]>
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> |
1da177e4 LT |
273 | |
274 | #include <net/icmp.h> | |
cf60af03 | 275 | #include <net/inet_common.h> |
1da177e4 LT |
276 | #include <net/tcp.h> |
277 | #include <net/xfrm.h> | |
278 | #include <net/ip.h> | |
9c55e01c | 279 | #include <net/sock.h> |
1da177e4 | 280 | |
7c0f6ba6 | 281 | #include <linux/uaccess.h> |
1da177e4 | 282 | #include <asm/ioctls.h> |
076bb0c8 | 283 | #include <net/busy_poll.h> |
1da177e4 | 284 | |
e8fce239 SL |
285 | #include <trace/events/tcp.h> |
286 | ||
95bd09eb ED |
287 | int sysctl_tcp_min_tso_segs __read_mostly = 2; |
288 | ||
f54b3111 ED |
289 | int sysctl_tcp_autocorking __read_mostly = 1; |
290 | ||
dd24c001 | 291 | struct percpu_counter tcp_orphan_count; |
0a5578cf ACM |
292 | EXPORT_SYMBOL_GPL(tcp_orphan_count); |
293 | ||
a4fe34bf | 294 | long sysctl_tcp_mem[3] __read_mostly; |
b8059ead DM |
295 | int sysctl_tcp_wmem[3] __read_mostly; |
296 | int sysctl_tcp_rmem[3] __read_mostly; | |
1da177e4 | 297 | |
a4fe34bf | 298 | EXPORT_SYMBOL(sysctl_tcp_mem); |
1da177e4 LT |
299 | EXPORT_SYMBOL(sysctl_tcp_rmem); |
300 | EXPORT_SYMBOL(sysctl_tcp_wmem); | |
301 | ||
8d987e5c | 302 | atomic_long_t tcp_memory_allocated; /* Current allocated memory. */ |
1da177e4 | 303 | EXPORT_SYMBOL(tcp_memory_allocated); |
1748376b ED |
304 | |
305 | /* | |
306 | * Current number of TCP sockets. | |
307 | */ | |
308 | struct percpu_counter tcp_sockets_allocated; | |
1da177e4 LT |
309 | EXPORT_SYMBOL(tcp_sockets_allocated); |
310 | ||
9c55e01c JA |
311 | /* |
312 | * TCP splice context | |
313 | */ | |
314 | struct tcp_splice_state { | |
315 | struct pipe_inode_info *pipe; | |
316 | size_t len; | |
317 | unsigned int flags; | |
318 | }; | |
319 | ||
1da177e4 LT |
320 | /* |
321 | * Pressure flag: try to collapse. | |
322 | * Technical note: it is used by multiple contexts non atomically. | |
3ab224be | 323 | * All the __sk_mem_schedule() is of this nature: accounting |
1da177e4 LT |
324 | * is strict, actions are advisory and have some latency. |
325 | */ | |
06044751 ED |
326 | unsigned long tcp_memory_pressure __read_mostly; |
327 | EXPORT_SYMBOL_GPL(tcp_memory_pressure); | |
1da177e4 | 328 | |
5c52ba17 | 329 | void tcp_enter_memory_pressure(struct sock *sk) |
1da177e4 | 330 | { |
06044751 ED |
331 | unsigned long val; |
332 | ||
333 | if (tcp_memory_pressure) | |
334 | return; | |
335 | val = jiffies; | |
336 | ||
337 | if (!val) | |
338 | val--; | |
339 | if (!cmpxchg(&tcp_memory_pressure, 0, val)) | |
4e673444 | 340 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES); |
1da177e4 | 341 | } |
06044751 ED |
342 | EXPORT_SYMBOL_GPL(tcp_enter_memory_pressure); |
343 | ||
344 | void tcp_leave_memory_pressure(struct sock *sk) | |
345 | { | |
346 | unsigned long val; | |
347 | ||
348 | if (!tcp_memory_pressure) | |
349 | return; | |
350 | val = xchg(&tcp_memory_pressure, 0); | |
351 | if (val) | |
352 | NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURESCHRONO, | |
353 | jiffies_to_msecs(jiffies - val)); | |
354 | } | |
355 | EXPORT_SYMBOL_GPL(tcp_leave_memory_pressure); | |
1da177e4 | 356 | |
b103cf34 JA |
357 | /* Convert seconds to retransmits based on initial and max timeout */ |
358 | static u8 secs_to_retrans(int seconds, int timeout, int rto_max) | |
359 | { | |
360 | u8 res = 0; | |
361 | ||
362 | if (seconds > 0) { | |
363 | int period = timeout; | |
364 | ||
365 | res = 1; | |
366 | while (seconds > period && res < 255) { | |
367 | res++; | |
368 | timeout <<= 1; | |
369 | if (timeout > rto_max) | |
370 | timeout = rto_max; | |
371 | period += timeout; | |
372 | } | |
373 | } | |
374 | return res; | |
375 | } | |
376 | ||
377 | /* Convert retransmits to seconds based on initial and max timeout */ | |
378 | static int retrans_to_secs(u8 retrans, int timeout, int rto_max) | |
379 | { | |
380 | int period = 0; | |
381 | ||
382 | if (retrans > 0) { | |
383 | period = timeout; | |
384 | while (--retrans) { | |
385 | timeout <<= 1; | |
386 | if (timeout > rto_max) | |
387 | timeout = rto_max; | |
388 | period += timeout; | |
389 | } | |
390 | } | |
391 | return period; | |
392 | } | |
393 | ||
0263598c WW |
394 | static u64 tcp_compute_delivery_rate(const struct tcp_sock *tp) |
395 | { | |
396 | u32 rate = READ_ONCE(tp->rate_delivered); | |
397 | u32 intv = READ_ONCE(tp->rate_interval_us); | |
398 | u64 rate64 = 0; | |
399 | ||
400 | if (rate && intv) { | |
401 | rate64 = (u64)rate * tp->mss_cache * USEC_PER_SEC; | |
402 | do_div(rate64, intv); | |
403 | } | |
404 | return rate64; | |
405 | } | |
406 | ||
900f65d3 NC |
407 | /* Address-family independent initialization for a tcp_sock. |
408 | * | |
409 | * NOTE: A lot of things set to zero explicitly by call to | |
410 | * sk_alloc() so need not be done here. | |
411 | */ | |
412 | void tcp_init_sock(struct sock *sk) | |
413 | { | |
414 | struct inet_connection_sock *icsk = inet_csk(sk); | |
415 | struct tcp_sock *tp = tcp_sk(sk); | |
416 | ||
9f5afeae | 417 | tp->out_of_order_queue = RB_ROOT; |
75c119af | 418 | sk->tcp_rtx_queue = RB_ROOT; |
900f65d3 | 419 | tcp_init_xmit_timers(sk); |
46d3ceab | 420 | INIT_LIST_HEAD(&tp->tsq_node); |
e2080072 | 421 | INIT_LIST_HEAD(&tp->tsorted_sent_queue); |
900f65d3 NC |
422 | |
423 | icsk->icsk_rto = TCP_TIMEOUT_INIT; | |
740b0f18 | 424 | tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT); |
ac9517fc | 425 | minmax_reset(&tp->rtt_min, tcp_jiffies32, ~0U); |
900f65d3 NC |
426 | |
427 | /* So many TCP implementations out there (incorrectly) count the | |
428 | * initial SYN frame in their delayed-ACK and congestion control | |
429 | * algorithms that we must have the following bandaid to talk | |
430 | * efficiently to them. -DaveM | |
431 | */ | |
432 | tp->snd_cwnd = TCP_INIT_CWND; | |
433 | ||
d7722e85 SHY |
434 | /* There's a bubble in the pipe until at least the first ACK. */ |
435 | tp->app_limited = ~0U; | |
436 | ||
900f65d3 NC |
437 | /* See draft-stevens-tcpca-spec-01 for discussion of the |
438 | * initialization of these values. | |
439 | */ | |
440 | tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; | |
441 | tp->snd_cwnd_clamp = ~0; | |
442 | tp->mss_cache = TCP_MSS_DEFAULT; | |
443 | ||
1043e25f | 444 | tp->reordering = sock_net(sk)->ipv4.sysctl_tcp_reordering; |
55d8694f | 445 | tcp_assign_congestion_control(sk); |
900f65d3 | 446 | |
ceaa1fef AV |
447 | tp->tsoffset = 0; |
448 | ||
900f65d3 NC |
449 | sk->sk_state = TCP_CLOSE; |
450 | ||
451 | sk->sk_write_space = sk_stream_write_space; | |
452 | sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); | |
453 | ||
454 | icsk->icsk_sync_mss = tcp_sync_mss; | |
455 | ||
900f65d3 NC |
456 | sk->sk_sndbuf = sysctl_tcp_wmem[1]; |
457 | sk->sk_rcvbuf = sysctl_tcp_rmem[1]; | |
458 | ||
900f65d3 | 459 | sk_sockets_allocated_inc(sk); |
900f65d3 NC |
460 | } |
461 | EXPORT_SYMBOL(tcp_init_sock); | |
462 | ||
27204aaa WW |
463 | void tcp_init_transfer(struct sock *sk, int bpf_op) |
464 | { | |
465 | struct inet_connection_sock *icsk = inet_csk(sk); | |
466 | ||
467 | tcp_mtup_init(sk); | |
468 | icsk->icsk_af_ops->rebuild_header(sk); | |
469 | tcp_init_metrics(sk); | |
470 | tcp_call_bpf(sk, bpf_op); | |
471 | tcp_init_congestion_control(sk); | |
472 | tcp_init_buffer_space(sk); | |
473 | } | |
474 | ||
4e8cc228 | 475 | static void tcp_tx_timestamp(struct sock *sk, u16 tsflags) |
4ed2d765 | 476 | { |
4e8cc228 ED |
477 | struct sk_buff *skb = tcp_write_queue_tail(sk); |
478 | ||
ad02c4f5 | 479 | if (tsflags && skb) { |
f066e2b0 | 480 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
6b084928 | 481 | struct tcp_skb_cb *tcb = TCP_SKB_CB(skb); |
4ed2d765 | 482 | |
c14ac945 | 483 | sock_tx_timestamp(sk, tsflags, &shinfo->tx_flags); |
0a2cf20c SHY |
484 | if (tsflags & SOF_TIMESTAMPING_TX_ACK) |
485 | tcb->txstamp_ack = 1; | |
486 | if (tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) | |
f066e2b0 WB |
487 | shinfo->tskey = TCP_SKB_CB(skb)->seq + skb->len - 1; |
488 | } | |
4ed2d765 WB |
489 | } |
490 | ||
1da177e4 LT |
491 | /* |
492 | * Wait for a TCP event. | |
493 | * | |
494 | * Note that we don't need to lock the socket, as the upper poll layers | |
495 | * take care of normal races (between the test and the event) and we don't | |
496 | * go look at any of the socket buffers directly. | |
497 | */ | |
498 | unsigned int tcp_poll(struct file *file, struct socket *sock, poll_table *wait) | |
499 | { | |
500 | unsigned int mask; | |
501 | struct sock *sk = sock->sk; | |
cf533ea5 | 502 | const struct tcp_sock *tp = tcp_sk(sk); |
00fd38d9 | 503 | int state; |
1da177e4 | 504 | |
c3f1dbaf DM |
505 | sock_rps_record_flow(sk); |
506 | ||
aa395145 | 507 | sock_poll_wait(file, sk_sleep(sk), wait); |
00fd38d9 ED |
508 | |
509 | state = sk_state_load(sk); | |
510 | if (state == TCP_LISTEN) | |
dc40c7bc | 511 | return inet_csk_listen_poll(sk); |
1da177e4 LT |
512 | |
513 | /* Socket is not locked. We are protected from async events | |
70efce27 WN |
514 | * by poll logic and correct handling of state changes |
515 | * made by other threads is impossible in any case. | |
1da177e4 LT |
516 | */ |
517 | ||
518 | mask = 0; | |
1da177e4 LT |
519 | |
520 | /* | |
521 | * POLLHUP is certainly not done right. But poll() doesn't | |
522 | * have a notion of HUP in just one direction, and for a | |
523 | * socket the read side is more interesting. | |
524 | * | |
525 | * Some poll() documentation says that POLLHUP is incompatible | |
526 | * with the POLLOUT/POLLWR flags, so somebody should check this | |
527 | * all. But careful, it tends to be safer to return too many | |
528 | * bits than too few, and you can easily break real applications | |
529 | * if you don't tell them that something has hung up! | |
530 | * | |
531 | * Check-me. | |
532 | * | |
533 | * Check number 1. POLLHUP is _UNMASKABLE_ event (see UNIX98 and | |
534 | * our fs/select.c). It means that after we received EOF, | |
535 | * poll always returns immediately, making impossible poll() on write() | |
536 | * in state CLOSE_WAIT. One solution is evident --- to set POLLHUP | |
537 | * if and only if shutdown has been made in both directions. | |
538 | * Actually, it is interesting to look how Solaris and DUX | |
70efce27 | 539 | * solve this dilemma. I would prefer, if POLLHUP were maskable, |
1da177e4 LT |
540 | * then we could set it on SND_SHUTDOWN. BTW examples given |
541 | * in Stevens' books assume exactly this behaviour, it explains | |
70efce27 | 542 | * why POLLHUP is incompatible with POLLOUT. --ANK |
1da177e4 LT |
543 | * |
544 | * NOTE. Check for TCP_CLOSE is added. The goal is to prevent | |
545 | * blocking on fresh not-connected or disconnected socket. --ANK | |
546 | */ | |
00fd38d9 | 547 | if (sk->sk_shutdown == SHUTDOWN_MASK || state == TCP_CLOSE) |
1da177e4 LT |
548 | mask |= POLLHUP; |
549 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
f348d70a | 550 | mask |= POLLIN | POLLRDNORM | POLLRDHUP; |
1da177e4 | 551 | |
8336886f | 552 | /* Connected or passive Fast Open socket? */ |
00fd38d9 ED |
553 | if (state != TCP_SYN_SENT && |
554 | (state != TCP_SYN_RECV || tp->fastopen_rsk)) { | |
c7004482 DM |
555 | int target = sock_rcvlowat(sk, 0, INT_MAX); |
556 | ||
557 | if (tp->urg_seq == tp->copied_seq && | |
558 | !sock_flag(sk, SOCK_URGINLINE) && | |
559 | tp->urg_data) | |
b634f875 | 560 | target++; |
c7004482 | 561 | |
c7004482 | 562 | if (tp->rcv_nxt - tp->copied_seq >= target) |
1da177e4 LT |
563 | mask |= POLLIN | POLLRDNORM; |
564 | ||
565 | if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { | |
64dc6130 | 566 | if (sk_stream_is_writeable(sk)) { |
1da177e4 LT |
567 | mask |= POLLOUT | POLLWRNORM; |
568 | } else { /* send SIGIO later */ | |
9cd3e072 | 569 | sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 LT |
570 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
571 | ||
572 | /* Race breaker. If space is freed after | |
573 | * wspace test but before the flags are set, | |
3c715127 | 574 | * IO signal will be lost. Memory barrier |
575 | * pairs with the input side. | |
1da177e4 | 576 | */ |
3c715127 | 577 | smp_mb__after_atomic(); |
64dc6130 | 578 | if (sk_stream_is_writeable(sk)) |
1da177e4 LT |
579 | mask |= POLLOUT | POLLWRNORM; |
580 | } | |
d84ba638 KM |
581 | } else |
582 | mask |= POLLOUT | POLLWRNORM; | |
1da177e4 LT |
583 | |
584 | if (tp->urg_data & TCP_URG_VALID) | |
585 | mask |= POLLPRI; | |
d68be71e | 586 | } else if (state == TCP_SYN_SENT && inet_sk(sk)->defer_connect) { |
19f6d3f3 WW |
587 | /* Active TCP fastopen socket with defer_connect |
588 | * Return POLLOUT so application can call write() | |
589 | * in order for kernel to generate SYN+data | |
590 | */ | |
591 | mask |= POLLOUT | POLLWRNORM; | |
1da177e4 | 592 | } |
a4d25803 TM |
593 | /* This barrier is coupled with smp_wmb() in tcp_reset() */ |
594 | smp_rmb(); | |
4ed2d765 | 595 | if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue)) |
a4d25803 TM |
596 | mask |= POLLERR; |
597 | ||
1da177e4 LT |
598 | return mask; |
599 | } | |
4bc2f18b | 600 | EXPORT_SYMBOL(tcp_poll); |
1da177e4 LT |
601 | |
602 | int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg) | |
603 | { | |
604 | struct tcp_sock *tp = tcp_sk(sk); | |
605 | int answ; | |
0e71c55c | 606 | bool slow; |
1da177e4 LT |
607 | |
608 | switch (cmd) { | |
609 | case SIOCINQ: | |
610 | if (sk->sk_state == TCP_LISTEN) | |
611 | return -EINVAL; | |
612 | ||
0e71c55c | 613 | slow = lock_sock_fast(sk); |
473bd239 | 614 | answ = tcp_inq(sk); |
0e71c55c | 615 | unlock_sock_fast(sk, slow); |
1da177e4 LT |
616 | break; |
617 | case SIOCATMARK: | |
618 | answ = tp->urg_data && tp->urg_seq == tp->copied_seq; | |
619 | break; | |
620 | case SIOCOUTQ: | |
621 | if (sk->sk_state == TCP_LISTEN) | |
622 | return -EINVAL; | |
623 | ||
624 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) | |
625 | answ = 0; | |
626 | else | |
627 | answ = tp->write_seq - tp->snd_una; | |
628 | break; | |
2f4e1b39 MS |
629 | case SIOCOUTQNSD: |
630 | if (sk->sk_state == TCP_LISTEN) | |
631 | return -EINVAL; | |
632 | ||
633 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) | |
634 | answ = 0; | |
635 | else | |
636 | answ = tp->write_seq - tp->snd_nxt; | |
637 | break; | |
1da177e4 LT |
638 | default: |
639 | return -ENOIOCTLCMD; | |
3ff50b79 | 640 | } |
1da177e4 LT |
641 | |
642 | return put_user(answ, (int __user *)arg); | |
643 | } | |
4bc2f18b | 644 | EXPORT_SYMBOL(tcp_ioctl); |
1da177e4 | 645 | |
1da177e4 LT |
646 | static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb) |
647 | { | |
4de075e0 | 648 | TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH; |
1da177e4 LT |
649 | tp->pushed_seq = tp->write_seq; |
650 | } | |
651 | ||
a2a385d6 | 652 | static inline bool forced_push(const struct tcp_sock *tp) |
1da177e4 LT |
653 | { |
654 | return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1)); | |
655 | } | |
656 | ||
f4a775d1 | 657 | static void skb_entail(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 658 | { |
9e412ba7 | 659 | struct tcp_sock *tp = tcp_sk(sk); |
352d4800 ACM |
660 | struct tcp_skb_cb *tcb = TCP_SKB_CB(skb); |
661 | ||
662 | skb->csum = 0; | |
663 | tcb->seq = tcb->end_seq = tp->write_seq; | |
4de075e0 | 664 | tcb->tcp_flags = TCPHDR_ACK; |
352d4800 | 665 | tcb->sacked = 0; |
f4a775d1 | 666 | __skb_header_release(skb); |
fe067e8a | 667 | tcp_add_write_queue_tail(sk, skb); |
3ab224be HA |
668 | sk->sk_wmem_queued += skb->truesize; |
669 | sk_mem_charge(sk, skb->truesize); | |
89ebd197 | 670 | if (tp->nonagle & TCP_NAGLE_PUSH) |
e905a9ed | 671 | tp->nonagle &= ~TCP_NAGLE_PUSH; |
6f021c62 ED |
672 | |
673 | tcp_slow_start_after_idle_check(sk); | |
1da177e4 LT |
674 | } |
675 | ||
afeca340 | 676 | static inline void tcp_mark_urg(struct tcp_sock *tp, int flags) |
1da177e4 | 677 | { |
33f5f57e | 678 | if (flags & MSG_OOB) |
1da177e4 | 679 | tp->snd_up = tp->write_seq; |
