| 1 |
1629 |
jcastillo |
/*
|
| 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: @(#)tcp.c 1.0.16 05/25/93
|
| 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. select
|
| 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 select() 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(), select() 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 selecting 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 : 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 stat 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 it's 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 |
|
|
* Eric Schenk : Changed receiver side silly window
|
| 198 |
|
|
* avoidance algorithm to BSD style
|
| 199 |
|
|
* algorithm. This doubles throughput
|
| 200 |
|
|
* against machines running Solaris,
|
| 201 |
|
|
* and seems to result in general
|
| 202 |
|
|
* improvement.
|
| 203 |
|
|
* Stefan Magdalinski : adjusted tcp_readable() to fix FIONREAD
|
| 204 |
|
|
* Willy Konynenberg : Transparent proxying support.
|
| 205 |
|
|
* Theodore Ts'o : Do secure TCP sequence numbers.
|
| 206 |
|
|
* David S. Miller : New socket lookup architecture for ISS.
|
| 207 |
|
|
* This code is dedicated to John Dyson.
|
| 208 |
|
|
* Elliot Poger : Added support for SO_BINDTODEVICE.
|
| 209 |
|
|
*
|
| 210 |
|
|
* To Fix:
|
| 211 |
|
|
* Fast path the code. Two things here - fix the window calculation
|
| 212 |
|
|
* so it doesn't iterate over the queue, also spot packets with no funny
|
| 213 |
|
|
* options arriving in order and process directly.
|
| 214 |
|
|
*
|
| 215 |
|
|
* Rewrite output state machine to use a single queue.
|
| 216 |
|
|
* Speed up input assembly algorithm.
|
| 217 |
|
|
* RFC1323 - PAWS and window scaling. PAWS is required for IPv6 so we
|
| 218 |
|
|
* could do with it working on IPv4
|
| 219 |
|
|
* User settable/learned rtt/max window/mtu
|
| 220 |
|
|
*
|
| 221 |
|
|
* Change the fundamental structure to a single send queue maintained
|
| 222 |
|
|
* by TCP (removing the bogus ip stuff [thus fixing mtu drops on
|
| 223 |
|
|
* active routes too]). Cut the queue off in tcp_retransmit/
|
| 224 |
|
|
* tcp_transmit.
|
| 225 |
|
|
* Change the receive queue to assemble as it goes. This lets us
|
| 226 |
|
|
* dispose of most of tcp_sequence, half of tcp_ack and chunks of
|
| 227 |
|
|
* tcp_data/tcp_read as well as the window shrink crud.
|
| 228 |
|
|
* Separate out duplicated code - tcp_alloc_skb, tcp_build_ack
|
| 229 |
|
|
* tcp_queue_skb seem obvious routines to extract.
|
| 230 |
|
|
*
|
| 231 |
|
|
* This program is free software; you can redistribute it and/or
|
| 232 |
|
|
* modify it under the terms of the GNU General Public License
|
| 233 |
|
|
* as published by the Free Software Foundation; either version
|
| 234 |
|
|
* 2 of the License, or(at your option) any later version.
|
| 235 |
|
|
*
|
| 236 |
|
|
* Description of States:
|
| 237 |
|
|
*
|
| 238 |
|
|
* TCP_SYN_SENT sent a connection request, waiting for ack
|
| 239 |
|
|
*
|
| 240 |
|
|
* TCP_SYN_RECV received a connection request, sent ack,
|
| 241 |
|
|
* waiting for final ack in three-way handshake.
|
| 242 |
|
|
*
|
| 243 |
|
|
* TCP_ESTABLISHED connection established
|
| 244 |
|
|
*
|
| 245 |
|
|
* TCP_FIN_WAIT1 our side has shutdown, waiting to complete
|
| 246 |
|
|
* transmission of remaining buffered data
|
| 247 |
|
|
*
|
| 248 |
|
|
* TCP_FIN_WAIT2 all buffered data sent, waiting for remote
|
| 249 |
|
|
* to shutdown
|
| 250 |
|
|
*
|
| 251 |
|
|
* TCP_CLOSING both sides have shutdown but we still have
|
| 252 |
|
|
* data we have to finish sending
|
| 253 |
|
|
*
|
| 254 |
|
|
* TCP_TIME_WAIT timeout to catch resent junk before entering
|
| 255 |
|
|
* closed, can only be entered from FIN_WAIT2
|
| 256 |
|
|
* or CLOSING. Required because the other end
|
| 257 |
|
|
* may not have gotten our last ACK causing it
|
| 258 |
|
|
* to retransmit the data packet (which we ignore)
|
| 259 |
|
|
*
|
| 260 |
|
|
* TCP_CLOSE_WAIT remote side has shutdown and is waiting for
|
| 261 |
|
|
* us to finish writing our data and to shutdown
|
| 262 |
|
|
* (we have to close() to move on to LAST_ACK)
|
| 263 |
|
|
*
|
| 264 |
|
|
* TCP_LAST_ACK out side has shutdown after remote has
|
| 265 |
|
|
* shutdown. There may still be data in our
|
| 266 |
|
|
* buffer that we have to finish sending
|
| 267 |
|
|
*
|
| 268 |
|
|
* TCP_CLOSE socket is finished
|
| 269 |
|
|
*/
|
| 270 |
|
|
|
| 271 |
|
|
/*
|
| 272 |
|
|
* RFC1122 status:
|
| 273 |
|
|
* NOTE: I'm not going to be doing comments in the code for this one except
|
| 274 |
|
|
* for violations and the like. tcp.c is just too big... If I say something
|
| 275 |
|
|
* "does?" or "doesn't?", it means I'm not sure, and will have to hash it out
|
| 276 |
|
|
* with Alan. -- MS 950903
|
| 277 |
|
|
*
|
| 278 |
|
|
* Use of PSH (4.2.2.2)
|
| 279 |
|
|
* MAY aggregate data sent without the PSH flag. (does)
|
| 280 |
|
|
* MAY queue data received without the PSH flag. (does)
|
| 281 |
|
|
* SHOULD collapse successive PSH flags when it packetizes data. (doesn't)
|
| 282 |
|
|
* MAY implement PSH on send calls. (doesn't, thus:)
|
| 283 |
|
|
* MUST NOT buffer data indefinitely (doesn't [1 second])
|
| 284 |
|
|
* MUST set PSH on last segment (does)
|
| 285 |
|
|
* MAY pass received PSH to application layer (doesn't)
|
| 286 |
|
|
* SHOULD send maximum-sized segment whenever possible. (almost always does)
|
| 287 |
|
|
*
|
| 288 |
|
|
* Window Size (4.2.2.3, 4.2.2.16)
|
| 289 |
|
|
* MUST treat window size as an unsigned number (does)
|
| 290 |
|
|
* SHOULD treat window size as a 32-bit number (does not)
|
| 291 |
|
|
* MUST NOT shrink window once it is offered (does not normally)
|
| 292 |
|
|
*
|
| 293 |
|
|
* Urgent Pointer (4.2.2.4)
|
| 294 |
|
|
* **MUST point urgent pointer to last byte of urgent data (not right
|
| 295 |
|
|
* after). (doesn't, to be like BSD)
|
| 296 |
|
|
* MUST inform application layer asynchronously of incoming urgent
|
| 297 |
|
|
* data. (does)
|
| 298 |
|
|
* MUST provide application with means of determining the amount of
|
| 299 |
|
|
* urgent data pending. (does)
|
| 300 |
|
|
* **MUST support urgent data sequence of arbitrary length. (doesn't, but
|
| 301 |
|
|
* it's sort of tricky to fix, as urg_ptr is a 16-bit quantity)
|
| 302 |
|
|
* [Follows BSD 1 byte of urgent data]
|
| 303 |
|
|
*
|
| 304 |
|
|
* TCP Options (4.2.2.5)
|
| 305 |
|
|
* MUST be able to receive TCP options in any segment. (does)
|
| 306 |
|
|
* MUST ignore unsupported options (does)
|
| 307 |
|
|
*
|
| 308 |
|
|
* Maximum Segment Size Option (4.2.2.6)
|
| 309 |
|
|
* MUST implement both sending and receiving MSS. (does)
|
| 310 |
|
|
* SHOULD send an MSS with every SYN where receive MSS != 536 (MAY send
|
| 311 |
|
|
* it always). (does, even when MSS == 536, which is legal)
|
| 312 |
|
|
* MUST assume MSS == 536 if no MSS received at connection setup (does)
|
| 313 |
|
|
* MUST calculate "effective send MSS" correctly:
|
| 314 |
|
|
* min(physical_MTU, remote_MSS+20) - sizeof(tcphdr) - sizeof(ipopts)
|
| 315 |
|
|
* (does - but allows operator override)
|
| 316 |
|
|
*
|
| 317 |
|
|
* TCP Checksum (4.2.2.7)
|
| 318 |
|
|
* MUST generate and check TCP checksum. (does)
|
| 319 |
|
|
*
|
| 320 |
|
|
* Initial Sequence Number Selection (4.2.2.8)
|
| 321 |
|
|
* MUST use the RFC 793 clock selection mechanism. (doesn't, but it's
|
| 322 |
|
|
* OK: RFC 793 specifies a 250KHz clock, while we use 1MHz, which is
|
| 323 |
|
|
* necessary for 10Mbps networks - and harder than BSD to spoof!)
|
| 324 |
|
|
*
|
| 325 |
|
|
* Simultaneous Open Attempts (4.2.2.10)
|
| 326 |
|
|
* MUST support simultaneous open attempts (does)
|
| 327 |
|
|
*
|
| 328 |
|
|
* Recovery from Old Duplicate SYN (4.2.2.11)
|
| 329 |
|
|
* MUST keep track of active vs. passive open (does)
|
| 330 |
|
|
*
|
| 331 |
|
|
* RST segment (4.2.2.12)
|
| 332 |
|
|
* SHOULD allow an RST segment to contain data (does, but doesn't do
|
| 333 |
|
|
* anything with it, which is standard)
|
| 334 |
|
|
*
|
| 335 |
|
|
* Closing a Connection (4.2.2.13)
|
| 336 |
|
|
* MUST inform application of whether connection was closed by RST or
|
| 337 |
|
|
* normal close. (does)
|
| 338 |
|
|
* MAY allow "half-duplex" close (treat connection as closed for the
|
| 339 |
|
|
* local app, even before handshake is done). (does)
|
| 340 |
|
|
* MUST linger in TIME_WAIT for 2 * MSL (does)
|
| 341 |
|
|
*
|
| 342 |
|
|
* Retransmission Timeout (4.2.2.15)
|
| 343 |
|
|
* MUST implement Jacobson's slow start and congestion avoidance
|
| 344 |
|
|
* stuff. (does)
|
| 345 |
|
|
*
|
| 346 |
|
|
* Probing Zero Windows (4.2.2.17)
|
| 347 |
|
|
* MUST support probing of zero windows. (does)
|
| 348 |
|
|
* MAY keep offered window closed indefinitely. (does)
|
| 349 |
|
|
* MUST allow remote window to stay closed indefinitely. (does)
|
| 350 |
|
|
*
|
| 351 |
|
|
* Passive Open Calls (4.2.2.18)
|
| 352 |
|
|
* MUST NOT let new passive open affect other connections. (doesn't)
|
| 353 |
|
|
* MUST support passive opens (LISTENs) concurrently. (does)
|
| 354 |
|
|
*
|
| 355 |
|
|
* Time to Live (4.2.2.19)
|
| 356 |
|
|
* MUST make TCP TTL configurable. (does - IP_TTL option)
|
| 357 |
|
|
*
|
| 358 |
|
|
* Event Processing (4.2.2.20)
|
| 359 |
|
|
* SHOULD queue out-of-order segments. (does)
|
| 360 |
|
|
* MUST aggregate ACK segments whenever possible. (does but badly)
|
| 361 |
|
|
*
|
| 362 |
|
|
* Retransmission Timeout Calculation (4.2.3.1)
|
| 363 |
|
|
* MUST implement Karn's algorithm and Jacobson's algorithm for RTO
|
| 364 |
|
|
* calculation. (does, or at least explains them in the comments 8*b)
|
| 365 |
|
|
* SHOULD initialize RTO to 0 and RTT to 3. (does)
|
| 366 |
|
|
*
|
| 367 |
|
|
* When to Send an ACK Segment (4.2.3.2)
|
| 368 |
|
|
* SHOULD implement delayed ACK. (does)
|
| 369 |
|
|
* MUST keep ACK delay < 0.5 sec. (does)
|
| 370 |
|
|
*
|
| 371 |
|
|
* When to Send a Window Update (4.2.3.3)
|
| 372 |
|
|
* MUST implement receiver-side SWS. (does)
|
| 373 |
|
|
*
|
| 374 |
|
|
* When to Send Data (4.2.3.4)
|
| 375 |
|
|
* MUST implement sender-side SWS. (does)
|
| 376 |
|
|
* SHOULD implement Nagle algorithm. (does)
|
| 377 |
|
|
*
|
| 378 |
|
|
* TCP Connection Failures (4.2.3.5)
|
| 379 |
|
|
* MUST handle excessive retransmissions "properly" (see the RFC). (does)
|
| 380 |
|
|
* SHOULD inform application layer of soft errors. (does)
|
| 381 |
|
|
*
|
| 382 |
|
|
* TCP Keep-Alives (4.2.3.6)
|
| 383 |
|
|
* MAY provide keep-alives. (does)
|
| 384 |
|
|
* MUST make keep-alives configurable on a per-connection basis. (does)
|
| 385 |
|
|
* MUST default to no keep-alives. (does)
|
| 386 |
|
|
* **MUST make keep-alive interval configurable. (doesn't)
|
| 387 |
|
|
* **MUST make default keep-alive interval > 2 hours. (doesn't)
|
| 388 |
|
|
* MUST NOT interpret failure to ACK keep-alive packet as dead
|
| 389 |
|
|
* connection. (doesn't)
|
| 390 |
|
|
* SHOULD send keep-alive with no data. (does)
|
| 391 |
|
|
*
|
| 392 |
|
|
* TCP Multihoming (4.2.3.7)
|
| 393 |
|
|
* MUST get source address from IP layer before sending first
|
| 394 |
|
|
* SYN. (does)
|
| 395 |
|
|
* MUST use same local address for all segments of a connection. (does)
|
| 396 |
|
|
*
|
| 397 |
|
|
* IP Options (4.2.3.8)
|
| 398 |
|
|
* MUST ignore unsupported IP options. (does)
|
| 399 |
|
|
* MAY support Time Stamp and Record Route. (does)
|
| 400 |
|
|
* MUST allow application to specify a source route. (does)
|
| 401 |
|
|
* MUST allow received Source Route option to set route for all future
|
| 402 |
|
|
* segments on this connection. (does not (security issues))
|
| 403 |
|
|
*
|
| 404 |
|
|
* ICMP messages (4.2.3.9)
|
| 405 |
|
|
* MUST act on ICMP errors. (does)
|
| 406 |
|
|
* MUST slow transmission upon receipt of a Source Quench. (does)
|
| 407 |
|
|
* MUST NOT abort connection upon receipt of soft Destination
|
| 408 |
|
|
* Unreachables (0, 1, 5), Time Exceededs and Parameter
|
| 409 |
|
|
* Problems. (doesn't)
|
| 410 |
|
|
* SHOULD report soft Destination Unreachables etc. to the
|
| 411 |
|
|
* application. (does)
|
| 412 |
|
|
* SHOULD abort connection upon receipt of hard Destination Unreachable
|
| 413 |
|
|
* messages (2, 3, 4). (does)
|
| 414 |
|
|
*
|
| 415 |
|
|
* Remote Address Validation (4.2.3.10)
|
| 416 |
|
|
* MUST reject as an error OPEN for invalid remote IP address. (does)
|
| 417 |
|
|
* MUST ignore SYN with invalid source address. (does)
|
| 418 |
|
|
* MUST silently discard incoming SYN for broadcast/multicast
|
| 419 |
|
|
* address. (does)
|
| 420 |
|
|
*
|
| 421 |
|
|
* Asynchronous Reports (4.2.4.1)
|
| 422 |
|
|
* MUST provide mechanism for reporting soft errors to application
|
| 423 |
|
|
* layer. (does)
|
| 424 |
|
|
*
|
| 425 |
|
|
* Type of Service (4.2.4.2)
|
| 426 |
|
|
* MUST allow application layer to set Type of Service. (does IP_TOS)
|
| 427 |
|
|
*
|
| 428 |
|
|
* (Whew. -- MS 950903)
|
| 429 |
|
|
**/
|
| 430 |
|
|
|
| 431 |
|
|
#include <linux/config.h>
|
| 432 |
|
|
#include <linux/types.h>
|
| 433 |
|
|
#include <linux/fcntl.h>
|
| 434 |
|
|
#include <linux/random.h>
|
| 435 |
|
|
|
| 436 |
|
|
#include <net/icmp.h>
|
| 437 |
|
|
#include <net/tcp.h>
|
| 438 |
|
|
|
| 439 |
|
|
#include <asm/segment.h>
|
| 440 |
|
|
|
| 441 |
|
|
unsigned long seq_offset;
|
| 442 |
|
|
struct tcp_mib tcp_statistics;
|
| 443 |
|
|
|
| 444 |
|
|
/* This is for sockets with full identity only. Sockets here will always
|
| 445 |
|
|
* be without wildcards and will have the following invariant:
|
| 446 |
|
|
* TCP_ESTABLISHED <= sk->state < TCP_CLOSE
|
| 447 |
|
|
*/
|
| 448 |
|
|
struct sock *tcp_established_hash[TCP_HTABLE_SIZE];
|
| 449 |
|
|
|
| 450 |
|
|
/* All sockets in TCP_LISTEN state will be in here. This is the only table
|
| 451 |
|
|
* where wildcard'd TCP sockets can exist. Hash function here is just local
|
| 452 |
|
|
* port number. XXX Fix or we'll lose with thousands of IP aliases...
