| 1 |
1275 |
phoenix |
/*
|
| 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 |
|
|
*
|
| 8 |
|
|
* Version: $Id: tcp.c,v 1.1.1.1 2004-04-15 01:13:34 phoenix Exp $
|
| 9 |
|
|
*
|
| 10 |
|
|
* Authors: Ross Biro, <bir7@leland.Stanford.Edu>
|
| 11 |
|
|
* Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
|
| 12 |
|
|
* Mark Evans, <evansmp@uhura.aston.ac.uk>
|
| 13 |
|
|
* Corey Minyard <wf-rch!minyard@relay.EU.net>
|
| 14 |
|
|
* Florian La Roche, <flla@stud.uni-sb.de>
|
| 15 |
|
|
* Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
|
| 16 |
|
|
* Linus Torvalds, <torvalds@cs.helsinki.fi>
|
| 17 |
|
|
* Alan Cox, <gw4pts@gw4pts.ampr.org>
|
| 18 |
|
|
* Matthew Dillon, <dillon@apollo.west.oic.com>
|
| 19 |
|
|
* Arnt Gulbrandsen, <agulbra@nvg.unit.no>
|
| 20 |
|
|
* Jorge Cwik, <jorge@laser.satlink.net>
|
| 21 |
|
|
*
|
| 22 |
|
|
* Fixes:
|
| 23 |
|
|
* Alan Cox : Numerous verify_area() calls
|
| 24 |
|
|
* Alan Cox : Set the ACK bit on a reset
|
| 25 |
|
|
* Alan Cox : Stopped it crashing if it closed while
|
| 26 |
|
|
* sk->inuse=1 and was trying to connect
|
| 27 |
|
|
* (tcp_err()).
|
| 28 |
|
|
* Alan Cox : All icmp error handling was broken
|
| 29 |
|
|
* pointers passed where wrong and the
|
| 30 |
|
|
* socket was looked up backwards. Nobody
|
| 31 |
|
|
* tested any icmp error code obviously.
|
| 32 |
|
|
* Alan Cox : tcp_err() now handled properly. It
|
| 33 |
|
|
* wakes people on errors. poll
|
| 34 |
|
|
* behaves and the icmp error race
|
| 35 |
|
|
* has gone by moving it into sock.c
|
| 36 |
|
|
* Alan Cox : tcp_send_reset() fixed to work for
|
| 37 |
|
|
* everything not just packets for
|
| 38 |
|
|
* unknown sockets.
|
| 39 |
|
|
* Alan Cox : tcp option processing.
|
| 40 |
|
|
* Alan Cox : Reset tweaked (still not 100%) [Had
|
| 41 |
|
|
* syn rule wrong]
|
| 42 |
|
|
* Herp Rosmanith : More reset fixes
|
| 43 |
|
|
* Alan Cox : No longer acks invalid rst frames.
|
| 44 |
|
|
* Acking any kind of RST is right out.
|
| 45 |
|
|
* Alan Cox : Sets an ignore me flag on an rst
|
| 46 |
|
|
* receive otherwise odd bits of prattle
|
| 47 |
|
|
* escape still
|
| 48 |
|
|
* Alan Cox : Fixed another acking RST frame bug.
|
| 49 |
|
|
* Should stop LAN workplace lockups.
|
| 50 |
|
|
* Alan Cox : Some tidyups using the new skb list
|
| 51 |
|
|
* facilities
|
| 52 |
|
|
* Alan Cox : sk->keepopen now seems to work
|
| 53 |
|
|
* Alan Cox : Pulls options out correctly on accepts
|
| 54 |
|
|
* Alan Cox : Fixed assorted sk->rqueue->next errors
|
| 55 |
|
|
* Alan Cox : PSH doesn't end a TCP read. Switched a
|
| 56 |
|
|
* bit to skb ops.
|
| 57 |
|
|
* Alan Cox : Tidied tcp_data to avoid a potential
|
| 58 |
|
|
* nasty.
|
| 59 |
|
|
* Alan Cox : Added some better commenting, as the
|
| 60 |
|
|
* tcp is hard to follow
|
| 61 |
|
|
* Alan Cox : Removed incorrect check for 20 * psh
|
| 62 |
|
|
* Michael O'Reilly : ack < copied bug fix.
|
| 63 |
|
|
* Johannes Stille : Misc tcp fixes (not all in yet).
|
| 64 |
|
|
* Alan Cox : FIN with no memory -> CRASH
|
| 65 |
|
|
* Alan Cox : Added socket option proto entries.
|
| 66 |
|
|
* Also added awareness of them to accept.
|
| 67 |
|
|
* Alan Cox : Added TCP options (SOL_TCP)
|
| 68 |
|
|
* Alan Cox : Switched wakeup calls to callbacks,
|
| 69 |
|
|
* so the kernel can layer network
|
| 70 |
|
|
* sockets.
|
| 71 |
|
|
* Alan Cox : Use ip_tos/ip_ttl settings.
|
| 72 |
|
|
* Alan Cox : Handle FIN (more) properly (we hope).
|
| 73 |
|
|
* Alan Cox : RST frames sent on unsynchronised
|
| 74 |
|
|
* state ack error.
|
| 75 |
|
|
* Alan Cox : Put in missing check for SYN bit.
|
| 76 |
|
|
* Alan Cox : Added tcp_select_window() aka NET2E
|
| 77 |
|
|
* window non shrink trick.
|
| 78 |
|
|
* Alan Cox : Added a couple of small NET2E timer
|
| 79 |
|
|
* fixes
|
| 80 |
|
|
* Charles Hedrick : TCP fixes
|
| 81 |
|
|
* Toomas Tamm : TCP window fixes
|
| 82 |
|
|
* Alan Cox : Small URG fix to rlogin ^C ack fight
|
| 83 |
|
|
* Charles Hedrick : Rewrote most of it to actually work
|
| 84 |
|
|
* Linus : Rewrote tcp_read() and URG handling
|
| 85 |
|
|
* completely
|
| 86 |
|
|
* Gerhard Koerting: Fixed some missing timer handling
|
| 87 |
|
|
* Matthew Dillon : Reworked TCP machine states as per RFC
|
| 88 |
|
|
* Gerhard Koerting: PC/TCP workarounds
|
| 89 |
|
|
* Adam Caldwell : Assorted timer/timing errors
|
| 90 |
|
|
* Matthew Dillon : Fixed another RST bug
|
| 91 |
|
|
* Alan Cox : Move to kernel side addressing changes.
|
| 92 |
|
|
* Alan Cox : Beginning work on TCP fastpathing
|
| 93 |
|
|
* (not yet usable)
|
| 94 |
|
|
* Arnt Gulbrandsen: Turbocharged tcp_check() routine.
|
| 95 |
|
|
* Alan Cox : TCP fast path debugging
|
| 96 |
|
|
* Alan Cox : Window clamping
|
| 97 |
|
|
* Michael Riepe : Bug in tcp_check()
|
| 98 |
|
|
* Matt Dillon : More TCP improvements and RST bug fixes
|
| 99 |
|
|
* Matt Dillon : Yet more small nasties remove from the
|
| 100 |
|
|
* TCP code (Be very nice to this man if
|
| 101 |
|
|
* tcp finally works 100%) 8)
|
| 102 |
|
|
* Alan Cox : BSD accept semantics.
|
| 103 |
|
|
* Alan Cox : Reset on closedown bug.
|
| 104 |
|
|
* Peter De Schrijver : ENOTCONN check missing in tcp_sendto().
|
| 105 |
|
|
* Michael Pall : Handle poll() after URG properly in
|
| 106 |
|
|
* all cases.
|
| 107 |
|
|
* Michael Pall : Undo the last fix in tcp_read_urg()
|
| 108 |
|
|
* (multi URG PUSH broke rlogin).
|
| 109 |
|
|
* Michael Pall : Fix the multi URG PUSH problem in
|
| 110 |
|
|
* tcp_readable(), poll() after URG
|
| 111 |
|
|
* works now.
|
| 112 |
|
|
* Michael Pall : recv(...,MSG_OOB) never blocks in the
|
| 113 |
|
|
* BSD api.
|
| 114 |
|
|
* Alan Cox : Changed the semantics of sk->socket to
|
| 115 |
|
|
* fix a race and a signal problem with
|
| 116 |
|
|
* accept() and async I/O.
|
| 117 |
|
|
* Alan Cox : Relaxed the rules on tcp_sendto().
|
| 118 |
|
|
* Yury Shevchuk : Really fixed accept() blocking problem.
|
| 119 |
|
|
* Craig I. Hagan : Allow for BSD compatible TIME_WAIT for
|
| 120 |
|
|
* clients/servers which listen in on
|
| 121 |
|
|
* fixed ports.
|
| 122 |
|
|
* Alan Cox : Cleaned the above up and shrank it to
|
| 123 |
|
|
* a sensible code size.
|
| 124 |
|
|
* Alan Cox : Self connect lockup fix.
|
| 125 |
|
|
* Alan Cox : No connect to multicast.
|
| 126 |
|
|
* Ross Biro : Close unaccepted children on master
|
| 127 |
|
|
* socket close.
|
| 128 |
|
|
* Alan Cox : Reset tracing code.
|
| 129 |
|
|
* Alan Cox : Spurious resets on shutdown.
|
| 130 |
|
|
* Alan Cox : Giant 15 minute/60 second timer error
|
| 131 |
|
|
* Alan Cox : Small whoops in polling before an
|
| 132 |
|
|
* accept.
|
| 133 |
|
|
* Alan Cox : Kept the state trace facility since
|
| 134 |
|
|
* it's handy for debugging.
|
| 135 |
|
|
* Alan Cox : More reset handler fixes.
|
| 136 |
|
|
* Alan Cox : Started rewriting the code based on
|
| 137 |
|
|
* the RFC's for other useful protocol
|
| 138 |
|
|
* references see: Comer, KA9Q NOS, and
|
| 139 |
|
|
* for a reference on the difference
|
| 140 |
|
|
* between specifications and how BSD
|
| 141 |
|
|
* works see the 4.4lite source.
|
| 142 |
|
|
* A.N.Kuznetsov : Don't time wait on completion of tidy
|
| 143 |
|
|
* close.
|
| 144 |
|
|
* Linus Torvalds : Fin/Shutdown & copied_seq changes.
|
| 145 |
|
|
* Linus Torvalds : Fixed BSD port reuse to work first syn
|
| 146 |
|
|
* Alan Cox : Reimplemented timers as per the RFC
|
| 147 |
|
|
* and using multiple timers for sanity.
|
| 148 |
|
|
* Alan Cox : Small bug fixes, and a lot of new
|
| 149 |
|
|
* comments.
|
| 150 |
|
|
* Alan Cox : Fixed dual reader crash by locking
|
| 151 |
|
|
* the buffers (much like datagram.c)
|
| 152 |
|
|
* Alan Cox : Fixed stuck sockets in probe. A probe
|
| 153 |
|
|
* now gets fed up of retrying without
|
| 154 |
|
|
* (even a no space) answer.
|
| 155 |
|
|
* Alan Cox : Extracted closing code better
|
| 156 |
|
|
* Alan Cox : Fixed the closing state machine to
|
| 157 |
|
|
* resemble the RFC.
|
| 158 |
|
|
* Alan Cox : More 'per spec' fixes.
|
| 159 |
|
|
* Jorge Cwik : Even faster checksumming.
|
| 160 |
|
|
* Alan Cox : tcp_data() doesn't ack illegal PSH
|
| 161 |
|
|
* only frames. At least one pc tcp stack
|
| 162 |
|
|
* generates them.
|
| 163 |
|
|
* Alan Cox : Cache last socket.
|
| 164 |
|
|
* Alan Cox : Per route irtt.
|
| 165 |
|
|
* Matt Day : poll()->select() match BSD precisely on error
|
| 166 |
|
|
* Alan Cox : New buffers
|
| 167 |
|
|
* Marc Tamsky : Various sk->prot->retransmits and
|
| 168 |
|
|
* sk->retransmits misupdating fixed.
|
| 169 |
|
|
* Fixed tcp_write_timeout: stuck close,
|
| 170 |
|
|
* and TCP syn retries gets used now.
|
| 171 |
|
|
* Mark Yarvis : In tcp_read_wakeup(), don't send an
|
| 172 |
|
|
* ack if state is TCP_CLOSED.
|
| 173 |
|
|
* Alan Cox : Look up device on a retransmit - routes may
|
| 174 |
|
|
* change. Doesn't yet cope with MSS shrink right
|
| 175 |
|
|
* but its a start!
|
| 176 |
|
|
* Marc Tamsky : Closing in closing fixes.
|
| 177 |
|
|
* Mike Shaver : RFC1122 verifications.
|
| 178 |
|
|
* Alan Cox : rcv_saddr errors.
|
| 179 |
|
|
* Alan Cox : Block double connect().
|
| 180 |
|
|
* Alan Cox : Small hooks for enSKIP.
|
| 181 |
|
|
* Alexey Kuznetsov: Path MTU discovery.
|
| 182 |
|
|
* Alan Cox : Support soft errors.
|
| 183 |
|
|
* Alan Cox : Fix MTU discovery pathological case
|
| 184 |
|
|
* when the remote claims no mtu!
|
| 185 |
|
|
* Marc Tamsky : TCP_CLOSE fix.
|
| 186 |
|
|
* Colin (G3TNE) : Send a reset on syn ack replies in
|
| 187 |
|
|
* window but wrong (fixes NT lpd problems)
|
| 188 |
|
|
* Pedro Roque : Better TCP window handling, delayed ack.
|
| 189 |
|
|
* Joerg Reuter : No modification of locked buffers in
|
| 190 |
|
|
* tcp_do_retransmit()
|
| 191 |
|
|
* Eric Schenk : Changed receiver side silly window
|
| 192 |
|
|
* avoidance algorithm to BSD style
|
| 193 |
|
|
* algorithm. This doubles throughput
|
| 194 |
|
|
* against machines running Solaris,
|
| 195 |
|
|
* and seems to result in general
|
| 196 |
|
|
* improvement.
|
| 197 |
|
|
* Stefan Magdalinski : adjusted tcp_readable() to fix FIONREAD
|
| 198 |
|
|
* Willy Konynenberg : Transparent proxying support.
|
| 199 |
|
|
* Mike McLagan : Routing by source
|
| 200 |
|
|
* Keith Owens : Do proper merging with partial SKB's in
|
| 201 |
|
|
* tcp_do_sendmsg to avoid burstiness.
|
| 202 |
|
|
* Eric Schenk : Fix fast close down bug with
|
| 203 |
|
|
* shutdown() followed by close().
|
| 204 |
|
|
* Andi Kleen : Make poll agree with SIGIO
|
| 205 |
|
|
* Salvatore Sanfilippo : Support SO_LINGER with linger == 1 and
|
| 206 |
|
|
* lingertime == 0 (RFC 793 ABORT Call)
|
| 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 |
|
|
|
| 248 |
|
|
#include <linux/config.h>
|
| 249 |
|
|
#include <linux/types.h>
|
| 250 |
|
|
#include <linux/fcntl.h>
|
| 251 |
|
|
#include <linux/poll.h>
|
| 252 |
|
|
#include <linux/init.h>
|
| 253 |
|
|
#include <linux/smp_lock.h>
|
| 254 |
|
|
#include <linux/fs.h>
|
| 255 |
|
|
#include <linux/random.h>
|
| 256 |
|
|
|
| 257 |
|
|
#include <net/icmp.h>
|
| 258 |
|
|
#include <net/tcp.h>
|
| 259 |
|
|
|
| 260 |
|
|
#include <asm/uaccess.h>
|
| 261 |
|
|
#include <asm/ioctls.h>
|
| 262 |
|
|
|
| 263 |
|
|
int sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT;
|
| 264 |
|
|
|
| 265 |
|
|
struct tcp_mib tcp_statistics[NR_CPUS*2];
|
| 266 |
|
|
|
| 267 |
|
|
kmem_cache_t *tcp_openreq_cachep;
|
| 268 |
|
|
kmem_cache_t *tcp_bucket_cachep;
|
| 269 |
|
|
kmem_cache_t *tcp_timewait_cachep;
|
| 270 |
|
|
|
| 271 |
|
|
atomic_t tcp_orphan_count = ATOMIC_INIT(0);
|
| 272 |
|
|
|
| 273 |
|
|
int sysctl_tcp_mem[3];
|
| 274 |
|
|
int sysctl_tcp_wmem[3] = { 4*1024, 16*1024, 128*1024 };
|
| 275 |
|
|
int sysctl_tcp_rmem[3] = { 4*1024, 87380, 87380*2 };
|
| 276 |
|
|
|
| 277 |
|
|
atomic_t tcp_memory_allocated; /* Current allocated memory. */
|
| 278 |
|
|
atomic_t tcp_sockets_allocated; /* Current number of TCP sockets. */
|
| 279 |
|
|
|
| 280 |
|
|
/* Pressure flag: try to collapse.
|
| 281 |
|
|
* Technical note: it is used by multiple contexts non atomically.
|
| 282 |
|
|
* All the tcp_mem_schedule() is of this nature: accounting
|
| 283 |
|
|
* is strict, actions are advisory and have some latency. */
|
| 284 |
|
|
int tcp_memory_pressure;
|
| 285 |
|
|
|
| 286 |
|
|
#define TCP_PAGES(amt) (((amt)+TCP_MEM_QUANTUM-1)/TCP_MEM_QUANTUM)
|
| 287 |
|
|
|
| 288 |
|
|
int tcp_mem_schedule(struct sock *sk, int size, int kind)
|
| 289 |
|
|
{
|
| 290 |
|
|
int amt = TCP_PAGES(size);
|
| 291 |
|
|
|
| 292 |
|
|
sk->forward_alloc += amt*TCP_MEM_QUANTUM;
|
| 293 |
|
|
atomic_add(amt, &tcp_memory_allocated);
|
| 294 |
|
|
|
| 295 |
|
|
/* Under limit. */
|
| 296 |
|
|
if (atomic_read(&tcp_memory_allocated) < sysctl_tcp_mem[0]) {
|
| 297 |
|
|
if (tcp_memory_pressure)
|
| 298 |
|
|
tcp_memory_pressure = 0;
|
| 299 |
|
|
return 1;
|
| 300 |
|
|
}
|
| 301 |
|
|
|
| 302 |
|
|
/* Over hard limit. */
|
| 303 |
|
|
if (atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]) {
|
| 304 |
|
|
tcp_enter_memory_pressure();
|
| 305 |
|
|
goto suppress_allocation;
|
| 306 |
|
|
}
|
| 307 |
|
|
|
| 308 |
|
|
/* Under pressure. */
|
| 309 |
|
|
if (atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[1])
|
| 310 |
|
|
tcp_enter_memory_pressure();
|
| 311 |
|
|
|
| 312 |
|
|
if (kind) {
|
| 313 |
|
|
if (atomic_read(&sk->rmem_alloc) < sysctl_tcp_rmem[0])
|
| 314 |
|
|
return 1;
|
| 315 |
|
|
} else {
|
| 316 |
|
|
if (sk->wmem_queued < sysctl_tcp_wmem[0])
|
| 317 |
|
|
return 1;
|
| 318 |
|
|
}
|
| 319 |
|
|
|
| 320 |
|
|
if (!tcp_memory_pressure ||
|
| 321 |
|
|
sysctl_tcp_mem[2] > atomic_read(&tcp_sockets_allocated)
|
| 322 |
|
|
* TCP_PAGES(sk->wmem_queued+atomic_read(&sk->rmem_alloc)+
|
| 323 |
|
|
sk->forward_alloc))
|
| 324 |
|
|
return 1;
|
| 325 |
|
|
|
| 326 |
|
|
suppress_allocation:
|
| 327 |
|
|
|
| 328 |
|
|
if (kind == 0) {
|
| 329 |
|
|
tcp_moderate_sndbuf(sk);
|
| 330 |
|
|
|
| 331 |
|
|
/* Fail only if socket is _under_ its sndbuf.
|
| 332 |
|
|
* In this case we cannot block, so that we have to fail.