1da177e4 LT |
680 | } |
681 | ||
f54b3111 | 682 | /* If a not yet filled skb is pushed, do not send it if |
a181ceb5 | 683 | * we have data packets in Qdisc or NIC queues : |
f54b3111 ED |
684 | * Because TX completion will happen shortly, it gives a chance |
685 | * to coalesce future sendmsg() payload into this skb, without | |
686 | * need for a timer, and with no latency trade off. | |
687 | * As packets containing data payload have a bigger truesize | |
a181ceb5 ED |
688 | * than pure acks (dataless) packets, the last checks prevent |
689 | * autocorking if we only have an ACK in Qdisc/NIC queues, | |
690 | * or if TX completion was delayed after we processed ACK packet. | |
f54b3111 ED |
691 | */ |
692 | static bool tcp_should_autocork(struct sock *sk, struct sk_buff *skb, | |
693 | int size_goal) | |
1da177e4 | 694 | { |
f54b3111 ED |
695 | return skb->len < size_goal && |
696 | sysctl_tcp_autocorking && | |
a181ceb5 | 697 | skb != tcp_write_queue_head(sk) && |
14afee4b | 698 | refcount_read(&sk->sk_wmem_alloc) > skb->truesize; |
f54b3111 ED |
699 | } |
700 | ||
701 | static void tcp_push(struct sock *sk, int flags, int mss_now, | |
702 | int nonagle, int size_goal) | |
703 | { | |
704 | struct tcp_sock *tp = tcp_sk(sk); | |
705 | struct sk_buff *skb; | |
afeca340 | 706 | |
f54b3111 | 707 | skb = tcp_write_queue_tail(sk); |
75c119af ED |
708 | if (!skb) |
709 | return; | |
f54b3111 ED |
710 | if (!(flags & MSG_MORE) || forced_push(tp)) |
711 | tcp_mark_push(tp, skb); | |
712 | ||
713 | tcp_mark_urg(tp, flags); | |
714 | ||
715 | if (tcp_should_autocork(sk, skb, size_goal)) { | |
716 | ||
717 | /* avoid atomic op if TSQ_THROTTLED bit is already set */ | |
7aa5470c | 718 | if (!test_bit(TSQ_THROTTLED, &sk->sk_tsq_flags)) { |
f54b3111 | 719 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAUTOCORKING); |
7aa5470c | 720 | set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags); |
f54b3111 | 721 | } |
a181ceb5 ED |
722 | /* It is possible TX completion already happened |
723 | * before we set TSQ_THROTTLED. | |
724 | */ | |
14afee4b | 725 | if (refcount_read(&sk->sk_wmem_alloc) > skb->truesize) |
a181ceb5 | 726 | return; |
1da177e4 | 727 | } |
f54b3111 ED |
728 | |
729 | if (flags & MSG_MORE) | |
730 | nonagle = TCP_NAGLE_CORK; | |
731 | ||
732 | __tcp_push_pending_frames(sk, mss_now, nonagle); | |
1da177e4 LT |
733 | } |
734 | ||
6ff7751d AB |
735 | static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, |
736 | unsigned int offset, size_t len) | |
9c55e01c JA |
737 | { |
738 | struct tcp_splice_state *tss = rd_desc->arg.data; | |
33966dd0 | 739 | int ret; |
9c55e01c | 740 | |
a60e3cc7 | 741 | ret = skb_splice_bits(skb, skb->sk, offset, tss->pipe, |
25869262 | 742 | min(rd_desc->count, len), tss->flags); |
33966dd0 WT |
743 | if (ret > 0) |
744 | rd_desc->count -= ret; | |
745 | return ret; | |
9c55e01c JA |
746 | } |
747 | ||
748 | static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss) | |
749 | { | |
750 | /* Store TCP splice context information in read_descriptor_t. */ | |
751 | read_descriptor_t rd_desc = { | |
752 | .arg.data = tss, | |
33966dd0 | 753 | .count = tss->len, |
9c55e01c JA |
754 | }; |
755 | ||
756 | return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv); | |
757 | } | |
758 | ||
759 | /** | |
760 | * tcp_splice_read - splice data from TCP socket to a pipe | |
761 | * @sock: socket to splice from | |
762 | * @ppos: position (not valid) | |
763 | * @pipe: pipe to splice to | |
764 | * @len: number of bytes to splice | |
765 | * @flags: splice modifier flags | |
766 | * | |
767 | * Description: | |
768 | * Will read pages from given socket and fill them into a pipe. | |
769 | * | |
770 | **/ | |
771 | ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos, | |
772 | struct pipe_inode_info *pipe, size_t len, | |
773 | unsigned int flags) | |
774 | { | |
775 | struct sock *sk = sock->sk; | |
776 | struct tcp_splice_state tss = { | |
777 | .pipe = pipe, | |
778 | .len = len, | |
779 | .flags = flags, | |
780 | }; | |
781 | long timeo; | |
782 | ssize_t spliced; | |
783 | int ret; | |
784 | ||
3a047bf8 | 785 | sock_rps_record_flow(sk); |
9c55e01c JA |
786 | /* |
787 | * We can't seek on a socket input | |
788 | */ | |
789 | if (unlikely(*ppos)) | |
790 | return -ESPIPE; | |
791 | ||
792 | ret = spliced = 0; | |
793 | ||
794 | lock_sock(sk); | |
795 | ||
42324c62 | 796 | timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK); |
9c55e01c JA |
797 | while (tss.len) { |
798 | ret = __tcp_splice_read(sk, &tss); | |
799 | if (ret < 0) | |
800 | break; | |
801 | else if (!ret) { | |
802 | if (spliced) | |
803 | break; | |
9c55e01c JA |
804 | if (sock_flag(sk, SOCK_DONE)) |
805 | break; | |
806 | if (sk->sk_err) { | |
807 | ret = sock_error(sk); | |
808 | break; | |
809 | } | |
810 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
811 | break; | |
812 | if (sk->sk_state == TCP_CLOSE) { | |
813 | /* | |
814 | * This occurs when user tries to read | |
815 | * from never connected socket. | |
816 | */ | |
817 | if (!sock_flag(sk, SOCK_DONE)) | |
818 | ret = -ENOTCONN; | |
819 | break; | |
820 | } | |
821 | if (!timeo) { | |
822 | ret = -EAGAIN; | |
823 | break; | |
824 | } | |
ccf7abb9 ED |
825 | /* if __tcp_splice_read() got nothing while we have |
826 | * an skb in receive queue, we do not want to loop. | |
827 | * This might happen with URG data. | |
828 | */ | |
829 | if (!skb_queue_empty(&sk->sk_receive_queue)) | |
830 | break; | |
dfbafc99 | 831 | sk_wait_data(sk, &timeo, NULL); |
9c55e01c JA |
832 | if (signal_pending(current)) { |
833 | ret = sock_intr_errno(timeo); | |
834 | break; | |
835 | } | |
836 | continue; | |
837 | } | |
838 | tss.len -= ret; | |
839 | spliced += ret; | |
840 | ||
33966dd0 WT |
841 | if (!timeo) |
842 | break; | |
9c55e01c JA |
843 | release_sock(sk); |
844 | lock_sock(sk); | |
845 | ||
846 | if (sk->sk_err || sk->sk_state == TCP_CLOSE || | |
33966dd0 | 847 | (sk->sk_shutdown & RCV_SHUTDOWN) || |
9c55e01c JA |
848 | signal_pending(current)) |
849 | break; | |
850 | } | |
851 | ||
852 | release_sock(sk); | |
853 | ||
854 | if (spliced) | |
855 | return spliced; | |
856 | ||
857 | return ret; | |
858 | } | |
4bc2f18b | 859 | EXPORT_SYMBOL(tcp_splice_read); |
9c55e01c | 860 | |
eb934478 ED |
861 | struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, |
862 | bool force_schedule) | |
f561d0f2 PE |
863 | { |
864 | struct sk_buff *skb; | |
865 | ||
866 | /* The TCP header must be at least 32-bit aligned. */ | |
867 | size = ALIGN(size, 4); | |
868 | ||
8e4d980a ED |
869 | if (unlikely(tcp_under_memory_pressure(sk))) |
870 | sk_mem_reclaim_partial(sk); | |
871 | ||
f561d0f2 | 872 | skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp); |
8e4d980a | 873 | if (likely(skb)) { |
eb934478 | 874 | bool mem_scheduled; |
8e4d980a | 875 | |
eb934478 ED |
876 | if (force_schedule) { |
877 | mem_scheduled = true; | |
8e4d980a ED |
878 | sk_forced_mem_schedule(sk, skb->truesize); |
879 | } else { | |
eb934478 | 880 | mem_scheduled = sk_wmem_schedule(sk, skb->truesize); |
8e4d980a | 881 | } |
eb934478 | 882 | if (likely(mem_scheduled)) { |
a21d4572 | 883 | skb_reserve(skb, sk->sk_prot->max_header); |
f561d0f2 PE |
884 | /* |
885 | * Make sure that we have exactly size bytes | |
886 | * available to the caller, no more, no less. | |
887 | */ | |
16fad69c | 888 | skb->reserved_tailroom = skb->end - skb->tail - size; |
e2080072 | 889 | INIT_LIST_HEAD(&skb->tcp_tsorted_anchor); |
f561d0f2 PE |
890 | return skb; |
891 | } | |
892 | __kfree_skb(skb); | |
893 | } else { | |
5c52ba17 | 894 | sk->sk_prot->enter_memory_pressure(sk); |
f561d0f2 PE |
895 | sk_stream_moderate_sndbuf(sk); |
896 | } | |
897 | return NULL; | |
898 | } | |
899 | ||
0c54b85f IJ |
900 | static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now, |
901 | int large_allowed) | |
902 | { | |
903 | struct tcp_sock *tp = tcp_sk(sk); | |
6c09fa09 | 904 | u32 new_size_goal, size_goal; |
605ad7f1 ED |
905 | |
906 | if (!large_allowed || !sk_can_gso(sk)) | |
907 | return mss_now; | |
908 | ||
6c09fa09 ED |
909 | /* Note : tcp_tso_autosize() will eventually split this later */ |
910 | new_size_goal = sk->sk_gso_max_size - 1 - MAX_TCP_HEADER; | |
605ad7f1 ED |
911 | new_size_goal = tcp_bound_to_half_wnd(tp, new_size_goal); |
912 | ||
913 | /* We try hard to avoid divides here */ | |
914 | size_goal = tp->gso_segs * mss_now; | |
915 | if (unlikely(new_size_goal < size_goal || | |
916 | new_size_goal >= size_goal + mss_now)) { | |
917 | tp->gso_segs = min_t(u16, new_size_goal / mss_now, | |
918 | sk->sk_gso_max_segs); | |
919 | size_goal = tp->gso_segs * mss_now; | |
0c54b85f IJ |
920 | } |
921 | ||
605ad7f1 | 922 | return max(size_goal, mss_now); |
0c54b85f IJ |
923 | } |
924 | ||
925 | static int tcp_send_mss(struct sock *sk, int *size_goal, int flags) | |
926 | { | |
927 | int mss_now; | |
928 | ||
929 | mss_now = tcp_current_mss(sk); | |
930 | *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB)); | |
931 | ||
932 | return mss_now; | |
933 | } | |
934 | ||
e3b5616a DW |
935 | ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, |
936 | size_t size, int flags) | |
1da177e4 LT |
937 | { |
938 | struct tcp_sock *tp = tcp_sk(sk); | |
c1b4a7e6 | 939 | int mss_now, size_goal; |
1da177e4 LT |
940 | int err; |
941 | ssize_t copied; | |
942 | long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); | |
943 | ||
8336886f JC |
944 | /* Wait for a connection to finish. One exception is TCP Fast Open |
945 | * (passive side) where data is allowed to be sent before a connection | |
946 | * is fully established. | |
947 | */ | |
948 | if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) && | |
949 | !tcp_passive_fastopen(sk)) { | |
686a5624 YM |
950 | err = sk_stream_wait_connect(sk, &timeo); |
951 | if (err != 0) | |
1da177e4 | 952 | goto out_err; |
8336886f | 953 | } |
1da177e4 | 954 | |
9cd3e072 | 955 | sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 | 956 | |
0c54b85f | 957 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
958 | copied = 0; |
959 | ||
960 | err = -EPIPE; | |
961 | if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) | |
0d6a775e | 962 | goto out_err; |
1da177e4 | 963 | |
64022d0b | 964 | while (size > 0) { |
fe067e8a | 965 | struct sk_buff *skb = tcp_write_queue_tail(sk); |
38ba0a65 | 966 | int copy, i; |
38ba0a65 | 967 | bool can_coalesce; |
1da177e4 | 968 | |
75c119af | 969 | if (!skb || (copy = size_goal - skb->len) <= 0 || |
c134ecb8 | 970 | !tcp_skb_can_collapse_to(skb)) { |
1da177e4 LT |
971 | new_segment: |
972 | if (!sk_stream_memory_free(sk)) | |
973 | goto wait_for_sndbuf; | |
974 | ||
eb934478 | 975 | skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation, |
75c119af | 976 | tcp_rtx_and_write_queues_empty(sk)); |
1da177e4 LT |
977 | if (!skb) |
978 | goto wait_for_memory; | |
979 | ||
9e412ba7 | 980 | skb_entail(sk, skb); |
c1b4a7e6 | 981 | copy = size_goal; |
1da177e4 LT |
982 | } |
983 | ||
984 | if (copy > size) | |
985 | copy = size; | |
986 | ||
987 | i = skb_shinfo(skb)->nr_frags; | |
988 | can_coalesce = skb_can_coalesce(skb, i, page, offset); | |
5f74f82e | 989 | if (!can_coalesce && i >= sysctl_max_skb_frags) { |
1da177e4 LT |
990 | tcp_mark_push(tp, skb); |
991 | goto new_segment; | |
992 | } | |
3ab224be | 993 | if (!sk_wmem_schedule(sk, copy)) |
1da177e4 | 994 | goto wait_for_memory; |
e905a9ed | 995 | |
1da177e4 | 996 | if (can_coalesce) { |
9e903e08 | 997 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); |
1da177e4 LT |
998 | } else { |
999 | get_page(page); | |
1000 | skb_fill_page_desc(skb, i, page, offset, copy); | |
1001 | } | |
c9af6db4 | 1002 | skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; |
cef401de | 1003 | |
1da177e4 LT |
1004 | skb->len += copy; |
1005 | skb->data_len += copy; | |
1006 | skb->truesize += copy; | |
1007 | sk->sk_wmem_queued += copy; | |
3ab224be | 1008 | sk_mem_charge(sk, copy); |
84fa7933 | 1009 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 LT |
1010 | tp->write_seq += copy; |
1011 | TCP_SKB_CB(skb)->end_seq += copy; | |
cd7d8498 | 1012 | tcp_skb_pcount_set(skb, 0); |
1da177e4 LT |
1013 | |
1014 | if (!copied) | |
4de075e0 | 1015 | TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH; |
1da177e4 LT |
1016 | |
1017 | copied += copy; | |
64022d0b | 1018 | offset += copy; |
686a5624 | 1019 | size -= copy; |
ad02c4f5 | 1020 | if (!size) |
1da177e4 LT |
1021 | goto out; |
1022 | ||
69d15067 | 1023 | if (skb->len < size_goal || (flags & MSG_OOB)) |
1da177e4 LT |
1024 | continue; |
1025 | ||
1026 | if (forced_push(tp)) { | |
1027 | tcp_mark_push(tp, skb); | |
9e412ba7 | 1028 | __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH); |
fe067e8a | 1029 | } else if (skb == tcp_send_head(sk)) |
1da177e4 LT |
1030 | tcp_push_one(sk, mss_now); |
1031 | continue; | |
1032 | ||
1033 | wait_for_sndbuf: | |
1034 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
1035 | wait_for_memory: | |
f54b3111 ED |
1036 | tcp_push(sk, flags & ~MSG_MORE, mss_now, |
1037 | TCP_NAGLE_PUSH, size_goal); | |
1da177e4 | 1038 | |
686a5624 YM |
1039 | err = sk_stream_wait_memory(sk, &timeo); |
1040 | if (err != 0) | |
1da177e4 LT |
1041 | goto do_error; |
1042 | ||
0c54b85f | 1043 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
1044 | } |
1045 | ||
1046 | out: | |
ad02c4f5 | 1047 | if (copied) { |
4e8cc228 | 1048 | tcp_tx_timestamp(sk, sk->sk_tsflags); |
ad02c4f5 SHY |
1049 | if (!(flags & MSG_SENDPAGE_NOTLAST)) |
1050 | tcp_push(sk, flags, mss_now, tp->nonagle, size_goal); | |
1051 | } | |
1da177e4 LT |
1052 | return copied; |
1053 | ||
1054 | do_error: | |
1055 | if (copied) | |
1056 | goto out; | |
1057 | out_err: | |
ce5ec440 | 1058 | /* make sure we wake any epoll edge trigger waiter */ |
b0f71bd3 FY |
1059 | if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && |
1060 | err == -EAGAIN)) { | |
ce5ec440 | 1061 | sk->sk_write_space(sk); |
b0f71bd3 FY |
1062 | tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED); |
1063 | } | |
1da177e4 LT |
1064 | return sk_stream_error(sk, flags, err); |
1065 | } | |
e3b5616a | 1066 | EXPORT_SYMBOL_GPL(do_tcp_sendpages); |
1da177e4 | 1067 | |
306b13eb TH |
1068 | int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, |
1069 | size_t size, int flags) | |
1da177e4 | 1070 | { |
1da177e4 | 1071 | if (!(sk->sk_route_caps & NETIF_F_SG) || |
9a49850d | 1072 | !sk_check_csum_caps(sk)) |
bd9dfc54 | 1073 | return sock_no_sendpage_locked(sk, page, offset, size, flags); |
1da177e4 | 1074 | |
d7722e85 SHY |
1075 | tcp_rate_check_app_limited(sk); /* is sending application-limited? */ |
1076 | ||
306b13eb TH |
1077 | return do_tcp_sendpages(sk, page, offset, size, flags); |
1078 | } | |
774c4673 | 1079 | EXPORT_SYMBOL_GPL(tcp_sendpage_locked); |
306b13eb TH |
1080 | |
1081 | int tcp_sendpage(struct sock *sk, struct page *page, int offset, | |
1082 | size_t size, int flags) | |
1083 | { | |
1084 | int ret; | |
1085 | ||
1086 | lock_sock(sk); | |
1087 | ret = tcp_sendpage_locked(sk, page, offset, size, flags); | |
1da177e4 | 1088 | release_sock(sk); |
306b13eb TH |
1089 | |
1090 | return ret; | |
1da177e4 | 1091 | } |
4bc2f18b | 1092 | EXPORT_SYMBOL(tcp_sendpage); |
1da177e4 | 1093 | |
3613b3db ED |
1094 | /* Do not bother using a page frag for very small frames. |
1095 | * But use this heuristic only for the first skb in write queue. | |
1096 | * | |
1097 | * Having no payload in skb->head allows better SACK shifting | |
1098 | * in tcp_shift_skb_data(), reducing sack/rack overhead, because | |
1099 | * write queue has less skbs. | |
1100 | * Each skb can hold up to MAX_SKB_FRAGS * 32Kbytes, or ~0.5 MB. | |
1101 | * This also speeds up tso_fragment(), since it wont fallback | |
1102 | * to tcp_fragment(). | |
1103 | */ | |
1104 | static int linear_payload_sz(bool first_skb) | |
1105 | { | |
1106 | if (first_skb) | |
