tcp_congestion_ops tcp_reno; |
1da177e4 LT |
3566 | |
3567 | static __initdata unsigned long thash_entries; | |
3568 | static int __init set_thash_entries(char *str) | |
3569 | { | |
413c27d8 EZ |
3570 | ssize_t ret; |
3571 | ||
1da177e4 LT |
3572 | if (!str) |
3573 | return 0; | |
413c27d8 EZ |
3574 | |
3575 | ret = kstrtoul(str, 0, &thash_entries); | |
3576 | if (ret) | |
3577 | return 0; | |
3578 | ||
1da177e4 LT |
3579 | return 1; |
3580 | } | |
3581 | __setup("thash_entries=", set_thash_entries); | |
3582 | ||
47d7a88c | 3583 | static void __init tcp_init_mem(void) |
4acb4190 | 3584 | { |
b66e91cc ED |
3585 | unsigned long limit = nr_free_buffer_pages() / 16; |
3586 | ||
4acb4190 | 3587 | limit = max(limit, 128UL); |
b66e91cc ED |
3588 | sysctl_tcp_mem[0] = limit / 4 * 3; /* 4.68 % */ |
3589 | sysctl_tcp_mem[1] = limit; /* 6.25 % */ | |
3590 | sysctl_tcp_mem[2] = sysctl_tcp_mem[0] * 2; /* 9.37 % */ | |
4acb4190 GC |
3591 | } |
3592 | ||
1da177e4 LT |
3593 | void __init tcp_init(void) |
3594 | { | |
b49960a0 | 3595 | int max_rshare, max_wshare, cnt; |
b2d3ea4a | 3596 | unsigned long limit; |
074b8517 | 3597 | unsigned int i; |
1da177e4 | 3598 | |
b2d3ea4a ED |
3599 | BUILD_BUG_ON(sizeof(struct tcp_skb_cb) > |
3600 | FIELD_SIZEOF(struct sk_buff, cb)); | |
1da177e4 | 3601 | |
908c7f19 TH |
3602 | percpu_counter_init(&tcp_sockets_allocated, 0, GFP_KERNEL); |
3603 | percpu_counter_init(&tcp_orphan_count, 0, GFP_KERNEL); | |
1946e672 | 3604 | inet_hashinfo_init(&tcp_hashinfo); |
27da6d37 MKL |
3605 | inet_hashinfo2_init(&tcp_hashinfo, "tcp_listen_portaddr_hash", |
3606 | thash_entries, 21, /* one slot per 2 MB*/ | |
3607 | 0, 64 * 1024); | |
6e04e021 ACM |
3608 | tcp_hashinfo.bind_bucket_cachep = |
3609 | kmem_cache_create("tcp_bind_bucket", | |
3610 | sizeof(struct inet_bind_bucket), 0, | |
20c2df83 | 3611 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3612 | |
1da177e4 LT |
3613 | /* Size and allocate the main established and bind bucket |
3614 | * hash tables. | |
3615 | * | |
3616 | * The methodology is similar to that of the buffer cache. | |
3617 | */ | |
6e04e021 | 3618 | tcp_hashinfo.ehash = |
1da177e4 | 3619 | alloc_large_system_hash("TCP established", |
0f7ff927 | 3620 | sizeof(struct inet_ehash_bucket), |
1da177e4 | 3621 | thash_entries, |
fd90b29d | 3622 | 17, /* one slot per 128 KB of memory */ |
9e950efa | 3623 | 0, |
1da177e4 | 3624 | NULL, |
f373b53b | 3625 | &tcp_hashinfo.ehash_mask, |
31fe62b9 | 3626 | 0, |
0ccfe618 | 3627 | thash_entries ? 0 : 512 * 1024); |
05dbc7b5 | 3628 | for (i = 0; i <= tcp_hashinfo.ehash_mask; i++) |
3ab5aee7 | 3629 | INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i); |
05dbc7b5 | 3630 | |
230140cf ED |
3631 | if (inet_ehash_locks_alloc(&tcp_hashinfo)) |
3632 | panic("TCP: failed to alloc ehash_locks"); | |
6e04e021 | 3633 | tcp_hashinfo.bhash = |
1da177e4 | 3634 | alloc_large_system_hash("TCP bind", |
0f7ff927 | 3635 | sizeof(struct inet_bind_hashbucket), |
f373b53b | 3636 | tcp_hashinfo.ehash_mask + 1, |
fd90b29d | 3637 | 17, /* one slot per 128 KB of memory */ |
9e950efa | 3638 | 0, |
6e04e021 | 3639 | &tcp_hashinfo.bhash_size, |
1da177e4 | 3640 | NULL, |
31fe62b9 | 3641 | 0, |
1da177e4 | 3642 | 64 * 1024); |
074b8517 | 3643 | tcp_hashinfo.bhash_size = 1U << tcp_hashinfo.bhash_size; |
6e04e021 ACM |
3644 | for (i = 0; i < tcp_hashinfo.bhash_size; i++) { |
3645 | spin_lock_init(&tcp_hashinfo.bhash[i].lock); | |
3646 | INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain); | |
1da177e4 LT |
3647 | } |
3648 | ||
c5ed63d6 ED |
3649 | |
3650 | cnt = tcp_hashinfo.ehash_mask + 1; | |
c5ed63d6 | 3651 | sysctl_tcp_max_orphans = cnt / 2; |
1da177e4 | 3652 | |
a4fe34bf | 3653 | tcp_init_mem(); |
c43b874d | 3654 | /* Set per-socket limits to no more than 1/128 the pressure threshold */ |
5fb84b14 | 3655 | limit = nr_free_buffer_pages() << (PAGE_SHIFT - 7); |
b49960a0 ED |
3656 | max_wshare = min(4UL*1024*1024, limit); |
3657 | max_rshare = min(6UL*1024*1024, limit); | |
7b4f4b5e | 3658 | |
356d1833 ED |
3659 | init_net.ipv4.sysctl_tcp_wmem[0] = SK_MEM_QUANTUM; |
3660 | init_net.ipv4.sysctl_tcp_wmem[1] = 16*1024; | |
3661 | init_net.ipv4.sysctl_tcp_wmem[2] = max(64*1024, max_wshare); | |
7b4f4b5e | 3662 | |
356d1833 ED |
3663 | init_net.ipv4.sysctl_tcp_rmem[0] = SK_MEM_QUANTUM; |
3664 | init_net.ipv4.sysctl_tcp_rmem[1] = 87380; | |
3665 | init_net.ipv4.sysctl_tcp_rmem[2] = max(87380, max_rshare); | |
1da177e4 | 3666 | |
afd46503 | 3667 | pr_info("Hash tables configured (established %u bind %u)\n", |
058bd4d2 | 3668 | tcp_hashinfo.ehash_mask + 1, tcp_hashinfo.bhash_size); |
317a76f9 | 3669 | |
1946e672 | 3670 | tcp_v4_init(); |
51c5d0c4 | 3671 | tcp_metrics_init(); |
55d8694f | 3672 | BUG_ON(tcp_register_congestion_control(&tcp_reno) != 0); |
46d3ceab | 3673 | tcp_tasklet_init(); |
1da177e4 | 3674 | } |