|
| 333 |
|
|
*/
|
| 334 |
|
|
if (sk->wmem_queued+size >= sk->sndbuf)
|
| 335 |
|
|
return 1;
|
| 336 |
|
|
}
|
| 337 |
|
|
|
| 338 |
|
|
/* Alas. Undo changes. */
|
| 339 |
|
|
sk->forward_alloc -= amt*TCP_MEM_QUANTUM;
|
| 340 |
|
|
atomic_sub(amt, &tcp_memory_allocated);
|
| 341 |
|
|
return 0;
|
| 342 |
|
|
}
|
| 343 |
|
|
|
| 344 |
|
|
void __tcp_mem_reclaim(struct sock *sk)
|
| 345 |
|
|
{
|
| 346 |
|
|
if (sk->forward_alloc >= TCP_MEM_QUANTUM) {
|
| 347 |
|
|
atomic_sub(sk->forward_alloc/TCP_MEM_QUANTUM, &tcp_memory_allocated);
|
| 348 |
|
|
sk->forward_alloc &= (TCP_MEM_QUANTUM-1);
|
| 349 |
|
|
if (tcp_memory_pressure &&
|
| 350 |
|
|
atomic_read(&tcp_memory_allocated) < sysctl_tcp_mem[0])
|
| 351 |
|
|
tcp_memory_pressure = 0;
|
| 352 |
|
|
}
|
| 353 |
|
|
}
|
| 354 |
|
|
|
| 355 |
|
|
void tcp_rfree(struct sk_buff *skb)
|
| 356 |
|
|
{
|
| 357 |
|
|
struct sock *sk = skb->sk;
|
| 358 |
|
|
|
| 359 |
|
|
atomic_sub(skb->truesize, &sk->rmem_alloc);
|
| 360 |
|
|
sk->forward_alloc += skb->truesize;
|
| 361 |
|
|
}
|
| 362 |
|
|
|
| 363 |
|
|
/*
|
| 364 |
|
|
* LISTEN is a special case for poll..
|
| 365 |
|
|
*/
|
| 366 |
|
|
static __inline__ unsigned int tcp_listen_poll(struct sock *sk, poll_table *wait)
|
| 367 |
|
|
{
|
| 368 |
|
|
return sk->tp_pinfo.af_tcp.accept_queue ? (POLLIN | POLLRDNORM) : 0;
|
| 369 |
|
|
}
|
| 370 |
|
|
|
| 371 |
|
|
/*
|
| 372 |
|
|
* Wait for a TCP event.
|
| 373 |
|
|
*
|
| 374 |
|
|
* Note that we don't need to lock the socket, as the upper poll layers
|
| 375 |
|
|
* take care of normal races (between the test and the event) and we don't
|
| 376 |
|
|
* go look at any of the socket buffers directly.
|
| 377 |
|
|
*/
|
| 378 |
|
|
unsigned int tcp_poll(struct file * file, struct socket *sock, poll_table *wait)
|
| 379 |
|
|
{
|
| 380 |
|
|
unsigned int mask;
|
| 381 |
|
|
struct sock *sk = sock->sk;
|
| 382 |
|
|
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
|
| 383 |
|
|
|
| 384 |
|
|
poll_wait(file, sk->sleep, wait);
|
| 385 |
|
|
if (sk->state == TCP_LISTEN)
|
| 386 |
|
|
return tcp_listen_poll(sk, wait);
|
| 387 |
|
|
|
| 388 |
|
|
/* Socket is not locked. We are protected from async events
|
| 389 |
|
|
by poll logic and correct handling of state changes
|
| 390 |
|
|
made by another threads is impossible in any case.
|
| 391 |
|
|
*/
|
| 392 |
|
|
|
| 393 |
|
|
mask = 0;
|
| 394 |
|
|
if (sk->err)
|
| 395 |
|
|
mask = POLLERR;
|
| 396 |
|
|
|
| 397 |
|
|
/*
|
| 398 |
|
|
* POLLHUP is certainly not done right. But poll() doesn't
|
| 399 |
|
|
* have a notion of HUP in just one direction, and for a
|
| 400 |
|
|
* socket the read side is more interesting.
|
| 401 |
|
|
*
|
| 402 |
|
|
* Some poll() documentation says that POLLHUP is incompatible
|
| 403 |
|
|
* with the POLLOUT/POLLWR flags, so somebody should check this
|
| 404 |
|
|
* all. But careful, it tends to be safer to return too many
|
| 405 |
|
|
* bits than too few, and you can easily break real applications
|
| 406 |
|
|
* if you don't tell them that something has hung up!
|
| 407 |
|
|
*
|
| 408 |
|
|
* Check-me.
|
| 409 |
|
|
*
|
| 410 |
|
|
* Check number 1. POLLHUP is _UNMASKABLE_ event (see UNIX98 and
|
| 411 |
|
|
* our fs/select.c). It means that after we received EOF,
|
| 412 |
|
|
* poll always returns immediately, making impossible poll() on write()
|
| 413 |
|
|
* in state CLOSE_WAIT. One solution is evident --- to set POLLHUP
|
| 414 |
|
|
* if and only if shutdown has been made in both directions.
|
| 415 |
|
|
* Actually, it is interesting to look how Solaris and DUX
|
| 416 |
|
|
* solve this dilemma. I would prefer, if PULLHUP were maskable,
|
| 417 |
|
|
* then we could set it on SND_SHUTDOWN. BTW examples given
|
| 418 |
|
|
* in Stevens' books assume exactly this behaviour, it explains
|
| 419 |
|
|
* why PULLHUP is incompatible with POLLOUT. --ANK
|
| 420 |
|
|
*
|
| 421 |
|
|
* NOTE. Check for TCP_CLOSE is added. The goal is to prevent
|
| 422 |
|
|
* blocking on fresh not-connected or disconnected socket. --ANK
|
| 423 |
|
|
*/
|
| 424 |
|
|
if (sk->shutdown == SHUTDOWN_MASK || sk->state == TCP_CLOSE)
|
| 425 |
|
|
mask |= POLLHUP;
|
| 426 |
|
|
if (sk->shutdown & RCV_SHUTDOWN)
|
| 427 |
|
|
mask |= POLLIN | POLLRDNORM;
|
| 428 |
|
|
|
| 429 |
|
|
/* Connected? */
|
| 430 |
|
|
if ((1 << sk->state) & ~(TCPF_SYN_SENT|TCPF_SYN_RECV)) {
|
| 431 |
|
|
/* Potential race condition. If read of tp below will
|
| 432 |
|
|
* escape above sk->state, we can be illegally awaken
|
| 433 |
|
|
* in SYN_* states. */
|
| 434 |
|
|
if ((tp->rcv_nxt != tp->copied_seq) &&
|
| 435 |
|
|
(tp->urg_seq != tp->copied_seq ||
|
| 436 |
|
|
tp->rcv_nxt != tp->copied_seq+1 ||
|
| 437 |
|
|
sk->urginline || !tp->urg_data))
|
| 438 |
|
|
mask |= POLLIN | POLLRDNORM;
|
| 439 |
|
|
|
| 440 |
|
|
if (!(sk->shutdown & SEND_SHUTDOWN)) {
|
| 441 |
|
|
if (tcp_wspace(sk) >= tcp_min_write_space(sk)) {
|
| 442 |
|
|
mask |= POLLOUT | POLLWRNORM;
|
| 443 |
|
|
} else { /* send SIGIO later */
|
| 444 |
|
|
set_bit(SOCK_ASYNC_NOSPACE, &sk->socket->flags);
|
| 445 |
|
|
set_bit(SOCK_NOSPACE, &sk->socket->flags);
|
| 446 |
|
|
|
| 447 |
|
|
/* Race breaker. If space is freed after
|
| 448 |
|
|
* wspace test but before the flags are set,
|
| 449 |
|
|
* IO signal will be lost.
|
| 450 |
|
|
*/
|
| 451 |
|
|
if (tcp_wspace(sk) >= tcp_min_write_space(sk))
|
| 452 |
|
|
mask |= POLLOUT | POLLWRNORM;
|
| 453 |
|
|
}
|
| 454 |
|
|
}
|
| 455 |
|
|
|
| 456 |
|
|
if (tp->urg_data & TCP_URG_VALID)
|
| 457 |
|
|
mask |= POLLPRI;
|
| 458 |
|
|
}
|
| 459 |
|
|
return mask;
|
| 460 |
|
|
}
|
| 461 |
|
|
|
| 462 |
|
|
/*
|
| 463 |
|
|
* TCP socket write_space callback.
|
| 464 |
|
|
*/
|
| 465 |
|
|
void tcp_write_space(struct sock *sk)
|
| 466 |
|
|
{
|
| 467 |
|
|
struct socket *sock = sk->socket;
|
| 468 |
|
|
|
| 469 |
|
|
if (tcp_wspace(sk) >= tcp_min_write_space(sk) && sock) {
|
| 470 |
|
|
clear_bit(SOCK_NOSPACE, &sock->flags);
|
| 471 |
|
|
|
| 472 |
|
|
if (sk->sleep && waitqueue_active(sk->sleep))
|
| 473 |
|
|
wake_up_interruptible(sk->sleep);
|
| 474 |
|
|
|
| 475 |
|
|
if (sock->fasync_list && !(sk->shutdown&SEND_SHUTDOWN))
|
| 476 |
|
|
sock_wake_async(sock, 2, POLL_OUT);
|
| 477 |
|
|
}
|
| 478 |
|
|
}
|
| 479 |
|
|
|
| 480 |
|
|
int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
|
| 481 |
|
|
{
|
| 482 |
|
|
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
|
| 483 |
|
|
int answ;
|
| 484 |
|
|
|
| 485 |
|
|
switch(cmd) {
|
| 486 |
|
|
case SIOCINQ:
|
| 487 |
|
|
if (sk->state == TCP_LISTEN)
|
| 488 |
|
|
return(-EINVAL);
|
| 489 |
|
|
|
| 490 |
|
|
lock_sock(sk);
|
| 491 |
|
|
if ((1<<sk->state) & (TCPF_SYN_SENT|TCPF_SYN_RECV))
|
| 492 |
|
|
answ = 0;
|
| 493 |
|
|
else if (sk->urginline || !tp->urg_data ||
|
| 494 |
|
|
before(tp->urg_seq,tp->copied_seq) ||
|
| 495 |
|
|
!before(tp->urg_seq,tp->rcv_nxt)) {
|
| 496 |
|
|
answ = tp->rcv_nxt - tp->copied_seq;
|
| 497 |
|
|
|
| 498 |
|
|
/* Subtract 1, if FIN is in queue. */
|
| 499 |
|
|
if (answ && !skb_queue_empty(&sk->receive_queue))
|
| 500 |
|
|
answ -= ((struct sk_buff*)sk->receive_queue.prev)->h.th->fin;
|
| 501 |
|
|
} else
|
| 502 |
|
|
answ = tp->urg_seq - tp->copied_seq;
|
| 503 |
|
|
release_sock(sk);
|
| 504 |
|
|
break;
|
| 505 |
|
|
case SIOCATMARK:
|
| 506 |
|
|
{
|
| 507 |
|
|
answ = tp->urg_data && tp->urg_seq == tp->copied_seq;
|
| 508 |
|
|
break;
|
| 509 |
|
|
}
|
| 510 |
|
|
case SIOCOUTQ:
|
| 511 |
|
|
if (sk->state == TCP_LISTEN)
|
| 512 |
|
|
return(-EINVAL);
|
| 513 |
|
|
|
| 514 |
|
|
if ((1<<sk->state) & (TCPF_SYN_SENT|TCPF_SYN_RECV))
|
| 515 |
|
|
answ = 0;
|
| 516 |
|
|
else
|
| 517 |
|
|
answ = tp->write_seq - tp->snd_una;
|
| 518 |
|
|
break;
|
| 519 |
|
|
default:
|
| 520 |
|
|
return(-ENOIOCTLCMD);
|
| 521 |
|
|
};
|
| 522 |
|
|
|
| 523 |
|
|
return put_user(answ, (int *)arg);
|
| 524 |
|
|
}
|
| 525 |
|
|
|
| 526 |
|
|
|
| 527 |
|
|
int tcp_listen_start(struct sock *sk)
|
| 528 |
|
|
{
|
| 529 |
|
|
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
|
| 530 |
|
|
struct tcp_listen_opt *lopt;
|
| 531 |
|
|
|
| 532 |
|
|
sk->max_ack_backlog = 0;
|
| 533 |
|
|
sk->ack_backlog = 0;
|
| 534 |
|
|
tp->accept_queue = tp->accept_queue_tail = NULL;
|
| 535 |
|
|
tp->syn_wait_lock = RW_LOCK_UNLOCKED;
|
| 536 |
|
|
tcp_delack_init(tp);
|
| 537 |
|
|
|
| 538 |
|
|
lopt = kmalloc(sizeof(struct tcp_listen_opt), GFP_KERNEL);
|
| 539 |
|
|
if (!lopt)
|
| 540 |
|
|
return -ENOMEM;
|
| 541 |
|
|
|
| 542 |
|
|
memset(lopt, 0, sizeof(struct tcp_listen_opt));
|
| 543 |
|
|
for (lopt->max_qlen_log = 6; ; lopt->max_qlen_log++)
|
| 544 |
|
|
if ((1<<lopt->max_qlen_log) >= sysctl_max_syn_backlog)
|
| 545 |
|
|
break;
|
| 546 |
|
|
get_random_bytes(&lopt->hash_rnd, 4);
|
| 547 |
|
|
|
| 548 |
|
|
write_lock_bh(&tp->syn_wait_lock);
|
| 549 |
|
|
tp->listen_opt = lopt;
|
| 550 |
|
|
write_unlock_bh(&tp->syn_wait_lock);
|
| 551 |
|
|
|
| 552 |
|
|
/* There is race window here: we announce ourselves listening,
|
| 553 |
|
|
* but this transition is still not validated by get_port().
|
| 554 |
|
|
* It is OK, because this socket enters to hash table only
|
| 555 |
|
|
* after validation is complete.
|
| 556 |
|
|
*/
|
| 557 |
|
|
sk->state = TCP_LISTEN;
|
| 558 |
|
|
if (sk->prot->get_port(sk, sk->num) == 0) {
|
| 559 |
|
|
sk->sport = htons(sk->num);
|
| 560 |
|
|
|
| 561 |
|
|
sk_dst_reset(sk);
|
| 562 |
|
|
sk->prot->hash(sk);
|
| 563 |
|
|
|
| 564 |
|
|
return 0;
|
| 565 |
|
|
}
|
| 566 |
|
|
|
| 567 |
|
|
sk->state = TCP_CLOSE;
|
| 568 |
|
|
write_lock_bh(&tp->syn_wait_lock);
|
| 569 |
|
|
tp->listen_opt = NULL;
|
| 570 |
|
|
write_unlock_bh(&tp->syn_wait_lock);
|
| 571 |
|
|
kfree(lopt);
|
| 572 |
|
|
return -EADDRINUSE;
|
| 573 |
|
|
}
|
| 574 |
|
|
|
| 575 |
|
|
/*
|
| 576 |
|
|
* This routine closes sockets which have been at least partially
|
| 577 |
|
|
* opened, but not yet accepted.
|
| 578 |
|
|
*/
|
| 579 |
|
|
|
| 580 |
|
|
static void tcp_listen_stop (struct sock *sk)
|
| 581 |
|
|
{
|
| 582 |
|
|
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
|
| 583 |
|
|
struct tcp_listen_opt *lopt = tp->listen_opt;
|
| 584 |
|
|
struct open_request *acc_req = tp->accept_queue;
|
| 585 |
|
|
struct open_request *req;
|
| 586 |
|
|
int i;
|
| 587 |
|
|
|
| 588 |
|
|
tcp_delete_keepalive_timer(sk);
|
| 589 |
|
|
|
| 590 |
|
|
/* make all the listen_opt local to us */
|
| 591 |
|
|
write_lock_bh(&tp->syn_wait_lock);
|
| 592 |
|
|
tp->listen_opt =NULL;
|
| 593 |
|
|
write_unlock_bh(&tp->syn_wait_lock);
|
| 594 |
|
|
tp->accept_queue = tp->accept_queue_tail = NULL;
|
| 595 |
|
|
|
| 596 |
|
|
if (lopt->qlen) {
|
| 597 |
|
|
for (i=0; i<TCP_SYNQ_HSIZE; i++) {
|
| 598 |
|
|
while ((req = lopt->syn_table[i]) != NULL) {
|
| 599 |
|
|
lopt->syn_table[i] = req->dl_next;
|
| 600 |
|
|
lopt->qlen--;
|
| 601 |
|
|
tcp_openreq_free(req);
|
| 602 |
|
|
|
| 603 |
|
|
/* Following specs, it would be better either to send FIN
|
| 604 |
|
|
* (and enter FIN-WAIT-1, it is normal close)
|
| 605 |
|
|
* or to send active reset (abort).
|
| 606 |
|
|
* Certainly, it is pretty dangerous while synflood, but it is
|
| 607 |
|
|
* bad justification for our negligence 8)
|
| 608 |
|
|
* To be honest, we are not able to make either
|
| 609 |
|
|
* of the variants now. --ANK
|
| 610 |
|
|
*/
|
| 611 |
|
|
}
|
| 612 |
|
|
}
|
| 613 |
|
|
}
|
| 614 |
|
|
BUG_TRAP(lopt->qlen == 0);
|
| 615 |
|
|
|
| 616 |
|
|
kfree(lopt);
|
| 617 |
|
|
|
| 618 |
|
|
while ((req=acc_req) != NULL) {
|
| 619 |
|
|
struct sock *child = req->sk;
|
| 620 |
|
|
|
| 621 |
|
|
acc_req = req->dl_next;
|
| 622 |
|
|
|
| 623 |
|
|
local_bh_disable();
|
| 624 |
|
|
bh_lock_sock(child);
|
| 625 |
|
|
BUG_TRAP(child->lock.users==0);
|
| 626 |
|
|
sock_hold(child);
|
| 627 |
|
|
|
| 628 |
|
|
tcp_disconnect(child, O_NONBLOCK);
|
| 629 |
|
|
|
| 630 |
|
|
sock_orphan(child);
|
| 631 |
|
|
|
| 632 |
|
|
atomic_inc(&tcp_orphan_count);
|
| 633 |
|
|
|
| 634 |
|
|
tcp_destroy_sock(child);
|
| 635 |
|
|
|
| 636 |
|
|
bh_unlock_sock(child);
|
| 637 |
|
|
local_bh_enable();
|
| 638 |
|
|
sock_put(child);
|
| 639 |
|
|
|
| 640 |
|
|
tcp_acceptq_removed(sk);
|
| 641 |
|
|
tcp_openreq_fastfree(req);
|
| 642 |
|
|
}
|
| 643 |
|
|
BUG_TRAP(sk->ack_backlog == 0);
|
| 644 |
|
|
}
|
| 645 |
|
|
|
| 646 |
|
|
/*
|
| 647 |
|
|
* Wait for a socket to get into the connected state
|
| 648 |
|
|
*
|
| 649 |
|
|
* Note: Must be called with the socket locked.