|
| 453 |
|
|
*/
|
| 454 |
|
|
struct sock *tcp_listening_hash[TCP_LHTABLE_SIZE];
|
| 455 |
|
|
|
| 456 |
|
|
/* Ok, let's try this, I give up, we do need a local binding
|
| 457 |
|
|
* TCP hash as well as the others for fast bind/connect.
|
| 458 |
|
|
*/
|
| 459 |
|
|
struct sock *tcp_bound_hash[TCP_BHTABLE_SIZE];
|
| 460 |
|
|
|
| 461 |
|
|
extern struct sock *tcp_v4_lookup(u32 saddr, u16 sport, u32 daddr, u16 dport);
|
| 462 |
|
|
|
| 463 |
|
|
static int tcp_v4_verify_bind(struct sock *sk, unsigned short snum)
|
| 464 |
|
|
{
|
| 465 |
|
|
struct sock *sk2;
|
| 466 |
|
|
int retval = 0, sk_reuse = sk->reuse;
|
| 467 |
|
|
|
| 468 |
|
|
SOCKHASH_LOCK();
|
| 469 |
|
|
sk2 = tcp_bound_hash[tcp_bhashfn(snum)];
|
| 470 |
|
|
for(; sk2 != NULL; sk2 = sk2->bind_next) {
|
| 471 |
|
|
if((sk2->num == snum) && (sk2 != sk)) {
|
| 472 |
|
|
unsigned char state = sk2->state;
|
| 473 |
|
|
int sk2_reuse = sk2->reuse;
|
| 474 |
|
|
|
| 475 |
|
|
/* Two sockets can be bound to the same port if they're
|
| 476 |
|
|
* bound to different interfaces... */
|
| 477 |
|
|
if (sk->bound_device != sk2->bound_device)
|
| 478 |
|
|
continue;
|
| 479 |
|
|
|
| 480 |
|
|
if(!sk2->rcv_saddr || !sk->rcv_saddr) {
|
| 481 |
|
|
if((!sk2_reuse) ||
|
| 482 |
|
|
(!sk_reuse) ||
|
| 483 |
|
|
(state == TCP_LISTEN)) {
|
| 484 |
|
|
retval = 1;
|
| 485 |
|
|
break;
|
| 486 |
|
|
}
|
| 487 |
|
|
} else if(sk2->rcv_saddr == sk->rcv_saddr) {
|
| 488 |
|
|
if((!sk_reuse) ||
|
| 489 |
|
|
(!sk2_reuse) ||
|
| 490 |
|
|
(state == TCP_LISTEN)) {
|
| 491 |
|
|
retval = 1;
|
| 492 |
|
|
break;
|
| 493 |
|
|
}
|
| 494 |
|
|
}
|
| 495 |
|
|
}
|
| 496 |
|
|
}
|
| 497 |
|
|
SOCKHASH_UNLOCK();
|
| 498 |
|
|
|
| 499 |
|
|
return retval;
|
| 500 |
|
|
}
|
| 501 |
|
|
|
| 502 |
|
|
static __inline__ int tcp_lport_inuse(int num)
|
| 503 |
|
|
{
|
| 504 |
|
|
struct sock *sk = tcp_bound_hash[tcp_bhashfn(num)];
|
| 505 |
|
|
|
| 506 |
|
|
for(; sk != NULL; sk = sk->bind_next) {
|
| 507 |
|
|
if(sk->num == num)
|
| 508 |
|
|
return 1;
|
| 509 |
|
|
}
|
| 510 |
|
|
return 0;
|
| 511 |
|
|
}
|
| 512 |
|
|
|
| 513 |
|
|
/* Find a "good" local port, this is family independant.
|
| 514 |
|
|
* There are several strategies working in unison here to
|
| 515 |
|
|
* get the best possible performance. The current socket
|
| 516 |
|
|
* load is kept track of, if it is zero there is a strong
|
| 517 |
|
|
* likely hood that there is a zero length chain we will
|
| 518 |
|
|
* find with a small amount of searching, else the load is
|
| 519 |
|
|
* what we shoot for when the chains all have at least
|
| 520 |
|
|
* one entry. The base helps us walk the chains in an
|
| 521 |
|
|
* order such that a good chain is found as quickly as possible. -DaveM
|
| 522 |
|
|
*/
|
| 523 |
|
|
unsigned short tcp_good_socknum(void)
|
| 524 |
|
|
{
|
| 525 |
|
|
static int start = PROT_SOCK;
|
| 526 |
|
|
static int binding_contour = 0;
|
| 527 |
|
|
int best = 0;
|
| 528 |
|
|
int size = 32767; /* a big num. */
|
| 529 |
|
|
int retval = 0, i, end, bc;
|
| 530 |
|
|
|
| 531 |
|
|
SOCKHASH_LOCK();
|
| 532 |
|
|
i = tcp_bhashfn(start);
|
| 533 |
|
|
end = i + TCP_BHTABLE_SIZE;
|
| 534 |
|
|
bc = binding_contour;
|
| 535 |
|
|
do {
|
| 536 |
|
|
struct sock *sk = tcp_bound_hash[i&(TCP_BHTABLE_SIZE-1)];
|
| 537 |
|
|
if(!sk) {
|
| 538 |
|
|
/* find the smallest value no smaller than start
|
| 539 |
|
|
* that has this hash value.
|
| 540 |
|
|
*/
|
| 541 |
|
|
retval = tcp_bhashnext(start-1,i&(TCP_BHTABLE_SIZE-1));
|
| 542 |
|
|
|
| 543 |
|
|
/* Check for decreasing load. */
|
| 544 |
|
|
if (bc != 0)
|
| 545 |
|
|
binding_contour = 0;
|
| 546 |
|
|
goto done;
|
| 547 |
|
|
} else {
|
| 548 |
|
|
int j = 0;
|
| 549 |
|
|
do { sk = sk->bind_next; } while (++j < size && sk);
|
| 550 |
|
|
if (j < size) {
|
| 551 |
|
|
best = i&(TCP_BHTABLE_SIZE-1);
|
| 552 |
|
|
size = j;
|
| 553 |
|
|
if (bc && size <= bc) {
|
| 554 |
|
|
i = best;
|
| 555 |
|
|
goto verify;
|
| 556 |
|
|
}
|
| 557 |
|
|
}
|
| 558 |
|
|
}
|
| 559 |
|
|
} while(++i != end);
|
| 560 |
|
|
i = best;
|
| 561 |
|
|
|
| 562 |
|
|
/* Socket load is increasing, adjust our load average. */
|
| 563 |
|
|
binding_contour = size;
|
| 564 |
|
|
verify:
|
| 565 |
|
|
if (size < binding_contour)
|
| 566 |
|
|
binding_contour = size;
|
| 567 |
|
|
|
| 568 |
|
|
retval = tcp_bhashnext(start-1,i);
|
| 569 |
|
|
|
| 570 |
|
|
best = retval; /* mark the starting point to avoid infinite loops */
|
| 571 |
|
|
while(tcp_lport_inuse(retval)) {
|
| 572 |
|
|
retval = tcp_bhashnext(retval,i);
|
| 573 |
|
|
if (retval > 32767) /* Upper bound */
|
| 574 |
|
|
retval = tcp_bhashnext(PROT_SOCK,i);
|
| 575 |
|
|
if (retval == best) {
|
| 576 |
|
|
/* This hash chain is full. No answer. */
|
| 577 |
|
|
retval = 0;
|
| 578 |
|
|
break;
|
| 579 |
|
|
}
|
| 580 |
|
|
}
|
| 581 |
|
|
|
| 582 |
|
|
done:
|
| 583 |
|
|
start = (retval + 1);
|
| 584 |
|
|
if (start > 32767 || start < PROT_SOCK)
|
| 585 |
|
|
start = PROT_SOCK;
|
| 586 |
|
|
SOCKHASH_UNLOCK();
|
| 587 |
|
|
|
| 588 |
|
|
return retval;
|
| 589 |
|
|
}
|
| 590 |
|
|
|
| 591 |
|
|
void tcp_v4_hash(struct sock *sk)
|
| 592 |
|
|
{
|
| 593 |
|
|
unsigned char state;
|
| 594 |
|
|
|
| 595 |
|
|
SOCKHASH_LOCK();
|
| 596 |
|
|
state = sk->state;
|
| 597 |
|
|
if(state != TCP_CLOSE || !sk->dead) {
|
| 598 |
|
|
struct sock **skp;
|
| 599 |
|
|
|
| 600 |
|
|
if(state == TCP_LISTEN)
|
| 601 |
|
|
skp = &tcp_listening_hash[tcp_sk_listen_hashfn(sk)];
|
| 602 |
|
|
else
|
| 603 |
|
|
skp = &tcp_established_hash[tcp_sk_hashfn(sk)];
|
| 604 |
|
|
|
| 605 |
|
|
if((sk->next = *skp) != NULL)
|
| 606 |
|
|
(*skp)->pprev = &sk->next;
|
| 607 |
|
|
*skp = sk;
|
| 608 |
|
|
sk->pprev = skp;
|
| 609 |
|
|
tcp_sk_bindify(sk);
|
| 610 |
|
|
}
|
| 611 |
|
|
SOCKHASH_UNLOCK();
|
| 612 |
|
|
}
|
| 613 |
|
|
|
| 614 |
|
|
void tcp_v4_unhash(struct sock *sk)
|
| 615 |
|
|
{
|
| 616 |
|
|
SOCKHASH_LOCK();
|
| 617 |
|
|
if(sk->pprev) {
|
| 618 |
|
|
if(sk->next)
|
| 619 |
|
|
sk->next->pprev = sk->pprev;
|
| 620 |
|
|
*sk->pprev = sk->next;
|
| 621 |
|
|
sk->pprev = NULL;
|
| 622 |
|
|
tcp_sk_unbindify(sk);
|
| 623 |
|
|
}
|
| 624 |
|
|
SOCKHASH_UNLOCK();
|
| 625 |
|
|
}
|
| 626 |
|
|
|
| 627 |
|
|
void tcp_v4_rehash(struct sock *sk)
|
| 628 |
|
|
{
|
| 629 |
|
|
unsigned char state;
|
| 630 |
|
|
|
| 631 |
|
|
SOCKHASH_LOCK();
|
| 632 |
|
|
state = sk->state;
|
| 633 |
|
|
if(sk->pprev) {
|
| 634 |
|
|
if(sk->next)
|
| 635 |
|
|
sk->next->pprev = sk->pprev;
|
| 636 |
|
|
*sk->pprev = sk->next;
|
| 637 |
|
|
sk->pprev = NULL;
|
| 638 |
|
|
tcp_sk_unbindify(sk);
|
| 639 |
|
|
}
|
| 640 |
|
|
if(state != TCP_CLOSE || !sk->dead) {
|
| 641 |
|
|
struct sock **skp;
|
| 642 |
|
|
|
| 643 |
|
|
if(state == TCP_LISTEN)
|
| 644 |
|
|
skp = &tcp_listening_hash[tcp_sk_listen_hashfn(sk)];
|
| 645 |
|
|
else
|
| 646 |
|
|
skp = &tcp_established_hash[tcp_sk_hashfn(sk)];
|
| 647 |
|
|
|
| 648 |
|
|
if((sk->next = *skp) != NULL)
|
| 649 |
|
|
(*skp)->pprev = &sk->next;
|
| 650 |
|
|
*skp = sk;
|
| 651 |
|
|
sk->pprev = skp;
|
| 652 |
|
|
tcp_sk_bindify(sk);
|
| 653 |
|
|
}
|
| 654 |
|
|
SOCKHASH_UNLOCK();
|
| 655 |
|
|
}
|
| 656 |
|
|
|
| 657 |
|
|
static void tcp_close(struct sock *sk, unsigned long timeout);
|
| 658 |
|
|
|
| 659 |
|
|
/*
|
| 660 |
|
|
* Find someone to 'accept'. Must be called with
|
| 661 |
|
|
* the socket locked or with interrupts disabled
|
| 662 |
|
|
*/
|
| 663 |
|
|
|
| 664 |
|
|
static struct sk_buff *tcp_find_established(struct sock *s)
|
| 665 |
|
|
{
|
| 666 |
|
|
struct sk_buff *p=skb_peek(&s->receive_queue);
|
| 667 |
|
|
if(p==NULL)
|
| 668 |
|
|
return NULL;
|
| 669 |
|
|
do
|
| 670 |
|
|
{
|
| 671 |
|
|
if(p->sk->state == TCP_ESTABLISHED || p->sk->state >= TCP_FIN_WAIT1)
|
| 672 |
|
|
return p;
|
| 673 |
|
|
p=p->next;
|
| 674 |
|
|
}
|
| 675 |
|
|
while(p!=(struct sk_buff *)&s->receive_queue);
|
| 676 |
|
|
return NULL;
|
| 677 |
|
|
}
|
| 678 |
|
|
|
| 679 |
|
|
/*
|
| 680 |
|
|
* This routine closes sockets which have been at least partially
|
| 681 |
|
|
* opened, but not yet accepted. Currently it is only called by
|
| 682 |
|
|
* tcp_close, and timeout mirrors the value there.
|
| 683 |
|
|
*/
|
| 684 |
|
|
|
| 685 |
|
|
static void tcp_close_pending (struct sock *sk)
|
| 686 |
|
|
{
|
| 687 |
|
|
struct sk_buff *skb;
|
| 688 |
|
|
|
| 689 |
|
|
while ((skb = skb_dequeue(&sk->receive_queue)) != NULL)
|
| 690 |
|
|
{
|
| 691 |
|
|
tcp_close(skb->sk, 0);
|
| 692 |
|
|
kfree_skb(skb, FREE_READ);
|
| 693 |
|
|
}
|
| 694 |
|
|
return;
|
| 695 |
|
|
}
|
| 696 |
|
|
|
| 697 |
|
|
/*
|
| 698 |
|
|
* Enter the time wait state.
|
| 699 |
|
|
*/
|
| 700 |
|
|
|
| 701 |
|
|
void tcp_time_wait(struct sock *sk)
|
| 702 |
|
|
{
|
| 703 |
|
|
tcp_set_state(sk,TCP_TIME_WAIT);
|
| 704 |
|
|
sk->shutdown = SHUTDOWN_MASK;
|
| 705 |
|
|
if (!sk->dead)
|
| 706 |
|
|
sk->state_change(sk);
|
| 707 |
|
|
tcp_reset_msl_timer(sk, TIME_CLOSE, TCP_TIMEWAIT_LEN);
|
| 708 |
|
|
}
|
| 709 |
|
|
|
| 710 |
|
|
|
| 711 |
|
|
/*
|
| 712 |
|
|
* This routine is called by the ICMP module when it gets some
|
| 713 |
|
|
* sort of error condition. If err < 0 then the socket should
|
| 714 |
|
|
* be closed and the error returned to the user. If err > 0
|
| 715 |
|
|
* it's just the icmp type << 8 | icmp code. After adjustment
|
| 716 |
|
|
* header points to the first 8 bytes of the tcp header. We need
|
| 717 |
|
|
* to find the appropriate port.