1107 | return SKB_WITH_OVERHEAD(2048 - MAX_TCP_HEADER); | |
1108 | return 0; | |
1109 | } | |
1110 | ||
1111 | static int select_size(const struct sock *sk, bool sg, bool first_skb) | |
1da177e4 | 1112 | { |
cf533ea5 | 1113 | const struct tcp_sock *tp = tcp_sk(sk); |
c1b4a7e6 | 1114 | int tmp = tp->mss_cache; |
1da177e4 | 1115 | |
def87cf4 | 1116 | if (sg) { |
f07d960d | 1117 | if (sk_can_gso(sk)) { |
3613b3db | 1118 | tmp = linear_payload_sz(first_skb); |
f07d960d | 1119 | } else { |
b4e26f5e DM |
1120 | int pgbreak = SKB_MAX_HEAD(MAX_TCP_HEADER); |
1121 | ||
1122 | if (tmp >= pgbreak && | |
1123 | tmp <= pgbreak + (MAX_SKB_FRAGS - 1) * PAGE_SIZE) | |
1124 | tmp = pgbreak; | |
1125 | } | |
1126 | } | |
1da177e4 | 1127 | |
1da177e4 LT |
1128 | return tmp; |
1129 | } | |
1130 | ||
cf60af03 YC |
1131 | void tcp_free_fastopen_req(struct tcp_sock *tp) |
1132 | { | |
00db4124 | 1133 | if (tp->fastopen_req) { |
cf60af03 YC |
1134 | kfree(tp->fastopen_req); |
1135 | tp->fastopen_req = NULL; | |
1136 | } | |
1137 | } | |
1138 | ||
f5ddcbbb ED |
1139 | static int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, |
1140 | int *copied, size_t size) | |
cf60af03 YC |
1141 | { |
1142 | struct tcp_sock *tp = tcp_sk(sk); | |
19f6d3f3 | 1143 | struct inet_sock *inet = inet_sk(sk); |
ba615f67 | 1144 | struct sockaddr *uaddr = msg->msg_name; |
cf60af03 YC |
1145 | int err, flags; |
1146 | ||
e1cfcbe8 | 1147 | if (!(sock_net(sk)->ipv4.sysctl_tcp_fastopen & TFO_CLIENT_ENABLE) || |
ba615f67 WW |
1148 | (uaddr && msg->msg_namelen >= sizeof(uaddr->sa_family) && |
1149 | uaddr->sa_family == AF_UNSPEC)) | |
cf60af03 | 1150 | return -EOPNOTSUPP; |
00db4124 | 1151 | if (tp->fastopen_req) |
cf60af03 YC |
1152 | return -EALREADY; /* Another Fast Open is in progress */ |
1153 | ||
1154 | tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request), | |
1155 | sk->sk_allocation); | |
51456b29 | 1156 | if (unlikely(!tp->fastopen_req)) |
cf60af03 YC |
1157 | return -ENOBUFS; |
1158 | tp->fastopen_req->data = msg; | |
f5ddcbbb | 1159 | tp->fastopen_req->size = size; |
cf60af03 | 1160 | |
19f6d3f3 WW |
1161 | if (inet->defer_connect) { |
1162 | err = tcp_connect(sk); | |
1163 | /* Same failure procedure as in tcp_v4/6_connect */ | |
1164 | if (err) { | |
1165 | tcp_set_state(sk, TCP_CLOSE); | |
1166 | inet->inet_dport = 0; | |
1167 | sk->sk_route_caps = 0; | |
1168 | } | |
1169 | } | |
cf60af03 | 1170 | flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0; |
ba615f67 | 1171 | err = __inet_stream_connect(sk->sk_socket, uaddr, |
3979ad7e | 1172 | msg->msg_namelen, flags, 1); |
7db92362 WW |
1173 | /* fastopen_req could already be freed in __inet_stream_connect |
1174 | * if the connection times out or gets rst | |
1175 | */ | |
1176 | if (tp->fastopen_req) { | |
1177 | *copied = tp->fastopen_req->copied; | |
1178 | tcp_free_fastopen_req(tp); | |
1179 | inet->defer_connect = 0; | |
1180 | } | |
cf60af03 YC |
1181 | return err; |
1182 | } | |
1183 | ||
306b13eb | 1184 | int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size) |
1da177e4 | 1185 | { |
1da177e4 | 1186 | struct tcp_sock *tp = tcp_sk(sk); |
f214f915 | 1187 | struct ubuf_info *uarg = NULL; |
1da177e4 | 1188 | struct sk_buff *skb; |
c14ac945 | 1189 | struct sockcm_cookie sockc; |
57be5bda AV |
1190 | int flags, err, copied = 0; |
1191 | int mss_now = 0, size_goal, copied_syn = 0; | |
d4011239 | 1192 | bool process_backlog = false; |
690e99c4 | 1193 | bool sg; |
1da177e4 LT |
1194 | long timeo; |
1195 | ||
1da177e4 | 1196 | flags = msg->msg_flags; |
f214f915 WB |
1197 | |
1198 | if (flags & MSG_ZEROCOPY && size) { | |
1199 | if (sk->sk_state != TCP_ESTABLISHED) { | |
1200 | err = -EINVAL; | |
1201 | goto out_err; | |
1202 | } | |
1203 | ||
75c119af | 1204 | skb = tcp_write_queue_tail(sk); |
f214f915 WB |
1205 | uarg = sock_zerocopy_realloc(sk, size, skb_zcopy(skb)); |
1206 | if (!uarg) { | |
1207 | err = -ENOBUFS; | |
1208 | goto out_err; | |
1209 | } | |
1210 | ||
f214f915 WB |
1211 | if (!(sk_check_csum_caps(sk) && sk->sk_route_caps & NETIF_F_SG)) |
1212 | uarg->zerocopy = 0; | |
1213 | } | |
1214 | ||
19f6d3f3 | 1215 | if (unlikely(flags & MSG_FASTOPEN || inet_sk(sk)->defer_connect)) { |
f5ddcbbb | 1216 | err = tcp_sendmsg_fastopen(sk, msg, &copied_syn, size); |
cf60af03 YC |
1217 | if (err == -EINPROGRESS && copied_syn > 0) |
1218 | goto out; | |
1219 | else if (err) | |
1220 | goto out_err; | |
cf60af03 YC |
1221 | } |
1222 | ||
1da177e4 LT |
1223 | timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); |
1224 | ||
d7722e85 SHY |
1225 | tcp_rate_check_app_limited(sk); /* is sending application-limited? */ |
1226 | ||
8336886f JC |
1227 | /* Wait for a connection to finish. One exception is TCP Fast Open |
1228 | * (passive side) where data is allowed to be sent before a connection | |
1229 | * is fully established. | |
1230 | */ | |
1231 | if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) && | |
1232 | !tcp_passive_fastopen(sk)) { | |
686a5624 YM |
1233 | err = sk_stream_wait_connect(sk, &timeo); |
1234 | if (err != 0) | |
cf60af03 | 1235 | goto do_error; |
8336886f | 1236 | } |
1da177e4 | 1237 | |
c0e88ff0 PE |
1238 | if (unlikely(tp->repair)) { |
1239 | if (tp->repair_queue == TCP_RECV_QUEUE) { | |
1240 | copied = tcp_send_rcvq(sk, msg, size); | |
5924f17a | 1241 | goto out_nopush; |
c0e88ff0 PE |
1242 | } |
1243 | ||
1244 | err = -EINVAL; | |
1245 | if (tp->repair_queue == TCP_NO_QUEUE) | |
1246 | goto out_err; | |
1247 | ||
1248 | /* 'common' sending to sendq */ | |
1249 | } | |
1250 | ||
c14ac945 SHY |
1251 | sockc.tsflags = sk->sk_tsflags; |
1252 | if (msg->msg_controllen) { | |
1253 | err = sock_cmsg_send(sk, msg, &sockc); | |
1254 | if (unlikely(err)) { | |
1255 | err = -EINVAL; | |
1256 | goto out_err; | |
1257 | } | |
1258 | } | |
1259 | ||
1da177e4 | 1260 | /* This should be in poll */ |
9cd3e072 | 1261 | sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 | 1262 | |
1da177e4 | 1263 | /* Ok commence sending. */ |
1da177e4 LT |
1264 | copied = 0; |
1265 | ||
d41a69f1 ED |
1266 | restart: |
1267 | mss_now = tcp_send_mss(sk, &size_goal, flags); | |
1268 | ||
1da177e4 LT |
1269 | err = -EPIPE; |
1270 | if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) | |
79d8665b | 1271 | goto do_error; |
1da177e4 | 1272 | |
690e99c4 | 1273 | sg = !!(sk->sk_route_caps & NETIF_F_SG); |
def87cf4 | 1274 | |
01e97e65 | 1275 | while (msg_data_left(msg)) { |
57be5bda AV |
1276 | int copy = 0; |
1277 | int max = size_goal; | |
1da177e4 | 1278 | |
57be5bda | 1279 | skb = tcp_write_queue_tail(sk); |
75c119af | 1280 | if (skb) { |
57be5bda AV |
1281 | if (skb->ip_summed == CHECKSUM_NONE) |
1282 | max = mss_now; | |
1283 | copy = max - skb->len; | |
cf60af03 | 1284 | } |
1da177e4 | 1285 | |
c134ecb8 | 1286 | if (copy <= 0 || !tcp_skb_can_collapse_to(skb)) { |
3613b3db ED |
1287 | bool first_skb; |
1288 | ||
1da177e4 | 1289 | new_segment: |
57be5bda AV |
1290 | /* Allocate new segment. If the interface is SG, |
1291 | * allocate skb fitting to single page. | |
1292 | */ | |
1293 | if (!sk_stream_memory_free(sk)) | |
1294 | goto wait_for_sndbuf; | |
1da177e4 | 1295 | |
d4011239 ED |
1296 | if (process_backlog && sk_flush_backlog(sk)) { |
1297 | process_backlog = false; | |
d41a69f1 | 1298 | goto restart; |
d4011239 | 1299 | } |
75c119af | 1300 | first_skb = tcp_rtx_and_write_queues_empty(sk); |
57be5bda | 1301 | skb = sk_stream_alloc_skb(sk, |
3613b3db | 1302 | select_size(sk, sg, first_skb), |
eb934478 | 1303 | sk->sk_allocation, |
3613b3db | 1304 | first_skb); |
57be5bda AV |
1305 | if (!skb) |
1306 | goto wait_for_memory; | |
1da177e4 | 1307 | |
d4011239 | 1308 | process_backlog = true; |
57be5bda AV |
1309 | /* |
1310 | * Check whether we can use HW checksum. | |
1311 | */ | |
9a49850d | 1312 | if (sk_check_csum_caps(sk)) |
57be5bda | 1313 | skb->ip_summed = CHECKSUM_PARTIAL; |
1da177e4 | 1314 | |
57be5bda AV |
1315 | skb_entail(sk, skb); |
1316 | copy = size_goal; | |
1317 | max = size_goal; | |
9d186cac | 1318 | |
57be5bda AV |
1319 | /* All packets are restored as if they have |
1320 | * already been sent. skb_mstamp isn't set to | |
1321 | * avoid wrong rtt estimation. | |
1322 | */ | |
1323 | if (tp->repair) | |
1324 | TCP_SKB_CB(skb)->sacked |= TCPCB_REPAIRED; | |
1325 | } | |
1da177e4 | 1326 | |
57be5bda | 1327 | /* Try to append data to the end of skb. */ |
01e97e65 AV |
1328 | if (copy > msg_data_left(msg)) |
1329 | copy = msg_data_left(msg); | |
57be5bda AV |
1330 | |
1331 | /* Where to copy to? */ | |
1332 | if (skb_availroom(skb) > 0) { | |
1333 | /* We have some space in skb head. Superb! */ | |
1334 | copy = min_t(int, copy, skb_availroom(skb)); | |
1335 | err = skb_add_data_nocache(sk, skb, &msg->msg_iter, copy); | |
1336 | if (err) | |
1337 | goto do_fault; | |
f214f915 | 1338 | } else if (!uarg || !uarg->zerocopy) { |
57be5bda AV |
1339 | bool merge = true; |
1340 | int i = skb_shinfo(skb)->nr_frags; | |
1341 | struct page_frag *pfrag = sk_page_frag(sk); | |
1342 | ||
1343 | if (!sk_page_frag_refill(sk, pfrag)) | |
1344 | goto wait_for_memory; | |
ef015786 | 1345 | |
57be5bda AV |
1346 | if (!skb_can_coalesce(skb, i, pfrag->page, |
1347 | pfrag->offset)) { | |
ac9e70b1 | 1348 | if (i >= sysctl_max_skb_frags || !sg) { |
57be5bda AV |
1349 | tcp_mark_push(tp, skb); |
1350 | goto new_segment; | |
1da177e4 | 1351 | } |
57be5bda | 1352 | merge = false; |
1da177e4 LT |
1353 | } |
1354 | ||
57be5bda AV |
1355 | copy = min_t(int, copy, pfrag->size - pfrag->offset); |
1356 | ||
1357 | if (!sk_wmem_schedule(sk, copy)) | |
1358 | goto wait_for_memory; | |
1da177e4 | 1359 | |
57be5bda AV |
1360 | err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb, |
1361 | pfrag->page, | |
1362 | pfrag->offset, | |
1363 | copy); | |
1364 | if (err) | |
1365 | goto do_error; | |
1da177e4 | 1366 | |
57be5bda AV |
1367 | /* Update the skb. */ |
1368 | if (merge) { | |
1369 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); | |
1370 | } else { | |
1371 | skb_fill_page_desc(skb, i, pfrag->page, | |
1372 | pfrag->offset, copy); | |
4e33e346 | 1373 | page_ref_inc(pfrag->page); |
4ed2d765 | 1374 | } |
57be5bda | 1375 | pfrag->offset += copy; |
f214f915 WB |
1376 | } else { |
1377 | err = skb_zerocopy_iter_stream(sk, skb, msg, copy, uarg); | |
1378 | if (err == -EMSGSIZE || err == -EEXIST) | |
1379 | goto new_segment; | |
1380 | if (err < 0) | |
1381 | goto do_error; | |
1382 | copy = err; | |
57be5bda AV |
1383 | } |
1384 | ||
1385 | if (!copied) | |
1386 | TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH; | |
1387 | ||
1388 | tp->write_seq += copy; | |
1389 | TCP_SKB_CB(skb)->end_seq += copy; | |
1390 | tcp_skb_pcount_set(skb, 0); | |
1da177e4 | 1391 | |
57be5bda | 1392 | copied += copy; |
01e97e65 | 1393 | if (!msg_data_left(msg)) { |
c134ecb8 MKL |
1394 | if (unlikely(flags & MSG_EOR)) |
1395 | TCP_SKB_CB(skb)->eor = 1; | |
57be5bda AV |
1396 | goto out; |
1397 | } | |
1da177e4 | 1398 | |
57be5bda | 1399 | if (skb->len < max || (flags & MSG_OOB) || unlikely(tp->repair)) |
1da177e4 LT |
1400 | continue; |
1401 | ||
57be5bda AV |
1402 | if (forced_push(tp)) { |
1403 | tcp_mark_push(tp, skb); | |
1404 | __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH); | |
1405 | } else if (skb == tcp_send_head(sk)) | |
1406 | tcp_push_one(sk, mss_now); | |
1407 | continue; | |
1408 | ||
1da177e4 | 1409 | wait_for_sndbuf: |
57be5bda | 1410 | set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
1da177e4 | 1411 | wait_for_memory: |
57be5bda AV |
1412 | if (copied) |
1413 | tcp_push(sk, flags & ~MSG_MORE, mss_now, | |
1414 | TCP_NAGLE_PUSH, size_goal); | |
1da177e4 | 1415 | |
686a5624 YM |
1416 | err = sk_stream_wait_memory(sk, &timeo); |
1417 | if (err != 0) | |
57be5bda | 1418 | goto do_error; |
1da177e4 | 1419 | |
57be5bda | 1420 | mss_now = tcp_send_mss(sk, &size_goal, flags); |
1da177e4 LT |
1421 | } |
1422 | ||
1423 | out: | |
ad02c4f5 | 1424 | if (copied) { |
4e8cc228 | 1425 | tcp_tx_timestamp(sk, sockc.tsflags); |
f54b3111 | 1426 | tcp_push(sk, flags, mss_now, tp->nonagle, size_goal); |
ad02c4f5 | 1427 | } |
5924f17a | 1428 | out_nopush: |
f214f915 | 1429 | sock_zerocopy_put(uarg); |
cf60af03 | 1430 | return copied + copied_syn; |
1da177e4 LT |
1431 | |
1432 | do_fault: | |
1433 | if (!skb->len) { | |
fe067e8a DM |
1434 | tcp_unlink_write_queue(skb, sk); |
1435 | /* It is the one place in all of TCP, except connection | |
1436 | * reset, where we can be unlinking the send_head. | |
1437 | */ | |
1438 | tcp_check_send_head(sk, skb); | |
3ab224be | 1439 | sk_wmem_free_skb(sk, skb); |
1da177e4 LT |
1440 | } |
1441 | ||
1442 | do_error: | |
cf60af03 | 1443 | if (copied + copied_syn) |
1da177e4 LT |
1444 | goto out; |
1445 | out_err: | |
f214f915 | 1446 | sock_zerocopy_put_abort(uarg); |
1da177e4 | 1447 | err = sk_stream_error(sk, flags, err); |
ce5ec440 | 1448 | /* make sure we wake any epoll edge trigger waiter */ |
b0f71bd3 FY |
1449 | if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && |
1450 | err == -EAGAIN)) { | |
ce5ec440 | 1451 | sk->sk_write_space(sk); |
b0f71bd3 FY |
1452 | tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED); |
1453 | } | |
1da177e4 LT |
1454 | return err; |
1455 | } | |
774c4673 | 1456 | EXPORT_SYMBOL_GPL(tcp_sendmsg_locked); |
306b13eb TH |
1457 | |
1458 | int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size) | |
1459 | { | |
1460 | int ret; | |
1461 | ||
1462 | lock_sock(sk); | |
1463 | ret = tcp_sendmsg_locked(sk, msg, size); | |
1464 | release_sock(sk); | |
1465 | ||
1466 | return ret; | |
1467 | } | |
4bc2f18b | 1468 | EXPORT_SYMBOL(tcp_sendmsg); |
1da177e4 LT |
1469 | |
1470 | /* | |
1471 | * Handle reading urgent data. BSD has very simple semantics for | |
1472 | * this, no blocking and very strange errors 8) | |
1473 | */ | |
1474 | ||
377f0a08 | 1475 | static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags) |
1da177e4 LT |
1476 | { |
1477 | struct tcp_sock *tp = tcp_sk(sk); | |
1478 | ||
1479 | /* No URG data to read. */ | |
1480 | if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data || | |
1481 | tp->urg_data == TCP_URG_READ) | |
1482 | return -EINVAL; /* Yes this is right ! */ | |
1483 | ||
1484 | if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE)) | |
1485 | return -ENOTCONN; | |
1486 | ||
1487 | if (tp->urg_data & TCP_URG_VALID) { | |
1488 | int err = 0; | |
1489 | char c = tp->urg_data; | |
1490 | ||
1491 | if (!(flags & MSG_PEEK)) | |
1492 | tp->urg_data = TCP_URG_READ; | |
1493 | ||
1494 | /* Read urgent data. */ | |
1495 | msg->msg_flags |= MSG_OOB; | |
1496 | ||
1497 | if (len > 0) { | |
1498 | if (!(flags & MSG_TRUNC)) | |
7eab8d9e | 1499 | err = memcpy_to_msg(msg, &c, 1); |
1da177e4 LT |
1500 | len = 1; |
1501 | } else | |
1502 | msg->msg_flags |= MSG_TRUNC; | |
1503 | ||
1504 | return err ? -EFAULT : len; | |
1505 | } | |
1506 | ||
1507 | if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN)) | |
1508 | return 0; | |
1509 | ||
1510 | /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and | |
1511 | * the available implementations agree in this case: | |
1512 | * this call should never block, independent of the | |
1513 | * blocking state of the socket. | |
1514 | * Mike <pall@rz.uni-karlsruhe.de> | |
1515 | */ | |
1516 | return -EAGAIN; | |
1517 | } | |
1518 | ||
c0e88ff0 PE |
1519 | static int tcp_peek_sndq(struct sock *sk, struct msghdr *msg, int len) |
1520 | { | |
1521 | struct sk_buff *skb; | |
1522 | int copied = 0, err = 0; | |
1523 | ||
1524 | /* XXX -- need to support SO_PEEK_OFF */ | |
1525 | ||
75c119af ED |
1526 | skb_rbtree_walk(skb, &sk->tcp_rtx_queue) { |
1527 | err = skb_copy_datagram_msg(skb, 0, msg, skb->len); | |
1528 | if (err) | |
1529 | return err; | |
1530 | copied += skb->len; | |
1531 | } | |
1532 | ||
c0e88ff0 | 1533 | skb_queue_walk(&sk->sk_write_queue, skb) { |