|
| 650 |
|
|
*/
|
| 651 |
|
|
static int wait_for_tcp_connect(struct sock * sk, int flags, long *timeo_p)
|
| 652 |
|
|
{
|
| 653 |
|
|
struct task_struct *tsk = current;
|
| 654 |
|
|
DECLARE_WAITQUEUE(wait, tsk);
|
| 655 |
|
|
|
| 656 |
|
|
while((1 << sk->state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
|
| 657 |
|
|
if(sk->err)
|
| 658 |
|
|
return sock_error(sk);
|
| 659 |
|
|
if((1 << sk->state) &
|
| 660 |
|
|
~(TCPF_SYN_SENT | TCPF_SYN_RECV))
|
| 661 |
|
|
return -EPIPE;
|
| 662 |
|
|
if(!*timeo_p)
|
| 663 |
|
|
return -EAGAIN;
|
| 664 |
|
|
if(signal_pending(tsk))
|
| 665 |
|
|
return sock_intr_errno(*timeo_p);
|
| 666 |
|
|
|
| 667 |
|
|
__set_task_state(tsk, TASK_INTERRUPTIBLE);
|
| 668 |
|
|
add_wait_queue(sk->sleep, &wait);
|
| 669 |
|
|
sk->tp_pinfo.af_tcp.write_pending++;
|
| 670 |
|
|
|
| 671 |
|
|
release_sock(sk);
|
| 672 |
|
|
*timeo_p = schedule_timeout(*timeo_p);
|
| 673 |
|
|
lock_sock(sk);
|
| 674 |
|
|
|
| 675 |
|
|
__set_task_state(tsk, TASK_RUNNING);
|
| 676 |
|
|
remove_wait_queue(sk->sleep, &wait);
|
| 677 |
|
|
sk->tp_pinfo.af_tcp.write_pending--;
|
| 678 |
|
|
}
|
| 679 |
|
|
return 0;
|
| 680 |
|
|
}
|
| 681 |
|
|
|
| 682 |
|
|
static inline int tcp_memory_free(struct sock *sk)
|
| 683 |
|
|
{
|
| 684 |
|
|
return sk->wmem_queued < sk->sndbuf;
|
| 685 |
|
|
}
|
| 686 |
|
|
|
| 687 |
|
|
/*
|
| 688 |
|
|
* Wait for more memory for a socket
|
| 689 |
|
|
*/
|
| 690 |
|
|
static int wait_for_tcp_memory(struct sock * sk, long *timeo)
|
| 691 |
|
|
{
|
| 692 |
|
|
int err = 0;
|
| 693 |
|
|
long vm_wait = 0;
|
| 694 |
|
|
long current_timeo = *timeo;
|
| 695 |
|
|
DECLARE_WAITQUEUE(wait, current);
|
| 696 |
|
|
|
| 697 |
|
|
if (tcp_memory_free(sk))
|
| 698 |
|
|
current_timeo = vm_wait = (net_random()%(HZ/5))+2;
|
| 699 |
|
|
|
| 700 |
|
|
add_wait_queue(sk->sleep, &wait);
|
| 701 |
|
|
for (;;) {
|
| 702 |
|
|
set_bit(SOCK_ASYNC_NOSPACE, &sk->socket->flags);
|
| 703 |
|
|
|
| 704 |
|
|
set_current_state(TASK_INTERRUPTIBLE);
|
| 705 |
|
|
|
| 706 |
|
|
if (sk->err || (sk->shutdown & SEND_SHUTDOWN))
|
| 707 |
|
|
goto do_error;
|
| 708 |
|
|
if (!*timeo)
|
| 709 |
|
|
goto do_nonblock;
|
| 710 |
|
|
if (signal_pending(current))
|
| 711 |
|
|
goto do_interrupted;
|
| 712 |
|
|
clear_bit(SOCK_ASYNC_NOSPACE, &sk->socket->flags);
|
| 713 |
|
|
if (tcp_memory_free(sk) && !vm_wait)
|
| 714 |
|
|
break;
|
| 715 |
|
|
|
| 716 |
|
|
set_bit(SOCK_NOSPACE, &sk->socket->flags);
|
| 717 |
|
|
sk->tp_pinfo.af_tcp.write_pending++;
|
| 718 |
|
|
release_sock(sk);
|
| 719 |
|
|
if (!tcp_memory_free(sk) || vm_wait)
|
| 720 |
|
|
current_timeo = schedule_timeout(current_timeo);
|
| 721 |
|
|
lock_sock(sk);
|
| 722 |
|
|
sk->tp_pinfo.af_tcp.write_pending--;
|
| 723 |
|
|
|
| 724 |
|
|
if (vm_wait) {
|
| 725 |
|
|
vm_wait -= current_timeo;
|
| 726 |
|
|
current_timeo = *timeo;
|
| 727 |
|
|
if (current_timeo != MAX_SCHEDULE_TIMEOUT &&
|
| 728 |
|
|
(current_timeo -= vm_wait) < 0)
|
| 729 |
|
|
current_timeo = 0;
|
| 730 |
|
|
vm_wait = 0;
|
| 731 |
|
|
}
|
| 732 |
|
|
*timeo = current_timeo;
|
| 733 |
|
|
}
|
| 734 |
|
|
out:
|
| 735 |
|
|
current->state = TASK_RUNNING;
|
| 736 |
|
|
remove_wait_queue(sk->sleep, &wait);
|
| 737 |
|
|
return err;
|
| 738 |
|
|
|
| 739 |
|
|
do_error:
|
| 740 |
|
|
err = -EPIPE;
|
| 741 |
|
|
goto out;
|
| 742 |
|
|
do_nonblock:
|
| 743 |
|
|
err = -EAGAIN;
|
| 744 |
|
|
goto out;
|
| 745 |
|
|
do_interrupted:
|
| 746 |
|
|
err = sock_intr_errno(*timeo);
|
| 747 |
|
|
goto out;
|
| 748 |
|
|
}
|
| 749 |
|
|
|
| 750 |
|
|
ssize_t do_tcp_sendpages(struct sock *sk, struct page **pages, int poffset, size_t psize, int flags);
|
| 751 |
|
|
|
| 752 |
|
|
static inline int
|
| 753 |
|
|
can_coalesce(struct sk_buff *skb, int i, struct page *page, int off)
|
| 754 |
|
|
{
|
| 755 |
|
|
if (i) {
|
| 756 |
|
|
skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
|
| 757 |
|
|
return page == frag->page &&
|
| 758 |
|
|
off == frag->page_offset+frag->size;
|
| 759 |
|
|
}
|
| 760 |
|
|
return 0;
|
| 761 |
|
|
}
|
| 762 |
|
|
|
| 763 |
|
|
static inline void
|
| 764 |
|
|
fill_page_desc(struct sk_buff *skb, int i, struct page *page, int off, int size)
|
| 765 |
|
|
{
|
| 766 |
|
|
skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
|
| 767 |
|
|
frag->page = page;
|
| 768 |
|
|
frag->page_offset = off;
|
| 769 |
|
|
frag->size = size;
|
| 770 |
|
|
skb_shinfo(skb)->nr_frags = i+1;
|
| 771 |
|
|
}
|
| 772 |
|
|
|
| 773 |
|
|
static inline void tcp_mark_push(struct tcp_opt *tp, struct sk_buff *skb)
|
| 774 |
|
|
{
|
| 775 |
|
|
TCP_SKB_CB(skb)->flags |= TCPCB_FLAG_PSH;
|
| 776 |
|
|
tp->pushed_seq = tp->write_seq;
|
| 777 |
|
|
}
|
| 778 |
|
|
|
| 779 |
|
|
static inline int forced_push(struct tcp_opt *tp)
|
| 780 |
|
|
{
|
| 781 |
|
|
return after(tp->write_seq, tp->pushed_seq + (tp->max_window>>1));
|
| 782 |
|
|
}
|
| 783 |
|
|
|
| 784 |
|
|
static inline void
|
| 785 |
|
|
skb_entail(struct sock *sk, struct tcp_opt *tp, struct sk_buff *skb)
|
| 786 |
|
|
{
|
| 787 |
|
|
skb->csum = 0;
|
| 788 |
|
|
TCP_SKB_CB(skb)->seq = tp->write_seq;
|
| 789 |
|
|
TCP_SKB_CB(skb)->end_seq = tp->write_seq;
|
| 790 |
|
|
TCP_SKB_CB(skb)->flags = TCPCB_FLAG_ACK;
|
| 791 |
|
|
TCP_SKB_CB(skb)->sacked = 0;
|
| 792 |
|
|
__skb_queue_tail(&sk->write_queue, skb);
|
| 793 |
|
|
tcp_charge_skb(sk, skb);
|
| 794 |
|
|
if (tp->send_head == NULL)
|
| 795 |
|
|
tp->send_head = skb;
|
| 796 |
|
|
}
|
| 797 |
|
|
|
| 798 |
|
|
static inline void
|
| 799 |
|
|
tcp_mark_urg(struct tcp_opt *tp, int flags, struct sk_buff *skb)
|
| 800 |
|
|
{
|
| 801 |
|
|
if (flags & MSG_OOB) {
|
| 802 |
|
|
tp->urg_mode = 1;
|
| 803 |
|
|
tp->snd_up = tp->write_seq;
|
| 804 |
|
|
TCP_SKB_CB(skb)->sacked |= TCPCB_URG;
|
| 805 |
|
|
}
|
| 806 |
|
|
}
|
| 807 |
|
|
|
| 808 |
|
|
static inline void
|
| 809 |
|
|
tcp_push(struct sock *sk, struct tcp_opt *tp, int flags, int mss_now, int nonagle)
|
| 810 |
|
|
{
|
| 811 |
|
|
if (tp->send_head) {
|
| 812 |
|
|
struct sk_buff *skb = sk->write_queue.prev;
|
| 813 |
|
|
if (!(flags&MSG_MORE) || forced_push(tp))
|
| 814 |
|
|
tcp_mark_push(tp, skb);
|
| 815 |
|
|
tcp_mark_urg(tp, flags, skb);
|
| 816 |
|
|
__tcp_push_pending_frames(sk, tp, mss_now, (flags&MSG_MORE) ? 2 : nonagle);
|
| 817 |
|
|
}
|
| 818 |
|
|
}
|
| 819 |
|
|
|
| 820 |
|
|
static int tcp_error(struct sock *sk, int flags, int err)
|
| 821 |
|
|
{
|
| 822 |
|
|
if (err == -EPIPE)
|
| 823 |
|
|
err = sock_error(sk) ? : -EPIPE;
|
| 824 |
|
|
if (err == -EPIPE && !(flags&MSG_NOSIGNAL))
|
| 825 |
|
|
send_sig(SIGPIPE, current, 0);
|
| 826 |
|
|
return err;
|
| 827 |
|
|
}
|
| 828 |
|
|
|
| 829 |
|
|
ssize_t do_tcp_sendpages(struct sock *sk, struct page **pages, int poffset, size_t psize, int flags)
|
| 830 |
|
|
{
|
| 831 |
|
|
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
|
| 832 |
|
|
int mss_now;
|
| 833 |
|
|
int err;
|
| 834 |
|
|
ssize_t copied;
|
| 835 |
|
|
long timeo = sock_sndtimeo(sk, flags&MSG_DONTWAIT);
|
| 836 |
|
|
|
| 837 |
|
|
/* Wait for a connection to finish. */
|
| 838 |
|
|
if ((1 << sk->state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))
|
| 839 |
|
|
if((err = wait_for_tcp_connect(sk, 0, &timeo)) != 0)
|
| 840 |
|
|
goto out_err;
|
| 841 |
|
|
|
| 842 |
|
|
clear_bit(SOCK_ASYNC_NOSPACE, &sk->socket->flags);
|
| 843 |
|
|
|
| 844 |
|
|
mss_now = tcp_current_mss(sk);
|
| 845 |
|
|
copied = 0;
|
| 846 |
|
|
|
| 847 |
|
|
err = -EPIPE;
|
| 848 |
|
|
if (sk->err || (sk->shutdown & SEND_SHUTDOWN))
|
| 849 |
|
|
goto do_error;
|
| 850 |
|
|
|
| 851 |
|
|
while (psize > 0) {
|
| 852 |
|
|
struct sk_buff *skb = sk->write_queue.prev;
|
| 853 |
|
|
int offset, size, copy, i;
|
| 854 |
|
|
struct page *page;
|
| 855 |
|
|
|
| 856 |
|
|
page = pages[poffset/PAGE_SIZE];
|
| 857 |
|
|
offset = poffset % PAGE_SIZE;
|
| 858 |
|
|
size = min_t(size_t, psize, PAGE_SIZE-offset);
|
| 859 |
|
|
|
| 860 |
|
|
if (tp->send_head==NULL || (copy = mss_now - skb->len) <= 0) {
|
| 861 |
|
|
new_segment:
|
| 862 |
|
|
if (!tcp_memory_free(sk))
|
| 863 |
|
|
goto wait_for_sndbuf;
|
| 864 |
|
|
|
| 865 |
|
|
skb = tcp_alloc_pskb(sk, 0, tp->mss_cache, sk->allocation);
|
| 866 |
|
|
if (skb == NULL)
|
| 867 |
|
|
goto wait_for_memory;
|
| 868 |
|
|
|
| 869 |
|
|
skb_entail(sk, tp, skb);
|
| 870 |
|
|
copy = mss_now;
|
| 871 |
|
|
}
|
| 872 |
|
|
|
| 873 |
|
|
if (copy > size)
|
| 874 |
|
|
copy = size;
|
| 875 |
|
|
|
| 876 |
|
|
i = skb_shinfo(skb)->nr_frags;
|
| 877 |
|
|
if (can_coalesce(skb, i, page, offset)) {
|
| 878 |
|
|
skb_shinfo(skb)->frags[i-1].size += copy;
|
| 879 |
|
|
} else if (i < MAX_SKB_FRAGS) {
|
| 880 |
|
|
get_page(page);
|
| 881 |
|
|
fill_page_desc(skb, i, page, offset, copy);
|
| 882 |
|
|
} else {
|
| 883 |
|
|
tcp_mark_push(tp, skb);
|
| 884 |
|
|
goto new_segment;
|
| 885 |
|
|
}
|
| 886 |
|
|
|
| 887 |
|
|
skb->len += copy;
|
| 888 |
|
|
skb->data_len += copy;
|
| 889 |
|
|
skb->ip_summed = CHECKSUM_HW;
|
| 890 |
|
|
tp->write_seq += copy;
|
| 891 |
|
|
TCP_SKB_CB(skb)->end_seq += copy;
|
| 892 |
|
|
|
| 893 |
|
|
if (!copied)
|
| 894 |
|
|
TCP_SKB_CB(skb)->flags &= ~TCPCB_FLAG_PSH;
|
| 895 |
|
|
|
| 896 |
|
|
copied += copy;
|
| 897 |
|
|
poffset += copy;
|
| 898 |
|
|
if (!(psize -= copy))
|
| 899 |
|
|
goto out;
|
| 900 |
|
|
|
| 901 |
|
|
if (skb->len != mss_now || (flags&MSG_OOB))
|
| 902 |
|
|
continue;
|
| 903 |
|
|
|
| 904 |
|
|
if (forced_push(tp)) {
|
| 905 |
|
|
tcp_mark_push(tp, skb);
|
| 906 |
|
|
__tcp_push_pending_frames(sk, tp, mss_now, 1);
|
| 907 |
|
|
} else if (skb == tp->send_head)
|
| 908 |
|
|
tcp_push_one(sk, mss_now);
|
| 909 |
|
|
continue;
|
| 910 |
|
|
|
| 911 |
|
|
wait_for_sndbuf:
|
| 912 |
|
|
set_bit(SOCK_NOSPACE, &sk->socket->flags);
|
| 913 |
|
|
wait_for_memory:
|
| 914 |
|
|
if (copied)
|
| 915 |
|
|
tcp_push(sk, tp, flags&~MSG_MORE, mss_now, 1);
|
| 916 |
|
|
|
| 917 |
|
|
if ((err = wait_for_tcp_memory(sk, &timeo)) != 0)
|
| 918 |
|
|
goto do_error;
|
| 919 |
|
|
|
| 920 |
|
|
mss_now = tcp_current_mss(sk);
|
| 921 |
|
|
}
|
| 922 |
|
|
|
| 923 |
|
|
out:
|
| 924 |
|
|
if (copied)
|
| 925 |
|
|
tcp_push(sk, tp, flags, mss_now, tp->nonagle);
|
| 926 |
|
|
return copied;
|
| 927 |
|
|
|
| 928 |
|
|
do_error:
|
| 929 |
|
|
if (copied)
|
| 930 |
|
|
goto out;
|
| 931 |
|
|
out_err:
|
| 932 |
|
|
return tcp_error(sk, flags, err);
|
| 933 |
|
|
}
|
| 934 |
|
|
|
| 935 |
|
|
ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
|
| 936 |
|
|
{
|
| 937 |
|
|
ssize_t res;
|
| 938 |
|
|
struct sock *sk = sock->sk;
|
| 939 |
|
|
|
| 940 |
|
|
#define TCP_ZC_CSUM_FLAGS (NETIF_F_IP_CSUM|NETIF_F_NO_CSUM|NETIF_F_HW_CSUM)
|
| 941 |
|
|
|
| 942 |
|
|
if (!(sk->route_caps & NETIF_F_SG) ||
|
| 943 |
|
|
!(sk->route_caps & TCP_ZC_CSUM_FLAGS))
|
| 944 |
|
|
return sock_no_sendpage(sock, page, offset, size, flags);
|
| 945 |
|
|
|
| 946 |
|
|
#undef TCP_ZC_CSUM_FLAGS
|
| 947 |
|
|
|
| 948 |
|
|
lock_sock(sk);
|
| 949 |
|
|
TCP_CHECK_TIMER(sk);
|
| 950 |
|
|
res = do_tcp_sendpages(sk, &page, offset, size, flags);
|
| 951 |
|
|
TCP_CHECK_TIMER(sk);
|
| 952 |
|
|
release_sock(sk);
|
| 953 |
|
|
return res;
|
| 954 |
|
|
}
|
| 955 |
|
|
|
| 956 |
|
|
#define TCP_PAGE(sk) (sk->tp_pinfo.af_tcp.sndmsg_page)
|
| 957 |
|
|
#define TCP_OFF(sk) (sk->tp_pinfo.af_tcp.sndmsg_off)
|
| 958 |
|
|
|
| 959 |
|
|
static inline int
|
| 960 |
|
|
tcp_copy_to_page(struct sock *sk, char *from, struct sk_buff *skb,
|
| 961 |
|
|
struct page *page, int off, int copy)
|
| 962 |
|
|
{
|
| 963 |
|
|
int err = 0;
|
| 964 |
|
|
unsigned int csum;
|
| 965 |
|
|
|
| 966 |
|
|
csum = csum_and_copy_from_user(from, page_address(page)+off,
|
| 967 |
|
|
copy, 0, &err);
|
| 968 |
|
|
if (!err) {
|
| 969 |
|
|
if (skb->ip_summed == CHECKSUM_NONE)
|
| 970 |
|
|
skb->csum = csum_block_add(skb->csum, csum, skb->len);
|
| 971 |
|
|
skb->len += copy;
|
| 972 |
|
|
skb->data_len += copy;
|
| 973 |
|
|
skb->truesize += copy;
|
| 974 |
|
|
sk->wmem_queued += copy;
|
| 975 |
|
|
sk->forward_alloc -= copy;
|
| 976 |
|
|
}
|
| 977 |
|
|
return err;
|
| 978 |
|
|
}
|
| 979 |
|
|
|
| 980 |
|
|
static inline int
|
| 981 |
|
|
skb_add_data(struct sk_buff *skb, char *from, int copy)
|
| 982 |
|
|
{
|
| 983 |
|
|
int err = 0;
|
| 984 |
|
|
unsigned int csum;
|
| 985 |
|
|
int off = skb->len;
|
| 986 |
|
|
|
| 987 |
|
|
csum = csum_and_copy_from_user(from, skb_put(skb, copy),
|
| 988 |
|
|
copy, 0, &err);
|
| 989 |
|
|
if (!err) {
|
| 990 |
|
|
skb->csum = csum_block_add(skb->csum, csum, off);
|
| 991 |
|
|
return 0;
|
| 992 |
|
|
}
|
| 993 |
|
|
|
| 994 |
|
|
__skb_trim(skb, off);
|
| 995 |
|
|
return -EFAULT;
|
| 996 |
|
|
}
|
| 997 |
|
|
|
| 998 |
|
|
static inline int select_size(struct sock *sk, struct tcp_opt *tp)
|
| 999 |
|
|
{
|
| 1000 |
|
|
int tmp = tp->mss_cache;
|
| 1001 |
|
|
|
| 1002 |
|
|
if (sk->route_caps&NETIF_F_SG) {
|
| 1003 |
|
|
int pgbreak = SKB_MAX_HEAD(MAX_TCP_HEADER);
|
| 1004 |
|
|
|
| 1005 |
|
|
if (tmp >= pgbreak && tmp <= pgbreak + (MAX_SKB_FRAGS-1)*PAGE_SIZE)
|
| 1006 |
|
|
tmp = pgbreak;
|
| 1007 |
|
|
}
|
| 1008 |
|
|
return tmp;
|
| 1009 |
|
|
}
|
| 1010 |
|
|
|
| 1011 |
|
|
int tcp_sendmsg(struct sock *sk, struct msghdr *msg, int size)
|
| 1012 |
|
|
{
|
| 1013 |
|
|
struct iovec *iov;
|
| 1014 |
|
|
struct tcp_opt *tp;
|
| 1015 |
|
|
struct sk_buff *skb;
|
| 1016 |
|
|
int iovlen, flags;
|
| 1017 |
|
|
int mss_now;
|
| 1018 |
|
|
int err, copied;
|
| 1019 |
|
|
long timeo;
|
| 1020 |
|
|
|
| 1021 |
|
|
tp = &(sk->tp_pinfo.af_tcp);
|
| 1022 |
|
|
|
| 1023 |
|
|
lock_sock(sk);
|
| 1024 |
|
|
TCP_CHECK_TIMER(sk);
|
| 1025 |
|
|
|
| 1026 |
|
|
flags = msg->msg_flags;
|
| 1027 |
|
|
timeo = sock_sndtimeo(sk, flags&MSG_DONTWAIT);
|
| 1028 |
|
|
|
| 1029 |
|
|
/* Wait for a connection to finish. */
|
| 1030 |
|
|
if ((1 << sk->state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))
|
| 1031 |
|
|
if((err = wait_for_tcp_connect(sk, flags, &timeo)) != 0)
|
| 1032 |
|
|
goto out_err;
|
| 1033 |
|
|
|
| 1034 |
|
|
/* This should be in poll */
|
| 1035 |
|
|
clear_bit(SOCK_ASYNC_NOSPACE, &sk->socket->flags);
|
| 1036 |
|
|
|
| 1037 |
|
|
mss_now = tcp_current_mss(sk);
|
| 1038 |
|
|
|
| 1039 |
|
|
/* Ok commence sending. */
|
| 1040 |
|
|
iovlen = msg->msg_iovlen;
|
| 1041 |
|
|
iov = msg->msg_iov;
|
| 1042 |
|
|
copied = 0;
|
| 1043 |
|
|
|
| 1044 |
|
|
err = -EPIPE;
|
| 1045 |
|
|
if (sk->err || (sk->shutdown&SEND_SHUTDOWN))
|
| 1046 |
|
|
goto do_error;
|
| 1047 |
|
|
|
| 1048 |
|
|
while (--iovlen >= 0) {
|
| 1049 |
|
|
int seglen=iov->iov_len;
|
| 1050 |
|
|
unsigned char * from=iov->iov_base;
|
| 1051 |
|
|
|
| 1052 |
|
|
iov++;
|
| 1053 |
|
|
|
| 1054 |
|
|
while (seglen > 0) {
|
| 1055 |
|
|
int copy;
|
| 1056 |
|
|
|
| 1057 |
|
|
skb = sk->write_queue.prev;
|
| 1058 |
|
|
|
| 1059 |
|
|
if (tp->send_head == NULL ||
|
| 1060 |
|
|
(copy = mss_now - skb->len) <= 0) {
|
| 1061 |
|
|
|
| 1062 |
|
|
new_segment:
|
| 1063 |
|
|
/* Allocate new segment. If the interface is SG,
|
| 1064 |
|
|
* allocate skb fitting to single page.