|
| 718 |
|
|
*/
|
| 719 |
|
|
|
| 720 |
|
|
void tcp_err(int type, int code, unsigned char *header, __u32 daddr,
|
| 721 |
|
|
__u32 saddr, struct inet_protocol *protocol, int len)
|
| 722 |
|
|
{
|
| 723 |
|
|
struct tcphdr *th = (struct tcphdr *)header;
|
| 724 |
|
|
struct sock *sk;
|
| 725 |
|
|
|
| 726 |
|
|
/*
|
| 727 |
|
|
* This one is _WRONG_. FIXME urgently.
|
| 728 |
|
|
*/
|
| 729 |
|
|
#ifndef CONFIG_NO_PATH_MTU_DISCOVERY
|
| 730 |
|
|
struct iphdr *iph=(struct iphdr *)(header-sizeof(struct iphdr));
|
| 731 |
|
|
#endif
|
| 732 |
|
|
th =(struct tcphdr *)header;
|
| 733 |
|
|
|
| 734 |
|
|
if (len < 8) /* NOT sizeof(struct tcphdr) */
|
| 735 |
|
|
return;
|
| 736 |
|
|
|
| 737 |
|
|
sk = tcp_v4_lookup(daddr, th->dest, saddr, th->source);
|
| 738 |
|
|
if (sk == NULL)
|
| 739 |
|
|
return;
|
| 740 |
|
|
|
| 741 |
|
|
if (type == ICMP_SOURCE_QUENCH)
|
| 742 |
|
|
{
|
| 743 |
|
|
/* Current practice says these frames are bad, plus the drops
|
| 744 |
|
|
will account right anyway. If we act on this we stall doubly */
|
| 745 |
|
|
return;
|
| 746 |
|
|
}
|
| 747 |
|
|
|
| 748 |
|
|
if (type == ICMP_PARAMETERPROB)
|
| 749 |
|
|
{
|
| 750 |
|
|
sk->err=EPROTO;
|
| 751 |
|
|
sk->error_report(sk);
|
| 752 |
|
|
}
|
| 753 |
|
|
|
| 754 |
|
|
#ifndef CONFIG_NO_PATH_MTU_DISCOVERY
|
| 755 |
|
|
if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
|
| 756 |
|
|
{
|
| 757 |
|
|
struct rtable * rt;
|
| 758 |
|
|
/*
|
| 759 |
|
|
* Ugly trick to pass MTU to protocol layer.
|
| 760 |
|
|
* Really we should add argument "info" to error handler.
|
| 761 |
|
|
*/
|
| 762 |
|
|
unsigned short new_mtu = ntohs(iph->id);
|
| 763 |
|
|
|
| 764 |
|
|
if ((rt = sk->ip_route_cache) != NULL)
|
| 765 |
|
|
if (rt->rt_mtu > new_mtu)
|
| 766 |
|
|
rt->rt_mtu = new_mtu;
|
| 767 |
|
|
|
| 768 |
|
|
/*
|
| 769 |
|
|
* FIXME::
|
| 770 |
|
|
* Not the nicest of fixes: Lose a MTU update if the socket is
|
| 771 |
|
|
* locked this instant. Not the right answer but will be best
|
| 772 |
|
|
* for the production fix. Make 2.1 work right!
|
| 773 |
|
|
*/
|
| 774 |
|
|
|
| 775 |
|
|
if (sk->mtu > new_mtu - sizeof(struct iphdr) - sizeof(struct tcphdr)
|
| 776 |
|
|
&& new_mtu > sizeof(struct iphdr)+sizeof(struct tcphdr) && !sk->users)
|
| 777 |
|
|
sk->mtu = new_mtu - sizeof(struct iphdr) - sizeof(struct tcphdr);
|
| 778 |
|
|
|
| 779 |
|
|
return;
|
| 780 |
|
|
}
|
| 781 |
|
|
#endif
|
| 782 |
|
|
|
| 783 |
|
|
/*
|
| 784 |
|
|
* If we've already connected we will keep trying
|
| 785 |
|
|
* until we time out, or the user gives up.
|
| 786 |
|
|
*/
|
| 787 |
|
|
|
| 788 |
|
|
if(code<=NR_ICMP_UNREACH)
|
| 789 |
|
|
{
|
| 790 |
|
|
if(icmp_err_convert[code].fatal || sk->state == TCP_SYN_SENT || sk->state == TCP_SYN_RECV)
|
| 791 |
|
|
{
|
| 792 |
|
|
sk->err = icmp_err_convert[code].errno;
|
| 793 |
|
|
if (sk->state == TCP_SYN_SENT || sk->state == TCP_SYN_RECV)
|
| 794 |
|
|
{
|
| 795 |
|
|
tcp_statistics.TcpAttemptFails++;
|
| 796 |
|
|
tcp_set_state(sk,TCP_CLOSE);
|
| 797 |
|
|
sk->error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
|
| 798 |
|
|
}
|
| 799 |
|
|
}
|
| 800 |
|
|
else /* Only an error on timeout */
|
| 801 |
|
|
sk->err_soft = icmp_err_convert[code].errno;
|
| 802 |
|
|
}
|
| 803 |
|
|
}
|
| 804 |
|
|
|
| 805 |
|
|
|
| 806 |
|
|
/*
|
| 807 |
|
|
* Walk down the receive queue counting readable data until we hit the end or we find a gap
|
| 808 |
|
|
* in the received data queue (ie a frame missing that needs sending to us). Not
|
| 809 |
|
|
* sorting using two queues as data arrives makes life so much harder.
|
| 810 |
|
|
*/
|
| 811 |
|
|
|
| 812 |
|
|
static int tcp_readable(struct sock *sk)
|
| 813 |
|
|
{
|
| 814 |
|
|
unsigned long counted;
|
| 815 |
|
|
unsigned long amount;
|
| 816 |
|
|
struct sk_buff *skb;
|
| 817 |
|
|
int sum;
|
| 818 |
|
|
unsigned long flags;
|
| 819 |
|
|
|
| 820 |
|
|
if(sk && sk->debug)
|
| 821 |
|
|
printk("tcp_readable: %p - ",sk);
|
| 822 |
|
|
|
| 823 |
|
|
save_flags(flags);
|
| 824 |
|
|
cli();
|
| 825 |
|
|
if (sk == NULL || (skb = skb_peek(&sk->receive_queue)) == NULL)
|
| 826 |
|
|
{
|
| 827 |
|
|
restore_flags(flags);
|
| 828 |
|
|
if(sk && sk->debug)
|
| 829 |
|
|
printk("empty\n");
|
| 830 |
|
|
return(0);
|
| 831 |
|
|
}
|
| 832 |
|
|
|
| 833 |
|
|
counted = sk->copied_seq; /* Where we are at the moment */
|
| 834 |
|
|
amount = 0;
|
| 835 |
|
|
|
| 836 |
|
|
/*
|
| 837 |
|
|
* Do until a push or until we are out of data.
|
| 838 |
|
|
*/
|
| 839 |
|
|
|
| 840 |
|
|
do
|
| 841 |
|
|
{
|
| 842 |
|
|
if (before(counted, skb->seq)) /* Found a hole so stops here */
|
| 843 |
|
|
break;
|
| 844 |
|
|
sum = skb->len - (counted - skb->seq); /* Length - header but start from where we are up to (avoid overlaps) */
|
| 845 |
|
|
if (skb->h.th->syn)
|
| 846 |
|
|
sum++;
|
| 847 |
|
|
if (sum > 0)
|
| 848 |
|
|
{ /* Add it up, move on */
|
| 849 |
|
|
amount += sum;
|
| 850 |
|
|
if (skb->h.th->syn)
|
| 851 |
|
|
amount--;
|
| 852 |
|
|
counted += sum;
|
| 853 |
|
|
}
|
| 854 |
|
|
/*
|
| 855 |
|
|
* Don't count urg data ... but do it in the right place!
|
| 856 |
|
|
* Consider: "old_data (ptr is here) URG PUSH data"
|
| 857 |
|
|
* The old code would stop at the first push because
|
| 858 |
|
|
* it counted the urg (amount==1) and then does amount--
|
| 859 |
|
|
* *after* the loop. This means tcp_readable() always
|
| 860 |
|
|
* returned zero if any URG PUSH was in the queue, even
|
| 861 |
|
|
* though there was normal data available. If we subtract
|
| 862 |
|
|
* the urg data right here, we even get it to work for more
|
| 863 |
|
|
* than one URG PUSH skb without normal data.
|
| 864 |
|
|
* This means that select() finally works now with urg data
|
| 865 |
|
|
* in the queue. Note that rlogin was never affected
|
| 866 |
|
|
* because it doesn't use select(); it uses two processes
|
| 867 |
|
|
* and a blocking read(). And the queue scan in tcp_read()
|
| 868 |
|
|
* was correct. Mike <pall@rz.uni-karlsruhe.de>
|
| 869 |
|
|
*/
|
| 870 |
|
|
if (skb->h.th->urg)
|
| 871 |
|
|
amount--; /* don't count urg data */
|
| 872 |
|
|
/* if (amount && skb->h.th->psh) break;*/
|
| 873 |
|
|
skb = skb->next;
|
| 874 |
|
|
}
|
| 875 |
|
|
while(skb != (struct sk_buff *)&sk->receive_queue);
|
| 876 |
|
|
|
| 877 |
|
|
restore_flags(flags);
|
| 878 |
|
|
if(sk->debug)
|
| 879 |
|
|
printk("got %lu bytes.\n",amount);
|
| 880 |
|
|
return(amount);
|
| 881 |
|
|
}
|
| 882 |
|
|
|
| 883 |
|
|
/*
|
| 884 |
|
|
* LISTEN is a special case for select..
|
| 885 |
|
|
*/
|
| 886 |
|
|
static int tcp_listen_select(struct sock *sk, int sel_type, select_table *wait)
|
| 887 |
|
|
{
|
| 888 |
|
|
if (sel_type == SEL_IN) {
|
| 889 |
|
|
struct sk_buff * skb;
|
| 890 |
|
|
|
| 891 |
|
|
lock_sock(sk);
|
| 892 |
|
|
skb = tcp_find_established(sk);
|
| 893 |
|
|
release_sock(sk);
|
| 894 |
|
|
if (skb)
|
| 895 |
|
|
return 1;
|
| 896 |
|
|
select_wait(sk->sleep,wait);
|
| 897 |
|
|
return 0;
|
| 898 |
|
|
}
|
| 899 |
|
|
return 0;
|
| 900 |
|
|
}
|
| 901 |
|
|
|
| 902 |
|
|
|
| 903 |
|
|
/*
|
| 904 |
|
|
* Wait for a TCP event.
|
| 905 |
|
|
*
|
| 906 |
|
|
* Note that we don't need to lock the socket, as the upper select layers
|
| 907 |
|
|
* take care of normal races (between the test and the event) and we don't
|
| 908 |
|
|
* go look at any of the socket buffers directly.
|
| 909 |
|
|
*/
|
| 910 |
|
|
static int tcp_select(struct sock *sk, int sel_type, select_table *wait)
|
| 911 |
|
|
{
|
| 912 |
|
|
if (sk->state == TCP_LISTEN)
|
| 913 |
|
|
return tcp_listen_select(sk, sel_type, wait);
|
| 914 |
|
|
|
| 915 |
|
|
switch(sel_type) {
|
| 916 |
|
|
case SEL_IN:
|
| 917 |
|
|
if (sk->err)
|
| 918 |
|
|
return 1;
|
| 919 |
|
|
if (sk->state == TCP_SYN_SENT || sk->state == TCP_SYN_RECV)
|
| 920 |
|
|
break;
|
| 921 |
|
|
|
| 922 |
|
|
if (sk->shutdown & RCV_SHUTDOWN)
|
| 923 |
|
|
return 1;
|
| 924 |
|
|
|
| 925 |
|
|
if (sk->acked_seq == sk->copied_seq)
|
| 926 |
|
|
break;
|
| 927 |
|
|
|
| 928 |
|
|
if (sk->urg_seq != sk->copied_seq ||
|
| 929 |
|
|
sk->acked_seq != sk->copied_seq+1 ||
|
| 930 |
|
|
sk->urginline || !sk->urg_data)
|
| 931 |
|
|
return 1;
|
| 932 |
|
|
break;
|
| 933 |
|
|
|
| 934 |
|
|
case SEL_OUT:
|
| 935 |
|
|
if (sk->err)
|
| 936 |
|
|
return 1;
|
| 937 |
|
|
if (sk->shutdown & SEND_SHUTDOWN)
|
| 938 |
|
|
return 0;
|
| 939 |
|
|
if (sk->state == TCP_SYN_SENT || sk->state == TCP_SYN_RECV)
|
| 940 |
|
|
break;
|
| 941 |
|
|
if (sk->wmem_alloc*2 > sk->sndbuf)
|
| 942 |
|
|
break;
|
| 943 |
|
|
return 1;
|
| 944 |
|
|
|
| 945 |
|
|
case SEL_EX:
|
| 946 |
|
|
if (sk->urg_data & URG_VALID)
|
| 947 |
|
|
return 1;
|
| 948 |
|
|
break;
|
| 949 |
|
|
}
|
| 950 |
|
|
select_wait(sk->sleep, wait);
|
| 951 |
|
|
return 0;
|
| 952 |
|
|
}
|
| 953 |
|
|
|
| 954 |
|
|
int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
|
| 955 |
|
|
{
|
| 956 |
|
|
int err;
|
| 957 |
|
|
switch(cmd)
|
| 958 |
|
|
{
|
| 959 |
|
|
|
| 960 |
|
|
case TIOCINQ:
|
| 961 |
|
|
#ifdef FIXME /* FIXME: */
|
| 962 |
|
|
case FIONREAD:
|
| 963 |
|
|
#endif
|
| 964 |
|
|
{
|
| 965 |
|
|
unsigned long amount;
|
| 966 |
|
|
|
| 967 |
|
|
if (sk->state == TCP_LISTEN)
|
| 968 |
|
|
return(-EINVAL);
|
| 969 |
|
|
|
| 970 |
|
|
lock_sock(sk);
|
| 971 |
|
|
amount = tcp_readable(sk);
|
| 972 |
|
|
release_sock(sk);
|
| 973 |
|
|
err=verify_area(VERIFY_WRITE,(void *)arg, sizeof(int));
|
| 974 |
|
|
if(err)
|
| 975 |
|
|
return err;
|
| 976 |
|
|
put_user(amount, (int *)arg);
|
| 977 |
|
|
return(0);
|
| 978 |
|
|
}
|
| 979 |
|
|
case SIOCATMARK:
|
| 980 |
|
|
{
|
| 981 |
|
|
int answ = sk->urg_data && sk->urg_seq == sk->copied_seq;
|
| 982 |
|
|
|
| 983 |
|
|
err = verify_area(VERIFY_WRITE,(void *) arg, sizeof(int));
|
| 984 |
|
|
if (err)
|
| 985 |
|
|
return err;
|
| 986 |
|
|
put_user(answ,(int *) arg);
|
| 987 |
|
|
return(0);
|
| 988 |
|
|
}
|
| 989 |
|
|
case TIOCOUTQ:
|
| 990 |
|
|
{
|
| 991 |
|
|
unsigned long amount;
|
| 992 |
|
|
|
| 993 |
|
|
if (sk->state == TCP_LISTEN) return(-EINVAL);
|
| 994 |
|
|
amount = sock_wspace(sk);
|
| 995 |
|
|
err=verify_area(VERIFY_WRITE,(void *)arg, sizeof(int));
|
| 996 |
|
|
if(err)
|
| 997 |
|
|
return err;
|
| 998 |
|
|
put_user(amount, (int *)arg);
|
| 999 |
|
|
return(0);
|
| 1000 |
|
|
}
|
| 1001 |
|
|
default:
|
| 1002 |
|
|
return(-EINVAL);
|
| 1003 |
|
|
}
|
| 1004 |
|
|
}
|
| 1005 |
|
|
|
| 1006 |
|
|
|
| 1007 |
|
|
/*
|
| 1008 |
|
|
* This routine computes a TCP checksum.