51f3d02b | 1534 | err = skb_copy_datagram_msg(skb, 0, msg, skb->len); |
c0e88ff0 PE |
1535 | if (err) |
1536 | break; | |
1537 | ||
1538 | copied += skb->len; | |
1539 | } | |
1540 | ||
1541 | return err ?: copied; | |
1542 | } | |
1543 | ||
1da177e4 LT |
1544 | /* Clean up the receive buffer for full frames taken by the user, |
1545 | * then send an ACK if necessary. COPIED is the number of bytes | |
1546 | * tcp_recvmsg has given to the user so far, it speeds up the | |
1547 | * calculation of whether or not we must ACK for the sake of | |
1548 | * a window update. | |
1549 | */ | |
3f334078 | 1550 | static void tcp_cleanup_rbuf(struct sock *sk, int copied) |
1da177e4 LT |
1551 | { |
1552 | struct tcp_sock *tp = tcp_sk(sk); | |
a2a385d6 | 1553 | bool time_to_ack = false; |
1da177e4 | 1554 | |
1da177e4 LT |
1555 | struct sk_buff *skb = skb_peek(&sk->sk_receive_queue); |
1556 | ||
d792c100 | 1557 | WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq), |
2af6fd8b | 1558 | "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n", |
d792c100 | 1559 | tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt); |
1da177e4 | 1560 | |
463c84b9 ACM |
1561 | if (inet_csk_ack_scheduled(sk)) { |
1562 | const struct inet_connection_sock *icsk = inet_csk(sk); | |
1da177e4 LT |
1563 | /* Delayed ACKs frequently hit locked sockets during bulk |
1564 | * receive. */ | |
463c84b9 | 1565 | if (icsk->icsk_ack.blocked || |
1da177e4 | 1566 | /* Once-per-two-segments ACK was not sent by tcp_input.c */ |
463c84b9 | 1567 | tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss || |
1da177e4 LT |
1568 | /* |
1569 | * If this read emptied read buffer, we send ACK, if | |
1570 | * connection is not bidirectional, user drained | |
1571 | * receive buffer and there was a small segment | |
1572 | * in queue. | |
1573 | */ | |
1ef9696c AK |
1574 | (copied > 0 && |
1575 | ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) || | |
1576 | ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) && | |
1577 | !icsk->icsk_ack.pingpong)) && | |
1578 | !atomic_read(&sk->sk_rmem_alloc))) | |
a2a385d6 | 1579 | time_to_ack = true; |
1da177e4 LT |
1580 | } |
1581 | ||
1582 | /* We send an ACK if we can now advertise a non-zero window | |
1583 | * which has been raised "significantly". | |
1584 | * | |
1585 | * Even if window raised up to infinity, do not send window open ACK | |
1586 | * in states, where we will not receive more. It is useless. | |
1587 | */ | |
1588 | if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) { | |
1589 | __u32 rcv_window_now = tcp_receive_window(tp); | |
1590 | ||
1591 | /* Optimize, __tcp_select_window() is not cheap. */ | |
1592 | if (2*rcv_window_now <= tp->window_clamp) { | |
1593 | __u32 new_window = __tcp_select_window(sk); | |
1594 | ||
1595 | /* Send ACK now, if this read freed lots of space | |
1596 | * in our buffer. Certainly, new_window is new window. | |
1597 | * We can advertise it now, if it is not less than current one. | |
1598 | * "Lots" means "at least twice" here. | |
1599 | */ | |
1600 | if (new_window && new_window >= 2 * rcv_window_now) | |
a2a385d6 | 1601 | time_to_ack = true; |
1da177e4 LT |
1602 | } |
1603 | } | |
1604 | if (time_to_ack) | |
1605 | tcp_send_ack(sk); | |
1606 | } | |
1607 | ||
f26845b4 | 1608 | static struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off) |
1da177e4 LT |
1609 | { |
1610 | struct sk_buff *skb; | |
1611 | u32 offset; | |
1612 | ||
f26845b4 | 1613 | while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) { |
1da177e4 | 1614 | offset = seq - TCP_SKB_CB(skb)->seq; |
9d691539 ED |
1615 | if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { |
1616 | pr_err_once("%s: found a SYN, please report !\n", __func__); | |
1da177e4 | 1617 | offset--; |
9d691539 | 1618 | } |
e11ecddf | 1619 | if (offset < skb->len || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) { |
1da177e4 LT |
1620 | *off = offset; |
1621 | return skb; | |
1622 | } | |
f26845b4 ED |
1623 | /* This looks weird, but this can happen if TCP collapsing |
1624 | * splitted a fat GRO packet, while we released socket lock | |
1625 | * in skb_splice_bits() | |
1626 | */ | |
7bced397 | 1627 | sk_eat_skb(sk, skb); |
1da177e4 LT |
1628 | } |
1629 | return NULL; | |
1630 | } | |
1631 | ||
1632 | /* | |
1633 | * This routine provides an alternative to tcp_recvmsg() for routines | |
1634 | * that would like to handle copying from skbuffs directly in 'sendfile' | |
1635 | * fashion. | |
1636 | * Note: | |
1637 | * - It is assumed that the socket was locked by the caller. | |
1638 | * - The routine does not block. | |
1639 | * - At present, there is no support for reading OOB data | |
1640 | * or for 'peeking' the socket using this routine | |
1641 | * (although both would be easy to implement). | |
1642 | */ | |
1643 | int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, | |
1644 | sk_read_actor_t recv_actor) | |
1645 | { | |
1646 | struct sk_buff *skb; | |
1647 | struct tcp_sock *tp = tcp_sk(sk); | |
1648 | u32 seq = tp->copied_seq; | |
1649 | u32 offset; | |
1650 | int copied = 0; | |
1651 | ||
1652 | if (sk->sk_state == TCP_LISTEN) | |
1653 | return -ENOTCONN; | |
1654 | while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) { | |
1655 | if (offset < skb->len) { | |
374e7b59 OP |
1656 | int used; |
1657 | size_t len; | |
1da177e4 LT |
1658 | |
1659 | len = skb->len - offset; | |
1660 | /* Stop reading if we hit a patch of urgent data */ | |
1661 | if (tp->urg_data) { | |
1662 | u32 urg_offset = tp->urg_seq - seq; | |
1663 | if (urg_offset < len) | |
1664 | len = urg_offset; | |
1665 | if (!len) | |
1666 | break; | |
1667 | } | |
1668 | used = recv_actor(desc, skb, offset, len); | |
ff905b1e | 1669 | if (used <= 0) { |
ddb61a57 JA |
1670 | if (!copied) |
1671 | copied = used; | |
1672 | break; | |
1673 | } else if (used <= len) { | |
1da177e4 LT |
1674 | seq += used; |
1675 | copied += used; | |
1676 | offset += used; | |
1677 | } | |
02275a2e | 1678 | /* If recv_actor drops the lock (e.g. TCP splice |
293ad604 OP |
1679 | * receive) the skb pointer might be invalid when |
1680 | * getting here: tcp_collapse might have deleted it | |
1681 | * while aggregating skbs from the socket queue. | |
1682 | */ | |
02275a2e WT |
1683 | skb = tcp_recv_skb(sk, seq - 1, &offset); |
1684 | if (!skb) | |
1da177e4 | 1685 | break; |
02275a2e WT |
1686 | /* TCP coalescing might have appended data to the skb. |
1687 | * Try to splice more frags | |
1688 | */ | |
1689 | if (offset + 1 != skb->len) | |
1690 | continue; | |
1da177e4 | 1691 | } |
e11ecddf | 1692 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) { |
7bced397 | 1693 | sk_eat_skb(sk, skb); |
1da177e4 LT |
1694 | ++seq; |
1695 | break; | |
1696 | } | |
7bced397 | 1697 | sk_eat_skb(sk, skb); |
1da177e4 LT |
1698 | if (!desc->count) |
1699 | break; | |
baff42ab | 1700 | tp->copied_seq = seq; |
1da177e4 LT |
1701 | } |
1702 | tp->copied_seq = seq; | |
1703 | ||
1704 | tcp_rcv_space_adjust(sk); | |
1705 | ||
1706 | /* Clean up data we have read: This will do ACK frames. */ | |
f26845b4 ED |
1707 | if (copied > 0) { |
1708 | tcp_recv_skb(sk, seq, &offset); | |
0e4b4992 | 1709 | tcp_cleanup_rbuf(sk, copied); |
f26845b4 | 1710 | } |
1da177e4 LT |
1711 | return copied; |
1712 | } | |
4bc2f18b | 1713 | EXPORT_SYMBOL(tcp_read_sock); |
1da177e4 | 1714 | |
32035585 TH |
1715 | int tcp_peek_len(struct socket *sock) |
1716 | { | |
1717 | return tcp_inq(sock->sk); | |
1718 | } | |
1719 | EXPORT_SYMBOL(tcp_peek_len); | |
1720 | ||
98aaa913 MM |
1721 | static void tcp_update_recv_tstamps(struct sk_buff *skb, |
1722 | struct scm_timestamping *tss) | |
1723 | { | |
1724 | if (skb->tstamp) | |
1725 | tss->ts[0] = ktime_to_timespec(skb->tstamp); | |
1726 | else | |
1727 | tss->ts[0] = (struct timespec) {0}; | |
1728 | ||
1729 | if (skb_hwtstamps(skb)->hwtstamp) | |
1730 | tss->ts[2] = ktime_to_timespec(skb_hwtstamps(skb)->hwtstamp); | |
1731 | else | |
1732 | tss->ts[2] = (struct timespec) {0}; | |
1733 | } | |
1734 | ||
1735 | /* Similar to __sock_recv_timestamp, but does not require an skb */ | |
1736 | void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, | |
1737 | struct scm_timestamping *tss) | |
1738 | { | |
1739 | struct timeval tv; | |
1740 | bool has_timestamping = false; | |
1741 | ||
1742 | if (tss->ts[0].tv_sec || tss->ts[0].tv_nsec) { | |
1743 | if (sock_flag(sk, SOCK_RCVTSTAMP)) { | |
1744 | if (sock_flag(sk, SOCK_RCVTSTAMPNS)) { | |
1745 | put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPNS, | |
1746 | sizeof(tss->ts[0]), &tss->ts[0]); | |
1747 | } else { | |
1748 | tv.tv_sec = tss->ts[0].tv_sec; | |
1749 | tv.tv_usec = tss->ts[0].tv_nsec / 1000; | |
1750 | ||
1751 | put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP, | |
1752 | sizeof(tv), &tv); | |
1753 | } | |
1754 | } | |
1755 | ||
1756 | if (sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) | |
1757 | has_timestamping = true; | |
1758 | else | |
1759 | tss->ts[0] = (struct timespec) {0}; | |
1760 | } | |
1761 | ||
1762 | if (tss->ts[2].tv_sec || tss->ts[2].tv_nsec) { | |
1763 | if (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) | |
1764 | has_timestamping = true; | |
1765 | else | |
1766 | tss->ts[2] = (struct timespec) {0}; | |
1767 | } | |
1768 | ||
1769 | if (has_timestamping) { | |
1770 | tss->ts[1] = (struct timespec) {0}; | |
1771 | put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING, | |
1772 | sizeof(*tss), tss); | |
1773 | } | |
1774 | } | |
1775 | ||
1da177e4 LT |
1776 | /* |
1777 | * This routine copies from a sock struct into the user buffer. | |
1778 | * | |
1779 | * Technical note: in 2.3 we work on _locked_ socket, so that | |
1780 | * tricks with *seq access order and skb->users are not required. | |
1781 | * Probably, code can be easily improved even more. | |
1782 | */ | |
1783 | ||
1b784140 YX |
1784 | int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, |
1785 | int flags, int *addr_len) | |
1da177e4 LT |
1786 | { |
1787 | struct tcp_sock *tp = tcp_sk(sk); | |
1788 | int copied = 0; | |
1789 | u32 peek_seq; | |
1790 | u32 *seq; | |
1791 | unsigned long used; | |
1792 | int err; | |
1793 | int target; /* Read at least this many bytes */ | |
1794 | long timeo; | |
dfbafc99 | 1795 | struct sk_buff *skb, *last; |
77527313 | 1796 | u32 urg_hole = 0; |
98aaa913 MM |
1797 | struct scm_timestamping tss; |
1798 | bool has_tss = false; | |
1da177e4 | 1799 | |
4ed2d765 | 1800 | if (unlikely(flags & MSG_ERRQUEUE)) |
f4713a3d | 1801 | return inet_recv_error(sk, msg, len, addr_len); |
4ed2d765 | 1802 | |
cbf55001 ET |
1803 | if (sk_can_busy_loop(sk) && skb_queue_empty(&sk->sk_receive_queue) && |
1804 | (sk->sk_state == TCP_ESTABLISHED)) | |
1805 | sk_busy_loop(sk, nonblock); | |
d30e383b | 1806 | |
1da177e4 LT |
1807 | lock_sock(sk); |
1808 | ||
1da177e4 LT |
1809 | err = -ENOTCONN; |
1810 | if (sk->sk_state == TCP_LISTEN) | |
1811 | goto out; | |
1812 | ||
1813 | timeo = sock_rcvtimeo(sk, nonblock); | |
1814 | ||
1815 | /* Urgent data needs to be handled specially. */ | |
1816 | if (flags & MSG_OOB) | |
1817 | goto recv_urg; | |
1818 | ||
c0e88ff0 PE |
1819 | if (unlikely(tp->repair)) { |
1820 | err = -EPERM; | |
1821 | if (!(flags & MSG_PEEK)) | |
1822 | goto out; | |
1823 | ||
1824 | if (tp->repair_queue == TCP_SEND_QUEUE) | |
1825 | goto recv_sndq; | |
1826 | ||
1827 | err = -EINVAL; | |
1828 | if (tp->repair_queue == TCP_NO_QUEUE) | |
1829 | goto out; | |
1830 | ||
1831 | /* 'common' recv queue MSG_PEEK-ing */ | |
1832 | } | |
1833 | ||
1da177e4 LT |
1834 | seq = &tp->copied_seq; |
1835 | if (flags & MSG_PEEK) { | |
1836 | peek_seq = tp->copied_seq; | |
1837 | seq = &peek_seq; | |
1838 | } | |
1839 | ||
1840 | target = sock_rcvlowat(sk, flags & MSG_WAITALL, len); | |
1841 | ||
1842 | do { | |
1da177e4 LT |
1843 | u32 offset; |
1844 | ||
1845 | /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */ | |
1846 | if (tp->urg_data && tp->urg_seq == *seq) { | |
1847 | if (copied) | |
1848 | break; | |
1849 | if (signal_pending(current)) { | |
1850 | copied = timeo ? sock_intr_errno(timeo) : -EAGAIN; | |
1851 | break; | |
1852 | } | |
1853 | } | |
1854 | ||
1855 | /* Next get a buffer. */ | |
1856 | ||
dfbafc99 | 1857 | last = skb_peek_tail(&sk->sk_receive_queue); |
91521944 | 1858 | skb_queue_walk(&sk->sk_receive_queue, skb) { |
dfbafc99 | 1859 | last = skb; |
1da177e4 LT |
1860 | /* Now that we have two receive queues this |
1861 | * shouldn't happen. | |
1862 | */ | |
d792c100 | 1863 | if (WARN(before(*seq, TCP_SKB_CB(skb)->seq), |
2af6fd8b JP |
1864 | "recvmsg bug: copied %X seq %X rcvnxt %X fl %X\n", |
1865 | *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, | |
1866 | flags)) | |
1da177e4 | 1867 | break; |
d792c100 | 1868 | |
1da177e4 | 1869 | offset = *seq - TCP_SKB_CB(skb)->seq; |
9d691539 ED |
1870 | if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { |
1871 | pr_err_once("%s: found a SYN, please report !\n", __func__); | |
1da177e4 | 1872 | offset--; |
9d691539 | 1873 | } |
1da177e4 LT |
1874 | if (offset < skb->len) |
1875 | goto found_ok_skb; | |
e11ecddf | 1876 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) |
1da177e4 | 1877 | goto found_fin_ok; |
2af6fd8b JP |
1878 | WARN(!(flags & MSG_PEEK), |
1879 | "recvmsg bug 2: copied %X seq %X rcvnxt %X fl %X\n", | |
1880 | *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags); | |
91521944 | 1881 | } |
1da177e4 LT |
1882 | |
1883 | /* Well, if we have backlog, try to process it now yet. */ | |
1884 | ||
1885 | if (copied >= target && !sk->sk_backlog.tail) | |
1886 | break; | |
1887 | ||
1888 | if (copied) { | |
1889 | if (sk->sk_err || | |
1890 | sk->sk_state == TCP_CLOSE || | |
1891 | (sk->sk_shutdown & RCV_SHUTDOWN) || | |
1892 | !timeo || | |
518a09ef | 1893 | signal_pending(current)) |
1da177e4 LT |
1894 | break; |
1895 | } else { | |
1896 | if (sock_flag(sk, SOCK_DONE)) | |
1897 | break; | |
1898 | ||
1899 | if (sk->sk_err) { | |
1900 | copied = sock_error(sk); | |
1901 | break; | |
1902 | } | |
1903 | ||
1904 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
1905 | break; | |
1906 | ||
1907 | if (sk->sk_state == TCP_CLOSE) { | |
1908 | if (!sock_flag(sk, SOCK_DONE)) { | |
1909 | /* This occurs when user tries to read | |
1910 | * from never connected socket. | |
1911 | */ | |
1912 | copied = -ENOTCONN; | |
1913 | break; | |
1914 | } | |
1915 | break; | |
1916 | } | |
1917 | ||
1918 | if (!timeo) { | |
1919 | copied = -EAGAIN; | |
1920 | break; | |
1921 | } | |
1922 | ||
1923 | if (signal_pending(current)) { | |
1924 | copied = sock_intr_errno(timeo); | |
1925 | break; | |
1926 | } | |
1927 | } | |
1928 | ||
0e4b4992 | 1929 | tcp_cleanup_rbuf(sk, copied); |
1da177e4 | 1930 | |
1da177e4 LT |
1931 | if (copied >= target) { |
1932 | /* Do not sleep, just process backlog. */ | |
1933 | release_sock(sk); | |
1934 | lock_sock(sk); | |
dfbafc99 SD |
1935 | } else { |
1936 | sk_wait_data(sk, &timeo, last); | |
1937 | } | |
1da177e4 | 1938 | |
77527313 IJ |
1939 | if ((flags & MSG_PEEK) && |
1940 | (peek_seq - copied - urg_hole != tp->copied_seq)) { | |
e87cc472 JP |
1941 | net_dbg_ratelimited("TCP(%s:%d): Application bug, race in MSG_PEEK\n", |
1942 | current->comm, | |
1943 | task_pid_nr(current)); | |
1da177e4 LT |
1944 | peek_seq = tp->copied_seq; |
1945 | } | |
1946 | continue; | |
1947 | ||
1948 | found_ok_skb: | |
1949 | /* Ok so how much can we use? */ | |
1950 | used = skb->len - offset; | |
1951 | if (len < used) | |
1952 | used = len; | |
1953 | ||
1954 | /* Do we have urgent data here? */ | |
1955 | if (tp->urg_data) { | |
1956 | u32 urg_offset = tp->urg_seq - *seq; | |
1957 | if (urg_offset < used) { | |
1958 | if (!urg_offset) { | |
1959 | if (!sock_flag(sk, SOCK_URGINLINE)) { | |
1960 | ++*seq; | |
77527313 | 1961 | urg_hole++; |
1da177e4 LT |
1962 | offset++; |
1963 | used--; | |
1964 | if (!used) | |