|
| 1065 |
|
|
*/
|
| 1066 |
|
|
if (!tcp_memory_free(sk))
|
| 1067 |
|
|
goto wait_for_sndbuf;
|
| 1068 |
|
|
|
| 1069 |
|
|
skb = tcp_alloc_pskb(sk, select_size(sk, tp), 0, sk->allocation);
|
| 1070 |
|
|
if (skb == NULL)
|
| 1071 |
|
|
goto wait_for_memory;
|
| 1072 |
|
|
|
| 1073 |
|
|
skb_entail(sk, tp, skb);
|
| 1074 |
|
|
copy = mss_now;
|
| 1075 |
|
|
}
|
| 1076 |
|
|
|
| 1077 |
|
|
/* Try to append data to the end of skb. */
|
| 1078 |
|
|
if (copy > seglen)
|
| 1079 |
|
|
copy = seglen;
|
| 1080 |
|
|
|
| 1081 |
|
|
/* Where to copy to? */
|
| 1082 |
|
|
if (skb_tailroom(skb) > 0) {
|
| 1083 |
|
|
/* We have some space in skb head. Superb! */
|
| 1084 |
|
|
if (copy > skb_tailroom(skb))
|
| 1085 |
|
|
copy = skb_tailroom(skb);
|
| 1086 |
|
|
if ((err = skb_add_data(skb, from, copy)) != 0)
|
| 1087 |
|
|
goto do_fault;
|
| 1088 |
|
|
} else {
|
| 1089 |
|
|
int merge = 0;
|
| 1090 |
|
|
int i = skb_shinfo(skb)->nr_frags;
|
| 1091 |
|
|
struct page *page = TCP_PAGE(sk);
|
| 1092 |
|
|
int off = TCP_OFF(sk);
|
| 1093 |
|
|
|
| 1094 |
|
|
if (can_coalesce(skb, i, page, off) && off != PAGE_SIZE) {
|
| 1095 |
|
|
/* We can extend the last page fragment. */
|
| 1096 |
|
|
merge = 1;
|
| 1097 |
|
|
} else if (i == MAX_SKB_FRAGS ||
|
| 1098 |
|
|
(i == 0 && !(sk->route_caps&NETIF_F_SG))) {
|
| 1099 |
|
|
/* Need to add new fragment and cannot
|
| 1100 |
|
|
* do this because interface is non-SG,
|
| 1101 |
|
|
* or because all the page slots are busy.
|
| 1102 |
|
|
*/
|
| 1103 |
|
|
tcp_mark_push(tp, skb);
|
| 1104 |
|
|
goto new_segment;
|
| 1105 |
|
|
} else if (page) {
|
| 1106 |
|
|
/* If page is cached, align
|
| 1107 |
|
|
* offset to L1 cache boundary
|
| 1108 |
|
|
*/
|
| 1109 |
|
|
off = (off+L1_CACHE_BYTES-1)&~(L1_CACHE_BYTES-1);
|
| 1110 |
|
|
if (off == PAGE_SIZE) {
|
| 1111 |
|
|
put_page(page);
|
| 1112 |
|
|
TCP_PAGE(sk) = page = NULL;
|
| 1113 |
|
|
}
|
| 1114 |
|
|
}
|
| 1115 |
|
|
|
| 1116 |
|
|
if (!page) {
|
| 1117 |
|
|
/* Allocate new cache page. */
|
| 1118 |
|
|
if (!(page=tcp_alloc_page(sk)))
|
| 1119 |
|
|
goto wait_for_memory;
|
| 1120 |
|
|
off = 0;
|
| 1121 |
|
|
}
|
| 1122 |
|
|
|
| 1123 |
|
|
if (copy > PAGE_SIZE-off)
|
| 1124 |
|
|
copy = PAGE_SIZE-off;
|
| 1125 |
|
|
|
| 1126 |
|
|
/* Time to copy data. We are close to the end! */
|
| 1127 |
|
|
err = tcp_copy_to_page(sk, from, skb, page, off, copy);
|
| 1128 |
|
|
if (err) {
|
| 1129 |
|
|
/* If this page was new, give it to the
|
| 1130 |
|
|
* socket so it does not get leaked.
|
| 1131 |
|
|
*/
|
| 1132 |
|
|
if (TCP_PAGE(sk) == NULL) {
|
| 1133 |
|
|
TCP_PAGE(sk) = page;
|
| 1134 |
|
|
TCP_OFF(sk) = 0;
|
| 1135 |
|
|
}
|
| 1136 |
|
|
goto do_error;
|
| 1137 |
|
|
}
|
| 1138 |
|
|
|
| 1139 |
|
|
/* Update the skb. */
|
| 1140 |
|
|
if (merge) {
|
| 1141 |
|
|
skb_shinfo(skb)->frags[i-1].size += copy;
|
| 1142 |
|
|
} else {
|
| 1143 |
|
|
fill_page_desc(skb, i, page, off, copy);
|
| 1144 |
|
|
if (TCP_PAGE(sk)) {
|
| 1145 |
|
|
get_page(page);
|
| 1146 |
|
|
} else if (off + copy < PAGE_SIZE) {
|
| 1147 |
|
|
get_page(page);
|
| 1148 |
|
|
TCP_PAGE(sk) = page;
|
| 1149 |
|
|
}
|
| 1150 |
|
|
}
|
| 1151 |
|
|
|
| 1152 |
|
|
TCP_OFF(sk) = off+copy;
|
| 1153 |
|
|
}
|
| 1154 |
|
|
|
| 1155 |
|
|
if (!copied)
|
| 1156 |
|
|
TCP_SKB_CB(skb)->flags &= ~TCPCB_FLAG_PSH;
|
| 1157 |
|
|
|
| 1158 |
|
|
tp->write_seq += copy;
|
| 1159 |
|
|
TCP_SKB_CB(skb)->end_seq += copy;
|
| 1160 |
|
|
|
| 1161 |
|
|
from += copy;
|
| 1162 |
|
|
copied += copy;
|
| 1163 |
|
|
if ((seglen -= copy) == 0 && iovlen == 0)
|
| 1164 |
|
|
goto out;
|
| 1165 |
|
|
|
| 1166 |
|
|
if (skb->len != mss_now || (flags&MSG_OOB))
|
| 1167 |
|
|
continue;
|
| 1168 |
|
|
|
| 1169 |
|
|
if (forced_push(tp)) {
|
| 1170 |
|
|
tcp_mark_push(tp, skb);
|
| 1171 |
|
|
__tcp_push_pending_frames(sk, tp, mss_now, 1);
|
| 1172 |
|
|
} else if (skb == tp->send_head)
|
| 1173 |
|
|
tcp_push_one(sk, mss_now);
|
| 1174 |
|
|
continue;
|
| 1175 |
|
|
|
| 1176 |
|
|
wait_for_sndbuf:
|
| 1177 |
|
|
set_bit(SOCK_NOSPACE, &sk->socket->flags);
|
| 1178 |
|
|
wait_for_memory:
|
| 1179 |
|
|
if (copied)
|
| 1180 |
|
|
tcp_push(sk, tp, flags&~MSG_MORE, mss_now, 1);
|
| 1181 |
|
|
|
| 1182 |
|
|
if ((err = wait_for_tcp_memory(sk, &timeo)) != 0)
|
| 1183 |
|
|
goto do_error;
|
| 1184 |
|
|
|
| 1185 |
|
|
mss_now = tcp_current_mss(sk);
|
| 1186 |
|
|
}
|
| 1187 |
|
|
}
|
| 1188 |
|
|
|
| 1189 |
|
|
out:
|
| 1190 |
|
|
if (copied)
|
| 1191 |
|
|
tcp_push(sk, tp, flags, mss_now, tp->nonagle);
|
| 1192 |
|
|
TCP_CHECK_TIMER(sk);
|
| 1193 |
|
|
release_sock(sk);
|
| 1194 |
|
|
return copied;
|
| 1195 |
|
|
|
| 1196 |
|
|
do_fault:
|
| 1197 |
|
|
if (skb->len == 0) {
|
| 1198 |
|
|
if (tp->send_head == skb)
|
| 1199 |
|
|
tp->send_head = NULL;
|
| 1200 |
|
|
__skb_unlink(skb, skb->list);
|
| 1201 |
|
|
tcp_free_skb(sk, skb);
|
| 1202 |
|
|
}
|
| 1203 |
|
|
|
| 1204 |
|
|
do_error:
|
| 1205 |
|
|
if (copied)
|
| 1206 |
|
|
goto out;
|
| 1207 |
|
|
out_err:
|
| 1208 |
|
|
err = tcp_error(sk, flags, err);
|
| 1209 |
|
|
TCP_CHECK_TIMER(sk);
|
| 1210 |
|
|
release_sock(sk);
|
| 1211 |
|
|
return err;
|
| 1212 |
|
|
}
|
| 1213 |
|
|
|
| 1214 |
|
|
/*
|
| 1215 |
|
|
* Handle reading urgent data. BSD has very simple semantics for
|
| 1216 |
|
|
* this, no blocking and very strange errors 8)
|
| 1217 |
|
|
*/
|
| 1218 |
|
|
|
| 1219 |
|
|
static int tcp_recv_urg(struct sock * sk, long timeo,
|
| 1220 |
|
|
struct msghdr *msg, int len, int flags,
|
| 1221 |
|
|
int *addr_len)
|
| 1222 |
|
|
{
|
| 1223 |
|
|
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
|
| 1224 |
|
|
|
| 1225 |
|
|
/* No URG data to read. */
|
| 1226 |
|
|
if (sk->urginline || !tp->urg_data || tp->urg_data == TCP_URG_READ)
|
| 1227 |
|
|
return -EINVAL; /* Yes this is right ! */
|
| 1228 |
|
|
|
| 1229 |
|
|
if (sk->state==TCP_CLOSE && !sk->done)
|
| 1230 |
|
|
return -ENOTCONN;
|
| 1231 |
|
|
|
| 1232 |
|
|
if (tp->urg_data & TCP_URG_VALID) {
|
| 1233 |
|
|
int err = 0;
|
| 1234 |
|
|
char c = tp->urg_data;
|
| 1235 |
|
|
|
| 1236 |
|
|
if (!(flags & MSG_PEEK))
|
| 1237 |
|
|
tp->urg_data = TCP_URG_READ;
|
| 1238 |
|
|
|
| 1239 |
|
|
/* Read urgent data. */
|
| 1240 |
|
|
msg->msg_flags|=MSG_OOB;
|
| 1241 |
|
|
|
| 1242 |
|
|
if(len>0) {
|
| 1243 |
|
|
if (!(flags & MSG_TRUNC))
|
| 1244 |
|
|
err = memcpy_toiovec(msg->msg_iov, &c, 1);
|
| 1245 |
|
|
len = 1;
|
| 1246 |
|
|
} else
|
| 1247 |
|
|
msg->msg_flags|=MSG_TRUNC;
|
| 1248 |
|
|
|
| 1249 |
|
|
return err ? -EFAULT : len;
|
| 1250 |
|
|
}
|
| 1251 |
|
|
|
| 1252 |
|
|
if (sk->state == TCP_CLOSE || (sk->shutdown & RCV_SHUTDOWN))
|
| 1253 |
|
|
return 0;
|
| 1254 |
|
|
|
| 1255 |
|
|
/* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and
|
| 1256 |
|
|
* the available implementations agree in this case:
|
| 1257 |
|
|
* this call should never block, independent of the
|
| 1258 |
|
|
* blocking state of the socket.
|
| 1259 |
|
|
* Mike <pall@rz.uni-karlsruhe.de>
|
| 1260 |
|
|
*/
|
| 1261 |
|
|
return -EAGAIN;
|
| 1262 |
|
|
}
|
| 1263 |
|
|
|
| 1264 |
|
|
/*
|
| 1265 |
|
|
* Release a skb if it is no longer needed. This routine
|
| 1266 |
|
|
* must be called with interrupts disabled or with the
|
| 1267 |
|
|
* socket locked so that the sk_buff queue operation is ok.
|
| 1268 |
|
|
*/
|
| 1269 |
|
|
|
| 1270 |
|
|
static inline void tcp_eat_skb(struct sock *sk, struct sk_buff * skb)
|
| 1271 |
|
|
{
|
| 1272 |
|
|
__skb_unlink(skb, &sk->receive_queue);
|
| 1273 |
|
|
__kfree_skb(skb);
|
| 1274 |
|
|
}
|
| 1275 |
|
|
|
| 1276 |
|
|
/* Clean up the receive buffer for full frames taken by the user,
|
| 1277 |
|
|
* then send an ACK if necessary. COPIED is the number of bytes
|
| 1278 |
|
|
* tcp_recvmsg has given to the user so far, it speeds up the
|
| 1279 |
|
|
* calculation of whether or not we must ACK for the sake of
|
| 1280 |
|
|
* a window update.
|
| 1281 |
|
|
*/
|
| 1282 |
|
|
static void cleanup_rbuf(struct sock *sk, int copied)
|
| 1283 |
|
|
{
|
| 1284 |
|
|
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
|
| 1285 |
|
|
int time_to_ack = 0;
|
| 1286 |
|
|
|
| 1287 |
|
|
#if TCP_DEBUG
|
| 1288 |
|
|
struct sk_buff *skb = skb_peek(&sk->receive_queue);
|
| 1289 |
|
|
|
| 1290 |
|
|
BUG_TRAP(skb==NULL || before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq));
|
| 1291 |
|
|
#endif
|
| 1292 |
|
|
|
| 1293 |
|
|
if (tcp_ack_scheduled(tp)) {
|
| 1294 |
|
|
/* Delayed ACKs frequently hit locked sockets during bulk receive. */
|
| 1295 |
|
|
if (tp->ack.blocked
|
| 1296 |
|
|
/* Once-per-two-segments ACK was not sent by tcp_input.c */
|
| 1297 |
|
|
|| tp->rcv_nxt - tp->rcv_wup > tp->ack.rcv_mss
|
| 1298 |
|
|
/*
|
| 1299 |
|
|
* If this read emptied read buffer, we send ACK, if
|
| 1300 |
|
|
* connection is not bidirectional, user drained
|
| 1301 |
|
|
* receive buffer and there was a small segment
|
| 1302 |
|
|
* in queue.
|
| 1303 |
|
|
*/
|
| 1304 |
|
|
|| (copied > 0 &&
|
| 1305 |
|
|
(tp->ack.pending&TCP_ACK_PUSHED) &&
|
| 1306 |
|
|
!tp->ack.pingpong &&
|
| 1307 |
|
|
atomic_read(&sk->rmem_alloc) == 0)) {
|
| 1308 |
|
|
time_to_ack = 1;
|
| 1309 |
|
|
}
|
| 1310 |
|
|
}
|
| 1311 |
|
|
|
| 1312 |
|
|
/* We send an ACK if we can now advertise a non-zero window
|
| 1313 |
|
|
* which has been raised "significantly".
|
| 1314 |
|
|
*
|
| 1315 |
|
|
* Even if window raised up to infinity, do not send window open ACK
|
| 1316 |
|
|
* in states, where we will not receive more. It is useless.
|
| 1317 |
|
|
*/
|
| 1318 |
|
|
if(copied > 0 && !time_to_ack && !(sk->shutdown&RCV_SHUTDOWN)) {
|
| 1319 |
|
|
__u32 rcv_window_now = tcp_receive_window(tp);
|
| 1320 |
|
|
|
| 1321 |
|
|
/* Optimize, __tcp_select_window() is not cheap. */
|
| 1322 |
|
|
if (2*rcv_window_now <= tp->window_clamp) {
|
| 1323 |
|
|
__u32 new_window = __tcp_select_window(sk);
|
| 1324 |
|
|
|
| 1325 |
|
|
/* Send ACK now, if this read freed lots of space
|
| 1326 |
|
|
* in our buffer. Certainly, new_window is new window.
|
| 1327 |
|
|
* We can advertise it now, if it is not less than current one.
|
| 1328 |
|
|
* "Lots" means "at least twice" here.
|
| 1329 |
|
|
*/
|
| 1330 |
|
|
if(new_window && new_window >= 2*rcv_window_now)
|
| 1331 |
|
|
time_to_ack = 1;
|
| 1332 |
|
|
}
|
| 1333 |
|
|
}
|
| 1334 |
|
|
if (time_to_ack)
|
| 1335 |
|
|
tcp_send_ack(sk);
|
| 1336 |
|
|
}
|
| 1337 |
|
|
|
| 1338 |
|
|
/* Now socket state including sk->err is changed only under lock,
|
| 1339 |
|
|
* hence we may omit checks after joining wait queue.