|
| 1009 |
|
|
*
|
| 1010 |
|
|
* Modified January 1995 from a go-faster DOS routine by
|
| 1011 |
|
|
* Jorge Cwik <jorge@laser.satlink.net>
|
| 1012 |
|
|
*/
|
| 1013 |
|
|
#undef DEBUG_TCP_CHECK
|
| 1014 |
|
|
void tcp_send_check(struct tcphdr *th, unsigned long saddr,
|
| 1015 |
|
|
unsigned long daddr, int len, struct sk_buff *skb)
|
| 1016 |
|
|
{
|
| 1017 |
|
|
#ifdef DEBUG_TCP_CHECK
|
| 1018 |
|
|
u16 check;
|
| 1019 |
|
|
#endif
|
| 1020 |
|
|
th->check = 0;
|
| 1021 |
|
|
th->check = tcp_check(th, len, saddr, daddr,
|
| 1022 |
|
|
csum_partial((char *)th,sizeof(*th),skb->csum));
|
| 1023 |
|
|
|
| 1024 |
|
|
#ifdef DEBUG_TCP_CHECK
|
| 1025 |
|
|
check = th->check;
|
| 1026 |
|
|
th->check = 0;
|
| 1027 |
|
|
th->check = tcp_check(th, len, saddr, daddr,
|
| 1028 |
|
|
csum_partial((char *)th,len,0));
|
| 1029 |
|
|
if (check != th->check) {
|
| 1030 |
|
|
static int count = 0;
|
| 1031 |
|
|
if (++count < 10) {
|
| 1032 |
|
|
printk("Checksum %x (%x) from %p\n", th->check, check,
|
| 1033 |
|
|
(&th)[-1]);
|
| 1034 |
|
|
printk("TCP=<off:%d a:%d s:%d f:%d>\n", th->doff*4, th->ack, th->syn, th->fin);
|
| 1035 |
|
|
}
|
| 1036 |
|
|
}
|
| 1037 |
|
|
#endif
|
| 1038 |
|
|
}
|
| 1039 |
|
|
|
| 1040 |
|
|
|
| 1041 |
|
|
/*
|
| 1042 |
|
|
* This routine builds a generic TCP header.
|
| 1043 |
|
|
*/
|
| 1044 |
|
|
|
| 1045 |
|
|
static inline int tcp_build_header(struct tcphdr *th, struct sock *sk, int push)
|
| 1046 |
|
|
{
|
| 1047 |
|
|
memcpy(th,(void *) &(sk->dummy_th), sizeof(*th));
|
| 1048 |
|
|
th->psh = (push == 0) ? 1 : 0;
|
| 1049 |
|
|
th->seq = htonl(sk->write_seq);
|
| 1050 |
|
|
th->ack_seq = htonl(sk->acked_seq);
|
| 1051 |
|
|
th->window = htons(tcp_select_window(sk));
|
| 1052 |
|
|
|
| 1053 |
|
|
return(sizeof(*th));
|
| 1054 |
|
|
}
|
| 1055 |
|
|
|
| 1056 |
|
|
/*
|
| 1057 |
|
|
* Wait for a socket to get into the connected state
|
| 1058 |
|
|
*/
|
| 1059 |
|
|
static void wait_for_tcp_connect(struct sock * sk)
|
| 1060 |
|
|
{
|
| 1061 |
|
|
release_sock(sk);
|
| 1062 |
|
|
cli();
|
| 1063 |
|
|
if (sk->state != TCP_ESTABLISHED && sk->state != TCP_CLOSE_WAIT && sk->err == 0)
|
| 1064 |
|
|
{
|
| 1065 |
|
|
interruptible_sleep_on(sk->sleep);
|
| 1066 |
|
|
}
|
| 1067 |
|
|
sti();
|
| 1068 |
|
|
lock_sock(sk);
|
| 1069 |
|
|
}
|
| 1070 |
|
|
|
| 1071 |
|
|
static inline int tcp_memory_free(struct sock *sk)
|
| 1072 |
|
|
{
|
| 1073 |
|
|
return sk->wmem_alloc < sk->sndbuf;
|
| 1074 |
|
|
}
|
| 1075 |
|
|
|
| 1076 |
|
|
/*
|
| 1077 |
|
|
* Wait for more memory for a socket
|
| 1078 |
|
|
*/
|
| 1079 |
|
|
static void wait_for_tcp_memory(struct sock * sk)
|
| 1080 |
|
|
{
|
| 1081 |
|
|
release_sock(sk);
|
| 1082 |
|
|
if (!tcp_memory_free(sk)) {
|
| 1083 |
|
|
struct wait_queue wait = { current, NULL };
|
| 1084 |
|
|
|
| 1085 |
|
|
sk->socket->flags &= ~SO_NOSPACE;
|
| 1086 |
|
|
add_wait_queue(sk->sleep, &wait);
|
| 1087 |
|
|
for (;;) {
|
| 1088 |
|
|
if (current->signal & ~current->blocked)
|
| 1089 |
|
|
break;
|
| 1090 |
|
|
current->state = TASK_INTERRUPTIBLE;
|
| 1091 |
|
|
if (tcp_memory_free(sk))
|
| 1092 |
|
|
break;
|
| 1093 |
|
|
if (sk->shutdown & SEND_SHUTDOWN)
|
| 1094 |
|
|
break;
|
| 1095 |
|
|
if (sk->err)
|
| 1096 |
|
|
break;
|
| 1097 |
|
|
schedule();
|
| 1098 |
|
|
}
|
| 1099 |
|
|
current->state = TASK_RUNNING;
|
| 1100 |
|
|
remove_wait_queue(sk->sleep, &wait);
|
| 1101 |
|
|
}
|
| 1102 |
|
|
lock_sock(sk);
|
| 1103 |
|
|
}
|
| 1104 |
|
|
|
| 1105 |
|
|
|
| 1106 |
|
|
/*
|
| 1107 |
|
|
* This routine copies from a user buffer into a socket,
|
| 1108 |
|
|
* and starts the transmit system.
|
| 1109 |
|
|
*/
|
| 1110 |
|
|
|
| 1111 |
|
|
static int do_tcp_sendmsg(struct sock *sk,
|
| 1112 |
|
|
int iovlen, struct iovec *iov,
|
| 1113 |
|
|
int len, int nonblock, int flags)
|
| 1114 |
|
|
{
|
| 1115 |
|
|
int copied = 0;
|
| 1116 |
|
|
struct device *dev = NULL;
|
| 1117 |
|
|
|
| 1118 |
|
|
/*
|
| 1119 |
|
|
* Wait for a connection to finish.
|
| 1120 |
|
|
*/
|
| 1121 |
|
|
while (sk->state != TCP_ESTABLISHED && sk->state != TCP_CLOSE_WAIT)
|
| 1122 |
|
|
{
|
| 1123 |
|
|
if (sk->err)
|
| 1124 |
|
|
return sock_error(sk);
|
| 1125 |
|
|
|
| 1126 |
|
|
if (sk->state != TCP_SYN_SENT && sk->state != TCP_SYN_RECV)
|
| 1127 |
|
|
{
|
| 1128 |
|
|
if (sk->keepopen)
|
| 1129 |
|
|
send_sig(SIGPIPE, current, 0);
|
| 1130 |
|
|
return -EPIPE;
|
| 1131 |
|
|
}
|
| 1132 |
|
|
|
| 1133 |
|
|
if (nonblock)
|
| 1134 |
|
|
return -EAGAIN;
|
| 1135 |
|
|
|
| 1136 |
|
|
if (current->signal & ~current->blocked)
|
| 1137 |
|
|
return -ERESTARTSYS;
|
| 1138 |
|
|
|
| 1139 |
|
|
wait_for_tcp_connect(sk);
|
| 1140 |
|
|
}
|
| 1141 |
|
|
|
| 1142 |
|
|
/*
|
| 1143 |
|
|
* Ok commence sending
|
| 1144 |
|
|
*/
|
| 1145 |
|
|
|
| 1146 |
|
|
while (--iovlen >= 0)
|
| 1147 |
|
|
{
|
| 1148 |
|
|
int seglen=iov->iov_len;
|
| 1149 |
|
|
unsigned char * from=iov->iov_base;
|
| 1150 |
|
|
iov++;
|
| 1151 |
|
|
|
| 1152 |
|
|
while(seglen > 0)
|
| 1153 |
|
|
{
|
| 1154 |
|
|
int copy, delay;
|
| 1155 |
|
|
int tmp;
|
| 1156 |
|
|
struct sk_buff *skb;
|
| 1157 |
|
|
|
| 1158 |
|
|
/*
|
| 1159 |
|
|
* Stop on errors
|
| 1160 |
|
|
*/
|
| 1161 |
|
|
if (sk->err)
|
| 1162 |
|
|
{
|
| 1163 |
|
|
if (copied)
|
| 1164 |
|
|
return copied;
|
| 1165 |
|
|
return sock_error(sk);
|
| 1166 |
|
|
}
|
| 1167 |
|
|
|
| 1168 |
|
|
/*
|
| 1169 |
|
|
* Make sure that we are established.
|
| 1170 |
|
|
*/
|
| 1171 |
|
|
if (sk->shutdown & SEND_SHUTDOWN)
|
| 1172 |
|
|
{
|
| 1173 |
|
|
if (copied)
|
| 1174 |
|
|
return copied;
|
| 1175 |
|
|
send_sig(SIGPIPE,current,0);
|
| 1176 |
|
|
return -EPIPE;
|
| 1177 |
|
|
}
|
| 1178 |
|
|
|
| 1179 |
|
|
/*
|
| 1180 |
|
|
* The following code can result in copy <= if sk->mss is ever
|
| 1181 |
|
|
* decreased. It shouldn't be. sk->mss is min(sk->mtu, sk->max_window).
|
| 1182 |
|
|
* sk->mtu is constant once SYN processing is finished. I.e. we
|
| 1183 |
|
|
* had better not get here until we've seen his SYN and at least one
|
| 1184 |
|
|
* valid ack. (The SYN sets sk->mtu and the ack sets sk->max_window.)
|
| 1185 |
|
|
* But ESTABLISHED should guarantee that. sk->max_window is by definition
|
| 1186 |
|
|
* non-decreasing. Note that any ioctl to set user_mss must be done
|
| 1187 |
|
|
* before the exchange of SYN's. If the initial ack from the other
|
| 1188 |
|
|
* end has a window of 0, max_window and thus mss will both be 0.
|
| 1189 |
|
|
*/
|
| 1190 |
|
|
|
| 1191 |
|
|
/*
|
| 1192 |
|
|
* Now we need to check if we have a half built packet.
|
| 1193 |
|
|
*/
|
| 1194 |
|
|
#ifndef CONFIG_NO_PATH_MTU_DISCOVERY
|
| 1195 |
|
|
/*
|
| 1196 |
|
|
* Really, we should rebuild all the queues...
|
| 1197 |
|
|
* It's difficult. Temporary hack is to send all
|
| 1198 |
|
|
* queued segments with allowed fragmentation.
|
| 1199 |
|
|
*/
|
| 1200 |
|
|
{
|
| 1201 |
|
|
/*
|
| 1202 |
|
|
* new_mss may be zero. That indicates
|
| 1203 |
|
|
* we don't have a window estimate for
|
| 1204 |
|
|
* the remote box yet.
|
| 1205 |
|
|
* -- AC
|
| 1206 |
|
|
*/
|
| 1207 |
|
|
|
| 1208 |
|
|
int new_mss = min(sk->mtu, sk->max_window);
|
| 1209 |
|
|
if (new_mss && new_mss < sk->mss)
|
| 1210 |
|
|
{
|
| 1211 |
|
|
tcp_send_partial(sk);
|
| 1212 |
|
|
sk->mss = new_mss;
|
| 1213 |
|
|
}
|
| 1214 |
|
|
}
|
| 1215 |
|
|
#endif
|
| 1216 |
|
|
|
| 1217 |
|
|
/*
|
| 1218 |
|
|
* If there is a partly filled frame we can fill
|
| 1219 |
|
|
* out.
|
| 1220 |
|
|
*/
|
| 1221 |
|
|
if ((skb = tcp_dequeue_partial(sk)) != NULL)
|
| 1222 |
|
|
{
|
| 1223 |
|
|
int tcp_size;
|
| 1224 |
|
|
|
| 1225 |
|
|
tcp_size = skb->tail - (unsigned char *)(skb->h.th + 1);
|
| 1226 |
|
|
|
| 1227 |
|
|
/* Add more stuff to the end of skb->len */
|
| 1228 |
|
|
if (!(flags & MSG_OOB))
|
| 1229 |
|
|
{
|
| 1230 |
|
|
copy = min(sk->mss - tcp_size, seglen);
|
| 1231 |
|
|
|
| 1232 |
|
|
/*
|
| 1233 |
|
|
* Now we may find the frame is as big, or too
|
| 1234 |
|
|
* big for our MSS. Thats all fine. It means the
|
| 1235 |
|
|
* MSS shrank (from an ICMP) after we allocated
|
| 1236 |
|
|
* this frame.
|
| 1237 |
|
|
*/
|
| 1238 |
|
|
|
| 1239 |
|
|
if (tcp_size >= sk->mss)
|
| 1240 |
|
|
{
|
| 1241 |
|
|
/*
|
| 1242 |
|
|
* Send the now forced complete frame out.
|
| 1243 |
|
|
*
|
| 1244 |
|
|
* Note for 2.1: The MSS reduce code ought to
|
| 1245 |
|
|
* flush any frames in partial that are now
|
| 1246 |
|
|
* full sized. Not serious, potential tiny
|
| 1247 |
|
|
* performance hit.
|
| 1248 |
|
|
*/
|
| 1249 |
|
|
tcp_send_skb(sk,skb);
|
| 1250 |
|
|
/*
|
| 1251 |
|
|
* Get a new buffer and try again.
|
| 1252 |
|
|
*/
|
| 1253 |
|
|
continue;
|
| 1254 |
|
|
}
|
| 1255 |
|
|
/*
|
| 1256 |
|
|
* Otherwise continue to fill the buffer.
|
| 1257 |
|
|
*/
|
| 1258 |
|
|
tcp_size += copy;
|
| 1259 |
|
|
memcpy_fromfs(skb_put(skb,copy), from, copy);
|
| 1260 |
|
|
skb->csum = csum_partial(skb->tail - tcp_size, tcp_size, 0);
|
| 1261 |
|
|
from += copy;
|
| 1262 |
|
|
copied += copy;
|
| 1263 |
|
|
len -= copy;
|
| 1264 |
|
|
sk->write_seq += copy;
|
| 1265 |
|
|
seglen -= copy;
|
| 1266 |
|
|
}
|
| 1267 |
|
|
/* If we have a full packet or a new OOB
|
| 1268 |
|
|
* message, we have to force this packet out.
|
| 1269 |
|
|
*/
|
| 1270 |
|
|
if (tcp_size >= sk->mss || (flags & MSG_OOB))
|
| 1271 |
|
|
tcp_send_skb(sk, skb);
|
| 1272 |
|
|
else
|
| 1273 |
|
|
tcp_enqueue_partial(skb, sk);
|
| 1274 |
|
|
continue;
|
| 1275 |
|
|
}
|
| 1276 |
|
|
|
| 1277 |
|
|
/*
|
| 1278 |
|
|
* We also need to worry about the window.
|
| 1279 |
|
|
* If window < 1/2 the maximum window we've seen from this
|
| 1280 |
|
|
* host, don't use it. This is sender side
|
| 1281 |
|
|
* silly window prevention, as specified in RFC1122.
|
| 1282 |
|
|
* (Note that this is different than earlier versions of
|
| 1283 |
|
|
* SWS prevention, e.g. RFC813.). What we actually do is
|
| 1284 |
|
|
* use the whole MSS. Since the results in the right
|
| 1285 |
|
|
* edge of the packet being outside the window, it will
|
| 1286 |
|
|
* be queued for later rather than sent.
