| 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_input.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: Eric Schenk : avoid multiple retransmissions in one
|
| 23 |
|
|
* : round trip timeout.
|
| 24 |
|
|
* Eric Schenk : tcp rst and syn cookies to deal
|
| 25 |
|
|
* : with synflooding attacks.
|
| 26 |
|
|
* Eric Schenk : If a simultaneous close occurred, and the
|
| 27 |
|
|
* connection was over an assymetric route, we
|
| 28 |
|
|
* would lose badly if we dropped our outgoing
|
| 29 |
|
|
* FIN because the route disappeared on us.
|
| 30 |
|
|
* We now handle this case correctly.
|
| 31 |
|
|
* Eric Schenk : Handle the case where a route changes, and
|
| 32 |
|
|
* thus the outgoing device does as well, when
|
| 33 |
|
|
* skb's are on the retransmit queue which still
|
| 34 |
|
|
* refer to the old obsolete destination.
|
| 35 |
|
|
* Elliot Poger : Added support for SO_BINDTODEVICE.
|
| 36 |
|
|
* Juan Jose Ciarlante : Added sock dynamic source address rewriting
|
| 37 |
|
|
* Alan Cox : Clear reserved fields - bug reported by
|
| 38 |
|
|
* J Hadi Salim
|
| 39 |
|
|
*/
|
| 40 |
|
|
|
| 41 |
|
|
#include <linux/config.h>
|
| 42 |
|
|
#include <net/tcp.h>
|
| 43 |
|
|
#include <linux/ip_fw.h>
|
| 44 |
|
|
#include <linux/firewall.h>
|
| 45 |
|
|
#include <linux/interrupt.h>
|
| 46 |
|
|
#ifdef CONFIG_RST_COOKIES
|
| 47 |
|
|
#include <linux/random.h>
|
| 48 |
|
|
#endif
|
| 49 |
|
|
|
| 50 |
|
|
/*
|
| 51 |
|
|
* RFC 1122 says:
|
| 52 |
|
|
*
|
| 53 |
|
|
* "the suggested [SWS] avoidance algorithm for the receiver is to keep
|
| 54 |
|
|
* RECV.NEXT + RCV.WIN fixed until:
|
| 55 |
|
|
* RCV.BUFF - RCV.USER - RCV.WINDOW >= min(1/2 RCV.BUFF, MSS)"
|
| 56 |
|
|
*
|
| 57 |
|
|
* We do BSD style SWS avoidance -- note that RFC1122 only says we
|
| 58 |
|
|
* must do silly window avoidance, it does not require that we use
|
| 59 |
|
|
* the suggested algorithm. Following BSD avoids breaking header
|
| 60 |
|
|
* prediction.
|
| 61 |
|
|
*
|
| 62 |
|
|
* The "rcvbuf" and "rmem_alloc" values are shifted by 1, because
|
| 63 |
|
|
* they also contain buffer handling overhead etc, so the window
|
| 64 |
|
|
* we actually use is essentially based on only half those values.