|
| 1340 |
|
|
* We check receive queue before schedule() only as optimization;
|
| 1341 |
|
|
* it is very likely that release_sock() added new data.
|
| 1342 |
|
|
*/
|
| 1343 |
|
|
|
| 1344 |
|
|
static long tcp_data_wait(struct sock *sk, long timeo)
|
| 1345 |
|
|
{
|
| 1346 |
|
|
DECLARE_WAITQUEUE(wait, current);
|
| 1347 |
|
|
|
| 1348 |
|
|
add_wait_queue(sk->sleep, &wait);
|
| 1349 |
|
|
|
| 1350 |
|
|
__set_current_state(TASK_INTERRUPTIBLE);
|
| 1351 |
|
|
|
| 1352 |
|
|
set_bit(SOCK_ASYNC_WAITDATA, &sk->socket->flags);
|
| 1353 |
|
|
release_sock(sk);
|
| 1354 |
|
|
|
| 1355 |
|
|
if (skb_queue_empty(&sk->receive_queue))
|
| 1356 |
|
|
timeo = schedule_timeout(timeo);
|
| 1357 |
|
|
|
| 1358 |
|
|
lock_sock(sk);
|
| 1359 |
|
|
clear_bit(SOCK_ASYNC_WAITDATA, &sk->socket->flags);
|
| 1360 |
|
|
|
| 1361 |
|
|
remove_wait_queue(sk->sleep, &wait);
|
| 1362 |
|
|
__set_current_state(TASK_RUNNING);
|
| 1363 |
|
|
return timeo;
|
| 1364 |
|
|
}
|
| 1365 |
|
|
|
| 1366 |
|
|
static void tcp_prequeue_process(struct sock *sk)
|
| 1367 |
|
|
{
|
| 1368 |
|
|
struct sk_buff *skb;
|
| 1369 |
|
|
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
|
| 1370 |
|
|
|
| 1371 |
|
|
net_statistics[smp_processor_id()*2+1].TCPPrequeued += skb_queue_len(&tp->ucopy.prequeue);
|
| 1372 |
|
|
|
| 1373 |
|
|
/* RX process wants to run with disabled BHs, though it is not necessary */
|
| 1374 |
|
|
local_bh_disable();
|
| 1375 |
|
|
while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
|
| 1376 |
|
|
sk->backlog_rcv(sk, skb);
|
| 1377 |
|
|
local_bh_enable();
|
| 1378 |
|
|
|
| 1379 |
|
|
/* Clear memory counter. */
|
| 1380 |
|
|
tp->ucopy.memory = 0;
|
| 1381 |
|
|
}
|
| 1382 |
|
|
|
| 1383 |
|
|
static inline
|
| 1384 |
|
|
struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off)
|
| 1385 |
|
|
{
|
| 1386 |
|
|
struct sk_buff *skb;
|
| 1387 |
|
|
u32 offset;
|
| 1388 |
|
|
|
| 1389 |
|
|
skb_queue_walk(&sk->receive_queue, skb) {
|
| 1390 |
|
|
offset = seq - TCP_SKB_CB(skb)->seq;
|
| 1391 |
|
|
if (skb->h.th->syn)
|
| 1392 |
|
|
offset--;
|
| 1393 |
|
|
if (offset < skb->len || skb->h.th->fin) {
|
| 1394 |
|
|
*off = offset;
|
| 1395 |
|
|
return skb;
|
| 1396 |
|
|
}
|
| 1397 |
|
|
}
|
| 1398 |
|
|
return NULL;
|
| 1399 |
|
|
}
|
| 1400 |
|
|
|
| 1401 |
|
|
/*
|
| 1402 |
|
|
* This routine provides an alternative to tcp_recvmsg() for routines
|
| 1403 |
|
|
* that would like to handle copying from skbuffs directly in 'sendfile'
|
| 1404 |
|
|
* fashion.
|
| 1405 |
|
|
* Note:
|
| 1406 |
|
|
* - It is assumed that the socket was locked by the caller.
|
| 1407 |
|
|
* - The routine does not block.
|
| 1408 |
|
|
* - At present, there is no support for reading OOB data
|
| 1409 |
|
|
* or for 'peeking' the socket using this routine
|
| 1410 |
|
|
* (although both would be easy to implement).
|
| 1411 |
|
|
*/
|
| 1412 |
|
|
int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
|
| 1413 |
|
|
sk_read_actor_t recv_actor)
|
| 1414 |
|
|
{
|
| 1415 |
|
|
struct sk_buff *skb;
|
| 1416 |
|
|
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
|
| 1417 |
|
|
u32 seq = tp->copied_seq;
|
| 1418 |
|
|
u32 offset;
|
| 1419 |
|
|
int copied = 0;
|
| 1420 |
|
|
|
| 1421 |
|
|
if (sk->state == TCP_LISTEN)
|
| 1422 |
|
|
return -ENOTCONN;
|
| 1423 |
|
|
while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) {
|
| 1424 |
|
|
if (offset < skb->len) {
|
| 1425 |
|
|
size_t used, len;
|
| 1426 |
|
|
|
| 1427 |
|
|
len = skb->len - offset;
|
| 1428 |
|
|
/* Stop reading if we hit a patch of urgent data */
|
| 1429 |
|
|
if (tp->urg_data) {
|
| 1430 |
|
|
u32 urg_offset = tp->urg_seq - seq;
|
| 1431 |
|
|
if (urg_offset < len)
|
| 1432 |
|
|
len = urg_offset;
|
| 1433 |
|
|
if (!len)
|
| 1434 |
|
|
break;
|
| 1435 |
|
|
}
|
| 1436 |
|
|
used = recv_actor(desc, skb, offset, len);
|
| 1437 |
|
|
if (used <= len) {
|
| 1438 |
|
|
seq += used;
|
| 1439 |
|
|
copied += used;
|
| 1440 |
|
|
offset += used;
|
| 1441 |
|
|
}
|
| 1442 |
|
|
if (offset != skb->len)
|
| 1443 |
|
|
break;
|
| 1444 |
|
|
}
|
| 1445 |
|
|
if (skb->h.th->fin) {
|
| 1446 |
|
|
tcp_eat_skb(sk, skb);
|
| 1447 |
|
|
++seq;
|
| 1448 |
|
|
break;
|
| 1449 |
|
|
}
|
| 1450 |
|
|
tcp_eat_skb(sk, skb);
|
| 1451 |
|
|
if (!desc->count)
|
| 1452 |
|
|
break;
|
| 1453 |
|
|
}
|
| 1454 |
|
|
tp->copied_seq = seq;
|
| 1455 |
|
|
/* Clean up data we have read: This will do ACK frames. */
|
| 1456 |
|
|
if (copied)
|
| 1457 |
|
|
cleanup_rbuf(sk, copied);
|
| 1458 |
|
|
return copied;
|
| 1459 |
|
|
}
|
| 1460 |
|
|
|
| 1461 |
|
|
/*
|
| 1462 |
|
|
* This routine copies from a sock struct into the user buffer.
|
| 1463 |
|
|
*
|
| 1464 |
|
|
* Technical note: in 2.3 we work on _locked_ socket, so that
|
| 1465 |
|
|
* tricks with *seq access order and skb->users are not required.
|
| 1466 |
|
|
* Probably, code can be easily improved even more.
|
| 1467 |
|
|
*/
|
| 1468 |
|
|
|
| 1469 |
|
|
int tcp_recvmsg(struct sock *sk, struct msghdr *msg,
|
| 1470 |
|
|
int len, int nonblock, int flags, int *addr_len)
|
| 1471 |
|
|
{
|
| 1472 |
|
|
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
|
| 1473 |
|
|
int copied = 0;
|
| 1474 |
|
|
u32 peek_seq;
|
| 1475 |
|
|
u32 *seq;
|
| 1476 |
|
|
unsigned long used;
|
| 1477 |
|
|
int err;
|
| 1478 |
|
|
int target; /* Read at least this many bytes */
|
| 1479 |
|
|
long timeo;
|
| 1480 |
|
|
struct task_struct *user_recv = NULL;
|
| 1481 |
|
|
|
| 1482 |
|
|
lock_sock(sk);
|
| 1483 |
|
|
|
| 1484 |
|
|
TCP_CHECK_TIMER(sk);
|
| 1485 |
|
|
|
| 1486 |
|
|
err = -ENOTCONN;
|
| 1487 |
|
|
if (sk->state == TCP_LISTEN)
|
| 1488 |
|
|
goto out;
|
| 1489 |
|
|
|
| 1490 |
|
|
timeo = sock_rcvtimeo(sk, nonblock);
|
| 1491 |
|
|
|
| 1492 |
|
|
/* Urgent data needs to be handled specially. */
|
| 1493 |
|
|
if (flags & MSG_OOB)
|
| 1494 |
|
|
goto recv_urg;
|
| 1495 |
|
|
|
| 1496 |
|
|
seq = &tp->copied_seq;
|
| 1497 |
|
|
if (flags & MSG_PEEK) {
|
| 1498 |
|
|
peek_seq = tp->copied_seq;
|
| 1499 |
|
|
seq = &peek_seq;
|
| 1500 |
|
|
}
|
| 1501 |
|
|
|
| 1502 |
|
|
target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
|
| 1503 |
|
|
|
| 1504 |
|
|
do {
|
| 1505 |
|
|
struct sk_buff * skb;
|
| 1506 |
|
|
u32 offset;
|
| 1507 |
|
|
|
| 1508 |
|
|
/* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */
|
| 1509 |
|
|
if (tp->urg_data && tp->urg_seq == *seq) {
|
| 1510 |
|
|
if (copied)
|
| 1511 |
|
|
break;
|
| 1512 |
|
|
if (signal_pending(current)) {
|
| 1513 |
|
|
copied = timeo ? sock_intr_errno(timeo) : -EAGAIN;
|
| 1514 |
|
|
break;
|
| 1515 |
|
|
}
|
| 1516 |
|
|
}
|
| 1517 |
|
|
|
| 1518 |
|
|
/* Next get a buffer. */
|
| 1519 |
|
|
|
| 1520 |
|
|
skb = skb_peek(&sk->receive_queue);
|
| 1521 |
|
|
do {
|
| 1522 |
|
|
if (!skb)
|
| 1523 |
|
|
break;
|
| 1524 |
|
|
|
| 1525 |
|
|
/* Now that we have two receive queues this
|
| 1526 |
|
|
* shouldn't happen.
|
| 1527 |
|
|
*/
|
| 1528 |
|
|
if (before(*seq, TCP_SKB_CB(skb)->seq)) {
|
| 1529 |
|
|
printk(KERN_INFO "recvmsg bug: copied %X seq %X\n",
|
| 1530 |
|
|
*seq, TCP_SKB_CB(skb)->seq);
|
| 1531 |
|
|
break;
|
| 1532 |
|
|
}
|
| 1533 |
|
|
offset = *seq - TCP_SKB_CB(skb)->seq;
|
| 1534 |
|
|
if (skb->h.th->syn)
|
| 1535 |
|
|
offset--;
|
| 1536 |
|
|
if (offset < skb->len)
|
| 1537 |
|
|
goto found_ok_skb;
|
| 1538 |
|
|
if (skb->h.th->fin)
|
| 1539 |
|
|
goto found_fin_ok;
|
| 1540 |
|
|
BUG_TRAP(flags&MSG_PEEK);
|
| 1541 |
|
|
skb = skb->next;
|
| 1542 |
|
|
} while (skb != (struct sk_buff *)&sk->receive_queue);
|
| 1543 |
|
|
|
| 1544 |
|
|
/* Well, if we have backlog, try to process it now yet. */
|
| 1545 |
|
|
|
| 1546 |
|
|
if (copied >= target && sk->backlog.tail == NULL)
|
| 1547 |
|
|
break;
|
| 1548 |
|
|
|
| 1549 |
|
|
if (copied) {
|
| 1550 |
|
|
if (sk->err ||
|
| 1551 |
|
|
sk->state == TCP_CLOSE ||
|
| 1552 |
|
|
(sk->shutdown & RCV_SHUTDOWN) ||
|
| 1553 |
|
|
!timeo ||
|
| 1554 |
|
|
signal_pending(current) ||
|
| 1555 |
|
|
(flags & MSG_PEEK))
|
| 1556 |
|
|
break;
|
| 1557 |
|
|
} else {
|
| 1558 |
|
|
if (sk->done)
|
| 1559 |
|
|
break;
|
| 1560 |
|
|
|
| 1561 |
|
|
if (sk->err) {
|
| 1562 |
|
|
copied = sock_error(sk);
|
| 1563 |
|
|
break;
|
| 1564 |
|
|
}
|
| 1565 |
|
|
|
| 1566 |
|
|
if (sk->shutdown & RCV_SHUTDOWN)
|
| 1567 |
|
|
break;
|
| 1568 |
|
|
|
| 1569 |
|
|
if (sk->state == TCP_CLOSE) {
|
| 1570 |
|
|
if (!sk->done) {
|
| 1571 |
|
|
/* This occurs when user tries to read
|
| 1572 |
|
|
* from never connected socket.
|
| 1573 |
|
|
*/
|
| 1574 |
|
|
copied = -ENOTCONN;
|
| 1575 |
|
|
break;
|
| 1576 |
|
|
}
|
| 1577 |
|
|
break;
|
| 1578 |
|
|
}
|
| 1579 |
|
|
|
| 1580 |
|
|
if (!timeo) {
|
| 1581 |
|
|
copied = -EAGAIN;
|
| 1582 |
|
|
break;
|
| 1583 |
|
|
}
|
| 1584 |
|
|
|
| 1585 |
|
|
if (signal_pending(current)) {
|
| 1586 |
|
|
copied = sock_intr_errno(timeo);
|
| 1587 |
|
|
break;
|
| 1588 |
|
|
}
|
| 1589 |
|
|
}
|
| 1590 |
|
|
|
| 1591 |
|
|
cleanup_rbuf(sk, copied);
|
| 1592 |
|
|
|
| 1593 |
|
|
if (tp->ucopy.task == user_recv) {
|
| 1594 |
|
|
/* Install new reader */
|
| 1595 |
|
|
if (user_recv == NULL && !(flags&(MSG_TRUNC|MSG_PEEK))) {
|
| 1596 |
|
|
user_recv = current;
|
| 1597 |
|
|
tp->ucopy.task = user_recv;
|
| 1598 |
|
|
tp->ucopy.iov = msg->msg_iov;
|
| 1599 |
|
|
}
|
| 1600 |
|
|
|
| 1601 |
|
|
tp->ucopy.len = len;
|
| 1602 |
|
|
|
| 1603 |
|
|
BUG_TRAP(tp->copied_seq == tp->rcv_nxt || (flags&(MSG_PEEK|MSG_TRUNC)));
|
| 1604 |
|
|
|
| 1605 |
|
|
/* Ugly... If prequeue is not empty, we have to
|
| 1606 |
|
|
* process it before releasing socket, otherwise
|
| 1607 |
|
|
* order will be broken at second iteration.
|
| 1608 |
|
|
* More elegant solution is required!!!
|
| 1609 |
|
|
*
|
| 1610 |
|
|
* Look: we have the following (pseudo)queues:
|
| 1611 |
|
|
*
|
| 1612 |
|
|
* 1. packets in flight
|
| 1613 |
|
|
* 2. backlog
|
| 1614 |
|
|
* 3. prequeue
|
| 1615 |
|
|
* 4. receive_queue
|
| 1616 |
|
|
*
|
| 1617 |
|
|
* Each queue can be processed only if the next ones
|
| 1618 |
|
|
* are empty. At this point we have empty receive_queue.
|
| 1619 |
|
|
* But prequeue _can_ be not empty after second iteration,
|
| 1620 |
|
|
* when we jumped to start of loop because backlog
|
| 1621 |
|
|
* processing added something to receive_queue.
|
| 1622 |
|
|
* We cannot release_sock(), because backlog contains
|
| 1623 |
|
|
* packets arrived _after_ prequeued ones.
|
| 1624 |
|
|
*
|
| 1625 |
|
|
* Shortly, algorithm is clear --- to process all
|
| 1626 |
|
|
* the queues in order. We could make it more directly,
|
| 1627 |
|
|
* requeueing packets from backlog to prequeue, if
|
| 1628 |
|
|
* is not empty. It is more elegant, but eats cycles,
|
| 1629 |
|
|
* unfortunately.