|
| 1287 |
|
|
*/
|
| 1288 |
|
|
|
| 1289 |
|
|
copy = sk->window_seq - sk->write_seq;
|
| 1290 |
|
|
if (copy <= 0 || copy < (sk->max_window >> 1) || copy > sk->mss)
|
| 1291 |
|
|
copy = sk->mss;
|
| 1292 |
|
|
if (copy > seglen)
|
| 1293 |
|
|
copy = seglen;
|
| 1294 |
|
|
if (copy <= 0)
|
| 1295 |
|
|
{
|
| 1296 |
|
|
printk(KERN_CRIT "TCP: **bug**: copy=%d, sk->mss=%d\n", copy, sk->mss);
|
| 1297 |
|
|
return -EFAULT;
|
| 1298 |
|
|
}
|
| 1299 |
|
|
|
| 1300 |
|
|
/*
|
| 1301 |
|
|
* We should really check the window here also.
|
| 1302 |
|
|
*/
|
| 1303 |
|
|
|
| 1304 |
|
|
delay = 0;
|
| 1305 |
|
|
tmp = copy + sk->prot->max_header + 15;
|
| 1306 |
|
|
/* If won't fill the current packet, and it's not an OOB message,
|
| 1307 |
|
|
* then we might want to delay to allow data in the later parts
|
| 1308 |
|
|
* of iov to fill this packet out. Note that if we aren't
|
| 1309 |
|
|
* Nagling or there are no packets currently out then the top
|
| 1310 |
|
|
* level code in tcp_sendmsg() will force any partial packets out
|
| 1311 |
|
|
* after we finish building the largest packets this write allows.
|
| 1312 |
|
|
*/
|
| 1313 |
|
|
if (copy < sk->mss && !(flags & MSG_OOB)) {
|
| 1314 |
|
|
tmp = tmp - copy + sk->mtu + 128;
|
| 1315 |
|
|
delay = 1;
|
| 1316 |
|
|
}
|
| 1317 |
|
|
skb = sock_wmalloc(sk, tmp, 0, GFP_KERNEL);
|
| 1318 |
|
|
|
| 1319 |
|
|
/*
|
| 1320 |
|
|
* If we didn't get any memory, we need to sleep.
|
| 1321 |
|
|
*/
|
| 1322 |
|
|
|
| 1323 |
|
|
if (skb == NULL)
|
| 1324 |
|
|
{
|
| 1325 |
|
|
sk->socket->flags |= SO_NOSPACE;
|
| 1326 |
|
|
if (nonblock)
|
| 1327 |
|
|
{
|
| 1328 |
|
|
if (copied)
|
| 1329 |
|
|
return copied;
|
| 1330 |
|
|
return -EAGAIN;
|
| 1331 |
|
|
}
|
| 1332 |
|
|
|
| 1333 |
|
|
if (current->signal & ~current->blocked)
|
| 1334 |
|
|
{
|
| 1335 |
|
|
if (copied)
|
| 1336 |
|
|
return copied;
|
| 1337 |
|
|
return -ERESTARTSYS;
|
| 1338 |
|
|
}
|
| 1339 |
|
|
|
| 1340 |
|
|
wait_for_tcp_memory(sk);
|
| 1341 |
|
|
continue;
|
| 1342 |
|
|
}
|
| 1343 |
|
|
|
| 1344 |
|
|
skb->sk = sk;
|
| 1345 |
|
|
skb->free = 0;
|
| 1346 |
|
|
skb->localroute = sk->localroute|(flags&MSG_DONTROUTE);
|
| 1347 |
|
|
|
| 1348 |
|
|
/*
|
| 1349 |
|
|
* FIXME: we need to optimize this.
|
| 1350 |
|
|
* Perhaps some hints here would be good.
|
| 1351 |
|
|
*/
|
| 1352 |
|
|
|
| 1353 |
|
|
tmp = sk->prot->build_header(skb, sk->saddr, sk->daddr, &dev,
|
| 1354 |
|
|
IPPROTO_TCP, sk->opt, skb->truesize,sk->ip_tos,sk->ip_ttl,&sk->ip_route_cache);
|
| 1355 |
|
|
if (tmp < 0 )
|
| 1356 |
|
|
{
|
| 1357 |
|
|
sock_wfree(sk, skb);
|
| 1358 |
|
|
if (copied)
|
| 1359 |
|
|
return(copied);
|
| 1360 |
|
|
return(tmp);
|
| 1361 |
|
|
}
|
| 1362 |
|
|
#ifndef CONFIG_NO_PATH_MTU_DISCOVERY
|
| 1363 |
|
|
skb->ip_hdr->frag_off |= htons(IP_DF);
|
| 1364 |
|
|
#endif
|
| 1365 |
|
|
skb->dev = dev;
|
| 1366 |
|
|
skb->h.th =(struct tcphdr *)skb_put(skb,sizeof(struct tcphdr));
|
| 1367 |
|
|
tmp = tcp_build_header(skb->h.th, sk, seglen-copy);
|
| 1368 |
|
|
if (tmp < 0)
|
| 1369 |
|
|
{
|
| 1370 |
|
|
sock_wfree(sk, skb);
|
| 1371 |
|
|
if (copied)
|
| 1372 |
|
|
return(copied);
|
| 1373 |
|
|
return(tmp);
|
| 1374 |
|
|
}
|
| 1375 |
|
|
|
| 1376 |
|
|
if (flags & MSG_OOB)
|
| 1377 |
|
|
{
|
| 1378 |
|
|
skb->h.th->urg = 1;
|
| 1379 |
|
|
skb->h.th->urg_ptr = ntohs(copy);
|
| 1380 |
|
|
}
|
| 1381 |
|
|
|
| 1382 |
|
|
skb->csum = csum_partial_copy_fromuser(from,
|
| 1383 |
|
|
skb_put(skb,copy), copy, 0);
|
| 1384 |
|
|
|
| 1385 |
|
|
from += copy;
|
| 1386 |
|
|
copied += copy;
|
| 1387 |
|
|
len -= copy;
|
| 1388 |
|
|
seglen -= copy;
|
| 1389 |
|
|
skb->free = 0;
|
| 1390 |
|
|
sk->write_seq += copy;
|
| 1391 |
|
|
|
| 1392 |
|
|
if (delay)
|
| 1393 |
|
|
{
|
| 1394 |
|
|
tcp_enqueue_partial(skb, sk);
|
| 1395 |
|
|
continue;
|
| 1396 |
|
|
}
|
| 1397 |
|
|
tcp_send_skb(sk, skb);
|
| 1398 |
|
|
}
|
| 1399 |
|
|
}
|
| 1400 |
|
|
sk->err = 0;
|
| 1401 |
|
|
|
| 1402 |
|
|
return copied;
|
| 1403 |
|
|
}
|
| 1404 |
|
|
|
| 1405 |
|
|
|
| 1406 |
|
|
static int tcp_sendmsg(struct sock *sk, struct msghdr *msg,
|
| 1407 |
|
|
int len, int nonblock, int flags)
|
| 1408 |
|
|
{
|
| 1409 |
|
|
int retval = -EINVAL;
|
| 1410 |
|
|
|
| 1411 |
|
|
/*
|
| 1412 |
|
|
* Do sanity checking for sendmsg/sendto/send
|
| 1413 |
|
|
*/
|
| 1414 |
|
|
|
| 1415 |
|
|
if (flags & ~(MSG_OOB|MSG_DONTROUTE))
|
| 1416 |
|
|
goto out;
|
| 1417 |
|
|
if (msg->msg_name) {
|
| 1418 |
|
|
struct sockaddr_in *addr=(struct sockaddr_in *)msg->msg_name;
|
| 1419 |
|
|
|
| 1420 |
|
|
if (msg->msg_namelen < sizeof(*addr))
|
| 1421 |
|
|
goto out;
|
| 1422 |
|
|
if (addr->sin_family && addr->sin_family != AF_INET)
|
| 1423 |
|
|
goto out;
|
| 1424 |
|
|
retval = -ENOTCONN;
|
| 1425 |
|
|
if(sk->state == TCP_CLOSE)
|
| 1426 |
|
|
goto out;
|
| 1427 |
|
|
retval = -EISCONN;
|
| 1428 |
|
|
if (addr->sin_port != sk->dummy_th.dest)
|
| 1429 |
|
|
goto out;
|
| 1430 |
|
|
if (addr->sin_addr.s_addr != sk->daddr)
|
| 1431 |
|
|
goto out;
|
| 1432 |
|
|
}
|
| 1433 |
|
|
|
| 1434 |
|
|
lock_sock(sk);
|
| 1435 |
|
|
retval = do_tcp_sendmsg(sk, msg->msg_iovlen, msg->msg_iov, len, nonblock, flags);
|
| 1436 |
|
|
|
| 1437 |
|
|
/*
|
| 1438 |
|
|
* Nagle's rule. Turn Nagle off with TCP_NODELAY for highly
|
| 1439 |
|
|
* interactive fast network servers. It's meant to be on and
|
| 1440 |
|
|
* it really improves the throughput though not the echo time
|
| 1441 |
|
|
* on my slow slip link - Alan
|
| 1442 |
|
|
*
|
| 1443 |
|
|
* If not nagling we can send on the before case too..
|
| 1444 |
|
|
*/
|
| 1445 |
|
|
|
| 1446 |
|
|
if (sk->partial) {
|
| 1447 |
|
|
if (!sk->packets_out ||
|
| 1448 |
|
|
(sk->nonagle && before(sk->write_seq , sk->window_seq))) {
|
| 1449 |
|
|
tcp_send_partial(sk);
|
| 1450 |
|
|
}
|
| 1451 |
|
|
}
|
| 1452 |
|
|
|
| 1453 |
|
|
release_sock(sk);
|
| 1454 |
|
|
|
| 1455 |
|
|
out:
|
| 1456 |
|
|
return retval;
|
| 1457 |
|
|
}
|
| 1458 |
|
|
|
| 1459 |
|
|
|
| 1460 |
|
|
/*
|
| 1461 |
|
|
* Send an ack if one is backlogged at this point.
|
| 1462 |
|
|
*/
|
| 1463 |
|
|
|
| 1464 |
|
|
void tcp_read_wakeup(struct sock *sk)
|
| 1465 |
|
|
{
|
| 1466 |
|
|
if (!sk->ack_backlog)
|
| 1467 |
|
|
return;
|
| 1468 |
|
|
|
| 1469 |
|
|
/*
|
| 1470 |
|
|
* If we're closed, don't send an ack, or we'll get a RST
|
| 1471 |
|
|
* from the closed destination.
|
| 1472 |
|
|
*/
|
| 1473 |
|
|
if ((sk->state == TCP_CLOSE) || (sk->state == TCP_TIME_WAIT))
|
| 1474 |
|
|
return;
|
| 1475 |
|
|
|
| 1476 |
|
|
tcp_send_ack(sk);
|
| 1477 |
|
|
}
|
| 1478 |
|
|
|
| 1479 |
|
|
|
| 1480 |
|
|
/*
|
| 1481 |
|
|
* Handle reading urgent data. BSD has very simple semantics for
|
| 1482 |
|
|
* this, no blocking and very strange errors 8)
|
| 1483 |
|
|
*/
|
| 1484 |
|
|
|
| 1485 |
|
|
static int tcp_recv_urg(struct sock * sk, int nonblock,
|
| 1486 |
|
|
struct msghdr *msg, int len, int flags, int *addr_len)
|
| 1487 |
|
|
{
|
| 1488 |
|
|
/*
|
| 1489 |
|
|
* No URG data to read
|
| 1490 |
|
|
*/
|
| 1491 |
|
|
if (sk->urginline || !sk->urg_data || sk->urg_data == URG_READ)
|
| 1492 |
|
|
return -EINVAL; /* Yes this is right ! */
|
| 1493 |
|
|
|
| 1494 |
|
|
if (sk->err)
|
| 1495 |
|
|
return sock_error(sk);
|
| 1496 |
|
|
|
| 1497 |
|
|
if (sk->state == TCP_CLOSE || sk->done)
|
| 1498 |
|
|
{
|
| 1499 |
|
|
if (!sk->done)
|
| 1500 |
|
|
{
|
| 1501 |
|
|
sk->done = 1;
|
| 1502 |
|
|
return 0;
|
| 1503 |
|
|
}
|
| 1504 |
|
|
return -ENOTCONN;
|
| 1505 |
|
|
}
|
| 1506 |
|
|
|
| 1507 |
|
|
if (sk->shutdown & RCV_SHUTDOWN)
|
| 1508 |
|
|
{
|
| 1509 |
|
|
sk->done = 1;
|
| 1510 |
|
|
return 0;
|
| 1511 |
|
|
}
|
| 1512 |
|
|
lock_sock(sk);
|
| 1513 |
|
|
if (sk->urg_data & URG_VALID)
|
| 1514 |
|
|
{
|
| 1515 |
|
|
char c = sk->urg_data;
|
| 1516 |
|
|
if (!(flags & MSG_PEEK))
|
| 1517 |
|
|
sk->urg_data = URG_READ;
|
| 1518 |
|
|
memcpy_toiovec(msg->msg_iov, &c, 1);
|
| 1519 |
|
|
if(msg->msg_name)
|
| 1520 |
|
|
{
|
| 1521 |
|
|
struct sockaddr_in *sin=(struct sockaddr_in *)msg->msg_name;
|
| 1522 |
|
|
sin->sin_family=AF_INET;
|
| 1523 |
|
|
sin->sin_addr.s_addr=sk->daddr;
|
| 1524 |
|
|
sin->sin_port=sk->dummy_th.dest;
|
| 1525 |
|
|
}
|
| 1526 |
|
|
if(addr_len)
|
| 1527 |
|
|
*addr_len=sizeof(struct sockaddr_in);
|
| 1528 |
|
|
release_sock(sk);
|
| 1529 |
|
|
return 1;
|
| 1530 |
|
|
}
|
| 1531 |
|
|
release_sock(sk);
|
| 1532 |
|
|
|
| 1533 |
|
|
/*
|
| 1534 |
|
|
* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and
|
| 1535 |
|
|
* the available implementations agree in this case:
|
| 1536 |
|
|
* this call should never block, independent of the
|
| 1537 |
|
|
* blocking state of the socket.
|
| 1538 |
|
|
* Mike <pall@rz.uni-karlsruhe.de>
|
| 1539 |
|
|
*/
|
| 1540 |
|
|
return -EAGAIN;
|
| 1541 |
|
|
}
|
| 1542 |
|
|
|
| 1543 |
|
|
/*
|
| 1544 |
|
|
* Release a skb if it is no longer needed. This routine
|
| 1545 |
|
|
* must be called with interrupts disabled or with the
|
| 1546 |
|
|
* socket locked so that the sk_buff queue operation is ok.
|
| 1547 |
|
|
*/
|
| 1548 |
|
|
|
| 1549 |
|
|
static inline void tcp_eat_skb(struct sock *sk, struct sk_buff * skb)
|
| 1550 |
|
|
{
|
| 1551 |
|
|
skb->sk = sk;
|
| 1552 |
|
|
__skb_unlink(skb, &sk->receive_queue);
|
| 1553 |
|
|
kfree_skb(skb, FREE_READ);
|
| 1554 |
|
|
}
|
| 1555 |
|
|
|
| 1556 |
|
|
/*
|
| 1557 |
|
|
* FIXME:
|
| 1558 |
|
|
* This routine frees used buffers.
|
| 1559 |
|
|
* It should consider sending an ACK to let the
|
| 1560 |
|
|
* other end know we now have a bigger window.
|
| 1561 |
|
|
*/
|
| 1562 |
|
|
|
| 1563 |
|
|
static void cleanup_rbuf(struct sock *sk)
|
| 1564 |
|
|
{
|
| 1565 |
|
|
/*
|
| 1566 |
|
|
* NOTE! The socket must be locked, so that we don't get
|
| 1567 |
|
|
* a messed-up receive queue.