1965 | goto skip_copy; | |
1966 | } | |
1967 | } else | |
1968 | used = urg_offset; | |
1969 | } | |
1970 | } | |
1971 | ||
1972 | if (!(flags & MSG_TRUNC)) { | |
51f3d02b | 1973 | err = skb_copy_datagram_msg(skb, offset, msg, used); |
7bced397 DW |
1974 | if (err) { |
1975 | /* Exception. Bailout! */ | |
1976 | if (!copied) | |
1977 | copied = -EFAULT; | |
1978 | break; | |
1da177e4 LT |
1979 | } |
1980 | } | |
1981 | ||
1982 | *seq += used; | |
1983 | copied += used; | |
1984 | len -= used; | |
1985 | ||
1986 | tcp_rcv_space_adjust(sk); | |
1987 | ||
1988 | skip_copy: | |
31770e34 | 1989 | if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) { |
1da177e4 | 1990 | tp->urg_data = 0; |
31770e34 FW |
1991 | tcp_fast_path_check(sk); |
1992 | } | |
1da177e4 LT |
1993 | if (used + offset < skb->len) |
1994 | continue; | |
1995 | ||
98aaa913 MM |
1996 | if (TCP_SKB_CB(skb)->has_rxtstamp) { |
1997 | tcp_update_recv_tstamps(skb, &tss); | |
1998 | has_tss = true; | |
1999 | } | |
e11ecddf | 2000 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) |
1da177e4 | 2001 | goto found_fin_ok; |
7bced397 DW |
2002 | if (!(flags & MSG_PEEK)) |
2003 | sk_eat_skb(sk, skb); | |
1da177e4 LT |
2004 | continue; |
2005 | ||
2006 | found_fin_ok: | |
2007 | /* Process the FIN. */ | |
2008 | ++*seq; | |
7bced397 DW |
2009 | if (!(flags & MSG_PEEK)) |
2010 | sk_eat_skb(sk, skb); | |
1da177e4 LT |
2011 | break; |
2012 | } while (len > 0); | |
2013 | ||
1da177e4 LT |
2014 | /* According to UNIX98, msg_name/msg_namelen are ignored |
2015 | * on connected socket. I was just happy when found this 8) --ANK | |
2016 | */ | |
2017 | ||
98aaa913 MM |
2018 | if (has_tss) |
2019 | tcp_recv_timestamp(msg, sk, &tss); | |
2020 | ||
1da177e4 | 2021 | /* Clean up data we have read: This will do ACK frames. */ |
0e4b4992 | 2022 | tcp_cleanup_rbuf(sk, copied); |
1da177e4 | 2023 | |
1da177e4 LT |
2024 | release_sock(sk); |
2025 | return copied; | |
2026 | ||
2027 | out: | |
1da177e4 LT |
2028 | release_sock(sk); |
2029 | return err; | |
2030 | ||
2031 | recv_urg: | |
377f0a08 | 2032 | err = tcp_recv_urg(sk, msg, len, flags); |
1da177e4 | 2033 | goto out; |
c0e88ff0 PE |
2034 | |
2035 | recv_sndq: | |
2036 | err = tcp_peek_sndq(sk, msg, len); | |
2037 | goto out; | |
1da177e4 | 2038 | } |
4bc2f18b | 2039 | EXPORT_SYMBOL(tcp_recvmsg); |
1da177e4 | 2040 | |
490d5046 IJ |
2041 | void tcp_set_state(struct sock *sk, int state) |
2042 | { | |
2043 | int oldstate = sk->sk_state; | |
2044 | ||
e8fce239 SL |
2045 | trace_tcp_set_state(sk, oldstate, state); |
2046 | ||
490d5046 IJ |
2047 | switch (state) { |
2048 | case TCP_ESTABLISHED: | |
2049 | if (oldstate != TCP_ESTABLISHED) | |
81cc8a75 | 2050 | TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB); |
490d5046 IJ |
2051 | break; |
2052 | ||
2053 | case TCP_CLOSE: | |
2054 | if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED) | |
81cc8a75 | 2055 | TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS); |
490d5046 IJ |
2056 | |
2057 | sk->sk_prot->unhash(sk); | |
2058 | if (inet_csk(sk)->icsk_bind_hash && | |
2059 | !(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) | |
ab1e0a13 | 2060 | inet_put_port(sk); |
490d5046 IJ |
2061 | /* fall through */ |
2062 | default: | |
5a5f3a8d | 2063 | if (oldstate == TCP_ESTABLISHED) |
74688e48 | 2064 | TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB); |
490d5046 IJ |
2065 | } |
2066 | ||
2067 | /* Change state AFTER socket is unhashed to avoid closed | |
2068 | * socket sitting in hash tables. | |
2069 | */ | |
00fd38d9 | 2070 | sk_state_store(sk, state); |
490d5046 IJ |
2071 | |
2072 | #ifdef STATE_TRACE | |
5a5f3a8d | 2073 | SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n", sk, statename[oldstate], statename[state]); |
490d5046 IJ |
2074 | #endif |
2075 | } | |
2076 | EXPORT_SYMBOL_GPL(tcp_set_state); | |
2077 | ||
1da177e4 LT |
2078 | /* |
2079 | * State processing on a close. This implements the state shift for | |
2080 | * sending our FIN frame. Note that we only send a FIN for some | |
2081 | * states. A shutdown() may have already sent the FIN, or we may be | |
2082 | * closed. | |
2083 | */ | |
2084 | ||
9b5b5cff | 2085 | static const unsigned char new_state[16] = { |
1da177e4 | 2086 | /* current state: new state: action: */ |
0980c1e3 ED |
2087 | [0 /* (Invalid) */] = TCP_CLOSE, |
2088 | [TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN, | |
2089 | [TCP_SYN_SENT] = TCP_CLOSE, | |
2090 | [TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN, | |
2091 | [TCP_FIN_WAIT1] = TCP_FIN_WAIT1, | |
2092 | [TCP_FIN_WAIT2] = TCP_FIN_WAIT2, | |
2093 | [TCP_TIME_WAIT] = TCP_CLOSE, | |
2094 | [TCP_CLOSE] = TCP_CLOSE, | |
2095 | [TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN, | |
2096 | [TCP_LAST_ACK] = TCP_LAST_ACK, | |
2097 | [TCP_LISTEN] = TCP_CLOSE, | |
2098 | [TCP_CLOSING] = TCP_CLOSING, | |
2099 | [TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */ | |
1da177e4 LT |
2100 | }; |
2101 | ||
2102 | static int tcp_close_state(struct sock *sk) | |
2103 | { | |
2104 | int next = (int)new_state[sk->sk_state]; | |
2105 | int ns = next & TCP_STATE_MASK; | |
2106 | ||
2107 | tcp_set_state(sk, ns); | |
2108 | ||
2109 | return next & TCP_ACTION_FIN; | |
2110 | } | |
2111 | ||
2112 | /* | |
2113 | * Shutdown the sending side of a connection. Much like close except | |
1f29b058 | 2114 | * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD). |
1da177e4 LT |
2115 | */ |
2116 | ||
2117 | void tcp_shutdown(struct sock *sk, int how) | |
2118 | { | |
2119 | /* We need to grab some memory, and put together a FIN, | |
2120 | * and then put it into the queue to be sent. | |
2121 | * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92. | |
2122 | */ | |
2123 | if (!(how & SEND_SHUTDOWN)) | |
2124 | return; | |
2125 | ||
2126 | /* If we've already sent a FIN, or it's a closed state, skip this. */ | |
2127 | if ((1 << sk->sk_state) & | |
2128 | (TCPF_ESTABLISHED | TCPF_SYN_SENT | | |
2129 | TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) { | |
2130 | /* Clear out any half completed packets. FIN if needed. */ | |
2131 | if (tcp_close_state(sk)) | |
2132 | tcp_send_fin(sk); | |
2133 | } | |
2134 | } | |
4bc2f18b | 2135 | EXPORT_SYMBOL(tcp_shutdown); |
1da177e4 | 2136 | |
efcdbf24 AS |
2137 | bool tcp_check_oom(struct sock *sk, int shift) |
2138 | { | |
2139 | bool too_many_orphans, out_of_socket_memory; | |
2140 | ||
2141 | too_many_orphans = tcp_too_many_orphans(sk, shift); | |
2142 | out_of_socket_memory = tcp_out_of_memory(sk); | |
2143 | ||
e87cc472 JP |
2144 | if (too_many_orphans) |
2145 | net_info_ratelimited("too many orphaned sockets\n"); | |
2146 | if (out_of_socket_memory) | |
2147 | net_info_ratelimited("out of memory -- consider tuning tcp_mem\n"); | |
efcdbf24 AS |
2148 | return too_many_orphans || out_of_socket_memory; |
2149 | } | |
2150 | ||
1da177e4 LT |
2151 | void tcp_close(struct sock *sk, long timeout) |
2152 | { | |
2153 | struct sk_buff *skb; | |
2154 | int data_was_unread = 0; | |
75c2d907 | 2155 | int state; |
1da177e4 LT |
2156 | |
2157 | lock_sock(sk); | |
2158 | sk->sk_shutdown = SHUTDOWN_MASK; | |
2159 | ||
2160 | if (sk->sk_state == TCP_LISTEN) { | |
2161 | tcp_set_state(sk, TCP_CLOSE); | |
2162 | ||
2163 | /* Special case. */ | |
0a5578cf | 2164 | inet_csk_listen_stop(sk); |
1da177e4 LT |
2165 | |
2166 | goto adjudge_to_death; | |
2167 | } | |
2168 | ||
2169 | /* We need to flush the recv. buffs. We do this only on the | |
2170 | * descriptor close, not protocol-sourced closes, because the | |
2171 | * reader process may not have drained the data yet! | |
2172 | */ | |
2173 | while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) { | |
e11ecddf ED |
2174 | u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq; |
2175 | ||
2176 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) | |
2177 | len--; | |
1da177e4 LT |
2178 | data_was_unread += len; |
2179 | __kfree_skb(skb); | |
2180 | } | |
2181 | ||
3ab224be | 2182 | sk_mem_reclaim(sk); |
1da177e4 | 2183 | |
565b7b2d KK |
2184 | /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */ |
2185 | if (sk->sk_state == TCP_CLOSE) | |
2186 | goto adjudge_to_death; | |
2187 | ||
65bb723c GR |
2188 | /* As outlined in RFC 2525, section 2.17, we send a RST here because |
2189 | * data was lost. To witness the awful effects of the old behavior of | |
2190 | * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk | |
2191 | * GET in an FTP client, suspend the process, wait for the client to | |
2192 | * advertise a zero window, then kill -9 the FTP client, wheee... | |
2193 | * Note: timeout is always zero in such a case. | |
1da177e4 | 2194 | */ |
ee995283 PE |
2195 | if (unlikely(tcp_sk(sk)->repair)) { |
2196 | sk->sk_prot->disconnect(sk, 0); | |
2197 | } else if (data_was_unread) { | |
1da177e4 | 2198 | /* Unread data was tossed, zap the connection. */ |
6aef70a8 | 2199 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE); |
1da177e4 | 2200 | tcp_set_state(sk, TCP_CLOSE); |
aa133076 | 2201 | tcp_send_active_reset(sk, sk->sk_allocation); |
1da177e4 LT |
2202 | } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) { |
2203 | /* Check zero linger _after_ checking for unread data. */ | |
2204 | sk->sk_prot->disconnect(sk, 0); | |
6aef70a8 | 2205 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONDATA); |
1da177e4 LT |
2206 | } else if (tcp_close_state(sk)) { |
2207 | /* We FIN if the application ate all the data before | |
2208 | * zapping the connection. | |
2209 | */ | |
2210 | ||
2211 | /* RED-PEN. Formally speaking, we have broken TCP state | |
2212 | * machine. State transitions: | |
2213 | * | |
2214 | * TCP_ESTABLISHED -> TCP_FIN_WAIT1 | |
2215 | * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible) | |
2216 | * TCP_CLOSE_WAIT -> TCP_LAST_ACK | |
2217 | * | |
2218 | * are legal only when FIN has been sent (i.e. in window), | |
2219 | * rather than queued out of window. Purists blame. | |
2220 | * | |
2221 | * F.e. "RFC state" is ESTABLISHED, | |
2222 | * if Linux state is FIN-WAIT-1, but FIN is still not sent. | |
2223 | * | |
2224 | * The visible declinations are that sometimes | |
2225 | * we enter time-wait state, when it is not required really | |
2226 | * (harmless), do not send active resets, when they are | |
2227 | * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when | |
2228 | * they look as CLOSING or LAST_ACK for Linux) | |
2229 | * Probably, I missed some more holelets. | |
2230 | * --ANK | |
8336886f JC |
2231 | * XXX (TFO) - To start off we don't support SYN+ACK+FIN |
2232 | * in a single packet! (May consider it later but will | |
2233 | * probably need API support or TCP_CORK SYN-ACK until | |
2234 | * data is written and socket is closed.) | |
1da177e4 LT |
2235 | */ |
2236 | tcp_send_fin(sk); | |
2237 | } | |
2238 | ||
2239 | sk_stream_wait_close(sk, timeout); | |
2240 | ||
2241 | adjudge_to_death: | |
75c2d907 HX |
2242 | state = sk->sk_state; |
2243 | sock_hold(sk); | |
2244 | sock_orphan(sk); | |
75c2d907 | 2245 | |
1da177e4 LT |
2246 | /* It is the last release_sock in its life. It will remove backlog. */ |
2247 | release_sock(sk); | |
2248 | ||
2249 | ||
2250 | /* Now socket is owned by kernel and we acquire BH lock | |
a777f715 | 2251 | * to finish close. No need to check for user refs. |
1da177e4 LT |
2252 | */ |
2253 | local_bh_disable(); | |
2254 | bh_lock_sock(sk); | |
547b792c | 2255 | WARN_ON(sock_owned_by_user(sk)); |
1da177e4 | 2256 | |
eb4dea58 HX |
2257 | percpu_counter_inc(sk->sk_prot->orphan_count); |
2258 | ||
75c2d907 HX |
2259 | /* Have we already been destroyed by a softirq or backlog? */ |
2260 | if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE) | |
2261 | goto out; | |
1da177e4 LT |
2262 | |
2263 | /* This is a (useful) BSD violating of the RFC. There is a | |
2264 | * problem with TCP as specified in that the other end could | |
2265 | * keep a socket open forever with no application left this end. | |
b10bd54c | 2266 | * We use a 1 minute timeout (about the same as BSD) then kill |
1da177e4 LT |
2267 | * our end. If they send after that then tough - BUT: long enough |
2268 | * that we won't make the old 4*rto = almost no time - whoops | |
2269 | * reset mistake. | |
2270 | * | |
2271 | * Nope, it was not mistake. It is really desired behaviour | |
2272 | * f.e. on http servers, when such sockets are useless, but | |
2273 | * consume significant resources. Let's do it with special | |
2274 | * linger2 option. --ANK | |
2275 | */ | |
2276 | ||
2277 | if (sk->sk_state == TCP_FIN_WAIT2) { | |
2278 | struct tcp_sock *tp = tcp_sk(sk); | |
2279 | if (tp->linger2 < 0) { | |
2280 | tcp_set_state(sk, TCP_CLOSE); | |
2281 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
02a1d6e7 | 2282 | __NET_INC_STATS(sock_net(sk), |
de0744af | 2283 | LINUX_MIB_TCPABORTONLINGER); |
1da177e4 | 2284 | } else { |
463c84b9 | 2285 | const int tmo = tcp_fin_time(sk); |
1da177e4 LT |
2286 | |
2287 | if (tmo > TCP_TIMEWAIT_LEN) { | |
52499afe DM |
2288 | inet_csk_reset_keepalive_timer(sk, |
2289 | tmo - TCP_TIMEWAIT_LEN); | |
1da177e4 | 2290 | } else { |
1da177e4 LT |
2291 | tcp_time_wait(sk, TCP_FIN_WAIT2, tmo); |
2292 | goto out; | |
2293 | } | |
2294 | } | |
2295 | } | |
2296 | if (sk->sk_state != TCP_CLOSE) { | |
3ab224be | 2297 | sk_mem_reclaim(sk); |
efcdbf24 | 2298 | if (tcp_check_oom(sk, 0)) { |
1da177e4 LT |
2299 | tcp_set_state(sk, TCP_CLOSE); |
2300 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
02a1d6e7 | 2301 | __NET_INC_STATS(sock_net(sk), |
de0744af | 2302 | LINUX_MIB_TCPABORTONMEMORY); |
1da177e4 LT |
2303 | } |
2304 | } | |
1da177e4 | 2305 | |
8336886f JC |
2306 | if (sk->sk_state == TCP_CLOSE) { |
2307 | struct request_sock *req = tcp_sk(sk)->fastopen_rsk; | |
2308 | /* We could get here with a non-NULL req if the socket is | |
2309 | * aborted (e.g., closed with unread data) before 3WHS | |
2310 | * finishes. | |
2311 | */ | |
00db4124 | 2312 | if (req) |
8336886f | 2313 | reqsk_fastopen_remove(sk, req, false); |
0a5578cf | 2314 | inet_csk_destroy_sock(sk); |
8336886f | 2315 | } |
1da177e4 LT |
2316 | /* Otherwise, socket is reprieved until protocol close. */ |
2317 | ||
2318 | out: | |
2319 | bh_unlock_sock(sk); | |
2320 | local_bh_enable(); | |
2321 | sock_put(sk); | |
2322 | } | |
4bc2f18b | 2323 | EXPORT_SYMBOL(tcp_close); |
1da177e4 LT |
2324 | |
2325 | /* These states need RST on ABORT according to RFC793 */ | |
2326 | ||
a2a385d6 | 2327 | static inline bool tcp_need_reset(int state) |
1da177e4 LT |
2328 | { |
2329 | return (1 << state) & | |
2330 | (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 | | |
2331 | TCPF_FIN_WAIT2 | TCPF_SYN_RECV); | |
2332 | } | |
2333 | ||
75c119af ED |
2334 | static void tcp_rtx_queue_purge(struct sock *sk) |
2335 | { | |
2336 | struct rb_node *p = rb_first(&sk->tcp_rtx_queue); | |
2337 | ||
2338 | while (p) { | |
2339 | struct sk_buff *skb = rb_to_skb(p); | |
2340 | ||
2341 | p = rb_next(p); | |
2342 | /* Since we are deleting whole queue, no need to | |
2343 | * list_del(&skb->tcp_tsorted_anchor) | |
2344 | */ | |
2345 | tcp_rtx_queue_unlink(skb, sk); | |
2346 | sk_wmem_free_skb(sk, skb); | |
2347 | } | |
2348 | } | |
2349 | ||
ac3f09ba ED |
2350 | void tcp_write_queue_purge(struct sock *sk) |
2351 | { | |
2352 | struct sk_buff *skb; | |
2353 | ||
2354 | tcp_chrono_stop(sk, TCP_CHRONO_BUSY); | |
2355 | while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) { | |
2356 | tcp_skb_tsorted_anchor_cleanup(skb); | |
2357 | sk_wmem_free_skb(sk, skb); | |
2358 | } | |
75c119af | 2359 | tcp_rtx_queue_purge(sk); |
ac3f09ba ED |
2360 | INIT_LIST_HEAD(&tcp_sk(sk)->tsorted_sent_queue); |
2361 | sk_mem_reclaim(sk); | |
2362 | tcp_clear_all_retrans_hints(tcp_sk(sk)); | |
2363 | } | |
2364 | ||
1da177e4 LT |
2365 | int tcp_disconnect(struct sock *sk, int flags) |
2366 | { | |
2367 | struct inet_sock *inet = inet_sk(sk); | |