|
| 1630 |
|
|
*/
|
| 1631 |
|
|
if (skb_queue_len(&tp->ucopy.prequeue))
|
| 1632 |
|
|
goto do_prequeue;
|
| 1633 |
|
|
|
| 1634 |
|
|
/* __ Set realtime policy in scheduler __ */
|
| 1635 |
|
|
}
|
| 1636 |
|
|
|
| 1637 |
|
|
if (copied >= target) {
|
| 1638 |
|
|
/* Do not sleep, just process backlog. */
|
| 1639 |
|
|
release_sock(sk);
|
| 1640 |
|
|
lock_sock(sk);
|
| 1641 |
|
|
} else {
|
| 1642 |
|
|
timeo = tcp_data_wait(sk, timeo);
|
| 1643 |
|
|
}
|
| 1644 |
|
|
|
| 1645 |
|
|
if (user_recv) {
|
| 1646 |
|
|
int chunk;
|
| 1647 |
|
|
|
| 1648 |
|
|
/* __ Restore normal policy in scheduler __ */
|
| 1649 |
|
|
|
| 1650 |
|
|
if ((chunk = len - tp->ucopy.len) != 0) {
|
| 1651 |
|
|
net_statistics[smp_processor_id()*2+1].TCPDirectCopyFromBacklog += chunk;
|
| 1652 |
|
|
len -= chunk;
|
| 1653 |
|
|
copied += chunk;
|
| 1654 |
|
|
}
|
| 1655 |
|
|
|
| 1656 |
|
|
if (tp->rcv_nxt == tp->copied_seq &&
|
| 1657 |
|
|
skb_queue_len(&tp->ucopy.prequeue)) {
|
| 1658 |
|
|
do_prequeue:
|
| 1659 |
|
|
tcp_prequeue_process(sk);
|
| 1660 |
|
|
|
| 1661 |
|
|
if ((chunk = len - tp->ucopy.len) != 0) {
|
| 1662 |
|
|
net_statistics[smp_processor_id()*2+1].TCPDirectCopyFromPrequeue += chunk;
|
| 1663 |
|
|
len -= chunk;
|
| 1664 |
|
|
copied += chunk;
|
| 1665 |
|
|
}
|
| 1666 |
|
|
}
|
| 1667 |
|
|
}
|
| 1668 |
|
|
if ((flags & MSG_PEEK) && peek_seq != tp->copied_seq) {
|
| 1669 |
|
|
if (net_ratelimit())
|
| 1670 |
|
|
printk(KERN_DEBUG "TCP(%s:%d): Application bug, race in MSG_PEEK.\n",
|
| 1671 |
|
|
current->comm, current->pid);
|
| 1672 |
|
|
peek_seq = tp->copied_seq;
|
| 1673 |
|
|
}
|
| 1674 |
|
|
continue;
|
| 1675 |
|
|
|
| 1676 |
|
|
found_ok_skb:
|
| 1677 |
|
|
/* Ok so how much can we use? */
|
| 1678 |
|
|
used = skb->len - offset;
|
| 1679 |
|
|
if (len < used)
|
| 1680 |
|
|
used = len;
|
| 1681 |
|
|
|
| 1682 |
|
|
/* Do we have urgent data here? */
|
| 1683 |
|
|
if (tp->urg_data) {
|
| 1684 |
|
|
u32 urg_offset = tp->urg_seq - *seq;
|
| 1685 |
|
|
if (urg_offset < used) {
|
| 1686 |
|
|
if (!urg_offset) {
|
| 1687 |
|
|
if (!sk->urginline) {
|
| 1688 |
|
|
++*seq;
|
| 1689 |
|
|
offset++;
|
| 1690 |
|
|
used--;
|
| 1691 |
|
|
if (!used)
|
| 1692 |
|
|
goto skip_copy;
|
| 1693 |
|
|
}
|
| 1694 |
|
|
} else
|
| 1695 |
|
|
used = urg_offset;
|
| 1696 |
|
|
}
|
| 1697 |
|
|
}
|
| 1698 |
|
|
|
| 1699 |
|
|
if (!(flags&MSG_TRUNC)) {
|
| 1700 |
|
|
err = skb_copy_datagram_iovec(skb, offset, msg->msg_iov, used);
|
| 1701 |
|
|
if (err) {
|
| 1702 |
|
|
/* Exception. Bailout! */
|
| 1703 |
|
|
if (!copied)
|
| 1704 |
|
|
copied = -EFAULT;
|
| 1705 |
|
|
break;
|
| 1706 |
|
|
}
|
| 1707 |
|
|
}
|
| 1708 |
|
|
|
| 1709 |
|
|
*seq += used;
|
| 1710 |
|
|
copied += used;
|
| 1711 |
|
|
len -= used;
|
| 1712 |
|
|
|
| 1713 |
|
|
skip_copy:
|
| 1714 |
|
|
if (tp->urg_data && after(tp->copied_seq,tp->urg_seq)) {
|
| 1715 |
|
|
tp->urg_data = 0;
|
| 1716 |
|
|
tcp_fast_path_check(sk, tp);
|
| 1717 |
|
|
}
|
| 1718 |
|
|
if (used + offset < skb->len)
|
| 1719 |
|
|
continue;
|
| 1720 |
|
|
|
| 1721 |
|
|
if (skb->h.th->fin)
|
| 1722 |
|
|
goto found_fin_ok;
|
| 1723 |
|
|
if (!(flags & MSG_PEEK))
|
| 1724 |
|
|
tcp_eat_skb(sk, skb);
|
| 1725 |
|
|
continue;
|
| 1726 |
|
|
|
| 1727 |
|
|
found_fin_ok:
|
| 1728 |
|
|
/* Process the FIN. */
|
| 1729 |
|
|
++*seq;
|
| 1730 |
|
|
if (!(flags & MSG_PEEK))
|
| 1731 |
|
|
tcp_eat_skb(sk, skb);
|
| 1732 |
|
|
break;
|
| 1733 |
|
|
} while (len > 0);
|
| 1734 |
|
|
|
| 1735 |
|
|
if (user_recv) {
|
| 1736 |
|
|
if (skb_queue_len(&tp->ucopy.prequeue)) {
|
| 1737 |
|
|
int chunk;
|
| 1738 |
|
|
|
| 1739 |
|
|
tp->ucopy.len = copied > 0 ? len : 0;
|
| 1740 |
|
|
|
| 1741 |
|
|
tcp_prequeue_process(sk);
|
| 1742 |
|
|
|
| 1743 |
|
|
if (copied > 0 && (chunk = len - tp->ucopy.len) != 0) {
|
| 1744 |
|
|
net_statistics[smp_processor_id()*2+1].TCPDirectCopyFromPrequeue += chunk;
|
| 1745 |
|
|
len -= chunk;
|
| 1746 |
|
|
copied += chunk;
|
| 1747 |
|
|
}
|
| 1748 |
|
|
}
|
| 1749 |
|
|
|
| 1750 |
|
|
tp->ucopy.task = NULL;
|
| 1751 |
|
|
tp->ucopy.len = 0;
|
| 1752 |
|
|
}
|
| 1753 |
|
|
|
| 1754 |
|
|
/* According to UNIX98, msg_name/msg_namelen are ignored
|
| 1755 |
|
|
* on connected socket. I was just happy when found this 8) --ANK
|
| 1756 |
|
|
*/
|
| 1757 |
|
|
|
| 1758 |
|
|
/* Clean up data we have read: This will do ACK frames. */
|
| 1759 |
|
|
cleanup_rbuf(sk, copied);
|
| 1760 |
|
|
|
| 1761 |
|
|
TCP_CHECK_TIMER(sk);
|
| 1762 |
|
|
release_sock(sk);
|
| 1763 |
|
|
return copied;
|
| 1764 |
|
|
|
| 1765 |
|
|
out:
|
| 1766 |
|
|
TCP_CHECK_TIMER(sk);
|
| 1767 |
|
|
release_sock(sk);
|
| 1768 |
|
|
return err;
|
| 1769 |
|
|
|
| 1770 |
|
|
recv_urg:
|
| 1771 |
|
|
err = tcp_recv_urg(sk, timeo, msg, len, flags, addr_len);
|
| 1772 |
|
|
goto out;
|
| 1773 |
|
|
}
|
| 1774 |
|
|
|
| 1775 |
|
|
/*
|
| 1776 |
|
|
* State processing on a close. This implements the state shift for
|
| 1777 |
|
|
* sending our FIN frame. Note that we only send a FIN for some
|
| 1778 |
|
|
* states. A shutdown() may have already sent the FIN, or we may be
|
| 1779 |
|
|
* closed.
|
| 1780 |
|
|
*/
|
| 1781 |
|
|
|
| 1782 |
|
|
static unsigned char new_state[16] = {
|
| 1783 |
|
|
/* current state: new state: action: */
|
| 1784 |
|
|
/* (Invalid) */ TCP_CLOSE,
|
| 1785 |
|
|
/* TCP_ESTABLISHED */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
|
| 1786 |
|
|
/* TCP_SYN_SENT */ TCP_CLOSE,
|
| 1787 |
|
|
/* TCP_SYN_RECV */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
|
| 1788 |
|
|
/* TCP_FIN_WAIT1 */ TCP_FIN_WAIT1,
|
| 1789 |
|
|
/* TCP_FIN_WAIT2 */ TCP_FIN_WAIT2,
|
| 1790 |
|
|
/* TCP_TIME_WAIT */ TCP_CLOSE,
|
| 1791 |
|
|
/* TCP_CLOSE */ TCP_CLOSE,
|
| 1792 |
|
|
/* TCP_CLOSE_WAIT */ TCP_LAST_ACK | TCP_ACTION_FIN,
|
| 1793 |
|
|
/* TCP_LAST_ACK */ TCP_LAST_ACK,
|
| 1794 |
|
|
/* TCP_LISTEN */ TCP_CLOSE,
|
| 1795 |
|
|
/* TCP_CLOSING */ TCP_CLOSING,
|
| 1796 |
|
|
};
|
| 1797 |
|
|
|
| 1798 |
|
|
static int tcp_close_state(struct sock *sk)
|
| 1799 |
|
|
{
|
| 1800 |
|
|
int next = (int) new_state[sk->state];
|
| 1801 |
|
|
int ns = (next & TCP_STATE_MASK);
|
| 1802 |
|
|
|
| 1803 |
|
|
tcp_set_state(sk, ns);
|
| 1804 |
|
|
|
| 1805 |
|
|
return (next & TCP_ACTION_FIN);
|
| 1806 |
|
|
}
|
| 1807 |
|
|
|
| 1808 |
|
|
/*
|
| 1809 |
|
|
* Shutdown the sending side of a connection. Much like close except
|
| 1810 |
|
|
* that we don't receive shut down or set sk->dead.
|
| 1811 |
|
|
*/
|
| 1812 |
|
|
|
| 1813 |
|
|
void tcp_shutdown(struct sock *sk, int how)
|
| 1814 |
|
|
{
|
| 1815 |
|
|
/* We need to grab some memory, and put together a FIN,
|
| 1816 |
|
|
* and then put it into the queue to be sent.
|
| 1817 |
|
|
* Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92.
|
| 1818 |
|
|
*/
|
| 1819 |
|
|
if (!(how & SEND_SHUTDOWN))
|
| 1820 |
|
|
return;
|
| 1821 |
|
|
|
| 1822 |
|
|
/* If we've already sent a FIN, or it's a closed state, skip this. */
|
| 1823 |
|
|
if ((1 << sk->state) &
|
| 1824 |
|
|
(TCPF_ESTABLISHED|TCPF_SYN_SENT|TCPF_SYN_RECV|TCPF_CLOSE_WAIT)) {
|
| 1825 |
|
|
/* Clear out any half completed packets. FIN if needed. */
|
| 1826 |
|
|
if (tcp_close_state(sk))
|
| 1827 |
|
|
tcp_send_fin(sk);
|
| 1828 |
|
|
}
|
| 1829 |
|
|
}
|
| 1830 |
|
|
|
| 1831 |
|
|
|
| 1832 |
|
|
/*
|
| 1833 |
|
|
* Return 1 if we still have things to send in our buffers.
|
| 1834 |
|
|
*/
|
| 1835 |
|
|
|
| 1836 |
|
|
static inline int closing(struct sock * sk)
|
| 1837 |
|
|
{
|
| 1838 |
|
|
return ((1 << sk->state) & (TCPF_FIN_WAIT1|TCPF_CLOSING|TCPF_LAST_ACK));
|
| 1839 |
|
|
}
|
| 1840 |
|
|
|
| 1841 |
|
|
static __inline__ void tcp_kill_sk_queues(struct sock *sk)
|
| 1842 |
|
|
{
|
| 1843 |
|
|
/* First the read buffer. */
|
| 1844 |
|
|
__skb_queue_purge(&sk->receive_queue);
|
| 1845 |
|
|
|
| 1846 |
|
|
/* Next, the error queue. */
|
| 1847 |
|
|
__skb_queue_purge(&sk->error_queue);
|
| 1848 |
|
|
|
| 1849 |
|
|
/* Next, the write queue. */
|
| 1850 |
|
|
BUG_TRAP(skb_queue_empty(&sk->write_queue));
|
| 1851 |
|
|
|
| 1852 |
|
|
/* Account for returned memory. */
|
| 1853 |
|
|
tcp_mem_reclaim(sk);
|
| 1854 |
|
|
|
| 1855 |
|
|
BUG_TRAP(sk->wmem_queued == 0);
|
| 1856 |
|
|
BUG_TRAP(sk->forward_alloc == 0);
|
| 1857 |
|
|
|
| 1858 |
|
|
/* It is _impossible_ for the backlog to contain anything
|
| 1859 |
|
|
* when we get here. All user references to this socket
|
| 1860 |
|
|
* have gone away, only the net layer knows can touch it.
|
| 1861 |
|
|
*/
|
| 1862 |
|
|
}
|
| 1863 |
|
|
|
| 1864 |
|
|
/*
|
| 1865 |
|
|
* At this point, there should be no process reference to this
|
| 1866 |
|
|
* socket, and thus no user references at all. Therefore we
|
| 1867 |
|
|
* can assume the socket waitqueue is inactive and nobody will
|
| 1868 |
|
|
* try to jump onto it.
|
| 1869 |
|
|
*/
|
| 1870 |
|
|
void tcp_destroy_sock(struct sock *sk)
|
| 1871 |
|
|
{
|
| 1872 |
|
|
BUG_TRAP(sk->state==TCP_CLOSE);
|
| 1873 |
|
|
BUG_TRAP(sk->dead);
|
| 1874 |
|
|
|
| 1875 |
|
|
/* It cannot be in hash table! */
|
| 1876 |
|
|
BUG_TRAP(sk->pprev==NULL);
|
| 1877 |
|
|
|
| 1878 |
|
|
/* If it has not 0 sk->num, it must be bound */
|
| 1879 |
|
|
BUG_TRAP(!sk->num || sk->prev!=NULL);
|
| 1880 |
|
|
|
| 1881 |
|
|
#ifdef TCP_DEBUG
|
| 1882 |
|
|
if (sk->zapped) {
|
| 1883 |
|
|
printk(KERN_DEBUG "TCP: double destroy sk=%p\n", sk);
|
| 1884 |
|
|
sock_hold(sk);
|
| 1885 |
|
|
}
|
| 1886 |
|
|
sk->zapped = 1;
|
| 1887 |
|
|
#endif
|
| 1888 |
|
|
|
| 1889 |
|
|
sk->prot->destroy(sk);
|
| 1890 |
|
|
|
| 1891 |
|
|
tcp_kill_sk_queues(sk);
|
| 1892 |
|
|
|
| 1893 |
|
|
#ifdef INET_REFCNT_DEBUG
|
| 1894 |
|
|
if (atomic_read(&sk->refcnt) != 1) {
|
| 1895 |
|
|
printk(KERN_DEBUG "Destruction TCP %p delayed, c=%d\n", sk, atomic_read(&sk->refcnt));
|
| 1896 |
|
|
}
|
| 1897 |
|
|
#endif
|
| 1898 |
|
|
|
| 1899 |
|
|
atomic_dec(&tcp_orphan_count);
|
| 1900 |
|
|
sock_put(sk);
|
| 1901 |
|
|
}
|
| 1902 |
|
|
|
| 1903 |
|
|
void tcp_close(struct sock *sk, long timeout)
|
| 1904 |
|
|
{
|
| 1905 |
|
|
struct sk_buff *skb;
|
| 1906 |
|
|
int data_was_unread = 0;
|
| 1907 |
|
|
|
| 1908 |
|
|
lock_sock(sk);
|
| 1909 |
|
|
sk->shutdown = SHUTDOWN_MASK;
|
| 1910 |
|
|
|
| 1911 |
|
|
if(sk->state == TCP_LISTEN) {
|
| 1912 |
|
|
tcp_set_state(sk, TCP_CLOSE);
|
| 1913 |
|
|
|
| 1914 |
|
|
/* Special case. */
|
| 1915 |
|
|
tcp_listen_stop(sk);
|
| 1916 |
|
|
|
| 1917 |
|
|
goto adjudge_to_death;
|
| 1918 |
|
|
}
|
| 1919 |
|
|
|
| 1920 |
|
|
/* We need to flush the recv. buffs. We do this only on the
|
| 1921 |
|
|
* descriptor close, not protocol-sourced closes, because the
|
| 1922 |
|
|
* reader process may not have drained the data yet!
|
| 1923 |
|
|
*/
|
| 1924 |
|
|
while((skb=__skb_dequeue(&sk->receive_queue))!=NULL) {
|
| 1925 |
|
|
u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq - skb->h.th->fin;
|
| 1926 |
|
|
data_was_unread += len;
|
| 1927 |
|
|
__kfree_skb(skb);
|
| 1928 |
|
|
}
|
| 1929 |
|
|
|
| 1930 |
|
|
tcp_mem_reclaim(sk);
|
| 1931 |
|
|
|
| 1932 |
|
|
/* As outlined in draft-ietf-tcpimpl-prob-03.txt, section
|
| 1933 |
|
|
* 3.10, we send a RST here because data was lost. To
|
| 1934 |
|
|
* witness the awful effects of the old behavior of always
|
| 1935 |
|
|
* doing a FIN, run an older 2.1.x kernel or 2.0.x, start
|
| 1936 |
|
|
* a bulk GET in an FTP client, suspend the process, wait
|
| 1937 |
|
|
* for the client to advertise a zero window, then kill -9
|
| 1938 |
|
|
* the FTP client, wheee... Note: timeout is always zero
|
| 1939 |
|
|
* in such a case.
|
| 1940 |
|
|
*/
|
| 1941 |
|
|
if(data_was_unread != 0) {
|
| 1942 |
|
|
/* Unread data was tossed, zap the connection. */
|
| 1943 |
|
|
NET_INC_STATS_USER(TCPAbortOnClose);
|
| 1944 |
|
|
tcp_set_state(sk, TCP_CLOSE);
|
| 1945 |
|
|
tcp_send_active_reset(sk, GFP_KERNEL);
|
| 1946 |
|
|
} else if (sk->linger && sk->lingertime==0) {
|
| 1947 |
|
|
/* Check zero linger _after_ checking for unread data. */
|
| 1948 |
|
|
sk->prot->disconnect(sk, 0);
|
| 1949 |
|
|
NET_INC_STATS_USER(TCPAbortOnData);
|
| 1950 |
|
|
} else if (tcp_close_state(sk)) {
|
| 1951 |
|
|
/* We FIN if the application ate all the data before
|
| 1952 |
|
|
* zapping the connection.
|
| 1953 |
|
|
*/
|
| 1954 |
|
|
|
| 1955 |
|
|
/* RED-PEN. Formally speaking, we have broken TCP state
|
| 1956 |
|
|
* machine. State transitions:
|
| 1957 |
|
|
*
|
| 1958 |
|
|
* TCP_ESTABLISHED -> TCP_FIN_WAIT1
|
| 1959 |
|
|
* TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible)
|
| 1960 |
|
|
* TCP_CLOSE_WAIT -> TCP_LAST_ACK
|
| 1961 |
|
|
*
|
| 1962 |
|
|
* are legal only when FIN has been sent (i.e. in window),
|
| 1963 |
|
|
* rather than queued out of window. Purists blame.
|
| 1964 |
|
|
*
|
| 1965 |
|
|
* F.e. "RFC state" is ESTABLISHED,
|
| 1966 |
|
|
* if Linux state is FIN-WAIT-1, but FIN is still not sent.
|
| 1967 |
|
|
*
|
| 1968 |
|
|
* The visible declinations are that sometimes
|
| 1969 |
|
|
* we enter time-wait state, when it is not required really
|
| 1970 |
|
|
* (harmless), do not send active resets, when they are
|
| 1971 |
|
|
* required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when
|
| 1972 |
|
|
* they look as CLOSING or LAST_ACK for Linux)
|
| 1973 |
|
|
* Probably, I missed some more holelets.
|
| 1974 |
|
|
* --ANK
|
| 1975 |
|
|
*/
|
| 1976 |
|
|
tcp_send_fin(sk);
|
| 1977 |
|
|
}
|
| 1978 |
|
|
|
| 1979 |
|
|
if (timeout) {
|
| 1980 |
|
|
struct task_struct *tsk = current;
|
| 1981 |
|
|
DECLARE_WAITQUEUE(wait, current);
|
| 1982 |
|
|
|
| 1983 |
|
|
add_wait_queue(sk->sleep, &wait);
|
| 1984 |
|
|
|
| 1985 |
|
|
do {
|
| 1986 |
|
|
set_current_state(TASK_INTERRUPTIBLE);
|
| 1987 |
|
|
if (!closing(sk))
|
| 1988 |
|
|
break;
|
| 1989 |
|
|
release_sock(sk);
|
| 1990 |
|
|
timeout = schedule_timeout(timeout);
|
| 1991 |
|
|
lock_sock(sk);
|
| 1992 |
|
|
} while (!signal_pending(tsk) && timeout);
|
| 1993 |
|
|
|
| 1994 |
|
|
tsk->state = TASK_RUNNING;
|
| 1995 |
|
|
remove_wait_queue(sk->sleep, &wait);
|
| 1996 |
|
|
}
|
| 1997 |
|
|
|
| 1998 |
|
|
adjudge_to_death:
|
| 1999 |
|
|
/* It is the last release_sock in its life. It will remove backlog. */
|
| 2000 |
|
|
release_sock(sk);
|
| 2001 |
|
|
|
| 2002 |
|
|
|
| 2003 |
|
|
/* Now socket is owned by kernel and we acquire BH lock
|
| 2004 |
|
|
to finish close. No need to check for user refs.
|
| 2005 |
|
|
*/
|
| 2006 |
|
|
local_bh_disable();
|
| 2007 |
|
|
bh_lock_sock(sk);
|
| 2008 |
|
|
BUG_TRAP(sk->lock.users==0);
|
| 2009 |
|
|
|
| 2010 |
|
|
sock_hold(sk);
|
| 2011 |
|
|
sock_orphan(sk);
|
| 2012 |
|
|
|
| 2013 |
|
|
/* This is a (useful) BSD violating of the RFC. There is a
|
| 2014 |
|
|
* problem with TCP as specified in that the other end could
|
| 2015 |
|
|
* keep a socket open forever with no application left this end.
|
| 2016 |
|
|
* We use a 3 minute timeout (about the same as BSD) then kill
|
| 2017 |
|
|
* our end. If they send after that then tough - BUT: long enough
|
| 2018 |
|
|
* that we won't make the old 4*rto = almost no time - whoops
|
| 2019 |
|
|
* reset mistake.