|
| 65 |
|
|
*/
|
| 66 |
|
|
int tcp_new_window(struct sock * sk)
|
| 67 |
|
|
{
|
| 68 |
|
|
unsigned long window = sk->window;
|
| 69 |
|
|
unsigned long minwin, maxwin;
|
| 70 |
|
|
unsigned long free_space;
|
| 71 |
|
|
|
| 72 |
|
|
/* Get minimum and maximum window values.. */
|
| 73 |
|
|
minwin = sk->mss;
|
| 74 |
|
|
if (!minwin)
|
| 75 |
|
|
minwin = sk->mtu;
|
| 76 |
|
|
if (!minwin) {
|
| 77 |
|
|
printk(KERN_DEBUG "tcp_new_window: mss fell to 0.\n");
|
| 78 |
|
|
minwin = 1;
|
| 79 |
|
|
}
|
| 80 |
|
|
maxwin = sk->window_clamp;
|
| 81 |
|
|
if (!maxwin)
|
| 82 |
|
|
maxwin = MAX_WINDOW;
|
| 83 |
|
|
|
| 84 |
|
|
if (minwin > maxwin/2)
|
| 85 |
|
|
minwin = maxwin/2;
|
| 86 |
|
|
|
| 87 |
|
|
/* Get current rcvbuf size.. */
|
| 88 |
|
|
free_space = sk->rcvbuf/2;
|
| 89 |
|
|
if (free_space < minwin) {
|
| 90 |
|
|
sk->rcvbuf = minwin*2;
|
| 91 |
|
|
free_space = minwin;
|
| 92 |
|
|
}
|
| 93 |
|
|
|
| 94 |
|
|
/* Check rcvbuf against used and minimum window */
|
| 95 |
|
|
free_space -= sk->rmem_alloc/2;
|
| 96 |
|
|
if ((long)(free_space - minwin) < 0) /* SWS avoidance */
|
| 97 |
|
|
return 0;
|
| 98 |
|
|
|
| 99 |
|
|
/* Try to avoid the divide and multiply if we can */
|
| 100 |
|
|
if (window <= free_space - minwin || window > free_space)
|
| 101 |
|
|
window = (free_space/minwin)*minwin;
|
| 102 |
|
|
|
| 103 |
|
|
if (window > maxwin)
|
| 104 |
|
|
window = maxwin;
|
| 105 |
|
|
return window;
|
| 106 |
|
|
}
|
| 107 |
|
|
|
| 108 |
|
|
/*
|
| 109 |
|
|
* Get rid of any delayed acks, we sent one already..
|
| 110 |
|
|
*/
|
| 111 |
|
|
static __inline__ void clear_delayed_acks(struct sock * sk)
|
| 112 |
|
|
{
|
| 113 |
|
|
sk->ack_timed = 0;
|
| 114 |
|
|
sk->ack_backlog = 0;
|
| 115 |
|
|
sk->bytes_rcv = 0;
|
| 116 |
|
|
del_timer(&sk->delack_timer);
|
| 117 |
|
|
}
|
| 118 |
|
|
|
| 119 |
|
|
/*
|
| 120 |
|
|
* This is the main buffer sending routine. We queue the buffer
|
| 121 |
|
|
* having checked it is sane seeming.
|
| 122 |
|
|
*/
|
| 123 |
|
|
|
| 124 |
|
|
void tcp_send_skb(struct sock *sk, struct sk_buff *skb)
|
| 125 |
|
|
{
|
| 126 |
|
|
int size;
|
| 127 |
|
|
struct tcphdr * th = skb->h.th;
|
| 128 |
|
|
|
| 129 |
|
|
/*
|
| 130 |
|
|
* length of packet (not counting length of pre-tcp headers)
|
| 131 |
|
|
*/
|
| 132 |
|
|
|
| 133 |
|
|
size = skb->len - ((unsigned char *) th - skb->data);
|
| 134 |
|
|
|
| 135 |
|
|
/*
|
| 136 |
|
|
* Sanity check it..
|
| 137 |
|
|
*/
|
| 138 |
|
|
|
| 139 |
|
|
if (size < sizeof(struct tcphdr) || size > skb->len)
|
| 140 |
|
|
{
|
| 141 |
|
|
printk(KERN_ERR "tcp_send_skb: bad skb (skb = %p, data = %p, th = %p, len = %lu)\n",
|
| 142 |
|
|
skb, skb->data, th, skb->len);
|
| 143 |
|
|
kfree_skb(skb, FREE_WRITE);
|
| 144 |
|
|
return;
|
| 145 |
|
|
}
|
| 146 |
|
|
|
| 147 |
|
|
/*
|
| 148 |
|
|
* If we have queued a header size packet.. (these crash a few
|
| 149 |
|
|
* tcp stacks if ack is not set)
|
| 150 |
|
|
*/
|
| 151 |
|
|
|
| 152 |
|
|
if (size == sizeof(struct tcphdr))
|
| 153 |
|
|
{
|
| 154 |
|
|
/* If it's got a syn or fin it's notionally included in the size..*/
|
| 155 |
|
|
if(!th->syn && !th->fin)
|
| 156 |
|
|
{
|
| 157 |
|
|
printk(KERN_ERR "tcp_send_skb: attempt to queue a bogon.\n");
|
| 158 |
|
|
kfree_skb(skb,FREE_WRITE);
|
| 159 |
|
|
return;
|
| 160 |
|
|
}
|
| 161 |
|
|
}
|
| 162 |
|
|
|
| 163 |
|
|
/*
|
| 164 |
|
|
* Jacobson recommends this in the appendix of his SIGCOMM'88 paper.