463c84b9 | 2368 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
2369 | struct tcp_sock *tp = tcp_sk(sk); |
2370 | int err = 0; | |
2371 | int old_state = sk->sk_state; | |
2372 | ||
2373 | if (old_state != TCP_CLOSE) | |
2374 | tcp_set_state(sk, TCP_CLOSE); | |
2375 | ||
2376 | /* ABORT function of RFC793 */ | |
2377 | if (old_state == TCP_LISTEN) { | |
0a5578cf | 2378 | inet_csk_listen_stop(sk); |
ee995283 PE |
2379 | } else if (unlikely(tp->repair)) { |
2380 | sk->sk_err = ECONNABORTED; | |
1da177e4 LT |
2381 | } else if (tcp_need_reset(old_state) || |
2382 | (tp->snd_nxt != tp->write_seq && | |
2383 | (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) { | |
caa20d9a | 2384 | /* The last check adjusts for discrepancy of Linux wrt. RFC |
1da177e4 LT |
2385 | * states |
2386 | */ | |
2387 | tcp_send_active_reset(sk, gfp_any()); | |
2388 | sk->sk_err = ECONNRESET; | |
2389 | } else if (old_state == TCP_SYN_SENT) | |
2390 | sk->sk_err = ECONNRESET; | |
2391 | ||
2392 | tcp_clear_xmit_timers(sk); | |
2393 | __skb_queue_purge(&sk->sk_receive_queue); | |
fe067e8a | 2394 | tcp_write_queue_purge(sk); |
cf1ef3f0 | 2395 | tcp_fastopen_active_disable_ofo_check(sk); |
9f5afeae | 2396 | skb_rbtree_purge(&tp->out_of_order_queue); |
1da177e4 | 2397 | |
c720c7e8 | 2398 | inet->inet_dport = 0; |
1da177e4 LT |
2399 | |
2400 | if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) | |
2401 | inet_reset_saddr(sk); | |
2402 | ||
2403 | sk->sk_shutdown = 0; | |
2404 | sock_reset_flag(sk, SOCK_DONE); | |
740b0f18 | 2405 | tp->srtt_us = 0; |
686a5624 YM |
2406 | tp->write_seq += tp->max_window + 2; |
2407 | if (tp->write_seq == 0) | |
1da177e4 | 2408 | tp->write_seq = 1; |
463c84b9 | 2409 | icsk->icsk_backoff = 0; |
1da177e4 | 2410 | tp->snd_cwnd = 2; |
6687e988 | 2411 | icsk->icsk_probes_out = 0; |
1da177e4 | 2412 | tp->packets_out = 0; |
0b6a05c1 | 2413 | tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; |
1da177e4 | 2414 | tp->snd_cwnd_cnt = 0; |
1fdf475a | 2415 | tp->window_clamp = 0; |
6687e988 | 2416 | tcp_set_ca_state(sk, TCP_CA_Open); |
1da177e4 | 2417 | tcp_clear_retrans(tp); |
463c84b9 | 2418 | inet_csk_delack_init(sk); |
499350a5 WW |
2419 | /* Initialize rcv_mss to TCP_MIN_MSS to avoid division by 0 |
2420 | * issue in __tcp_select_window() | |
2421 | */ | |
2422 | icsk->icsk_ack.rcv_mss = TCP_MIN_MSS; | |
b40b4f79 | 2423 | memset(&tp->rx_opt, 0, sizeof(tp->rx_opt)); |
1da177e4 | 2424 | __sk_dst_reset(sk); |
d747a7a5 WC |
2425 | dst_release(sk->sk_rx_dst); |
2426 | sk->sk_rx_dst = NULL; | |
17c3060b | 2427 | tcp_saved_syn_free(tp); |
1da177e4 | 2428 | |
7db92362 WW |
2429 | /* Clean up fastopen related fields */ |
2430 | tcp_free_fastopen_req(tp); | |
2431 | inet->defer_connect = 0; | |
2432 | ||
c720c7e8 | 2433 | WARN_ON(inet->inet_num && !icsk->icsk_bind_hash); |
1da177e4 LT |
2434 | |
2435 | sk->sk_error_report(sk); | |
2436 | return err; | |
2437 | } | |
4bc2f18b | 2438 | EXPORT_SYMBOL(tcp_disconnect); |
1da177e4 | 2439 | |
a2a385d6 | 2440 | static inline bool tcp_can_repair_sock(const struct sock *sk) |
ee995283 | 2441 | { |
52e804c6 | 2442 | return ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN) && |
319b0534 | 2443 | (sk->sk_state != TCP_LISTEN); |
ee995283 PE |
2444 | } |
2445 | ||
b1ed4c4f AV |
2446 | static int tcp_repair_set_window(struct tcp_sock *tp, char __user *optbuf, int len) |
2447 | { | |
2448 | struct tcp_repair_window opt; | |
2449 | ||
2450 | if (!tp->repair) | |
2451 | return -EPERM; | |
2452 | ||
2453 | if (len != sizeof(opt)) | |
2454 | return -EINVAL; | |
2455 | ||
2456 | if (copy_from_user(&opt, optbuf, sizeof(opt))) | |
2457 | return -EFAULT; | |
2458 | ||
2459 | if (opt.max_window < opt.snd_wnd) | |
2460 | return -EINVAL; | |
2461 | ||
2462 | if (after(opt.snd_wl1, tp->rcv_nxt + opt.rcv_wnd)) | |
2463 | return -EINVAL; | |
2464 | ||
2465 | if (after(opt.rcv_wup, tp->rcv_nxt)) | |
2466 | return -EINVAL; | |
2467 | ||
2468 | tp->snd_wl1 = opt.snd_wl1; | |
2469 | tp->snd_wnd = opt.snd_wnd; | |
2470 | tp->max_window = opt.max_window; | |
2471 | ||
2472 | tp->rcv_wnd = opt.rcv_wnd; | |
2473 | tp->rcv_wup = opt.rcv_wup; | |
2474 | ||
2475 | return 0; | |
2476 | } | |
2477 | ||
15e56515 | 2478 | static int tcp_repair_options_est(struct sock *sk, |
de248a75 | 2479 | struct tcp_repair_opt __user *optbuf, unsigned int len) |
b139ba4e | 2480 | { |
15e56515 | 2481 | struct tcp_sock *tp = tcp_sk(sk); |
de248a75 | 2482 | struct tcp_repair_opt opt; |
b139ba4e | 2483 | |
de248a75 PE |
2484 | while (len >= sizeof(opt)) { |
2485 | if (copy_from_user(&opt, optbuf, sizeof(opt))) | |
b139ba4e PE |
2486 | return -EFAULT; |
2487 | ||
2488 | optbuf++; | |
de248a75 | 2489 | len -= sizeof(opt); |
b139ba4e | 2490 | |
de248a75 PE |
2491 | switch (opt.opt_code) { |
2492 | case TCPOPT_MSS: | |
2493 | tp->rx_opt.mss_clamp = opt.opt_val; | |
15e56515 | 2494 | tcp_mtup_init(sk); |
b139ba4e | 2495 | break; |
de248a75 | 2496 | case TCPOPT_WINDOW: |
bc26ccd8 AV |
2497 | { |
2498 | u16 snd_wscale = opt.opt_val & 0xFFFF; | |
2499 | u16 rcv_wscale = opt.opt_val >> 16; | |
2500 | ||
589c49cb | 2501 | if (snd_wscale > TCP_MAX_WSCALE || rcv_wscale > TCP_MAX_WSCALE) |
bc26ccd8 | 2502 | return -EFBIG; |
b139ba4e | 2503 | |
bc26ccd8 AV |
2504 | tp->rx_opt.snd_wscale = snd_wscale; |
2505 | tp->rx_opt.rcv_wscale = rcv_wscale; | |
2506 | tp->rx_opt.wscale_ok = 1; | |
2507 | } | |
b139ba4e | 2508 | break; |
b139ba4e | 2509 | case TCPOPT_SACK_PERM: |
de248a75 PE |
2510 | if (opt.opt_val != 0) |
2511 | return -EINVAL; | |
2512 | ||
b139ba4e PE |
2513 | tp->rx_opt.sack_ok |= TCP_SACK_SEEN; |
2514 | if (sysctl_tcp_fack) | |
2515 | tcp_enable_fack(tp); | |
2516 | break; | |
2517 | case TCPOPT_TIMESTAMP: | |
de248a75 PE |
2518 | if (opt.opt_val != 0) |
2519 | return -EINVAL; | |
2520 | ||
b139ba4e PE |
2521 | tp->rx_opt.tstamp_ok = 1; |
2522 | break; | |
2523 | } | |
2524 | } | |
2525 | ||
2526 | return 0; | |
2527 | } | |
2528 | ||
1da177e4 LT |
2529 | /* |
2530 | * Socket option code for TCP. | |
2531 | */ | |
3fdadf7d | 2532 | static int do_tcp_setsockopt(struct sock *sk, int level, |
b7058842 | 2533 | int optname, char __user *optval, unsigned int optlen) |
1da177e4 LT |
2534 | { |
2535 | struct tcp_sock *tp = tcp_sk(sk); | |
463c84b9 | 2536 | struct inet_connection_sock *icsk = inet_csk(sk); |
1e579caa | 2537 | struct net *net = sock_net(sk); |
1da177e4 LT |
2538 | int val; |
2539 | int err = 0; | |
2540 | ||
e56fb50f WAS |
2541 | /* These are data/string values, all the others are ints */ |
2542 | switch (optname) { | |
2543 | case TCP_CONGESTION: { | |
5f8ef48d SH |
2544 | char name[TCP_CA_NAME_MAX]; |
2545 | ||
2546 | if (optlen < 1) | |
2547 | return -EINVAL; | |
2548 | ||
2549 | val = strncpy_from_user(name, optval, | |
4fdb78d3 | 2550 | min_t(long, TCP_CA_NAME_MAX-1, optlen)); |
5f8ef48d SH |
2551 | if (val < 0) |
2552 | return -EFAULT; | |
2553 | name[val] = 0; | |
2554 | ||
2555 | lock_sock(sk); | |
ebfa00c5 | 2556 | err = tcp_set_congestion_control(sk, name, true, true); |
5f8ef48d SH |
2557 | release_sock(sk); |
2558 | return err; | |
2559 | } | |
734942cc DW |
2560 | case TCP_ULP: { |
2561 | char name[TCP_ULP_NAME_MAX]; | |
2562 | ||
2563 | if (optlen < 1) | |
2564 | return -EINVAL; | |
2565 | ||
2566 | val = strncpy_from_user(name, optval, | |
2567 | min_t(long, TCP_ULP_NAME_MAX - 1, | |
2568 | optlen)); | |
2569 | if (val < 0) | |
2570 | return -EFAULT; | |
2571 | name[val] = 0; | |
2572 | ||
2573 | lock_sock(sk); | |
2574 | err = tcp_set_ulp(sk, name); | |
2575 | release_sock(sk); | |
2576 | return err; | |
2577 | } | |
1fba70e5 YC |
2578 | case TCP_FASTOPEN_KEY: { |
2579 | __u8 key[TCP_FASTOPEN_KEY_LENGTH]; | |
2580 | ||
2581 | if (optlen != sizeof(key)) | |
2582 | return -EINVAL; | |
2583 | ||
2584 | if (copy_from_user(key, optval, optlen)) | |
2585 | return -EFAULT; | |
2586 | ||
2587 | return tcp_fastopen_reset_cipher(net, sk, key, sizeof(key)); | |
2588 | } | |
e56fb50f WAS |
2589 | default: |
2590 | /* fallthru */ | |
2591 | break; | |
ccbd6a5a | 2592 | } |
5f8ef48d | 2593 | |
1da177e4 LT |
2594 | if (optlen < sizeof(int)) |
2595 | return -EINVAL; | |
2596 | ||
2597 | if (get_user(val, (int __user *)optval)) | |
2598 | return -EFAULT; | |
2599 | ||
2600 | lock_sock(sk); | |
2601 | ||
2602 | switch (optname) { | |
2603 | case TCP_MAXSEG: | |
2604 | /* Values greater than interface MTU won't take effect. However | |
2605 | * at the point when this call is done we typically don't yet | |
a777f715 RC |
2606 | * know which interface is going to be used |
2607 | */ | |
cfc62d87 | 2608 | if (val && (val < TCP_MIN_MSS || val > MAX_TCP_WINDOW)) { |
1da177e4 LT |
2609 | err = -EINVAL; |
2610 | break; | |
2611 | } | |
2612 | tp->rx_opt.user_mss = val; | |
2613 | break; | |
2614 | ||
2615 | case TCP_NODELAY: | |
2616 | if (val) { | |
2617 | /* TCP_NODELAY is weaker than TCP_CORK, so that | |
2618 | * this option on corked socket is remembered, but | |
2619 | * it is not activated until cork is cleared. | |
2620 | * | |
2621 | * However, when TCP_NODELAY is set we make | |
2622 | * an explicit push, which overrides even TCP_CORK | |
2623 | * for currently queued segments. | |
2624 | */ | |
2625 | tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH; | |
9e412ba7 | 2626 | tcp_push_pending_frames(sk); |
1da177e4 LT |
2627 | } else { |
2628 | tp->nonagle &= ~TCP_NAGLE_OFF; | |
2629 | } | |
2630 | break; | |
2631 | ||
36e31b0a AP |
2632 | case TCP_THIN_LINEAR_TIMEOUTS: |
2633 | if (val < 0 || val > 1) | |
2634 | err = -EINVAL; | |
2635 | else | |
2636 | tp->thin_lto = val; | |
2637 | break; | |
2638 | ||
7e380175 AP |
2639 | case TCP_THIN_DUPACK: |
2640 | if (val < 0 || val > 1) | |
2641 | err = -EINVAL; | |
7e380175 AP |
2642 | break; |
2643 | ||
ee995283 PE |
2644 | case TCP_REPAIR: |
2645 | if (!tcp_can_repair_sock(sk)) | |
2646 | err = -EPERM; | |
2647 | else if (val == 1) { | |
2648 | tp->repair = 1; | |
2649 | sk->sk_reuse = SK_FORCE_REUSE; | |
2650 | tp->repair_queue = TCP_NO_QUEUE; | |
2651 | } else if (val == 0) { | |
2652 | tp->repair = 0; | |
2653 | sk->sk_reuse = SK_NO_REUSE; | |
2654 | tcp_send_window_probe(sk); | |
2655 | } else | |
2656 | err = -EINVAL; | |
2657 | ||
2658 | break; | |
2659 | ||
2660 | case TCP_REPAIR_QUEUE: | |
2661 | if (!tp->repair) | |
2662 | err = -EPERM; | |
2663 | else if (val < TCP_QUEUES_NR) | |
2664 | tp->repair_queue = val; | |
2665 | else | |
2666 | err = -EINVAL; | |
2667 | break; | |
2668 | ||
2669 | case TCP_QUEUE_SEQ: | |
2670 | if (sk->sk_state != TCP_CLOSE) | |
2671 | err = -EPERM; | |
2672 | else if (tp->repair_queue == TCP_SEND_QUEUE) | |
2673 | tp->write_seq = val; | |
2674 | else if (tp->repair_queue == TCP_RECV_QUEUE) | |
2675 | tp->rcv_nxt = val; | |
2676 | else | |
2677 | err = -EINVAL; | |
2678 | break; | |
2679 | ||
b139ba4e PE |
2680 | case TCP_REPAIR_OPTIONS: |
2681 | if (!tp->repair) | |
2682 | err = -EINVAL; | |
2683 | else if (sk->sk_state == TCP_ESTABLISHED) | |
15e56515 | 2684 | err = tcp_repair_options_est(sk, |
de248a75 PE |
2685 | (struct tcp_repair_opt __user *)optval, |
2686 | optlen); | |
b139ba4e PE |
2687 | else |
2688 | err = -EPERM; | |
2689 | break; | |
2690 | ||
1da177e4 LT |
2691 | case TCP_CORK: |
2692 | /* When set indicates to always queue non-full frames. | |
2693 | * Later the user clears this option and we transmit | |
2694 | * any pending partial frames in the queue. This is | |
2695 | * meant to be used alongside sendfile() to get properly | |
2696 | * filled frames when the user (for example) must write | |
2697 | * out headers with a write() call first and then use | |
2698 | * sendfile to send out the data parts. | |
2699 | * | |
2700 | * TCP_CORK can be set together with TCP_NODELAY and it is | |
2701 | * stronger than TCP_NODELAY. | |
2702 | */ | |
2703 | if (val) { | |
2704 | tp->nonagle |= TCP_NAGLE_CORK; | |
2705 | } else { | |
2706 | tp->nonagle &= ~TCP_NAGLE_CORK; | |
2707 | if (tp->nonagle&TCP_NAGLE_OFF) | |
2708 | tp->nonagle |= TCP_NAGLE_PUSH; | |
9e412ba7 | 2709 | tcp_push_pending_frames(sk); |
1da177e4 LT |
2710 | } |
2711 | break; | |
2712 | ||
2713 | case TCP_KEEPIDLE: | |
2714 | if (val < 1 || val > MAX_TCP_KEEPIDLE) | |
2715 | err = -EINVAL; | |
2716 | else { | |
2717 | tp->keepalive_time = val * HZ; | |
2718 | if (sock_flag(sk, SOCK_KEEPOPEN) && | |
2719 | !((1 << sk->sk_state) & | |
2720 | (TCPF_CLOSE | TCPF_LISTEN))) { | |
6c37e5de | 2721 | u32 elapsed = keepalive_time_elapsed(tp); |
1da177e4 LT |
2722 | if (tp->keepalive_time > elapsed) |
2723 | elapsed = tp->keepalive_time - elapsed; | |
2724 | else | |
2725 | elapsed = 0; | |
463c84b9 | 2726 | inet_csk_reset_keepalive_timer(sk, elapsed); |
1da177e4 LT |
2727 | } |
2728 | } | |
2729 | break; | |
2730 | case TCP_KEEPINTVL: | |
2731 | if (val < 1 || val > MAX_TCP_KEEPINTVL) | |
2732 | err = -EINVAL; | |
2733 | else | |
2734 | tp->keepalive_intvl = val * HZ; | |
2735 | break; | |
2736 | case TCP_KEEPCNT: | |
2737 | if (val < 1 || val > MAX_TCP_KEEPCNT) | |
2738 | err = -EINVAL; | |
2739 | else | |
2740 | tp->keepalive_probes = val; | |
2741 | break; | |
2742 | case TCP_SYNCNT: | |
2743 | if (val < 1 || val > MAX_TCP_SYNCNT) | |
2744 | err = -EINVAL; | |
2745 | else | |
463c84b9 | 2746 | icsk->icsk_syn_retries = val; |
1da177e4 LT |
2747 | break; |
2748 | ||
cd8ae852 ED |
2749 | case TCP_SAVE_SYN: |
2750 | if (val < 0 || val > 1) | |
2751 | err = -EINVAL; | |
2752 | else | |
2753 | tp->save_syn = val; | |
2754 | break; | |
2755 | ||
1da177e4 LT |
2756 | case TCP_LINGER2: |
2757 | if (val < 0) | |
2758 | tp->linger2 = -1; | |
1e579caa | 2759 | else if (val > net->ipv4.sysctl_tcp_fin_timeout / HZ) |
1da177e4 LT |
2760 | tp->linger2 = 0; |
2761 | else | |
2762 | tp->linger2 = val * HZ; | |
2763 | break; | |
2764 | ||
2765 | case TCP_DEFER_ACCEPT: | |
b103cf34 JA |
2766 | /* Translate value in seconds to number of retransmits */ |
2767 | icsk->icsk_accept_queue.rskq_defer_accept = | |
2768 | secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ, | |
2769 | TCP_RTO_MAX / HZ); | |
1da177e4 LT |
2770 | break; |
2771 | ||
2772 | case TCP_WINDOW_CLAMP: | |
2773 | if (!val) { | |
2774 | if (sk->sk_state != TCP_CLOSE) { | |
2775 | err = -EINVAL; | |
2776 | break; | |
2777 | } | |
2778 | tp->window_clamp = 0; | |
2779 | } else | |
2780 | tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ? | |
2781 | SOCK_MIN_RCVBUF / 2 : val; | |
2782 | break; | |
2783 | ||
2784 | case TCP_QUICKACK: | |
2785 | if (!val) { | |
463c84b9 | 2786 | icsk->icsk_ack.pingpong = 1; |
1da177e4 | 2787 | } else { |
463c84b9 | 2788 | icsk->icsk_ack.pingpong = 0; |
1da177e4 LT |
2789 | if ((1 << sk->sk_state) & |
2790 | (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) && | |
463c84b9 ACM |
2791 | inet_csk_ack_scheduled(sk)) { |
2792 | icsk->icsk_ack.pending |= ICSK_ACK_PUSHED; | |
0e4b4992 | 2793 | tcp_cleanup_rbuf(sk, 1); |
1da177e4 | 2794 | if (!(val & 1)) |
463c84b9 | 2795 | icsk->icsk_ack.pingpong = 1; |
1da177e4 LT |
2796 | } |
2797 | } | |
2798 | break; | |
2799 | ||
cfb6eeb4 YH |
2800 | #ifdef CONFIG_TCP_MD5SIG |
2801 | case TCP_MD5SIG: | |
8917a777 | 2802 | case TCP_MD5SIG_EXT: |
cfb6eeb4 | 2803 | /* Read the IP->Key mappings from userspace */ |
8917a777 | 2804 | err = tp->af_specific->md5_parse(sk, optname, optval, optlen); |
cfb6eeb4 YH |
2805 | break; |
2806 | #endif | |
dca43c75 | 2807 | case TCP_USER_TIMEOUT: |
b248230c | 2808 | /* Cap the max time in ms TCP will retry or probe the window |
dca43c75 JC |
2809 | * before giving up and aborting (ETIMEDOUT) a connection. |
2810 | */ | |
42493570 HL |
2811 | if (val < 0) |
2812 | err = -EINVAL; | |
2813 | else | |
2814 | icsk->icsk_user_timeout = msecs_to_jiffies(val); | |
dca43c75 | 2815 | break; |
8336886f JC |
2816 | |
2817 | case TCP_FASTOPEN: | |
2818 | if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE | | |
dfea2aa6 | 2819 | TCPF_LISTEN))) { |