|
| 2020 |
|
|
*
|
| 2021 |
|
|
* Nope, it was not mistake. It is really desired behaviour
|
| 2022 |
|
|
* f.e. on http servers, when such sockets are useless, but
|
| 2023 |
|
|
* consume significant resources. Let's do it with special
|
| 2024 |
|
|
* linger2 option. --ANK
|
| 2025 |
|
|
*/
|
| 2026 |
|
|
|
| 2027 |
|
|
if (sk->state == TCP_FIN_WAIT2) {
|
| 2028 |
|
|
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
|
| 2029 |
|
|
if (tp->linger2 < 0) {
|
| 2030 |
|
|
tcp_set_state(sk, TCP_CLOSE);
|
| 2031 |
|
|
tcp_send_active_reset(sk, GFP_ATOMIC);
|
| 2032 |
|
|
NET_INC_STATS_BH(TCPAbortOnLinger);
|
| 2033 |
|
|
} else {
|
| 2034 |
|
|
int tmo = tcp_fin_time(tp);
|
| 2035 |
|
|
|
| 2036 |
|
|
if (tmo > TCP_TIMEWAIT_LEN) {
|
| 2037 |
|
|
tcp_reset_keepalive_timer(sk, tcp_fin_time(tp));
|
| 2038 |
|
|
} else {
|
| 2039 |
|
|
atomic_inc(&tcp_orphan_count);
|
| 2040 |
|
|
tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
|
| 2041 |
|
|
goto out;
|
| 2042 |
|
|
}
|
| 2043 |
|
|
}
|
| 2044 |
|
|
}
|
| 2045 |
|
|
if (sk->state != TCP_CLOSE) {
|
| 2046 |
|
|
tcp_mem_reclaim(sk);
|
| 2047 |
|
|
if (atomic_read(&tcp_orphan_count) > sysctl_tcp_max_orphans ||
|
| 2048 |
|
|
(sk->wmem_queued > SOCK_MIN_SNDBUF &&
|
| 2049 |
|
|
atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2])) {
|
| 2050 |
|
|
if (net_ratelimit())
|
| 2051 |
|
|
printk(KERN_INFO "TCP: too many of orphaned sockets\n");
|
| 2052 |
|
|
tcp_set_state(sk, TCP_CLOSE);
|
| 2053 |
|
|
tcp_send_active_reset(sk, GFP_ATOMIC);
|
| 2054 |
|
|
NET_INC_STATS_BH(TCPAbortOnMemory);
|
| 2055 |
|
|
}
|
| 2056 |
|
|
}
|
| 2057 |
|
|
atomic_inc(&tcp_orphan_count);
|
| 2058 |
|
|
|
| 2059 |
|
|
if (sk->state == TCP_CLOSE)
|
| 2060 |
|
|
tcp_destroy_sock(sk);
|
| 2061 |
|
|
/* Otherwise, socket is reprieved until protocol close. */
|
| 2062 |
|
|
|
| 2063 |
|
|
out:
|
| 2064 |
|
|
bh_unlock_sock(sk);
|
| 2065 |
|
|
local_bh_enable();
|
| 2066 |
|
|
sock_put(sk);
|
| 2067 |
|
|
}
|
| 2068 |
|
|
|
| 2069 |
|
|
/* These states need RST on ABORT according to RFC793 */
|
| 2070 |
|
|
|
| 2071 |
|
|
static inline int tcp_need_reset(int state)
|
| 2072 |
|
|
{
|
| 2073 |
|
|
return ((1 << state) &
|
| 2074 |
|
|
(TCPF_ESTABLISHED|TCPF_CLOSE_WAIT|TCPF_FIN_WAIT1|
|
| 2075 |
|
|
TCPF_FIN_WAIT2|TCPF_SYN_RECV));
|
| 2076 |
|
|
}
|
| 2077 |
|
|
|
| 2078 |
|
|
int tcp_disconnect(struct sock *sk, int flags)
|
| 2079 |
|
|
{
|
| 2080 |
|
|
struct tcp_opt *tp = &sk->tp_pinfo.af_tcp;
|
| 2081 |
|
|
int old_state;
|
| 2082 |
|
|
int err = 0;
|
| 2083 |
|
|
|
| 2084 |
|
|
old_state = sk->state;
|
| 2085 |
|
|
if (old_state != TCP_CLOSE)
|
| 2086 |
|
|
tcp_set_state(sk, TCP_CLOSE);
|
| 2087 |
|
|
|
| 2088 |
|
|
/* ABORT function of RFC793 */
|
| 2089 |
|
|
if (old_state == TCP_LISTEN) {
|
| 2090 |
|
|
tcp_listen_stop(sk);
|
| 2091 |
|
|
} else if (tcp_need_reset(old_state) ||
|
| 2092 |
|
|
(tp->snd_nxt != tp->write_seq &&
|
| 2093 |
|
|
(1<<old_state)&(TCPF_CLOSING|TCPF_LAST_ACK))) {
|
| 2094 |
|
|
/* The last check adjusts for discrepance of Linux wrt. RFC
|
| 2095 |
|
|
* states
|
| 2096 |
|
|
*/
|
| 2097 |
|
|
tcp_send_active_reset(sk, gfp_any());
|
| 2098 |
|
|
sk->err = ECONNRESET;
|
| 2099 |
|
|
} else if (old_state == TCP_SYN_SENT)
|
| 2100 |
|
|
sk->err = ECONNRESET;
|
| 2101 |
|
|
|
| 2102 |
|
|
tcp_clear_xmit_timers(sk);
|
| 2103 |
|
|
__skb_queue_purge(&sk->receive_queue);
|
| 2104 |
|
|
tcp_writequeue_purge(sk);
|
| 2105 |
|
|
__skb_queue_purge(&tp->out_of_order_queue);
|
| 2106 |
|
|
|
| 2107 |
|
|
sk->dport = 0;
|
| 2108 |
|
|
|
| 2109 |
|
|
if (!(sk->userlocks&SOCK_BINDADDR_LOCK)) {
|
| 2110 |
|
|
sk->rcv_saddr = 0;
|
| 2111 |
|
|
sk->saddr = 0;
|
| 2112 |
|
|
#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
|
| 2113 |
|
|
memset(&sk->net_pinfo.af_inet6.saddr, 0, 16);
|
| 2114 |
|
|
memset(&sk->net_pinfo.af_inet6.rcv_saddr, 0, 16);
|
| 2115 |
|
|
#endif
|
| 2116 |
|
|
}
|
| 2117 |
|
|
|
| 2118 |
|
|
sk->shutdown = 0;
|
| 2119 |
|
|
sk->done = 0;
|
| 2120 |
|
|
tp->srtt = 0;
|
| 2121 |
|
|
if ((tp->write_seq += tp->max_window+2) == 0)
|
| 2122 |
|
|
tp->write_seq = 1;
|
| 2123 |
|
|
tp->backoff = 0;
|
| 2124 |
|
|
tp->snd_cwnd = 2;
|
| 2125 |
|
|
tp->probes_out = 0;
|
| 2126 |
|
|
tp->packets_out = 0;
|
| 2127 |
|
|
tp->snd_ssthresh = 0x7fffffff;
|
| 2128 |
|
|
tp->snd_cwnd_cnt = 0;
|
| 2129 |
|
|
tp->ca_state = TCP_CA_Open;
|
| 2130 |
|
|
tcp_clear_retrans(tp);
|
| 2131 |
|
|
tcp_delack_init(tp);
|
| 2132 |
|
|
tp->send_head = NULL;
|
| 2133 |
|
|
tp->saw_tstamp = 0;
|
| 2134 |
|
|
tcp_sack_reset(tp);
|
| 2135 |
|
|
__sk_dst_reset(sk);
|
| 2136 |
|
|
|
| 2137 |
|
|
BUG_TRAP(!sk->num || sk->prev);
|
| 2138 |
|
|
|
| 2139 |
|
|
sk->error_report(sk);
|
| 2140 |
|
|
return err;
|
| 2141 |
|
|
}
|
| 2142 |
|
|
|
| 2143 |
|
|
/*
|
| 2144 |
|
|
* Wait for an incoming connection, avoid race
|
| 2145 |
|
|
* conditions. This must be called with the socket locked.
|
| 2146 |
|
|
*/
|
| 2147 |
|
|
static int wait_for_connect(struct sock * sk, long timeo)
|
| 2148 |
|
|
{
|
| 2149 |
|
|
DECLARE_WAITQUEUE(wait, current);
|
| 2150 |
|
|
int err;
|
| 2151 |
|
|
|
| 2152 |
|
|
/*
|
| 2153 |
|
|
* True wake-one mechanism for incoming connections: only
|
| 2154 |
|
|
* one process gets woken up, not the 'whole herd'.
|
| 2155 |
|
|
* Since we do not 'race & poll' for established sockets
|
| 2156 |
|
|
* anymore, the common case will execute the loop only once.
|
| 2157 |
|
|
*
|
| 2158 |
|
|
* Subtle issue: "add_wait_queue_exclusive()" will be added
|
| 2159 |
|
|
* after any current non-exclusive waiters, and we know that
|
| 2160 |
|
|
* it will always _stay_ after any new non-exclusive waiters
|
| 2161 |
|
|
* because all non-exclusive waiters are added at the
|
| 2162 |
|
|
* beginning of the wait-queue. As such, it's ok to "drop"
|
| 2163 |
|
|
* our exclusiveness temporarily when we get woken up without
|
| 2164 |
|
|
* having to remove and re-insert us on the wait queue.
|
| 2165 |
|
|
*/
|
| 2166 |
|
|
add_wait_queue_exclusive(sk->sleep, &wait);
|
| 2167 |
|
|
for (;;) {
|
| 2168 |
|
|
current->state = TASK_INTERRUPTIBLE;
|
| 2169 |
|
|
release_sock(sk);
|
| 2170 |
|
|
if (sk->tp_pinfo.af_tcp.accept_queue == NULL)
|
| 2171 |
|
|
timeo = schedule_timeout(timeo);
|
| 2172 |
|
|
lock_sock(sk);
|
| 2173 |
|
|
err = 0;
|
| 2174 |
|
|
if (sk->tp_pinfo.af_tcp.accept_queue)
|
| 2175 |
|
|
break;
|
| 2176 |
|
|
err = -EINVAL;
|
| 2177 |
|
|
if (sk->state != TCP_LISTEN)
|
| 2178 |
|
|
break;
|
| 2179 |
|
|
err = sock_intr_errno(timeo);
|
| 2180 |
|
|
if (signal_pending(current))
|
| 2181 |
|
|
break;
|
| 2182 |
|
|
err = -EAGAIN;
|
| 2183 |
|
|
if (!timeo)
|
| 2184 |
|
|
break;
|
| 2185 |
|
|
}
|
| 2186 |
|
|
current->state = TASK_RUNNING;
|
| 2187 |
|
|
remove_wait_queue(sk->sleep, &wait);
|
| 2188 |
|
|
return err;
|
| 2189 |
|
|
}
|
| 2190 |
|
|
|
| 2191 |
|
|
/*
|
| 2192 |
|
|
* This will accept the next outstanding connection.
|
| 2193 |
|
|
*/
|
| 2194 |
|
|
|
| 2195 |
|
|
struct sock *tcp_accept(struct sock *sk, int flags, int *err)
|
| 2196 |
|
|
{
|
| 2197 |
|
|
struct tcp_opt *tp = &sk->tp_pinfo.af_tcp;
|
| 2198 |
|
|
struct open_request *req;
|
| 2199 |
|
|
struct sock *newsk;
|
| 2200 |
|
|
int error;
|
| 2201 |
|
|
|
| 2202 |
|
|
lock_sock(sk);
|
| 2203 |
|
|
|
| 2204 |
|
|
/* We need to make sure that this socket is listening,
|
| 2205 |
|
|
* and that it has something pending.
|
| 2206 |
|
|
*/
|
| 2207 |
|
|
error = -EINVAL;
|
| 2208 |
|
|
if (sk->state != TCP_LISTEN)
|
| 2209 |
|
|
goto out;
|
| 2210 |
|
|
|
| 2211 |
|
|
/* Find already established connection */
|
| 2212 |
|
|
if (!tp->accept_queue) {
|
| 2213 |
|
|
long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
|
| 2214 |
|
|
|
| 2215 |
|
|
/* If this is a non blocking socket don't sleep */
|
| 2216 |
|
|
error = -EAGAIN;
|
| 2217 |
|
|
if (!timeo)
|
| 2218 |
|
|
goto out;
|
| 2219 |
|
|
|
| 2220 |
|
|
error = wait_for_connect(sk, timeo);
|
| 2221 |
|
|
if (error)
|
| 2222 |
|
|
goto out;
|
| 2223 |
|
|
}
|
| 2224 |
|
|
|
| 2225 |
|
|
req = tp->accept_queue;
|
| 2226 |
|
|
if ((tp->accept_queue = req->dl_next) == NULL)
|
| 2227 |
|
|
tp->accept_queue_tail = NULL;
|
| 2228 |
|
|
|
| 2229 |
|
|
newsk = req->sk;
|
| 2230 |
|
|
tcp_acceptq_removed(sk);
|
| 2231 |
|
|
tcp_openreq_fastfree(req);
|
| 2232 |
|
|
BUG_TRAP(newsk->state != TCP_SYN_RECV);
|
| 2233 |
|
|
release_sock(sk);
|
| 2234 |
|
|
return newsk;
|
| 2235 |
|
|
|
| 2236 |
|
|
out:
|
| 2237 |
|
|
release_sock(sk);
|
| 2238 |
|
|
*err = error;
|
| 2239 |
|
|
return NULL;
|
| 2240 |
|
|
}
|
| 2241 |
|
|
|
| 2242 |
|
|
/*
|
| 2243 |
|
|
* Socket option code for TCP.
|
| 2244 |
|
|
*/
|
| 2245 |
|
|
|
| 2246 |
|
|
int tcp_setsockopt(struct sock *sk, int level, int optname, char *optval,
|
| 2247 |
|
|
int optlen)
|
| 2248 |
|
|
{
|
| 2249 |
|
|
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
|
| 2250 |
|
|
int val;
|
| 2251 |
|
|
int err = 0;
|
| 2252 |
|
|
|
| 2253 |
|
|
if (level != SOL_TCP)
|
| 2254 |
|
|
return tp->af_specific->setsockopt(sk, level, optname,
|
| 2255 |
|
|
optval, optlen);
|
| 2256 |
|
|
|
| 2257 |
|
|
if(optlen<sizeof(int))
|
| 2258 |
|
|
return -EINVAL;
|
| 2259 |
|
|
|
| 2260 |
|
|
if (get_user(val, (int *)optval))
|
| 2261 |
|
|
return -EFAULT;
|
| 2262 |
|
|
|
| 2263 |
|
|
lock_sock(sk);
|
| 2264 |
|
|
|
| 2265 |
|
|
switch(optname) {
|
| 2266 |
|
|
case TCP_MAXSEG:
|
| 2267 |
|
|
/* values greater than interface MTU won't take effect. however at
|
| 2268 |
|
|
* the point when this call is done we typically don't yet know
|
| 2269 |
|
|
* which interface is going to be used
|
| 2270 |
|
|
*/
|
| 2271 |
|
|
if(val < 8 || val > MAX_TCP_WINDOW) {
|
| 2272 |
|
|
err = -EINVAL;
|
| 2273 |
|
|
break;
|
| 2274 |
|
|
}
|
| 2275 |
|
|
tp->user_mss = val;
|
| 2276 |
|
|
break;
|
| 2277 |
|
|
|
| 2278 |
|
|
case TCP_NODELAY:
|
| 2279 |
|
|
/* You cannot try to use this and TCP_CORK in
|
| 2280 |
|
|
* tandem, so let the user know.
|
| 2281 |
|
|
*/
|
| 2282 |
|
|
if (tp->nonagle == 2) {
|
| 2283 |
|
|
err = -EINVAL;
|
| 2284 |
|
|
break;
|
| 2285 |
|
|
}
|
| 2286 |
|
|
tp->nonagle = (val == 0) ? 0 : 1;
|
| 2287 |
|
|
if (val)
|
| 2288 |
|
|
tcp_push_pending_frames(sk, tp);
|
| 2289 |
|
|
break;
|
| 2290 |
|
|
|
| 2291 |
|
|
case TCP_CORK:
|
| 2292 |
|
|
/* When set indicates to always queue non-full frames.
|
| 2293 |
|
|
* Later the user clears this option and we transmit
|
| 2294 |
|
|
* any pending partial frames in the queue. This is
|
| 2295 |
|
|
* meant to be used alongside sendfile() to get properly
|
| 2296 |
|
|
* filled frames when the user (for example) must write
|
| 2297 |
|
|
* out headers with a write() call first and then use
|
| 2298 |
|
|
* sendfile to send out the data parts.
|
| 2299 |
|
|
*
|
| 2300 |
|
|
* You cannot try to use TCP_NODELAY and this mechanism
|
| 2301 |
|
|
* at the same time, so let the user know.
|
| 2302 |
|
|
*/
|
| 2303 |
|
|
if (tp->nonagle == 1) {
|
| 2304 |
|
|
err = -EINVAL;
|
| 2305 |
|
|
break;
|
| 2306 |
|
|
}
|
| 2307 |
|
|
if (val != 0) {
|
| 2308 |
|
|
tp->nonagle = 2;
|
| 2309 |
|
|
} else {
|
| 2310 |
|
|
tp->nonagle = 0;
|
| 2311 |
|
|
|
| 2312 |
|
|
tcp_push_pending_frames(sk, tp);
|
| 2313 |
|
|
}
|
| 2314 |
|
|
break;
|
| 2315 |
|
|
|
| 2316 |
|
|
case TCP_KEEPIDLE:
|
| 2317 |
|
|
if (val < 1 || val > MAX_TCP_KEEPIDLE)
|
| 2318 |
|
|
err = -EINVAL;
|
| 2319 |
|
|
else {
|
| 2320 |
|
|
tp->keepalive_time = val * HZ;
|
| 2321 |
|
|
if (sk->keepopen && !((1<<sk->state)&(TCPF_CLOSE|TCPF_LISTEN))) {
|
| 2322 |
|
|
__u32 elapsed = tcp_time_stamp - tp->rcv_tstamp;
|
| 2323 |
|
|
if (tp->keepalive_time > elapsed)
|
| 2324 |
|
|
elapsed = tp->keepalive_time - elapsed;
|
| 2325 |
|
|
else
|
| 2326 |
|
|
elapsed = 0;
|
| 2327 |
|
|
tcp_reset_keepalive_timer(sk, elapsed);
|
| 2328 |
|
|
}
|
| 2329 |
|
|
}
|
| 2330 |
|
|
break;
|
| 2331 |
|
|
case TCP_KEEPINTVL:
|
| 2332 |
|
|
if (val < 1 || val > MAX_TCP_KEEPINTVL)
|
| 2333 |
|
|
err = -EINVAL;
|
| 2334 |
|
|
else
|
| 2335 |
|
|
tp->keepalive_intvl = val * HZ;
|
| 2336 |
|
|
break;
|
| 2337 |
|
|
case TCP_KEEPCNT:
|
| 2338 |
|
|
if (val < 1 || val > MAX_TCP_KEEPCNT)
|
| 2339 |
|
|
err = -EINVAL;
|
| 2340 |
|
|
else
|
| 2341 |
|
|
tp->keepalive_probes = val;
|
| 2342 |
|
|
break;
|
| 2343 |
|
|
case TCP_SYNCNT:
|
| 2344 |
|
|
if (val < 1 || val > MAX_TCP_SYNCNT)
|
| 2345 |
|
|
err = -EINVAL;
|
| 2346 |
|
|
else
|
| 2347 |
|
|
tp->syn_retries = val;
|
| 2348 |
|
|
break;
|
| 2349 |
|
|
|
| 2350 |
|
|
case TCP_LINGER2:
|
| 2351 |
|
|
if (val < 0)
|
| 2352 |
|
|
tp->linger2 = -1;
|
| 2353 |
|
|
else if (val > sysctl_tcp_fin_timeout/HZ)
|
| 2354 |
|
|
tp->linger2 = 0;
|
| 2355 |
|
|
else
|
| 2356 |
|
|
tp->linger2 = val*HZ;
|
| 2357 |
|
|
break;
|
| 2358 |
|
|
|
| 2359 |
|
|
case TCP_DEFER_ACCEPT:
|
| 2360 |
|
|
tp->defer_accept = 0;
|
| 2361 |
|
|
if (val > 0) {
|
| 2362 |
|
|
/* Translate value in seconds to number of retransmits */
|
| 2363 |
|
|
while (tp->defer_accept < 32 && val > ((TCP_TIMEOUT_INIT/HZ)<<tp->defer_accept))
|
| 2364 |
|
|
tp->defer_accept++;
|
| 2365 |
|
|
tp->defer_accept++;
|
| 2366 |
|
|
}
|
| 2367 |
|
|
break;
|
| 2368 |
|
|
|
| 2369 |
|
|
case TCP_WINDOW_CLAMP:
|
| 2370 |
|
|
if (val==0) {
|
| 2371 |
|
|
if (sk->state != TCP_CLOSE) {
|
| 2372 |
|
|
err = -EINVAL;
|
| 2373 |
|
|
break;
|
| 2374 |
|
|
}
|
| 2375 |
|
|
tp->window_clamp = 0;
|
| 2376 |
|
|
} else {
|
| 2377 |
|
|
tp->window_clamp = val<SOCK_MIN_RCVBUF/2 ?