|
| 1568 |
|
|
*/
|
| 1569 |
|
|
while (!skb_queue_empty(&sk->receive_queue)) {
|
| 1570 |
|
|
struct sk_buff *skb = sk->receive_queue.next;
|
| 1571 |
|
|
if (!skb->used || skb->users)
|
| 1572 |
|
|
break;
|
| 1573 |
|
|
tcp_eat_skb(sk, skb);
|
| 1574 |
|
|
}
|
| 1575 |
|
|
|
| 1576 |
|
|
/*
|
| 1577 |
|
|
* Tell the world if we raised the window.
|
| 1578 |
|
|
*/
|
| 1579 |
|
|
if (tcp_raise_window(sk))
|
| 1580 |
|
|
tcp_send_ack(sk);
|
| 1581 |
|
|
}
|
| 1582 |
|
|
|
| 1583 |
|
|
|
| 1584 |
|
|
/*
|
| 1585 |
|
|
* This routine copies from a sock struct into the user buffer.
|
| 1586 |
|
|
*/
|
| 1587 |
|
|
|
| 1588 |
|
|
static int tcp_recvmsg(struct sock *sk, struct msghdr *msg,
|
| 1589 |
|
|
int len, int nonblock, int flags, int *addr_len)
|
| 1590 |
|
|
{
|
| 1591 |
|
|
struct wait_queue wait = { current, NULL };
|
| 1592 |
|
|
int copied = 0;
|
| 1593 |
|
|
u32 peek_seq;
|
| 1594 |
|
|
volatile u32 *seq; /* So gcc doesn't overoptimise */
|
| 1595 |
|
|
unsigned long used;
|
| 1596 |
|
|
|
| 1597 |
|
|
/*
|
| 1598 |
|
|
* This error should be checked.
|
| 1599 |
|
|
*/
|
| 1600 |
|
|
|
| 1601 |
|
|
if (sk->state == TCP_LISTEN)
|
| 1602 |
|
|
return -ENOTCONN;
|
| 1603 |
|
|
|
| 1604 |
|
|
/*
|
| 1605 |
|
|
* Urgent data needs to be handled specially.
|
| 1606 |
|
|
*/
|
| 1607 |
|
|
|
| 1608 |
|
|
if (flags & MSG_OOB)
|
| 1609 |
|
|
return tcp_recv_urg(sk, nonblock, msg, len, flags, addr_len);
|
| 1610 |
|
|
|
| 1611 |
|
|
/*
|
| 1612 |
|
|
* Copying sequence to update. This is volatile to handle
|
| 1613 |
|
|
* the multi-reader case neatly (memcpy_to/fromfs might be
|
| 1614 |
|
|
* inline and thus not flush cached variables otherwise).
|
| 1615 |
|
|
*/
|
| 1616 |
|
|
|
| 1617 |
|
|
peek_seq = sk->copied_seq;
|
| 1618 |
|
|
seq = &sk->copied_seq;
|
| 1619 |
|
|
if (flags & MSG_PEEK)
|
| 1620 |
|
|
seq = &peek_seq;
|
| 1621 |
|
|
|
| 1622 |
|
|
add_wait_queue(sk->sleep, &wait);
|
| 1623 |
|
|
lock_sock(sk);
|
| 1624 |
|
|
while (len > 0)
|
| 1625 |
|
|
{
|
| 1626 |
|
|
struct sk_buff * skb;
|
| 1627 |
|
|
u32 offset;
|
| 1628 |
|
|
|
| 1629 |
|
|
/*
|
| 1630 |
|
|
* Are we at urgent data? Stop if we have read anything.
|
| 1631 |
|
|
*/
|
| 1632 |
|
|
|
| 1633 |
|
|
if (copied && sk->urg_data && sk->urg_seq == *seq)
|
| 1634 |
|
|
break;
|
| 1635 |
|
|
|
| 1636 |
|
|
/*
|
| 1637 |
|
|
* We need to check signals first, to get correct SIGURG
|
| 1638 |
|
|
* handling.
|
| 1639 |
|
|
*/
|
| 1640 |
|
|
if (current->signal & ~current->blocked) {
|
| 1641 |
|
|
if (copied)
|
| 1642 |
|
|
break;
|
| 1643 |
|
|
copied = -ERESTARTSYS;
|
| 1644 |
|
|
if (nonblock)
|
| 1645 |
|
|
copied = -EAGAIN;
|
| 1646 |
|
|
break;
|
| 1647 |
|
|
}
|
| 1648 |
|
|
|
| 1649 |
|
|
/*
|
| 1650 |
|
|
* Next get a buffer.
|
| 1651 |
|
|
*/
|
| 1652 |
|
|
|
| 1653 |
|
|
current->state = TASK_INTERRUPTIBLE;
|
| 1654 |
|
|
|
| 1655 |
|
|
skb = sk->receive_queue.next;
|
| 1656 |
|
|
while (skb != (struct sk_buff *)&sk->receive_queue)
|
| 1657 |
|
|
{
|
| 1658 |
|
|
if (before(*seq, skb->seq))
|
| 1659 |
|
|
break;
|
| 1660 |
|
|
offset = *seq - skb->seq;
|
| 1661 |
|
|
if (skb->h.th->syn)
|
| 1662 |
|
|
offset--;
|
| 1663 |
|
|
if (offset < skb->len)
|
| 1664 |
|
|
goto found_ok_skb;
|
| 1665 |
|
|
if (skb->h.th->fin)
|
| 1666 |
|
|
goto found_fin_ok;
|
| 1667 |
|
|
if (!(flags & MSG_PEEK))
|
| 1668 |
|
|
skb->used = 1;
|
| 1669 |
|
|
skb = skb->next;
|
| 1670 |
|
|
}
|
| 1671 |
|
|
|
| 1672 |
|
|
if (copied)
|
| 1673 |
|
|
break;
|
| 1674 |
|
|
|
| 1675 |
|
|
if (sk->err && !(flags&MSG_PEEK))
|
| 1676 |
|
|
{
|
| 1677 |
|
|
copied = sock_error(sk);
|
| 1678 |
|
|
break;
|
| 1679 |
|
|
}
|
| 1680 |
|
|
|
| 1681 |
|
|
if (sk->state == TCP_CLOSE)
|
| 1682 |
|
|
{
|
| 1683 |
|
|
if (!sk->done)
|
| 1684 |
|
|
{
|
| 1685 |
|
|
sk->done = 1;
|
| 1686 |
|
|
break;
|
| 1687 |
|
|
}
|
| 1688 |
|
|
copied = -ENOTCONN;
|
| 1689 |
|
|
break;
|
| 1690 |
|
|
}
|
| 1691 |
|
|
|
| 1692 |
|
|
if (sk->shutdown & RCV_SHUTDOWN)
|
| 1693 |
|
|
{
|
| 1694 |
|
|
sk->done = 1;
|
| 1695 |
|
|
break;
|
| 1696 |
|
|
}
|
| 1697 |
|
|
|
| 1698 |
|
|
if (nonblock)
|
| 1699 |
|
|
{
|
| 1700 |
|
|
copied = -EAGAIN;
|
| 1701 |
|
|
break;
|
| 1702 |
|
|
}
|
| 1703 |
|
|
|
| 1704 |
|
|
cleanup_rbuf(sk);
|
| 1705 |
|
|
release_sock(sk);
|
| 1706 |
|
|
sk->socket->flags |= SO_WAITDATA;
|
| 1707 |
|
|
schedule();
|
| 1708 |
|
|
sk->socket->flags &= ~SO_WAITDATA;
|
| 1709 |
|
|
lock_sock(sk);
|
| 1710 |
|
|
continue;
|
| 1711 |
|
|
|
| 1712 |
|
|
found_ok_skb:
|
| 1713 |
|
|
/*
|
| 1714 |
|
|
* Lock the buffer. We can be fairly relaxed as
|
| 1715 |
|
|
* an interrupt will never steal a buffer we are
|
| 1716 |
|
|
* using unless I've missed something serious in
|
| 1717 |
|
|
* tcp_data.
|
| 1718 |
|
|
*/
|
| 1719 |
|
|
|
| 1720 |
|
|
skb->users++;
|
| 1721 |
|
|
|
| 1722 |
|
|
/*
|
| 1723 |
|
|
* Ok so how much can we use ?
|
| 1724 |
|
|
*/
|
| 1725 |
|
|
|
| 1726 |
|
|
used = skb->len - offset;
|
| 1727 |
|
|
if (len < used)
|
| 1728 |
|
|
used = len;
|
| 1729 |
|
|
/*
|
| 1730 |
|
|
* Do we have urgent data here?
|
| 1731 |
|
|
*/
|
| 1732 |
|
|
|
| 1733 |
|
|
if (sk->urg_data)
|
| 1734 |
|
|
{
|
| 1735 |
|
|
u32 urg_offset = sk->urg_seq - *seq;
|
| 1736 |
|
|
if (urg_offset < used)
|
| 1737 |
|
|
{
|
| 1738 |
|
|
if (!urg_offset)
|
| 1739 |
|
|
{
|
| 1740 |
|
|
if (!sk->urginline)
|
| 1741 |
|
|
{
|
| 1742 |
|
|
++*seq;
|
| 1743 |
|
|
offset++;
|
| 1744 |
|
|
used--;
|
| 1745 |
|
|
}
|
| 1746 |
|
|
}
|
| 1747 |
|
|
else
|
| 1748 |
|
|
used = urg_offset;
|
| 1749 |
|
|
}
|
| 1750 |
|
|
}
|
| 1751 |
|
|
|
| 1752 |
|
|
/*
|
| 1753 |
|
|
* Copy it - We _MUST_ update *seq first so that we
|
| 1754 |
|
|
* don't ever double read when we have dual readers
|
| 1755 |
|
|
*/
|
| 1756 |
|
|
|
| 1757 |
|
|
*seq += used;
|
| 1758 |
|
|
|
| 1759 |
|
|
/*
|
| 1760 |
|
|
* This memcpy_tofs can sleep. If it sleeps and we
|
| 1761 |
|
|
* do a second read it relies on the skb->users to avoid
|
| 1762 |
|
|
* a crash when cleanup_rbuf() gets called.
|
| 1763 |
|
|
*/
|
| 1764 |
|
|
|
| 1765 |
|
|
memcpy_toiovec(msg->msg_iov,((unsigned char *)skb->h.th) +
|
| 1766 |
|
|
skb->h.th->doff*4 + offset, used);
|
| 1767 |
|
|
copied += used;
|
| 1768 |
|
|
len -= used;
|
| 1769 |
|
|
|
| 1770 |
|
|
/*
|
| 1771 |
|
|
* We now will not sleep again until we are finished
|
| 1772 |
|
|
* with skb. Sorry if you are doing the SMP port
|
| 1773 |
|
|
* but you'll just have to fix it neatly ;)
|
| 1774 |
|
|
*/
|
| 1775 |
|
|
|
| 1776 |
|
|
skb->users --;
|
| 1777 |
|
|
|
| 1778 |
|
|
if (after(sk->copied_seq,sk->urg_seq))
|
| 1779 |
|
|
sk->urg_data = 0;
|
| 1780 |
|
|
if (used + offset < skb->len)
|
| 1781 |
|
|
continue;
|
| 1782 |
|
|
|
| 1783 |
|
|
/*
|
| 1784 |
|
|
* Process the FIN.
|
| 1785 |
|
|
*/
|
| 1786 |
|
|
|
| 1787 |
|
|
if (skb->h.th->fin)
|
| 1788 |
|
|
goto found_fin_ok;
|
| 1789 |
|
|
if (flags & MSG_PEEK)
|
| 1790 |
|
|
continue;
|
| 1791 |
|
|
skb->used = 1;
|
| 1792 |
|
|
if (!skb->users)
|
| 1793 |
|
|
tcp_eat_skb(sk, skb);
|
| 1794 |
|
|
continue;
|
| 1795 |
|
|
|
| 1796 |
|
|
found_fin_ok:
|
| 1797 |
|
|
++*seq;
|
| 1798 |
|
|
if (flags & MSG_PEEK)
|
| 1799 |
|
|
break;
|
| 1800 |
|
|
|
| 1801 |
|
|
/*
|
| 1802 |
|
|
* All is done
|
| 1803 |
|
|
*/
|
| 1804 |
|
|
|
| 1805 |
|
|
skb->used = 1;
|
| 1806 |
|
|
sk->shutdown |= RCV_SHUTDOWN;
|
| 1807 |
|
|
break;
|
| 1808 |
|
|
|
| 1809 |
|
|
}
|
| 1810 |
|
|
|
| 1811 |
|
|
if(copied>0 && msg->msg_name)
|
| 1812 |
|
|
{
|
| 1813 |
|
|
struct sockaddr_in *sin=(struct sockaddr_in *)msg->msg_name;
|
| 1814 |
|
|
sin->sin_family=AF_INET;
|
| 1815 |
|
|
sin->sin_addr.s_addr=sk->daddr;
|
| 1816 |
|
|
sin->sin_port=sk->dummy_th.dest;
|
| 1817 |
|
|
}
|
| 1818 |
|
|
if(addr_len)
|
| 1819 |
|
|
*addr_len=sizeof(struct sockaddr_in);
|
| 1820 |
|
|
|
| 1821 |
|
|
remove_wait_queue(sk->sleep, &wait);
|
| 1822 |
|
|
current->state = TASK_RUNNING;
|
| 1823 |
|
|
|
| 1824 |
|
|
/* Clean up data we have read: This will do ACK frames */
|
| 1825 |
|
|
cleanup_rbuf(sk);
|
| 1826 |
|
|
release_sock(sk);
|
| 1827 |
|
|
return copied;
|
| 1828 |
|
|
}
|
| 1829 |
|
|
|
| 1830 |
|
|
|
| 1831 |
|
|
|
| 1832 |
|
|
/*
|
| 1833 |
|
|
* State processing on a close. This implements the state shift for
|
| 1834 |
|
|
* sending our FIN frame. Note that we only send a FIN for some
|
| 1835 |
|
|
* states. A shutdown() may have already sent the FIN, or we may be
|
| 1836 |
|
|
* closed.
|
| 1837 |
|
|
*/
|
| 1838 |
|
|
|
| 1839 |
|
|
static int tcp_close_state(struct sock *sk, int dead)
|
| 1840 |
|
|
{
|
| 1841 |
|
|
int ns=TCP_CLOSE;
|
| 1842 |
|
|
int send_fin=0;
|
| 1843 |
|
|
switch(sk->state)
|
| 1844 |
|
|
{
|
| 1845 |
|
|
case TCP_SYN_SENT: /* No SYN back, no FIN needed */
|
| 1846 |
|
|
break;
|
| 1847 |
|
|
case TCP_SYN_RECV:
|
| 1848 |
|
|
case TCP_ESTABLISHED: /* Closedown begin */
|
| 1849 |
|
|
ns=TCP_FIN_WAIT1;
|
| 1850 |
|
|
send_fin=1;
|
| 1851 |
|
|
break;
|
| 1852 |
|
|
case TCP_FIN_WAIT1: /* Already closing, or FIN sent: no change */
|
| 1853 |
|
|
case TCP_FIN_WAIT2:
|
| 1854 |
|
|
case TCP_CLOSING:
|
| 1855 |
|
|
ns=sk->state;
|
| 1856 |
|
|
break;
|
| 1857 |
|
|
case TCP_CLOSE:
|
| 1858 |
|
|
case TCP_LISTEN:
|
| 1859 |
|
|
break;
|
| 1860 |
|
|
case TCP_LAST_ACK: /* Could have shutdown() then close().
|
| 1861 |
|
|
Be careful not to send double fin. */
|
| 1862 |
|
|
ns=TCP_LAST_ACK;
|
| 1863 |
|
|
break;
|
| 1864 |
|
|
case TCP_CLOSE_WAIT: /* They have FIN'd us. We send our FIN and
|
| 1865 |
|
|
wait only for the ACK */
|
| 1866 |
|
|
ns=TCP_LAST_ACK;
|
| 1867 |
|
|
send_fin=1;
|
| 1868 |
|
|
}
|
| 1869 |
|
|
|
| 1870 |
|
|
tcp_set_state(sk,ns);
|
| 1871 |
|
|
|
| 1872 |
|
|
/*
|
| 1873 |
|
|
* This is a (useful) BSD violating of the RFC. There is a
|
| 1874 |
|
|
* problem with TCP as specified in that the other end could
|
| 1875 |
|
|
* keep a socket open forever with no application left this end.