43713848 | 2820 | tcp_fastopen_init_key_once(net); |
dfea2aa6 | 2821 | |
0536fcc0 | 2822 | fastopen_queue_tune(sk, val); |
dfea2aa6 | 2823 | } else { |
8336886f | 2824 | err = -EINVAL; |
dfea2aa6 | 2825 | } |
8336886f | 2826 | break; |
19f6d3f3 WW |
2827 | case TCP_FASTOPEN_CONNECT: |
2828 | if (val > 1 || val < 0) { | |
2829 | err = -EINVAL; | |
e1cfcbe8 | 2830 | } else if (net->ipv4.sysctl_tcp_fastopen & TFO_CLIENT_ENABLE) { |
19f6d3f3 WW |
2831 | if (sk->sk_state == TCP_CLOSE) |
2832 | tp->fastopen_connect = val; | |
2833 | else | |
2834 | err = -EINVAL; | |
2835 | } else { | |
2836 | err = -EOPNOTSUPP; | |
2837 | } | |
2838 | break; | |
93be6ce0 AV |
2839 | case TCP_TIMESTAMP: |
2840 | if (!tp->repair) | |
2841 | err = -EPERM; | |
2842 | else | |
9a568de4 | 2843 | tp->tsoffset = val - tcp_time_stamp_raw(); |
93be6ce0 | 2844 | break; |
b1ed4c4f AV |
2845 | case TCP_REPAIR_WINDOW: |
2846 | err = tcp_repair_set_window(tp, optval, optlen); | |
2847 | break; | |
c9bee3b7 ED |
2848 | case TCP_NOTSENT_LOWAT: |
2849 | tp->notsent_lowat = val; | |
2850 | sk->sk_write_space(sk); | |
2851 | break; | |
1da177e4 LT |
2852 | default: |
2853 | err = -ENOPROTOOPT; | |
2854 | break; | |
3ff50b79 SH |
2855 | } |
2856 | ||
1da177e4 LT |
2857 | release_sock(sk); |
2858 | return err; | |
2859 | } | |
2860 | ||
3fdadf7d | 2861 | int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval, |
b7058842 | 2862 | unsigned int optlen) |
3fdadf7d | 2863 | { |
cf533ea5 | 2864 | const struct inet_connection_sock *icsk = inet_csk(sk); |
3fdadf7d DM |
2865 | |
2866 | if (level != SOL_TCP) | |
2867 | return icsk->icsk_af_ops->setsockopt(sk, level, optname, | |
2868 | optval, optlen); | |
2869 | return do_tcp_setsockopt(sk, level, optname, optval, optlen); | |
2870 | } | |
4bc2f18b | 2871 | EXPORT_SYMBOL(tcp_setsockopt); |
3fdadf7d DM |
2872 | |
2873 | #ifdef CONFIG_COMPAT | |
543d9cfe | 2874 | int compat_tcp_setsockopt(struct sock *sk, int level, int optname, |
b7058842 | 2875 | char __user *optval, unsigned int optlen) |
3fdadf7d | 2876 | { |
dec73ff0 ACM |
2877 | if (level != SOL_TCP) |
2878 | return inet_csk_compat_setsockopt(sk, level, optname, | |
2879 | optval, optlen); | |
3fdadf7d DM |
2880 | return do_tcp_setsockopt(sk, level, optname, optval, optlen); |
2881 | } | |
543d9cfe | 2882 | EXPORT_SYMBOL(compat_tcp_setsockopt); |
3fdadf7d DM |
2883 | #endif |
2884 | ||
efd90174 FY |
2885 | static void tcp_get_info_chrono_stats(const struct tcp_sock *tp, |
2886 | struct tcp_info *info) | |
2887 | { | |
2888 | u64 stats[__TCP_CHRONO_MAX], total = 0; | |
2889 | enum tcp_chrono i; | |
2890 | ||
2891 | for (i = TCP_CHRONO_BUSY; i < __TCP_CHRONO_MAX; ++i) { | |
2892 | stats[i] = tp->chrono_stat[i - 1]; | |
2893 | if (i == tp->chrono_type) | |
628174cc | 2894 | stats[i] += tcp_jiffies32 - tp->chrono_start; |
efd90174 FY |
2895 | stats[i] *= USEC_PER_SEC / HZ; |
2896 | total += stats[i]; | |
2897 | } | |
2898 | ||
2899 | info->tcpi_busy_time = total; | |
2900 | info->tcpi_rwnd_limited = stats[TCP_CHRONO_RWND_LIMITED]; | |
2901 | info->tcpi_sndbuf_limited = stats[TCP_CHRONO_SNDBUF_LIMITED]; | |
2902 | } | |
2903 | ||
1da177e4 | 2904 | /* Return information about state of tcp endpoint in API format. */ |
0df48c26 | 2905 | void tcp_get_info(struct sock *sk, struct tcp_info *info) |
1da177e4 | 2906 | { |
35ac838a | 2907 | const struct tcp_sock *tp = tcp_sk(sk); /* iff sk_type == SOCK_STREAM */ |
463c84b9 | 2908 | const struct inet_connection_sock *icsk = inet_csk(sk); |
0263598c | 2909 | u32 now; |
ff5d7497 | 2910 | u64 rate64; |
67db3e4b | 2911 | bool slow; |
fad9dfef | 2912 | u32 rate; |
1da177e4 LT |
2913 | |
2914 | memset(info, 0, sizeof(*info)); | |
35ac838a CG |
2915 | if (sk->sk_type != SOCK_STREAM) |
2916 | return; | |
1da177e4 | 2917 | |
00fd38d9 ED |
2918 | info->tcpi_state = sk_state_load(sk); |
2919 | ||
ccbf3bfa ED |
2920 | /* Report meaningful fields for all TCP states, including listeners */ |
2921 | rate = READ_ONCE(sk->sk_pacing_rate); | |
2922 | rate64 = rate != ~0U ? rate : ~0ULL; | |
f522a5fc | 2923 | info->tcpi_pacing_rate = rate64; |
ccbf3bfa ED |
2924 | |
2925 | rate = READ_ONCE(sk->sk_max_pacing_rate); | |
2926 | rate64 = rate != ~0U ? rate : ~0ULL; | |
f522a5fc | 2927 | info->tcpi_max_pacing_rate = rate64; |
ccbf3bfa ED |
2928 | |
2929 | info->tcpi_reordering = tp->reordering; | |
2930 | info->tcpi_snd_cwnd = tp->snd_cwnd; | |
2931 | ||
2932 | if (info->tcpi_state == TCP_LISTEN) { | |
2933 | /* listeners aliased fields : | |
2934 | * tcpi_unacked -> Number of children ready for accept() | |
2935 | * tcpi_sacked -> max backlog | |
2936 | */ | |
2937 | info->tcpi_unacked = sk->sk_ack_backlog; | |
2938 | info->tcpi_sacked = sk->sk_max_ack_backlog; | |
2939 | return; | |
2940 | } | |
b369e7fd ED |
2941 | |
2942 | slow = lock_sock_fast(sk); | |
2943 | ||
6687e988 | 2944 | info->tcpi_ca_state = icsk->icsk_ca_state; |
463c84b9 | 2945 | info->tcpi_retransmits = icsk->icsk_retransmits; |
6687e988 | 2946 | info->tcpi_probes = icsk->icsk_probes_out; |
463c84b9 | 2947 | info->tcpi_backoff = icsk->icsk_backoff; |
1da177e4 LT |
2948 | |
2949 | if (tp->rx_opt.tstamp_ok) | |
2950 | info->tcpi_options |= TCPI_OPT_TIMESTAMPS; | |
e60402d0 | 2951 | if (tcp_is_sack(tp)) |
1da177e4 LT |
2952 | info->tcpi_options |= TCPI_OPT_SACK; |
2953 | if (tp->rx_opt.wscale_ok) { | |
2954 | info->tcpi_options |= TCPI_OPT_WSCALE; | |
2955 | info->tcpi_snd_wscale = tp->rx_opt.snd_wscale; | |
2956 | info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale; | |
e905a9ed | 2957 | } |
1da177e4 | 2958 | |
b5c5693b | 2959 | if (tp->ecn_flags & TCP_ECN_OK) |
1da177e4 | 2960 | info->tcpi_options |= TCPI_OPT_ECN; |
b5c5693b ED |
2961 | if (tp->ecn_flags & TCP_ECN_SEEN) |
2962 | info->tcpi_options |= TCPI_OPT_ECN_SEEN; | |
6f73601e YC |
2963 | if (tp->syn_data_acked) |
2964 | info->tcpi_options |= TCPI_OPT_SYN_DATA; | |
1da177e4 | 2965 | |
463c84b9 ACM |
2966 | info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto); |
2967 | info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato); | |
c1b4a7e6 | 2968 | info->tcpi_snd_mss = tp->mss_cache; |
463c84b9 | 2969 | info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss; |
1da177e4 | 2970 | |
ccbf3bfa ED |
2971 | info->tcpi_unacked = tp->packets_out; |
2972 | info->tcpi_sacked = tp->sacked_out; | |
2973 | ||
1da177e4 LT |
2974 | info->tcpi_lost = tp->lost_out; |
2975 | info->tcpi_retrans = tp->retrans_out; | |
2976 | info->tcpi_fackets = tp->fackets_out; | |
2977 | ||
d635fbe2 | 2978 | now = tcp_jiffies32; |
1da177e4 | 2979 | info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime); |
463c84b9 | 2980 | info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime); |
1da177e4 LT |
2981 | info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp); |
2982 | ||
d83d8461 | 2983 | info->tcpi_pmtu = icsk->icsk_pmtu_cookie; |
1da177e4 | 2984 | info->tcpi_rcv_ssthresh = tp->rcv_ssthresh; |
740b0f18 ED |
2985 | info->tcpi_rtt = tp->srtt_us >> 3; |
2986 | info->tcpi_rttvar = tp->mdev_us >> 2; | |
1da177e4 | 2987 | info->tcpi_snd_ssthresh = tp->snd_ssthresh; |
1da177e4 | 2988 | info->tcpi_advmss = tp->advmss; |
1da177e4 | 2989 | |
645f4c6f | 2990 | info->tcpi_rcv_rtt = tp->rcv_rtt_est.rtt_us >> 3; |
1da177e4 LT |
2991 | info->tcpi_rcv_space = tp->rcvq_space.space; |
2992 | ||
2993 | info->tcpi_total_retrans = tp->total_retrans; | |
977cb0ec | 2994 | |
f522a5fc ED |
2995 | info->tcpi_bytes_acked = tp->bytes_acked; |
2996 | info->tcpi_bytes_received = tp->bytes_received; | |
67db3e4b | 2997 | info->tcpi_notsent_bytes = max_t(int, 0, tp->write_seq - tp->snd_nxt); |
efd90174 | 2998 | tcp_get_info_chrono_stats(tp, info); |
67db3e4b | 2999 | |
2efd055c MRL |
3000 | info->tcpi_segs_out = tp->segs_out; |
3001 | info->tcpi_segs_in = tp->segs_in; | |
cd9b2660 | 3002 | |
cd9b2660 | 3003 | info->tcpi_min_rtt = tcp_min_rtt(tp); |
a44d6eac MKL |
3004 | info->tcpi_data_segs_in = tp->data_segs_in; |
3005 | info->tcpi_data_segs_out = tp->data_segs_out; | |
eb8329e0 YC |
3006 | |
3007 | info->tcpi_delivery_rate_app_limited = tp->rate_app_limited ? 1 : 0; | |
0263598c WW |
3008 | rate64 = tcp_compute_delivery_rate(tp); |
3009 | if (rate64) | |
f522a5fc | 3010 | info->tcpi_delivery_rate = rate64; |
b369e7fd | 3011 | unlock_sock_fast(sk, slow); |
1da177e4 | 3012 | } |
1da177e4 LT |
3013 | EXPORT_SYMBOL_GPL(tcp_get_info); |
3014 | ||
1c885808 FY |
3015 | struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk) |
3016 | { | |
3017 | const struct tcp_sock *tp = tcp_sk(sk); | |
3018 | struct sk_buff *stats; | |
3019 | struct tcp_info info; | |
bb7c19f9 WW |
3020 | u64 rate64; |
3021 | u32 rate; | |
1c885808 | 3022 | |
bb7c19f9 WW |
3023 | stats = alloc_skb(7 * nla_total_size_64bit(sizeof(u64)) + |
3024 | 3 * nla_total_size(sizeof(u32)) + | |
3025 | 2 * nla_total_size(sizeof(u8)), GFP_ATOMIC); | |
1c885808 FY |
3026 | if (!stats) |
3027 | return NULL; | |
3028 | ||
3029 | tcp_get_info_chrono_stats(tp, &info); | |
3030 | nla_put_u64_64bit(stats, TCP_NLA_BUSY, | |
3031 | info.tcpi_busy_time, TCP_NLA_PAD); | |
3032 | nla_put_u64_64bit(stats, TCP_NLA_RWND_LIMITED, | |
3033 | info.tcpi_rwnd_limited, TCP_NLA_PAD); | |
3034 | nla_put_u64_64bit(stats, TCP_NLA_SNDBUF_LIMITED, | |
3035 | info.tcpi_sndbuf_limited, TCP_NLA_PAD); | |
7e98102f YC |
3036 | nla_put_u64_64bit(stats, TCP_NLA_DATA_SEGS_OUT, |
3037 | tp->data_segs_out, TCP_NLA_PAD); | |
3038 | nla_put_u64_64bit(stats, TCP_NLA_TOTAL_RETRANS, | |
3039 | tp->total_retrans, TCP_NLA_PAD); | |
bb7c19f9 WW |
3040 | |
3041 | rate = READ_ONCE(sk->sk_pacing_rate); | |
3042 | rate64 = rate != ~0U ? rate : ~0ULL; | |
3043 | nla_put_u64_64bit(stats, TCP_NLA_PACING_RATE, rate64, TCP_NLA_PAD); | |
3044 | ||
3045 | rate64 = tcp_compute_delivery_rate(tp); | |
3046 | nla_put_u64_64bit(stats, TCP_NLA_DELIVERY_RATE, rate64, TCP_NLA_PAD); | |
3047 | ||
3048 | nla_put_u32(stats, TCP_NLA_SND_CWND, tp->snd_cwnd); | |
3049 | nla_put_u32(stats, TCP_NLA_REORDERING, tp->reordering); | |
3050 | nla_put_u32(stats, TCP_NLA_MIN_RTT, tcp_min_rtt(tp)); | |
3051 | ||
3052 | nla_put_u8(stats, TCP_NLA_RECUR_RETRANS, inet_csk(sk)->icsk_retransmits); | |
3053 | nla_put_u8(stats, TCP_NLA_DELIVERY_RATE_APP_LMT, !!tp->rate_app_limited); | |
1c885808 FY |
3054 | return stats; |
3055 | } | |
3056 | ||
3fdadf7d DM |
3057 | static int do_tcp_getsockopt(struct sock *sk, int level, |
3058 | int optname, char __user *optval, int __user *optlen) | |
1da177e4 | 3059 | { |
295f7324 | 3060 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 3061 | struct tcp_sock *tp = tcp_sk(sk); |
6fa25166 | 3062 | struct net *net = sock_net(sk); |
1da177e4 LT |
3063 | int val, len; |
3064 | ||
1da177e4 LT |
3065 | if (get_user(len, optlen)) |
3066 | return -EFAULT; | |
3067 | ||
3068 | len = min_t(unsigned int, len, sizeof(int)); | |
3069 | ||
3070 | if (len < 0) | |
3071 | return -EINVAL; | |
3072 | ||
3073 | switch (optname) { | |
3074 | case TCP_MAXSEG: | |
c1b4a7e6 | 3075 | val = tp->mss_cache; |
1da177e4 LT |
3076 | if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) |
3077 | val = tp->rx_opt.user_mss; | |
5e6a3ce6 PE |
3078 | if (tp->repair) |
3079 | val = tp->rx_opt.mss_clamp; | |
1da177e4 LT |
3080 | break; |
3081 | case TCP_NODELAY: | |
3082 | val = !!(tp->nonagle&TCP_NAGLE_OFF); | |
3083 | break; | |
3084 | case TCP_CORK: | |
3085 | val = !!(tp->nonagle&TCP_NAGLE_CORK); | |
3086 | break; | |
3087 | case TCP_KEEPIDLE: | |
df19a626 | 3088 | val = keepalive_time_when(tp) / HZ; |
1da177e4 LT |
3089 | break; |
3090 | case TCP_KEEPINTVL: | |
df19a626 | 3091 | val = keepalive_intvl_when(tp) / HZ; |
1da177e4 LT |
3092 | break; |
3093 | case TCP_KEEPCNT: | |
df19a626 | 3094 | val = keepalive_probes(tp); |
1da177e4 LT |
3095 | break; |
3096 | case TCP_SYNCNT: | |
6fa25166 | 3097 | val = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries; |
1da177e4 LT |
3098 | break; |
3099 | case TCP_LINGER2: | |
3100 | val = tp->linger2; | |
3101 | if (val >= 0) | |
1e579caa | 3102 | val = (val ? : net->ipv4.sysctl_tcp_fin_timeout) / HZ; |
1da177e4 LT |
3103 | break; |
3104 | case TCP_DEFER_ACCEPT: | |
b103cf34 JA |
3105 | val = retrans_to_secs(icsk->icsk_accept_queue.rskq_defer_accept, |
3106 | TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ); | |
1da177e4 LT |
3107 | break; |
3108 | case TCP_WINDOW_CLAMP: | |
3109 | val = tp->window_clamp; | |
3110 | break; | |
3111 | case TCP_INFO: { | |
3112 | struct tcp_info info; | |
3113 | ||
3114 | if (get_user(len, optlen)) | |
3115 | return -EFAULT; | |
3116 | ||
3117 | tcp_get_info(sk, &info); | |
3118 | ||
3119 | len = min_t(unsigned int, len, sizeof(info)); | |
3120 | if (put_user(len, optlen)) | |
3121 | return -EFAULT; | |
3122 | if (copy_to_user(optval, &info, len)) | |
3123 | return -EFAULT; | |
3124 | return 0; | |
3125 | } | |
6e9250f5 ED |
3126 | case TCP_CC_INFO: { |
3127 | const struct tcp_congestion_ops *ca_ops; | |
3128 | union tcp_cc_info info; | |
3129 | size_t sz = 0; | |
3130 | int attr; | |
3131 | ||
3132 | if (get_user(len, optlen)) | |
3133 | return -EFAULT; | |
3134 | ||
3135 | ca_ops = icsk->icsk_ca_ops; | |
3136 | if (ca_ops && ca_ops->get_info) | |
3137 | sz = ca_ops->get_info(sk, ~0U, &attr, &info); | |
3138 | ||
3139 | len = min_t(unsigned int, len, sz); | |
3140 | if (put_user(len, optlen)) | |
3141 | return -EFAULT; | |
3142 | if (copy_to_user(optval, &info, len)) | |
3143 | return -EFAULT; | |
3144 | return 0; | |
3145 | } | |
1da177e4 | 3146 | case TCP_QUICKACK: |
295f7324 | 3147 | val = !icsk->icsk_ack.pingpong; |
1da177e4 | 3148 | break; |
5f8ef48d SH |
3149 | |
3150 | case TCP_CONGESTION: | |
3151 | if (get_user(len, optlen)) | |
3152 | return -EFAULT; | |
3153 | len = min_t(unsigned int, len, TCP_CA_NAME_MAX); | |
3154 | if (put_user(len, optlen)) | |
3155 | return -EFAULT; | |
6687e988 | 3156 | if (copy_to_user(optval, icsk->icsk_ca_ops->name, len)) |
5f8ef48d SH |
3157 | return -EFAULT; |
3158 | return 0; | |
e56fb50f | 3159 | |
734942cc DW |
3160 | case TCP_ULP: |
3161 | if (get_user(len, optlen)) | |
3162 | return -EFAULT; | |
3163 | len = min_t(unsigned int, len, TCP_ULP_NAME_MAX); | |
d97af30f DW |
3164 | if (!icsk->icsk_ulp_ops) { |
3165 | if (put_user(0, optlen)) | |
3166 | return -EFAULT; | |
3167 | return 0; | |
3168 | } | |
734942cc DW |
3169 | if (put_user(len, optlen)) |
3170 | return -EFAULT; | |
3171 | if (copy_to_user(optval, icsk->icsk_ulp_ops->name, len)) | |
3172 | return -EFAULT; | |
3173 | return 0; | |
3174 | ||
1fba70e5 YC |
3175 | case TCP_FASTOPEN_KEY: { |
3176 | __u8 key[TCP_FASTOPEN_KEY_LENGTH]; | |
3177 | struct tcp_fastopen_context *ctx; | |
3178 | ||
3179 | if (get_user(len, optlen)) | |
3180 | return -EFAULT; | |
3181 | ||
3182 | rcu_read_lock(); | |
3183 | ctx = rcu_dereference(icsk->icsk_accept_queue.fastopenq.ctx); | |
3184 | if (ctx) | |
3185 | memcpy(key, ctx->key, sizeof(key)); | |
3186 | else | |
3187 | len = 0; | |
3188 | rcu_read_unlock(); | |
3189 | ||
3190 | len = min_t(unsigned int, len, sizeof(key)); | |
3191 | if (put_user(len, optlen)) | |
3192 | return -EFAULT; | |
3193 | if (copy_to_user(optval, key, len)) | |
3194 | return -EFAULT; | |
3195 | return 0; | |
3196 | } | |
3c0fef0b JH |
3197 | case TCP_THIN_LINEAR_TIMEOUTS: |
3198 | val = tp->thin_lto; | |
3199 | break; | |
4a7f6009 | 3200 | |
3c0fef0b | 3201 | case TCP_THIN_DUPACK: |
4a7f6009 | 3202 | val = 0; |
3c0fef0b | 3203 | break; |
dca43c75 | 3204 | |
ee995283 PE |
3205 | case TCP_REPAIR: |
3206 | val = tp->repair; | |
3207 | break; | |
3208 | ||
3209 | case TCP_REPAIR_QUEUE: | |
3210 | if (tp->repair) | |
3211 | val = tp->repair_queue; | |
3212 | else | |
3213 | return -EINVAL; | |
3214 | break; | |
3215 | ||
b1ed4c4f AV |
3216 | case TCP_REPAIR_WINDOW: { |
3217 | struct tcp_repair_window opt; | |
3218 | ||
3219 | if (get_user(len, optlen)) | |
3220 | return -EFAULT; | |
3221 | ||
3222 | if (len != sizeof(opt)) | |