|
| 165 |
|
|
* The idea is to do a slow start again if we haven't been doing
|
| 166 |
|
|
* anything for a long time, in which case we have no reason to
|
| 167 |
|
|
* believe that our congestion window is still correct.
|
| 168 |
|
|
*/
|
| 169 |
|
|
if (sk->send_head == 0 && (jiffies - sk->idletime) > sk->rto) {
|
| 170 |
|
|
sk->cong_window = 1;
|
| 171 |
|
|
sk->cong_count = 0;
|
| 172 |
|
|
}
|
| 173 |
|
|
|
| 174 |
|
|
/*
|
| 175 |
|
|
* Actual processing.
|
| 176 |
|
|
*/
|
| 177 |
|
|
|
| 178 |
|
|
tcp_statistics.TcpOutSegs++;
|
| 179 |
|
|
skb->seq = ntohl(th->seq);
|
| 180 |
|
|
skb->end_seq = skb->seq + size - 4*th->doff + th->fin;
|
| 181 |
|
|
|
| 182 |
|
|
/*
|
| 183 |
|
|
* We must queue if
|
| 184 |
|
|
*
|
| 185 |
|
|
* a) The right edge of this frame exceeds the window
|
| 186 |
|
|
* b) We are retransmitting (Nagle's rule)
|
| 187 |
|
|
* c) We have too many packets 'in flight'
|
| 188 |
|
|
*/
|
| 189 |
|
|
|
| 190 |
|
|
if (after(skb->end_seq, sk->window_seq) ||
|
| 191 |
|
|
(sk->retransmits && sk->ip_xmit_timeout == TIME_WRITE) ||
|
| 192 |
|
|
sk->packets_out >= sk->cong_window)
|
| 193 |
|
|
{
|
| 194 |
|
|
/* checksum will be supplied by tcp_write_xmit. So
|
| 195 |
|
|
* we shouldn't need to set it at all. I'm being paranoid */
|
| 196 |
|
|
th->check = 0;
|
| 197 |
|
|
if (skb->next != NULL)
|
| 198 |
|
|
{
|
| 199 |
|
|
printk(KERN_ERR "tcp_send_partial: next != NULL\n");
|
| 200 |
|
|
skb_unlink(skb);
|
| 201 |
|
|
}
|
| 202 |
|
|
skb_queue_tail(&sk->write_queue, skb);
|
| 203 |
|
|
|
| 204 |
|
|
if (before(sk->window_seq, sk->write_queue.next->end_seq) &&
|
| 205 |
|
|
sk->send_head == NULL && sk->ack_backlog == 0)
|
| 206 |
|
|
tcp_reset_xmit_timer(sk, TIME_PROBE0, sk->rto);
|
| 207 |
|
|
}
|
| 208 |
|
|
else
|
| 209 |
|
|
{
|
| 210 |
|
|
/*
|
| 211 |
|
|
* This is going straight out
|
| 212 |
|
|
*/
|
| 213 |
|
|
clear_delayed_acks(sk);
|
| 214 |
|
|
th->ack_seq = htonl(sk->acked_seq);
|
| 215 |
|
|
th->window = htons(tcp_select_window(sk));
|
| 216 |
|
|
|
| 217 |
|
|
tcp_send_check(th, sk->saddr, sk->daddr, size, skb);
|
| 218 |
|
|
|
| 219 |
|
|
sk->sent_seq = sk->write_seq;
|
| 220 |
|
|
|
| 221 |
|
|
/*
|
| 222 |
|
|
* This is mad. The tcp retransmit queue is put together
|
| 223 |
|
|
* by the ip layer. This causes half the problems with
|
| 224 |
|
|
* unroutable FIN's and other things.