|
| 1876 |
|
|
* We use a 3 minute timeout (about the same as BSD) then kill
|
| 1877 |
|
|
* our end. If they send after that then tough - BUT: long enough
|
| 1878 |
|
|
* that we won't make the old 4*rto = almost no time - whoops
|
| 1879 |
|
|
* reset mistake.
|
| 1880 |
|
|
*/
|
| 1881 |
|
|
if(dead && ns==TCP_FIN_WAIT2)
|
| 1882 |
|
|
{
|
| 1883 |
|
|
int timer_active=del_timer(&sk->timer);
|
| 1884 |
|
|
if(timer_active)
|
| 1885 |
|
|
add_timer(&sk->timer);
|
| 1886 |
|
|
else
|
| 1887 |
|
|
tcp_reset_msl_timer(sk, TIME_CLOSE, TCP_FIN_TIMEOUT);
|
| 1888 |
|
|
}
|
| 1889 |
|
|
|
| 1890 |
|
|
return send_fin;
|
| 1891 |
|
|
}
|
| 1892 |
|
|
|
| 1893 |
|
|
/*
|
| 1894 |
|
|
* Shutdown the sending side of a connection. Much like close except
|
| 1895 |
|
|
* that we don't receive shut down or set sk->dead.
|
| 1896 |
|
|
*/
|
| 1897 |
|
|
|
| 1898 |
|
|
void tcp_shutdown(struct sock *sk, int how)
|
| 1899 |
|
|
{
|
| 1900 |
|
|
/*
|
| 1901 |
|
|
* We need to grab some memory, and put together a FIN,
|
| 1902 |
|
|
* and then put it into the queue to be sent.
|
| 1903 |
|
|
* Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92.
|
| 1904 |
|
|
*/
|
| 1905 |
|
|
|
| 1906 |
|
|
if (!(how & SEND_SHUTDOWN))
|
| 1907 |
|
|
return;
|
| 1908 |
|
|
|
| 1909 |
|
|
/*
|
| 1910 |
|
|
* If we've already sent a FIN, or it's a closed state
|
| 1911 |
|
|
*/
|
| 1912 |
|
|
|
| 1913 |
|
|
if (sk->state == TCP_FIN_WAIT1 ||
|
| 1914 |
|
|
sk->state == TCP_FIN_WAIT2 ||
|
| 1915 |
|
|
sk->state == TCP_CLOSING ||
|
| 1916 |
|
|
sk->state == TCP_LAST_ACK ||
|
| 1917 |
|
|
sk->state == TCP_TIME_WAIT ||
|
| 1918 |
|
|
sk->state == TCP_CLOSE ||
|
| 1919 |
|
|
sk->state == TCP_LISTEN
|
| 1920 |
|
|
)
|
| 1921 |
|
|
{
|
| 1922 |
|
|
return;
|
| 1923 |
|
|
}
|
| 1924 |
|
|
lock_sock(sk);
|
| 1925 |
|
|
|
| 1926 |
|
|
/*
|
| 1927 |
|
|
* flag that the sender has shutdown
|
| 1928 |
|
|
*/
|
| 1929 |
|
|
|
| 1930 |
|
|
sk->shutdown |= SEND_SHUTDOWN;
|
| 1931 |
|
|
|
| 1932 |
|
|
/*
|
| 1933 |
|
|
* Clear out any half completed packets.
|
| 1934 |
|
|
*/
|
| 1935 |
|
|
|
| 1936 |
|
|
if (sk->partial)
|
| 1937 |
|
|
tcp_send_partial(sk);
|
| 1938 |
|
|
|
| 1939 |
|
|
/*
|
| 1940 |
|
|
* FIN if needed
|
| 1941 |
|
|
*/
|
| 1942 |
|
|
|
| 1943 |
|
|
if (tcp_close_state(sk,0))
|
| 1944 |
|
|
tcp_send_fin(sk);
|
| 1945 |
|
|
|
| 1946 |
|
|
release_sock(sk);
|
| 1947 |
|
|
}
|
| 1948 |
|
|
|
| 1949 |
|
|
|
| 1950 |
|
|
/*
|
| 1951 |
|
|
* Return 1 if we still have things to send in our buffers.
|
| 1952 |
|
|
*/
|
| 1953 |
|
|
|
| 1954 |
|
|
static inline int closing(struct sock * sk)
|
| 1955 |
|
|
{
|
| 1956 |
|
|
switch (sk->state) {
|
| 1957 |
|
|
case TCP_FIN_WAIT1:
|
| 1958 |
|
|
case TCP_CLOSING:
|
| 1959 |
|
|
case TCP_LAST_ACK:
|
| 1960 |
|
|
return 1;
|
| 1961 |
|
|
}
|
| 1962 |
|
|
return 0;
|
| 1963 |
|
|
}
|
| 1964 |
|
|
|
| 1965 |
|
|
|
| 1966 |
|
|
static void tcp_close(struct sock *sk, unsigned long timeout)
|
| 1967 |
|
|
{
|
| 1968 |
|
|
struct sk_buff *skb;
|
| 1969 |
|
|
|
| 1970 |
|
|
/*
|
| 1971 |
|
|
* We need to grab some memory, and put together a FIN,
|
| 1972 |
|
|
* and then put it into the queue to be sent.
|
| 1973 |
|
|
*/
|
| 1974 |
|
|
|
| 1975 |
|
|
lock_sock(sk);
|
| 1976 |
|
|
|
| 1977 |
|
|
if(sk->state == TCP_LISTEN)
|
| 1978 |
|
|
{
|
| 1979 |
|
|
/*
|
| 1980 |
|
|
* Special case
|
| 1981 |
|
|
*/
|
| 1982 |
|
|
tcp_set_state(sk, TCP_CLOSE);
|
| 1983 |
|
|
/*
|
| 1984 |
|
|
* Our children must die before we do now that
|
| 1985 |
|
|
* sk->listening exists. It was right anyway but
|
| 1986 |
|
|
* dont break this assumption.
|
| 1987 |
|
|
*/
|
| 1988 |
|
|
tcp_close_pending(sk);
|
| 1989 |
|
|
release_sock(sk);
|
| 1990 |
|
|
sk->dead = 1;
|
| 1991 |
|
|
tcp_v4_unhash(sk);
|
| 1992 |
|
|
return;
|
| 1993 |
|
|
}
|
| 1994 |
|
|
|
| 1995 |
|
|
sk->keepopen = 1;
|
| 1996 |
|
|
sk->shutdown = SHUTDOWN_MASK;
|
| 1997 |
|
|
|
| 1998 |
|
|
if (!sk->dead)
|
| 1999 |
|
|
sk->state_change(sk);
|
| 2000 |
|
|
|
| 2001 |
|
|
/*
|
| 2002 |
|
|
* We need to flush the recv. buffs. We do this only on the
|
| 2003 |
|
|
* descriptor close, not protocol-sourced closes, because the
|
| 2004 |
|
|
* reader process may not have drained the data yet!
|
| 2005 |
|
|
*/
|
| 2006 |
|
|
|
| 2007 |
|
|
while((skb=skb_dequeue(&sk->receive_queue))!=NULL)
|
| 2008 |
|
|
kfree_skb(skb, FREE_READ);
|
| 2009 |
|
|
|
| 2010 |
|
|
/*
|
| 2011 |
|
|
* Get rid off any half-completed packets.
|
| 2012 |
|
|
*/
|
| 2013 |
|
|
|
| 2014 |
|
|
if (sk->partial)
|
| 2015 |
|
|
tcp_send_partial(sk);
|
| 2016 |
|
|
|
| 2017 |
|
|
/*
|
| 2018 |
|
|
* Timeout is not the same thing - however the code likes
|
| 2019 |
|
|
* to send both the same way (sigh).
|
| 2020 |
|
|
*/
|
| 2021 |
|
|
|
| 2022 |
|
|
if (tcp_close_state(sk,1)==1)
|
| 2023 |
|
|
{
|
| 2024 |
|
|
tcp_send_fin(sk);
|
| 2025 |
|
|
}
|
| 2026 |
|
|
|
| 2027 |
|
|
if (timeout) {
|
| 2028 |
|
|
cli();
|
| 2029 |
|
|
release_sock(sk);
|
| 2030 |
|
|
current->timeout = timeout;
|
| 2031 |
|
|
while(closing(sk) && current->timeout)
|
| 2032 |
|
|
{
|
| 2033 |
|
|
interruptible_sleep_on(sk->sleep);
|
| 2034 |
|
|
if (current->signal & ~current->blocked)
|
| 2035 |
|
|
{
|
| 2036 |
|
|
break;
|
| 2037 |
|
|
}
|
| 2038 |
|
|
}
|
| 2039 |
|
|
current->timeout=0;
|
| 2040 |
|
|
lock_sock(sk);
|
| 2041 |
|
|
sti();
|
| 2042 |
|
|
}
|
| 2043 |
|
|
|
| 2044 |
|
|
/* Now that the socket is dead, if we are in the FIN_WAIT2 state
|
| 2045 |
|
|
* we may need to set up a timer.
|
| 2046 |
|
|
*/
|
| 2047 |
|
|
if (sk->state==TCP_FIN_WAIT2)
|
| 2048 |
|
|
{
|
| 2049 |
|
|
int timer_active=del_timer(&sk->timer);
|
| 2050 |
|
|
if(timer_active)
|
| 2051 |
|
|
add_timer(&sk->timer);
|
| 2052 |
|
|
else
|
| 2053 |
|
|
tcp_reset_msl_timer(sk, TIME_CLOSE, TCP_FIN_TIMEOUT);
|
| 2054 |
|
|
}
|
| 2055 |
|
|
|
| 2056 |
|
|
release_sock(sk);
|
| 2057 |
|
|
|
| 2058 |
|
|
if(sk->state == TCP_CLOSE)
|
| 2059 |
|
|
tcp_v4_unhash(sk);
|
| 2060 |
|
|
|
| 2061 |
|
|
sk->dead = 1;
|
| 2062 |
|
|
}
|
| 2063 |
|
|
|
| 2064 |
|
|
|
| 2065 |
|
|
/*
|
| 2066 |
|
|
* Wait for a incoming connection, avoid race
|
| 2067 |
|
|
* conditions. This must be called with the socket
|
| 2068 |
|
|
* locked.
|
| 2069 |
|
|
*/
|
| 2070 |
|
|
static struct sk_buff * wait_for_connect(struct sock * sk)
|
| 2071 |
|
|
{
|
| 2072 |
|
|
struct wait_queue wait = { current, NULL };
|
| 2073 |
|
|
struct sk_buff * skb = NULL;
|
| 2074 |
|
|
|
| 2075 |
|
|
add_wait_queue(sk->sleep, &wait);
|
| 2076 |
|
|
for (;;) {
|
| 2077 |
|
|
current->state = TASK_INTERRUPTIBLE;
|
| 2078 |
|
|
end_bh_atomic();
|
| 2079 |
|
|
release_sock(sk);
|
| 2080 |
|
|
schedule();
|
| 2081 |
|
|
lock_sock(sk);
|
| 2082 |
|
|
start_bh_atomic();
|
| 2083 |
|
|
skb = tcp_find_established(sk);
|
| 2084 |
|
|
if (skb)
|
| 2085 |
|
|
break;
|
| 2086 |
|
|
if (current->signal & ~current->blocked)
|
| 2087 |
|
|
break;
|
| 2088 |
|
|
}
|
| 2089 |
|
|
remove_wait_queue(sk->sleep, &wait);
|
| 2090 |
|
|
return skb;
|
| 2091 |
|
|
}
|
| 2092 |
|
|
|
| 2093 |
|
|
/*
|
| 2094 |
|
|
* This will accept the next outstanding connection.
|
| 2095 |
|
|
*
|
| 2096 |
|
|
* Be careful about race conditions here - this is subtle.
|
| 2097 |
|
|
*/
|
| 2098 |
|
|
|
| 2099 |
|
|
static struct sock *tcp_accept(struct sock *sk, int flags)
|
| 2100 |
|
|
{
|
| 2101 |
|
|
int error;
|
| 2102 |
|
|
struct sk_buff *skb;
|
| 2103 |
|
|
struct sock *newsk = NULL;
|
| 2104 |
|
|
|
| 2105 |
|
|
/*
|
| 2106 |
|
|
* We need to make sure that this socket is listening,
|
| 2107 |
|
|
* and that it has something pending.
|
| 2108 |
|
|
*/
|
| 2109 |
|
|
|
| 2110 |
|
|
error = EINVAL;
|
| 2111 |
|
|
if (sk->state != TCP_LISTEN)
|
| 2112 |
|
|
goto no_listen;
|
| 2113 |
|
|
|
| 2114 |
|
|
lock_sock(sk);start_bh_atomic();
|
| 2115 |
|
|
|
| 2116 |
|
|
skb = tcp_find_established(sk);
|
| 2117 |
|
|
if (skb) {
|
| 2118 |
|
|
got_new_connect:
|
| 2119 |
|
|
__skb_unlink(skb, &sk->receive_queue);
|
| 2120 |
|
|
newsk = skb->sk;
|
| 2121 |
|
|
kfree_skb(skb, FREE_READ);
|
| 2122 |
|
|
sk->ack_backlog--;
|
| 2123 |
|
|
error = 0;
|
| 2124 |
|
|
out:
|
| 2125 |
|
|
end_bh_atomic();
|
| 2126 |
|
|
release_sock(sk);
|
| 2127 |
|
|
no_listen:
|
| 2128 |
|
|
sk->err = error;
|
| 2129 |
|
|
return newsk;
|
| 2130 |
|
|
}
|
| 2131 |
|
|
|
| 2132 |
|
|
error = EAGAIN;
|
| 2133 |
|
|
if (flags & O_NONBLOCK)
|
| 2134 |
|
|
goto out;
|
| 2135 |
|
|
skb = wait_for_connect(sk);
|
| 2136 |
|
|
if (skb)
|
| 2137 |
|
|
goto got_new_connect;
|
| 2138 |
|
|
error = ERESTARTSYS;
|
| 2139 |
|
|
goto out;
|
| 2140 |
|
|
}
|
| 2141 |
|
|
|
| 2142 |
|
|
/*
|
| 2143 |
|
|
* Check that a TCP address is unique, don't allow multiple
|
| 2144 |
|
|
* connects to/from the same address
|
| 2145 |
|
|
*/
|
| 2146 |
|
|
static int tcp_unique_address(u32 saddr, u16 snum, u32 daddr, u16 dnum)
|
| 2147 |
|
|
{
|
| 2148 |
|
|
int retval = 1, hashent = tcp_hashfn(saddr, snum, daddr, dnum);
|
| 2149 |
|
|
struct sock * sk;
|
| 2150 |
|
|
|
| 2151 |
|
|
/* Make sure we are allowed to connect here.
|
| 2152 |
|
|
* But freeze the hash while we snoop around.
|
| 2153 |
|
|
*/
|
| 2154 |
|
|
SOCKHASH_LOCK();
|
| 2155 |
|
|
sk = tcp_established_hash[hashent];
|
| 2156 |
|
|
for (; sk != NULL; sk = sk->next) {
|
| 2157 |
|
|
if(sk->daddr == daddr && /* remote address */
|
| 2158 |
|
|
sk->dummy_th.dest == dnum && /* remote port */
|
| 2159 |
|
|
sk->num == snum && /* local port */
|
| 2160 |
|
|
sk->saddr == saddr) { /* local address */
|
| 2161 |
|
|
retval = 0;
|
| 2162 |
|
|
break;
|
| 2163 |
|
|
}
|
| 2164 |
|
|
}
|
| 2165 |
|
|
SOCKHASH_UNLOCK();
|
| 2166 |
|
|
return retval;
|
| 2167 |
|
|
}
|
| 2168 |
|
|
|
| 2169 |
|
|
|
| 2170 |
|
|
/*
|
| 2171 |
|
|
* This will initiate an outgoing connection.