3223 | return -EINVAL; | |
3224 | ||
3225 | if (!tp->repair) | |
3226 | return -EPERM; | |
3227 | ||
3228 | opt.snd_wl1 = tp->snd_wl1; | |
3229 | opt.snd_wnd = tp->snd_wnd; | |
3230 | opt.max_window = tp->max_window; | |
3231 | opt.rcv_wnd = tp->rcv_wnd; | |
3232 | opt.rcv_wup = tp->rcv_wup; | |
3233 | ||
3234 | if (copy_to_user(optval, &opt, len)) | |
3235 | return -EFAULT; | |
3236 | return 0; | |
3237 | } | |
ee995283 PE |
3238 | case TCP_QUEUE_SEQ: |
3239 | if (tp->repair_queue == TCP_SEND_QUEUE) | |
3240 | val = tp->write_seq; | |
3241 | else if (tp->repair_queue == TCP_RECV_QUEUE) | |
3242 | val = tp->rcv_nxt; | |
3243 | else | |
3244 | return -EINVAL; | |
3245 | break; | |
3246 | ||
dca43c75 JC |
3247 | case TCP_USER_TIMEOUT: |
3248 | val = jiffies_to_msecs(icsk->icsk_user_timeout); | |
3249 | break; | |
1536e285 KN |
3250 | |
3251 | case TCP_FASTOPEN: | |
0536fcc0 | 3252 | val = icsk->icsk_accept_queue.fastopenq.max_qlen; |
1536e285 KN |
3253 | break; |
3254 | ||
19f6d3f3 WW |
3255 | case TCP_FASTOPEN_CONNECT: |
3256 | val = tp->fastopen_connect; | |
3257 | break; | |
3258 | ||
93be6ce0 | 3259 | case TCP_TIMESTAMP: |
9a568de4 | 3260 | val = tcp_time_stamp_raw() + tp->tsoffset; |
93be6ce0 | 3261 | break; |
c9bee3b7 ED |
3262 | case TCP_NOTSENT_LOWAT: |
3263 | val = tp->notsent_lowat; | |
3264 | break; | |
cd8ae852 ED |
3265 | case TCP_SAVE_SYN: |
3266 | val = tp->save_syn; | |
3267 | break; | |
3268 | case TCP_SAVED_SYN: { | |
3269 | if (get_user(len, optlen)) | |
3270 | return -EFAULT; | |
3271 | ||
3272 | lock_sock(sk); | |
3273 | if (tp->saved_syn) { | |
aea0929e EM |
3274 | if (len < tp->saved_syn[0]) { |
3275 | if (put_user(tp->saved_syn[0], optlen)) { | |
3276 | release_sock(sk); | |
3277 | return -EFAULT; | |
3278 | } | |
3279 | release_sock(sk); | |
3280 | return -EINVAL; | |
3281 | } | |
3282 | len = tp->saved_syn[0]; | |
cd8ae852 ED |
3283 | if (put_user(len, optlen)) { |
3284 | release_sock(sk); | |
3285 | return -EFAULT; | |
3286 | } | |
3287 | if (copy_to_user(optval, tp->saved_syn + 1, len)) { | |
3288 | release_sock(sk); | |
3289 | return -EFAULT; | |
3290 | } | |
3291 | tcp_saved_syn_free(tp); | |
3292 | release_sock(sk); | |
3293 | } else { | |
3294 | release_sock(sk); | |
3295 | len = 0; | |
3296 | if (put_user(len, optlen)) | |
3297 | return -EFAULT; | |
3298 | } | |
3299 | return 0; | |
3300 | } | |
1da177e4 LT |
3301 | default: |
3302 | return -ENOPROTOOPT; | |
3ff50b79 | 3303 | } |
1da177e4 LT |
3304 | |
3305 | if (put_user(len, optlen)) | |
3306 | return -EFAULT; | |
3307 | if (copy_to_user(optval, &val, len)) | |
3308 | return -EFAULT; | |
3309 | return 0; | |
3310 | } | |
3311 | ||
3fdadf7d DM |
3312 | int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval, |
3313 | int __user *optlen) | |
3314 | { | |
3315 | struct inet_connection_sock *icsk = inet_csk(sk); | |
3316 | ||
3317 | if (level != SOL_TCP) | |
3318 | return icsk->icsk_af_ops->getsockopt(sk, level, optname, | |
3319 | optval, optlen); | |
3320 | return do_tcp_getsockopt(sk, level, optname, optval, optlen); | |
3321 | } | |
4bc2f18b | 3322 | EXPORT_SYMBOL(tcp_getsockopt); |
3fdadf7d DM |
3323 | |
3324 | #ifdef CONFIG_COMPAT | |
543d9cfe ACM |
3325 | int compat_tcp_getsockopt(struct sock *sk, int level, int optname, |
3326 | char __user *optval, int __user *optlen) | |
3fdadf7d | 3327 | { |
dec73ff0 ACM |
3328 | if (level != SOL_TCP) |
3329 | return inet_csk_compat_getsockopt(sk, level, optname, | |
3330 | optval, optlen); | |
3fdadf7d DM |
3331 | return do_tcp_getsockopt(sk, level, optname, optval, optlen); |
3332 | } | |
543d9cfe | 3333 | EXPORT_SYMBOL(compat_tcp_getsockopt); |
3fdadf7d | 3334 | #endif |
1da177e4 | 3335 | |
cfb6eeb4 | 3336 | #ifdef CONFIG_TCP_MD5SIG |
349ce993 | 3337 | static DEFINE_PER_CPU(struct tcp_md5sig_pool, tcp_md5sig_pool); |
71cea17e | 3338 | static DEFINE_MUTEX(tcp_md5sig_mutex); |
349ce993 | 3339 | static bool tcp_md5sig_pool_populated = false; |
cfb6eeb4 | 3340 | |
71cea17e | 3341 | static void __tcp_alloc_md5sig_pool(void) |
cfb6eeb4 | 3342 | { |
cf80e0e4 | 3343 | struct crypto_ahash *hash; |
cfb6eeb4 | 3344 | int cpu; |
cfb6eeb4 | 3345 | |
cf80e0e4 | 3346 | hash = crypto_alloc_ahash("md5", 0, CRYPTO_ALG_ASYNC); |
70477371 | 3347 | if (IS_ERR(hash)) |
cf80e0e4 HX |
3348 | return; |
3349 | ||
cfb6eeb4 | 3350 | for_each_possible_cpu(cpu) { |
19689e38 | 3351 | void *scratch = per_cpu(tcp_md5sig_pool, cpu).scratch; |
cf80e0e4 | 3352 | struct ahash_request *req; |
cfb6eeb4 | 3353 | |
19689e38 ED |
3354 | if (!scratch) { |
3355 | scratch = kmalloc_node(sizeof(union tcp_md5sum_block) + | |
3356 | sizeof(struct tcphdr), | |
3357 | GFP_KERNEL, | |
3358 | cpu_to_node(cpu)); | |
3359 | if (!scratch) | |
3360 | return; | |
3361 | per_cpu(tcp_md5sig_pool, cpu).scratch = scratch; | |
3362 | } | |
cf80e0e4 HX |
3363 | if (per_cpu(tcp_md5sig_pool, cpu).md5_req) |
3364 | continue; | |
3365 | ||
3366 | req = ahash_request_alloc(hash, GFP_KERNEL); | |
3367 | if (!req) | |
3368 | return; | |
3369 | ||
3370 | ahash_request_set_callback(req, 0, NULL, NULL); | |
3371 | ||
3372 | per_cpu(tcp_md5sig_pool, cpu).md5_req = req; | |
cfb6eeb4 | 3373 | } |
349ce993 ED |
3374 | /* before setting tcp_md5sig_pool_populated, we must commit all writes |
3375 | * to memory. See smp_rmb() in tcp_get_md5sig_pool() | |
71cea17e ED |
3376 | */ |
3377 | smp_wmb(); | |
349ce993 | 3378 | tcp_md5sig_pool_populated = true; |
cfb6eeb4 YH |
3379 | } |
3380 | ||
71cea17e | 3381 | bool tcp_alloc_md5sig_pool(void) |
cfb6eeb4 | 3382 | { |
349ce993 | 3383 | if (unlikely(!tcp_md5sig_pool_populated)) { |
71cea17e ED |
3384 | mutex_lock(&tcp_md5sig_mutex); |
3385 | ||
349ce993 | 3386 | if (!tcp_md5sig_pool_populated) |
71cea17e ED |
3387 | __tcp_alloc_md5sig_pool(); |
3388 | ||
3389 | mutex_unlock(&tcp_md5sig_mutex); | |
cfb6eeb4 | 3390 | } |
349ce993 | 3391 | return tcp_md5sig_pool_populated; |
cfb6eeb4 | 3392 | } |
cfb6eeb4 YH |
3393 | EXPORT_SYMBOL(tcp_alloc_md5sig_pool); |
3394 | ||
35790c04 ED |
3395 | |
3396 | /** | |
3397 | * tcp_get_md5sig_pool - get md5sig_pool for this user | |
3398 | * | |
3399 | * We use percpu structure, so if we succeed, we exit with preemption | |
3400 | * and BH disabled, to make sure another thread or softirq handling | |
3401 | * wont try to get same context. | |
3402 | */ | |
3403 | struct tcp_md5sig_pool *tcp_get_md5sig_pool(void) | |
cfb6eeb4 | 3404 | { |
35790c04 | 3405 | local_bh_disable(); |
cfb6eeb4 | 3406 | |
349ce993 ED |
3407 | if (tcp_md5sig_pool_populated) { |
3408 | /* coupled with smp_wmb() in __tcp_alloc_md5sig_pool() */ | |
3409 | smp_rmb(); | |
3410 | return this_cpu_ptr(&tcp_md5sig_pool); | |
3411 | } | |
35790c04 ED |
3412 | local_bh_enable(); |
3413 | return NULL; | |
3414 | } | |
3415 | EXPORT_SYMBOL(tcp_get_md5sig_pool); | |
cfb6eeb4 | 3416 | |
49a72dfb | 3417 | int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp, |
cf533ea5 | 3418 | const struct sk_buff *skb, unsigned int header_len) |
49a72dfb AL |
3419 | { |
3420 | struct scatterlist sg; | |
3421 | const struct tcphdr *tp = tcp_hdr(skb); | |
cf80e0e4 | 3422 | struct ahash_request *req = hp->md5_req; |
95c96174 ED |
3423 | unsigned int i; |
3424 | const unsigned int head_data_len = skb_headlen(skb) > header_len ? | |
3425 | skb_headlen(skb) - header_len : 0; | |
49a72dfb | 3426 | const struct skb_shared_info *shi = skb_shinfo(skb); |
d7fd1b57 | 3427 | struct sk_buff *frag_iter; |
49a72dfb AL |
3428 | |
3429 | sg_init_table(&sg, 1); | |
3430 | ||
3431 | sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len); | |
cf80e0e4 HX |
3432 | ahash_request_set_crypt(req, &sg, NULL, head_data_len); |
3433 | if (crypto_ahash_update(req)) | |
49a72dfb AL |
3434 | return 1; |
3435 | ||
3436 | for (i = 0; i < shi->nr_frags; ++i) { | |
3437 | const struct skb_frag_struct *f = &shi->frags[i]; | |
54d27fcb ED |
3438 | unsigned int offset = f->page_offset; |
3439 | struct page *page = skb_frag_page(f) + (offset >> PAGE_SHIFT); | |
3440 | ||
3441 | sg_set_page(&sg, page, skb_frag_size(f), | |
3442 | offset_in_page(offset)); | |
cf80e0e4 HX |
3443 | ahash_request_set_crypt(req, &sg, NULL, skb_frag_size(f)); |
3444 | if (crypto_ahash_update(req)) | |
49a72dfb AL |
3445 | return 1; |
3446 | } | |
3447 | ||
d7fd1b57 ED |
3448 | skb_walk_frags(skb, frag_iter) |
3449 | if (tcp_md5_hash_skb_data(hp, frag_iter, 0)) | |
3450 | return 1; | |
3451 | ||
49a72dfb AL |
3452 | return 0; |
3453 | } | |
49a72dfb AL |
3454 | EXPORT_SYMBOL(tcp_md5_hash_skb_data); |
3455 | ||
cf533ea5 | 3456 | int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, const struct tcp_md5sig_key *key) |
49a72dfb AL |
3457 | { |
3458 | struct scatterlist sg; | |
3459 | ||
3460 | sg_init_one(&sg, key->key, key->keylen); | |
cf80e0e4 HX |
3461 | ahash_request_set_crypt(hp->md5_req, &sg, NULL, key->keylen); |
3462 | return crypto_ahash_update(hp->md5_req); | |
49a72dfb | 3463 | } |
49a72dfb AL |
3464 | EXPORT_SYMBOL(tcp_md5_hash_key); |
3465 | ||
cfb6eeb4 YH |
3466 | #endif |
3467 | ||
4ac02bab AK |
3468 | void tcp_done(struct sock *sk) |
3469 | { | |
8336886f JC |
3470 | struct request_sock *req = tcp_sk(sk)->fastopen_rsk; |
3471 | ||
5a5f3a8d | 3472 | if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV) |
c10d9310 | 3473 | TCP_INC_STATS(sock_net(sk), TCP_MIB_ATTEMPTFAILS); |
4ac02bab AK |
3474 | |
3475 | tcp_set_state(sk, TCP_CLOSE); | |
3476 | tcp_clear_xmit_timers(sk); | |
00db4124 | 3477 | if (req) |
8336886f | 3478 | reqsk_fastopen_remove(sk, req, false); |
4ac02bab AK |
3479 | |
3480 | sk->sk_shutdown = SHUTDOWN_MASK; | |
3481 | ||
3482 | if (!sock_flag(sk, SOCK_DEAD)) | |
3483 | sk->sk_state_change(sk); | |
3484 | else | |
3485 | inet_csk_destroy_sock(sk); | |
3486 | } | |
3487 | EXPORT_SYMBOL_GPL(tcp_done); | |
3488 | ||
c1e64e29 LC |
3489 | int tcp_abort(struct sock *sk, int err) |
3490 | { | |
3491 | if (!sk_fullsock(sk)) { | |
07f6f4a3 ED |
3492 | if (sk->sk_state == TCP_NEW_SYN_RECV) { |
3493 | struct request_sock *req = inet_reqsk(sk); | |
3494 | ||
3495 | local_bh_disable(); | |
3496 | inet_csk_reqsk_queue_drop_and_put(req->rsk_listener, | |
3497 | req); | |
3498 | local_bh_enable(); | |
3499 | return 0; | |
3500 | } | |
c1e64e29 LC |
3501 | return -EOPNOTSUPP; |
3502 | } | |
3503 | ||
3504 | /* Don't race with userspace socket closes such as tcp_close. */ | |
3505 | lock_sock(sk); | |
3506 | ||
2010b93e LC |
3507 | if (sk->sk_state == TCP_LISTEN) { |
3508 | tcp_set_state(sk, TCP_CLOSE); | |
3509 | inet_csk_listen_stop(sk); | |
3510 | } | |
3511 | ||
c1e64e29 LC |
3512 | /* Don't race with BH socket closes such as inet_csk_listen_stop. */ |
3513 | local_bh_disable(); | |
3514 | bh_lock_sock(sk); | |
3515 | ||
3516 | if (!sock_flag(sk, SOCK_DEAD)) { | |
3517 | sk->sk_err = err; | |
3518 | /* This barrier is coupled with smp_rmb() in tcp_poll() */ | |
3519 | smp_wmb(); | |
3520 | sk->sk_error_report(sk); | |
3521 | if (tcp_need_reset(sk->sk_state)) | |
3522 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
3523 | tcp_done(sk); | |
3524 | } | |
3525 | ||
3526 | bh_unlock_sock(sk); | |
3527 | local_bh_enable(); | |
3528 | release_sock(sk); | |
c1e64e29 LC |
3529 | return 0; |
3530 | } | |
3531 | EXPORT_SYMBOL_GPL(tcp_abort); | |
3532 | ||
5f8ef48d | 3533 | extern struct tcp_congestion_ops tcp_reno; |
1da177e4 LT |
3534 | |
3535 | static __initdata unsigned long thash_entries; | |
3536 | static int __init set_thash_entries(char *str) | |
3537 | { | |
413c27d8 EZ |
3538 | ssize_t ret; |
3539 | ||
1da177e4 LT |
3540 | if (!str) |
3541 | return 0; | |
413c27d8 EZ |
3542 | |
3543 | ret = kstrtoul(str, 0, &thash_entries); | |
3544 | if (ret) | |
3545 | return 0; | |
3546 | ||
1da177e4 LT |
3547 | return 1; |
3548 | } | |
3549 | __setup("thash_entries=", set_thash_entries); | |
3550 | ||
47d7a88c | 3551 | static void __init tcp_init_mem(void) |
4acb4190 | 3552 | { |
b66e91cc ED |
3553 | unsigned long limit = nr_free_buffer_pages() / 16; |
3554 | ||
4acb4190 | 3555 | limit = max(limit, 128UL); |
b66e91cc ED |
3556 | sysctl_tcp_mem[0] = limit / 4 * 3; /* 4.68 % */ |
3557 | sysctl_tcp_mem[1] = limit; /* 6.25 % */ | |
3558 | sysctl_tcp_mem[2] = sysctl_tcp_mem[0] * 2; /* 9.37 % */ | |
4acb4190 GC |
3559 | } |
3560 | ||
1da177e4 LT |
3561 | void __init tcp_init(void) |
3562 | { | |
b49960a0 | 3563 | int max_rshare, max_wshare, cnt; |
b2d3ea4a | 3564 | unsigned long limit; |
074b8517 | 3565 | unsigned int i; |
1da177e4 | 3566 | |
b2d3ea4a ED |
3567 | BUILD_BUG_ON(sizeof(struct tcp_skb_cb) > |
3568 | FIELD_SIZEOF(struct sk_buff, cb)); | |
1da177e4 | 3569 | |
908c7f19 TH |
3570 | percpu_counter_init(&tcp_sockets_allocated, 0, GFP_KERNEL); |
3571 | percpu_counter_init(&tcp_orphan_count, 0, GFP_KERNEL); | |
1946e672 | 3572 | inet_hashinfo_init(&tcp_hashinfo); |
6e04e021 ACM |
3573 | tcp_hashinfo.bind_bucket_cachep = |
3574 | kmem_cache_create("tcp_bind_bucket", | |
3575 | sizeof(struct inet_bind_bucket), 0, | |
20c2df83 | 3576 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3577 | |
1da177e4 LT |
3578 | /* Size and allocate the main established and bind bucket |
3579 | * hash tables. | |
3580 | * | |
3581 | * The methodology is similar to that of the buffer cache. | |
3582 | */ | |
6e04e021 | 3583 | tcp_hashinfo.ehash = |
1da177e4 | 3584 | alloc_large_system_hash("TCP established", |
0f7ff927 | 3585 | sizeof(struct inet_ehash_bucket), |
1da177e4 | 3586 | thash_entries, |
fd90b29d | 3587 | 17, /* one slot per 128 KB of memory */ |
9e950efa | 3588 | 0, |
1da177e4 | 3589 | NULL, |
f373b53b | 3590 | &tcp_hashinfo.ehash_mask, |
31fe62b9 | 3591 | 0, |
0ccfe618 | 3592 | thash_entries ? 0 : 512 * 1024); |
05dbc7b5 | 3593 | for (i = 0; i <= tcp_hashinfo.ehash_mask; i++) |
3ab5aee7 | 3594 | INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i); |
05dbc7b5 | 3595 | |
230140cf ED |
3596 | if (inet_ehash_locks_alloc(&tcp_hashinfo)) |
3597 | panic("TCP: failed to alloc ehash_locks"); | |
6e04e021 | 3598 | tcp_hashinfo.bhash = |
1da177e4 | 3599 | alloc_large_system_hash("TCP bind", |
0f7ff927 | 3600 | sizeof(struct inet_bind_hashbucket), |
f373b53b | 3601 | tcp_hashinfo.ehash_mask + 1, |
fd90b29d | 3602 | 17, /* one slot per 128 KB of memory */ |
9e950efa | 3603 | 0, |
6e04e021 | 3604 | &tcp_hashinfo.bhash_size, |
1da177e4 | 3605 | NULL, |
31fe62b9 | 3606 | 0, |
1da177e4 | 3607 | 64 * 1024); |
074b8517 | 3608 | tcp_hashinfo.bhash_size = 1U << tcp_hashinfo.bhash_size; |
6e04e021 ACM |
3609 | for (i = 0; i < tcp_hashinfo.bhash_size; i++) { |
3610 | spin_lock_init(&tcp_hashinfo.bhash[i].lock); | |
3611 | INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain); | |
1da177e4 LT |
3612 | } |
3613 | ||
c5ed63d6 ED |
3614 | |
3615 | cnt = tcp_hashinfo.ehash_mask + 1; | |
c5ed63d6 | 3616 | sysctl_tcp_max_orphans = cnt / 2; |
1da177e4 | 3617 | |
a4fe34bf | 3618 | tcp_init_mem(); |
c43b874d | 3619 | /* Set per-socket limits to no more than 1/128 the pressure threshold */ |
5fb84b14 | 3620 | limit = nr_free_buffer_pages() << (PAGE_SHIFT - 7); |
b49960a0 ED |
3621 | max_wshare = min(4UL*1024*1024, limit); |
3622 | max_rshare = min(6UL*1024*1024, limit); | |
7b4f4b5e | 3623 | |
3ab224be | 3624 | sysctl_tcp_wmem[0] = SK_MEM_QUANTUM; |
7b4f4b5e | 3625 | sysctl_tcp_wmem[1] = 16*1024; |
b49960a0 | 3626 | sysctl_tcp_wmem[2] = max(64*1024, max_wshare); |
7b4f4b5e | 3627 | |
3ab224be | 3628 | sysctl_tcp_rmem[0] = SK_MEM_QUANTUM; |
7b4f4b5e | 3629 | sysctl_tcp_rmem[1] = 87380; |
b49960a0 | 3630 | sysctl_tcp_rmem[2] = max(87380, max_rshare); |
1da177e4 | 3631 | |
afd46503 | 3632 | pr_info("Hash tables configured (established %u bind %u)\n", |
058bd4d2 | 3633 | tcp_hashinfo.ehash_mask + 1, tcp_hashinfo.bhash_size); |
317a76f9 | 3634 | |
1946e672 | 3635 | tcp_v4_init(); |
51c5d0c4 | 3636 | tcp_metrics_init(); |
55d8694f | 3637 | BUG_ON(tcp_register_congestion_control(&tcp_reno) != 0); |
46d3ceab | 3638 | tcp_tasklet_init(); |
1da177e4 | 3639 | } |