|
| 2378 |
|
|
SOCK_MIN_RCVBUF/2 : val;
|
| 2379 |
|
|
}
|
| 2380 |
|
|
break;
|
| 2381 |
|
|
|
| 2382 |
|
|
case TCP_QUICKACK:
|
| 2383 |
|
|
if (!val) {
|
| 2384 |
|
|
tp->ack.pingpong = 1;
|
| 2385 |
|
|
} else {
|
| 2386 |
|
|
tp->ack.pingpong = 0;
|
| 2387 |
|
|
if ((1<<sk->state)&(TCPF_ESTABLISHED|TCPF_CLOSE_WAIT) &&
|
| 2388 |
|
|
tcp_ack_scheduled(tp)) {
|
| 2389 |
|
|
tp->ack.pending |= TCP_ACK_PUSHED;
|
| 2390 |
|
|
cleanup_rbuf(sk, 1);
|
| 2391 |
|
|
if (!(val & 1))
|
| 2392 |
|
|
tp->ack.pingpong = 1;
|
| 2393 |
|
|
}
|
| 2394 |
|
|
}
|
| 2395 |
|
|
break;
|
| 2396 |
|
|
|
| 2397 |
|
|
default:
|
| 2398 |
|
|
err = -ENOPROTOOPT;
|
| 2399 |
|
|
break;
|
| 2400 |
|
|
};
|
| 2401 |
|
|
release_sock(sk);
|
| 2402 |
|
|
return err;
|
| 2403 |
|
|
}
|
| 2404 |
|
|
|
| 2405 |
|
|
int tcp_getsockopt(struct sock *sk, int level, int optname, char *optval,
|
| 2406 |
|
|
int *optlen)
|
| 2407 |
|
|
{
|
| 2408 |
|
|
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
|
| 2409 |
|
|
int val, len;
|
| 2410 |
|
|
|
| 2411 |
|
|
if(level != SOL_TCP)
|
| 2412 |
|
|
return tp->af_specific->getsockopt(sk, level, optname,
|
| 2413 |
|
|
optval, optlen);
|
| 2414 |
|
|
|
| 2415 |
|
|
if(get_user(len,optlen))
|
| 2416 |
|
|
return -EFAULT;
|
| 2417 |
|
|
|
| 2418 |
|
|
len = min_t(unsigned int, len, sizeof(int));
|
| 2419 |
|
|
|
| 2420 |
|
|
if(len < 0)
|
| 2421 |
|
|
return -EINVAL;
|
| 2422 |
|
|
|
| 2423 |
|
|
switch(optname) {
|
| 2424 |
|
|
case TCP_MAXSEG:
|
| 2425 |
|
|
val = tp->mss_cache;
|
| 2426 |
|
|
if (val == 0 && ((1<<sk->state)&(TCPF_CLOSE|TCPF_LISTEN)))
|
| 2427 |
|
|
val = tp->user_mss;
|
| 2428 |
|
|
break;
|
| 2429 |
|
|
case TCP_NODELAY:
|
| 2430 |
|
|
val = (tp->nonagle == 1);
|
| 2431 |
|
|
break;
|
| 2432 |
|
|
case TCP_CORK:
|
| 2433 |
|
|
val = (tp->nonagle == 2);
|
| 2434 |
|
|
break;
|
| 2435 |
|
|
case TCP_KEEPIDLE:
|
| 2436 |
|
|
val = (tp->keepalive_time ? : sysctl_tcp_keepalive_time)/HZ;
|
| 2437 |
|
|
break;
|
| 2438 |
|
|
case TCP_KEEPINTVL:
|
| 2439 |
|
|
val = (tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl)/HZ;
|
| 2440 |
|
|
break;
|
| 2441 |
|
|
case TCP_KEEPCNT:
|
| 2442 |
|
|
val = tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
|
| 2443 |
|
|
break;
|
| 2444 |
|
|
case TCP_SYNCNT:
|
| 2445 |
|
|
val = tp->syn_retries ? : sysctl_tcp_syn_retries;
|
| 2446 |
|
|
break;
|
| 2447 |
|
|
case TCP_LINGER2:
|
| 2448 |
|
|
val = tp->linger2;
|
| 2449 |
|
|
if (val >= 0)
|
| 2450 |
|
|
val = (val ? : sysctl_tcp_fin_timeout)/HZ;
|
| 2451 |
|
|
break;
|
| 2452 |
|
|
case TCP_DEFER_ACCEPT:
|
| 2453 |
|
|
val = tp->defer_accept == 0 ? 0 : ((TCP_TIMEOUT_INIT/HZ)<<(tp->defer_accept-1));
|
| 2454 |
|
|
break;
|
| 2455 |
|
|
case TCP_WINDOW_CLAMP:
|
| 2456 |
|
|
val = tp->window_clamp;
|
| 2457 |
|
|
break;
|
| 2458 |
|
|
case TCP_INFO:
|
| 2459 |
|
|
{
|
| 2460 |
|
|
struct tcp_info info;
|
| 2461 |
|
|
u32 now = tcp_time_stamp;
|
| 2462 |
|
|
|
| 2463 |
|
|
if(get_user(len,optlen))
|
| 2464 |
|
|
return -EFAULT;
|
| 2465 |
|
|
info.tcpi_state = sk->state;
|
| 2466 |
|
|
info.tcpi_ca_state = tp->ca_state;
|
| 2467 |
|
|
info.tcpi_retransmits = tp->retransmits;
|
| 2468 |
|
|
info.tcpi_probes = tp->probes_out;
|
| 2469 |
|
|
info.tcpi_backoff = tp->backoff;
|
| 2470 |
|
|
info.tcpi_options = 0;
|
| 2471 |
|
|
if (tp->tstamp_ok)
|
| 2472 |
|
|
info.tcpi_options |= TCPI_OPT_TIMESTAMPS;
|
| 2473 |
|
|
if (tp->sack_ok)
|
| 2474 |
|
|
info.tcpi_options |= TCPI_OPT_SACK;
|
| 2475 |
|
|
if (tp->wscale_ok) {
|
| 2476 |
|
|
info.tcpi_options |= TCPI_OPT_WSCALE;
|
| 2477 |
|
|
info.tcpi_snd_wscale = tp->snd_wscale;
|
| 2478 |
|
|
info.tcpi_rcv_wscale = tp->rcv_wscale;
|
| 2479 |
|
|
} else {
|
| 2480 |
|
|
info.tcpi_snd_wscale = 0;
|
| 2481 |
|
|
info.tcpi_rcv_wscale = 0;
|
| 2482 |
|
|
}
|
| 2483 |
|
|
if (tp->ecn_flags&TCP_ECN_OK)
|
| 2484 |
|
|
info.tcpi_options |= TCPI_OPT_ECN;
|
| 2485 |
|
|
|
| 2486 |
|
|
info.tcpi_rto = (1000000*tp->rto)/HZ;
|
| 2487 |
|
|
info.tcpi_ato = (1000000*tp->ack.ato)/HZ;
|
| 2488 |
|
|
info.tcpi_snd_mss = tp->mss_cache;
|
| 2489 |
|
|
info.tcpi_rcv_mss = tp->ack.rcv_mss;
|
| 2490 |
|
|
|
| 2491 |
|
|
info.tcpi_unacked = tp->packets_out;
|
| 2492 |
|
|
info.tcpi_sacked = tp->sacked_out;
|
| 2493 |
|
|
info.tcpi_lost = tp->lost_out;
|
| 2494 |
|
|
info.tcpi_retrans = tp->retrans_out;
|
| 2495 |
|
|
info.tcpi_fackets = tp->fackets_out;
|
| 2496 |
|
|
|
| 2497 |
|
|
info.tcpi_last_data_sent = ((now - tp->lsndtime)*1000)/HZ;
|
| 2498 |
|
|
info.tcpi_last_ack_sent = 0;
|
| 2499 |
|
|
info.tcpi_last_data_recv = ((now - tp->ack.lrcvtime)*1000)/HZ;
|
| 2500 |
|
|
info.tcpi_last_ack_recv = ((now - tp->rcv_tstamp)*1000)/HZ;
|
| 2501 |
|
|
|
| 2502 |
|
|
info.tcpi_pmtu = tp->pmtu_cookie;
|
| 2503 |
|
|
info.tcpi_rcv_ssthresh = tp->rcv_ssthresh;
|
| 2504 |
|
|
info.tcpi_rtt = ((1000000*tp->srtt)/HZ)>>3;
|
| 2505 |
|
|
info.tcpi_rttvar = ((1000000*tp->mdev)/HZ)>>2;
|
| 2506 |
|
|
info.tcpi_snd_ssthresh = tp->snd_ssthresh;
|
| 2507 |
|
|
info.tcpi_snd_cwnd = tp->snd_cwnd;
|
| 2508 |
|
|
info.tcpi_advmss = tp->advmss;
|
| 2509 |
|
|
info.tcpi_reordering = tp->reordering;
|
| 2510 |
|
|
|
| 2511 |
|
|
len = min_t(unsigned int, len, sizeof(info));
|
| 2512 |
|
|
if(put_user(len, optlen))
|
| 2513 |
|
|
return -EFAULT;
|
| 2514 |
|
|
if(copy_to_user(optval, &info,len))
|
| 2515 |
|
|
return -EFAULT;
|
| 2516 |
|
|
return 0;
|
| 2517 |
|
|
}
|
| 2518 |
|
|
case TCP_QUICKACK:
|
| 2519 |
|
|
val = !tp->ack.pingpong;
|
| 2520 |
|
|
break;
|
| 2521 |
|
|
default:
|
| 2522 |
|
|
return -ENOPROTOOPT;
|
| 2523 |
|
|
};
|
| 2524 |
|
|
|
| 2525 |
|
|
if(put_user(len, optlen))
|
| 2526 |
|
|
return -EFAULT;
|
| 2527 |
|
|
if(copy_to_user(optval, &val,len))
|
| 2528 |
|
|
return -EFAULT;
|
| 2529 |
|
|
return 0;
|
| 2530 |
|
|
}
|
| 2531 |
|
|
|
| 2532 |
|
|
|
| 2533 |
|
|
extern void __skb_cb_too_small_for_tcp(int, int);
|
| 2534 |
|
|
extern void tcpdiag_init(void);
|
| 2535 |
|
|
|
| 2536 |
|
|
void __init tcp_init(void)
|
| 2537 |
|
|
{
|
| 2538 |
|
|
struct sk_buff *skb = NULL;
|
| 2539 |
|
|
unsigned long goal;
|
| 2540 |
|
|
int order, i;
|
| 2541 |
|
|
|
| 2542 |
|
|
if(sizeof(struct tcp_skb_cb) > sizeof(skb->cb))
|
| 2543 |
|
|
__skb_cb_too_small_for_tcp(sizeof(struct tcp_skb_cb),
|
| 2544 |
|
|
sizeof(skb->cb));
|
| 2545 |
|
|
|
| 2546 |
|
|
tcp_openreq_cachep = kmem_cache_create("tcp_open_request",
|
| 2547 |
|
|
sizeof(struct open_request),
|
| 2548 |
|
|
0, SLAB_HWCACHE_ALIGN,
|
| 2549 |
|
|
NULL, NULL);
|
| 2550 |
|
|
if(!tcp_openreq_cachep)
|
| 2551 |
|
|
panic("tcp_init: Cannot alloc open_request cache.");
|
| 2552 |
|
|
|
| 2553 |
|
|
tcp_bucket_cachep = kmem_cache_create("tcp_bind_bucket",
|
| 2554 |
|
|
sizeof(struct tcp_bind_bucket),
|
| 2555 |
|
|
0, SLAB_HWCACHE_ALIGN,
|
| 2556 |
|
|
NULL, NULL);
|
| 2557 |
|
|
if(!tcp_bucket_cachep)
|
| 2558 |
|
|
panic("tcp_init: Cannot alloc tcp_bind_bucket cache.");
|
| 2559 |
|
|
|
| 2560 |
|
|
tcp_timewait_cachep = kmem_cache_create("tcp_tw_bucket",
|
| 2561 |
|
|
sizeof(struct tcp_tw_bucket),
|
| 2562 |
|
|
0, SLAB_HWCACHE_ALIGN,
|
| 2563 |
|
|
NULL, NULL);
|
| 2564 |
|
|
if(!tcp_timewait_cachep)
|
| 2565 |
|
|
panic("tcp_init: Cannot alloc tcp_tw_bucket cache.");
|
| 2566 |
|
|
|
| 2567 |
|
|
/* Size and allocate the main established and bind bucket
|
| 2568 |
|
|
* hash tables.
|
| 2569 |
|
|
*
|
| 2570 |
|
|
* The methodology is similar to that of the buffer cache.
|
| 2571 |
|
|
*/
|
| 2572 |
|
|
if (num_physpages >= (128 * 1024))
|
| 2573 |
|
|
goal = num_physpages >> (21 - PAGE_SHIFT);
|
| 2574 |
|
|
else
|
| 2575 |
|
|
goal = num_physpages >> (23 - PAGE_SHIFT);
|
| 2576 |
|
|
|
| 2577 |
|
|
for(order = 0; (1UL << order) < goal; order++)
|
| 2578 |
|
|
;
|
| 2579 |
|
|
do {
|
| 2580 |
|
|
tcp_ehash_size = (1UL << order) * PAGE_SIZE /
|
| 2581 |
|
|
sizeof(struct tcp_ehash_bucket);
|
| 2582 |
|
|
tcp_ehash_size >>= 1;
|
| 2583 |
|
|
while (tcp_ehash_size & (tcp_ehash_size-1))
|
| 2584 |
|
|
tcp_ehash_size--;
|
| 2585 |
|
|
tcp_ehash = (struct tcp_ehash_bucket *)
|
| 2586 |
|
|
__get_free_pages(GFP_ATOMIC, order);
|
| 2587 |
|
|
} while (tcp_ehash == NULL && --order > 0);
|
| 2588 |
|
|
|
| 2589 |
|
|
if (!tcp_ehash)
|
| 2590 |
|
|
panic("Failed to allocate TCP established hash table\n");
|
| 2591 |
|
|
for (i = 0; i < (tcp_ehash_size<<1); i++) {
|
| 2592 |
|
|
tcp_ehash[i].lock = RW_LOCK_UNLOCKED;
|
| 2593 |
|
|
tcp_ehash[i].chain = NULL;
|
| 2594 |
|
|
}
|
| 2595 |
|
|
|
| 2596 |
|
|
do {
|
| 2597 |
|
|
tcp_bhash_size = (1UL << order) * PAGE_SIZE /
|
| 2598 |
|
|
sizeof(struct tcp_bind_hashbucket);
|
| 2599 |
|
|
if ((tcp_bhash_size > (64 * 1024)) && order > 0)
|
| 2600 |
|
|
continue;
|
| 2601 |
|
|
tcp_bhash = (struct tcp_bind_hashbucket *)
|
| 2602 |
|
|
__get_free_pages(GFP_ATOMIC, order);
|
| 2603 |
|
|
} while (tcp_bhash == NULL && --order >= 0);
|
| 2604 |
|
|
|
| 2605 |
|
|
if (!tcp_bhash)
|
| 2606 |
|
|
panic("Failed to allocate TCP bind hash table\n");
|
| 2607 |
|
|
for (i = 0; i < tcp_bhash_size; i++) {
|
| 2608 |
|
|
tcp_bhash[i].lock = SPIN_LOCK_UNLOCKED;
|
| 2609 |
|
|
tcp_bhash[i].chain = NULL;
|
| 2610 |
|
|
}
|
| 2611 |
|
|
|
| 2612 |
|
|
/* Try to be a bit smarter and adjust defaults depending
|
| 2613 |
|
|
* on available memory.
|
| 2614 |
|
|
*/
|
| 2615 |
|
|
if (order > 4) {
|
| 2616 |
|
|
sysctl_local_port_range[0] = 32768;
|
| 2617 |
|
|
sysctl_local_port_range[1] = 61000;
|
| 2618 |
|
|
sysctl_tcp_max_tw_buckets = 180000;
|
| 2619 |
|
|
sysctl_tcp_max_orphans = 4096<<(order-4);
|
| 2620 |
|
|
sysctl_max_syn_backlog = 1024;
|
| 2621 |
|
|
} else if (order < 3) {
|
| 2622 |
|
|
sysctl_local_port_range[0] = 1024*(3-order);
|
| 2623 |
|
|
sysctl_tcp_max_tw_buckets >>= (3-order);
|
| 2624 |
|
|
sysctl_tcp_max_orphans >>= (3-order);
|
| 2625 |
|
|
sysctl_max_syn_backlog = 128;
|
| 2626 |
|
|
}
|
| 2627 |
|
|
tcp_port_rover = sysctl_local_port_range[0] - 1;
|
| 2628 |
|
|
|
| 2629 |
|
|
sysctl_tcp_mem[0] = 768<<order;
|
| 2630 |
|
|
sysctl_tcp_mem[1] = 1024<<order;
|
| 2631 |
|
|
sysctl_tcp_mem[2] = 1536<<order;
|
| 2632 |
|
|
if (sysctl_tcp_mem[2] - sysctl_tcp_mem[1] > 512)
|
| 2633 |
|
|
sysctl_tcp_mem[1] = sysctl_tcp_mem[2] - 512;
|
| 2634 |
|
|
if (sysctl_tcp_mem[1] - sysctl_tcp_mem[0] > 512)
|
| 2635 |
|
|
sysctl_tcp_mem[0] = sysctl_tcp_mem[1] - 512;
|
| 2636 |
|
|
|
| 2637 |
|
|
if (order < 3) {
|
| 2638 |
|
|
sysctl_tcp_wmem[2] = 64*1024;
|
| 2639 |
|
|
sysctl_tcp_rmem[0] = PAGE_SIZE;
|
| 2640 |
|
|
sysctl_tcp_rmem[1] = 43689;
|
| 2641 |
|
|
sysctl_tcp_rmem[2] = 2*43689;
|
| 2642 |
|
|
}
|
| 2643 |
|
|
|
| 2644 |
|
|
printk(KERN_INFO "TCP: Hash tables configured (established %d bind %d)\n",
|
| 2645 |
|
|
tcp_ehash_size<<1, tcp_bhash_size);
|
| 2646 |
|
|
|
| 2647 |
|
|
(void) tcp_mib_init();
|
| 2648 |
|
|
tcpdiag_init();
|
| 2649 |
|
|
}
|