|
| 2172 |
|
|
*/
|
| 2173 |
|
|
|
| 2174 |
|
|
static int tcp_connect(struct sock *sk, struct sockaddr_in *usin, int addr_len)
|
| 2175 |
|
|
{
|
| 2176 |
|
|
struct sk_buff *buff;
|
| 2177 |
|
|
struct device *dev=NULL;
|
| 2178 |
|
|
unsigned char *ptr;
|
| 2179 |
|
|
int tmp;
|
| 2180 |
|
|
int atype;
|
| 2181 |
|
|
struct tcphdr *t1;
|
| 2182 |
|
|
struct rtable *rt;
|
| 2183 |
|
|
|
| 2184 |
|
|
if (sk->state != TCP_CLOSE)
|
| 2185 |
|
|
return(-EISCONN);
|
| 2186 |
|
|
|
| 2187 |
|
|
/*
|
| 2188 |
|
|
* Don't allow a double connect.
|
| 2189 |
|
|
*/
|
| 2190 |
|
|
|
| 2191 |
|
|
if(sk->daddr)
|
| 2192 |
|
|
return -EINVAL;
|
| 2193 |
|
|
|
| 2194 |
|
|
if (addr_len < 8)
|
| 2195 |
|
|
return(-EINVAL);
|
| 2196 |
|
|
|
| 2197 |
|
|
if (usin->sin_family && usin->sin_family != AF_INET)
|
| 2198 |
|
|
return(-EAFNOSUPPORT);
|
| 2199 |
|
|
|
| 2200 |
|
|
/*
|
| 2201 |
|
|
* connect() to INADDR_ANY means loopback (BSD'ism).
|
| 2202 |
|
|
*/
|
| 2203 |
|
|
|
| 2204 |
|
|
if (usin->sin_addr.s_addr==INADDR_ANY)
|
| 2205 |
|
|
usin->sin_addr.s_addr=ip_my_addr();
|
| 2206 |
|
|
|
| 2207 |
|
|
/*
|
| 2208 |
|
|
* Don't want a TCP connection going to a broadcast address
|
| 2209 |
|
|
*/
|
| 2210 |
|
|
|
| 2211 |
|
|
if ((atype=ip_chk_addr(usin->sin_addr.s_addr)) == IS_BROADCAST || atype==IS_MULTICAST)
|
| 2212 |
|
|
return -ENETUNREACH;
|
| 2213 |
|
|
|
| 2214 |
|
|
if (!tcp_unique_address(sk->saddr, sk->num, usin->sin_addr.s_addr, usin->sin_port))
|
| 2215 |
|
|
return -EADDRNOTAVAIL;
|
| 2216 |
|
|
|
| 2217 |
|
|
lock_sock(sk);
|
| 2218 |
|
|
sk->daddr = usin->sin_addr.s_addr;
|
| 2219 |
|
|
|
| 2220 |
|
|
sk->rcv_ack_cnt = 1;
|
| 2221 |
|
|
sk->err = 0;
|
| 2222 |
|
|
sk->dummy_th.dest = usin->sin_port;
|
| 2223 |
|
|
|
| 2224 |
|
|
buff = sock_wmalloc(sk,MAX_SYN_SIZE,0, GFP_KERNEL);
|
| 2225 |
|
|
if (buff == NULL)
|
| 2226 |
|
|
{
|
| 2227 |
|
|
release_sock(sk);
|
| 2228 |
|
|
return(-ENOMEM);
|
| 2229 |
|
|
}
|
| 2230 |
|
|
buff->sk = sk;
|
| 2231 |
|
|
buff->free = 0;
|
| 2232 |
|
|
buff->localroute = sk->localroute;
|
| 2233 |
|
|
|
| 2234 |
|
|
/* If this socket is bound to a particular device, make sure we use it. */
|
| 2235 |
|
|
dev = sk->bound_device;
|
| 2236 |
|
|
|
| 2237 |
|
|
/*
|
| 2238 |
|
|
* Put in the IP header and routing stuff.
|
| 2239 |
|
|
*/
|
| 2240 |
|
|
|
| 2241 |
|
|
tmp = sk->prot->build_header(buff, sk->saddr, sk->daddr, &dev,
|
| 2242 |
|
|
IPPROTO_TCP, sk->opt, MAX_SYN_SIZE,sk->ip_tos,sk->ip_ttl,&sk->ip_route_cache);
|
| 2243 |
|
|
if (tmp < 0)
|
| 2244 |
|
|
{
|
| 2245 |
|
|
sock_wfree(sk, buff);
|
| 2246 |
|
|
release_sock(sk);
|
| 2247 |
|
|
return(-ENETUNREACH);
|
| 2248 |
|
|
}
|
| 2249 |
|
|
if ((rt = sk->ip_route_cache) != NULL && !sk->saddr)
|
| 2250 |
|
|
sk->saddr = rt->rt_src;
|
| 2251 |
|
|
sk->rcv_saddr = sk->saddr;
|
| 2252 |
|
|
|
| 2253 |
|
|
/*
|
| 2254 |
|
|
* Set up our outgoing TCP sequence number
|
| 2255 |
|
|
*/
|
| 2256 |
|
|
sk->write_seq = secure_tcp_sequence_number(sk->saddr, sk->daddr,
|
| 2257 |
|
|
sk->dummy_th.source,
|
| 2258 |
|
|
usin->sin_port);
|
| 2259 |
|
|
sk->window_seq = sk->write_seq;
|
| 2260 |
|
|
sk->rcv_ack_seq = sk->write_seq -1;
|
| 2261 |
|
|
|
| 2262 |
|
|
t1 = (struct tcphdr *) skb_put(buff,sizeof(struct tcphdr));
|
| 2263 |
|
|
|
| 2264 |
|
|
memcpy(t1,(void *)&(sk->dummy_th), sizeof(*t1));
|
| 2265 |
|
|
buff->seq = sk->write_seq++;
|
| 2266 |
|
|
t1->seq = htonl(buff->seq);
|
| 2267 |
|
|
sk->sent_seq = sk->write_seq;
|
| 2268 |
|
|
buff->end_seq = sk->write_seq;
|
| 2269 |
|
|
t1->ack = 0;
|
| 2270 |
|
|
t1->window = 2;
|
| 2271 |
|
|
t1->syn = 1;
|
| 2272 |
|
|
t1->doff = 6;
|
| 2273 |
|
|
/* use 512 or whatever user asked for */
|
| 2274 |
|
|
|
| 2275 |
|
|
if(rt!=NULL && (rt->rt_flags&RTF_WINDOW))
|
| 2276 |
|
|
sk->window_clamp=rt->rt_window;
|
| 2277 |
|
|
else
|
| 2278 |
|
|
sk->window_clamp=0;
|
| 2279 |
|
|
|
| 2280 |
|
|
if (sk->user_mss)
|
| 2281 |
|
|
sk->mtu = sk->user_mss;
|
| 2282 |
|
|
else if (rt)
|
| 2283 |
|
|
sk->mtu = rt->rt_mtu - sizeof(struct iphdr) - sizeof(struct tcphdr);
|
| 2284 |
|
|
else
|
| 2285 |
|
|
sk->mtu = 576 - sizeof(struct iphdr) - sizeof(struct tcphdr);
|
| 2286 |
|
|
|
| 2287 |
|
|
/*
|
| 2288 |
|
|
* but not bigger than device MTU
|
| 2289 |
|
|
*/
|
| 2290 |
|
|
|
| 2291 |
|
|
sk->mtu = min(sk->mtu, dev->mtu - sizeof(struct iphdr) - sizeof(struct tcphdr));
|
| 2292 |
|
|
|
| 2293 |
|
|
/* Must check it here, just to be absolutely safe. If we end up
|
| 2294 |
|
|
* with an sk->mtu of zero, we can thus end up with an sk->mss
|
| 2295 |
|
|
* of zero, which causes us to bomb out in tcp_do_sendmsg. -DaveM
|
| 2296 |
|
|
*/
|
| 2297 |
|
|
if(sk->mtu < 32)
|
| 2298 |
|
|
sk->mtu = 32; /* Sanity limit */
|
| 2299 |
|
|
|
| 2300 |
|
|
/*
|
| 2301 |
|
|
* Put in the TCP options to say MTU.
|
| 2302 |
|
|
*/
|
| 2303 |
|
|
|
| 2304 |
|
|
ptr = skb_put(buff,4);
|
| 2305 |
|
|
ptr[0] = 2;
|
| 2306 |
|
|
ptr[1] = 4;
|
| 2307 |
|
|
ptr[2] = (sk->mtu) >> 8;
|
| 2308 |
|
|
ptr[3] = (sk->mtu) & 0xff;
|
| 2309 |
|
|
buff->csum = csum_partial(ptr, 4, 0);
|
| 2310 |
|
|
tcp_send_check(t1, sk->saddr, sk->daddr,
|
| 2311 |
|
|
sizeof(struct tcphdr) + 4, buff);
|
| 2312 |
|
|
|
| 2313 |
|
|
tcp_set_state(sk,TCP_SYN_SENT);
|
| 2314 |
|
|
|
| 2315 |
|
|
/* Socket identity change complete, no longer
|
| 2316 |
|
|
* in TCP_CLOSE, so rehash.
|
| 2317 |
|
|
*/
|
| 2318 |
|
|
tcp_v4_rehash(sk);
|
| 2319 |
|
|
|
| 2320 |
|
|
if(rt&&rt->rt_flags&RTF_IRTT)
|
| 2321 |
|
|
sk->rto = rt->rt_irtt;
|
| 2322 |
|
|
else
|
| 2323 |
|
|
sk->rto = TCP_TIMEOUT_INIT;
|
| 2324 |
|
|
sk->delack_timer.function = tcp_delack_timer;
|
| 2325 |
|
|
sk->delack_timer.data = (unsigned long) sk;
|
| 2326 |
|
|
sk->retransmit_timer.function = tcp_retransmit_timer;
|
| 2327 |
|
|
sk->retransmit_timer.data = (unsigned long)sk;
|
| 2328 |
|
|
sk->retransmits = 0;
|
| 2329 |
|
|
sk->prot->queue_xmit(sk, dev, buff, 0);
|
| 2330 |
|
|
tcp_reset_xmit_timer(sk, TIME_WRITE, sk->rto);
|
| 2331 |
|
|
tcp_statistics.TcpActiveOpens++;
|
| 2332 |
|
|
tcp_statistics.TcpOutSegs++;
|
| 2333 |
|
|
|
| 2334 |
|
|
release_sock(sk);
|
| 2335 |
|
|
return(0);
|
| 2336 |
|
|
}
|
| 2337 |
|
|
|
| 2338 |
|
|
/*
|
| 2339 |
|
|
* Socket option code for TCP.
|
| 2340 |
|
|
*/
|
| 2341 |
|
|
|
| 2342 |
|
|
int tcp_setsockopt(struct sock *sk, int level, int optname, char *optval, int optlen)
|
| 2343 |
|
|
{
|
| 2344 |
|
|
int val,err;
|
| 2345 |
|
|
|
| 2346 |
|
|
if(level!=SOL_TCP)
|
| 2347 |
|
|
return ip_setsockopt(sk,level,optname,optval,optlen);
|
| 2348 |
|
|
|
| 2349 |
|
|
if (optval == NULL)
|
| 2350 |
|
|
return(-EINVAL);
|
| 2351 |
|
|
|
| 2352 |
|
|
err=verify_area(VERIFY_READ, optval, sizeof(int));
|
| 2353 |
|
|
if(err)
|
| 2354 |
|
|
return err;
|
| 2355 |
|
|
|
| 2356 |
|
|
val = get_user((int *)optval);
|
| 2357 |
|
|
|
| 2358 |
|
|
switch(optname)
|
| 2359 |
|
|
{
|
| 2360 |
|
|
case TCP_MAXSEG:
|
| 2361 |
|
|
/*
|
| 2362 |
|
|
* values greater than interface MTU won't take effect. however at
|
| 2363 |
|
|
* the point when this call is done we typically don't yet know
|
| 2364 |
|
|
* which interface is going to be used
|
| 2365 |
|
|
*/
|
| 2366 |
|
|
if(val<1||val>MAX_WINDOW)
|
| 2367 |
|
|
return -EINVAL;
|
| 2368 |
|
|
sk->user_mss=val;
|
| 2369 |
|
|
return 0;
|
| 2370 |
|
|
case TCP_NODELAY:
|
| 2371 |
|
|
sk->nonagle=(val==0)?0:1;
|
| 2372 |
|
|
return 0;
|
| 2373 |
|
|
default:
|
| 2374 |
|
|
return(-ENOPROTOOPT);
|
| 2375 |
|
|
}
|
| 2376 |
|
|
}
|
| 2377 |
|
|
|
| 2378 |
|
|
int tcp_getsockopt(struct sock *sk, int level, int optname, char *optval, int *optlen)
|
| 2379 |
|
|
{
|
| 2380 |
|
|
int val,err;
|
| 2381 |
|
|
|
| 2382 |
|
|
if(level!=SOL_TCP)
|
| 2383 |
|
|
return ip_getsockopt(sk,level,optname,optval,optlen);
|
| 2384 |
|
|
|
| 2385 |
|
|
switch(optname)
|
| 2386 |
|
|
{
|
| 2387 |
|
|
case TCP_MAXSEG:
|
| 2388 |
|
|
val=sk->user_mss;
|
| 2389 |
|
|
break;
|
| 2390 |
|
|
case TCP_NODELAY:
|
| 2391 |
|
|
val=sk->nonagle;
|
| 2392 |
|
|
break;
|
| 2393 |
|
|
default:
|
| 2394 |
|
|
return(-ENOPROTOOPT);
|
| 2395 |
|
|
}
|
| 2396 |
|
|
err=verify_area(VERIFY_WRITE, optlen, sizeof(int));
|
| 2397 |
|
|
if(err)
|
| 2398 |
|
|
return err;
|
| 2399 |
|
|
put_user(sizeof(int),(int *) optlen);
|
| 2400 |
|
|
|
| 2401 |
|
|
err=verify_area(VERIFY_WRITE, optval, sizeof(int));
|
| 2402 |
|
|
if(err)
|
| 2403 |
|
|
return err;
|
| 2404 |
|
|
put_user(val,(int *)optval);
|
| 2405 |
|
|
|
| 2406 |
|
|
return(0);
|
| 2407 |
|
|
}
|
| 2408 |
|
|
|
| 2409 |
|
|
|
| 2410 |
|
|
struct proto tcp_prot = {
|
| 2411 |
|
|
(struct sock *)&tcp_prot, /* sklist_next */
|
| 2412 |
|
|
(struct sock *)&tcp_prot, /* sklist_prev */
|
| 2413 |
|
|
tcp_close, /* close */
|
| 2414 |
|
|
ip_build_header, /* build_header */
|
| 2415 |
|
|
tcp_connect, /* connect */
|
| 2416 |
|
|
tcp_accept, /* accept */
|
| 2417 |
|
|
ip_queue_xmit, /* queue_xmit */
|
| 2418 |
|
|
tcp_retransmit, /* retransmit */
|
| 2419 |
|
|
tcp_write_wakeup, /* write_wakeup */
|
| 2420 |
|
|
tcp_read_wakeup, /* read_wakeup */
|
| 2421 |
|
|
tcp_rcv, /* rcv */
|
| 2422 |
|
|
tcp_select, /* select */
|
| 2423 |
|
|
tcp_ioctl, /* ioctl */
|
| 2424 |
|
|
NULL, /* init */
|
| 2425 |
|
|
tcp_shutdown, /* shutdown */
|
| 2426 |
|
|
tcp_setsockopt, /* setsockopt */
|
| 2427 |
|
|
tcp_getsockopt, /* getsockopt */
|
| 2428 |
|
|
tcp_sendmsg, /* sendmsg */
|
| 2429 |
|
|
tcp_recvmsg, /* recvmsg */
|
| 2430 |
|
|
NULL, /* bind */
|
| 2431 |
|
|
tcp_v4_hash, /* hash */
|
| 2432 |
|
|
tcp_v4_unhash, /* unhash */
|
| 2433 |
|
|
tcp_v4_rehash, /* rehash */
|
| 2434 |
|
|
tcp_good_socknum, /* good_socknum */
|
| 2435 |
|
|
tcp_v4_verify_bind, /* verify_bind */
|
| 2436 |
|
|
128, /* max_header */
|
| 2437 |
|
|
0, /* retransmits */
|
| 2438 |
|
|
"TCP", /* name */
|
| 2439 |
|
|
0, /* inuse */
|
| 2440 |
|
|
|
| 2441 |
|
|
};
|