|
| 225 |
|
|
*/
|
| 226 |
|
|
|
| 227 |
|
|
sk->prot->queue_xmit(sk, skb->dev, skb, 0);
|
| 228 |
|
|
|
| 229 |
|
|
/*
|
| 230 |
|
|
* Set for next retransmit based on expected ACK time
|
| 231 |
|
|
* of the first packet in the resend queue.
|
| 232 |
|
|
* This is no longer a window behind.
|
| 233 |
|
|
*/
|
| 234 |
|
|
|
| 235 |
|
|
tcp_reset_xmit_timer(sk, TIME_WRITE, sk->rto);
|
| 236 |
|
|
}
|
| 237 |
|
|
}
|
| 238 |
|
|
|
| 239 |
|
|
/*
|
| 240 |
|
|
* Locking problems lead us to a messy situation where we can have
|
| 241 |
|
|
* multiple partially complete buffers queued up. This is really bad
|
| 242 |
|
|
* as we don't want to be sending partial buffers. Fix this with
|
| 243 |
|
|
* a semaphore or similar to lock tcp_write per socket.
|
| 244 |
|
|
*
|
| 245 |
|
|
* These routines are pretty self descriptive.
|
| 246 |
|
|
*/
|
| 247 |
|
|
|
| 248 |
|
|
struct sk_buff * tcp_dequeue_partial(struct sock * sk)
|
| 249 |
|
|
{
|
| 250 |
|
|
struct sk_buff * skb;
|
| 251 |
|
|
unsigned long flags;
|
| 252 |
|
|
|
| 253 |
|
|
save_flags(flags);
|
| 254 |
|
|
cli();
|
| 255 |
|
|
skb = sk->partial;
|
| 256 |
|
|
if (skb) {
|
| 257 |
|
|
sk->partial = NULL;
|
| 258 |
|
|
del_timer(&sk->partial_timer);
|
| 259 |
|
|
}
|
| 260 |
|
|
restore_flags(flags);
|
| 261 |
|
|
return skb;
|
| 262 |
|
|
}
|
| 263 |
|
|
|
| 264 |
|
|
/*
|
| 265 |
|
|
* Empty the partial queue
|
| 266 |
|
|
*/
|
| 267 |
|
|
|
| 268 |
|
|
void tcp_send_partial(struct sock *sk)
|
| 269 |
|
|
{
|
| 270 |
|
|
struct sk_buff *skb;
|
| 271 |
|
|
|
| 272 |
|
|
if (sk == NULL)
|
| 273 |
|
|
return;
|
| 274 |
|
|
while ((skb = tcp_dequeue_partial(sk)) != NULL)
|
| 275 |
|
|
tcp_send_skb(sk, skb);
|
| 276 |
|
|
}
|
| 277 |
|
|
|
| 278 |
|
|
/*
|
| 279 |
|
|
* Queue a partial frame
|
| 280 |
|
|
*/
|
| 281 |
|
|
|
| 282 |
|
|
void tcp_enqueue_partial(struct sk_buff * skb, struct sock * sk)
|
| 283 |
|
|
{
|
| 284 |
|
|
struct sk_buff * tmp;
|
| 285 |
|
|
unsigned long flags;
|
| 286 |
|
|
|
| 287 |
|
|
save_flags(flags);
|
| 288 |
|
|
cli();
|
| 289 |
|
|
tmp = sk->partial;
|
| 290 |
|
|
if (tmp)
|
| 291 |
|
|
del_timer(&sk->partial_timer);
|
| 292 |
|
|
sk->partial = skb;
|
| 293 |
|
|
init_timer(&sk->partial_timer);
|
| 294 |
|
|
/*
|
| 295 |
|
|
* Wait up to 30 second for the buffer to fill.
|
| 296 |
|
|
* ( I have no idea why this timer is here!
|
| 297 |
|
|
* It seems to be sillyness for interactive response. Linus?
|
| 298 |
|
|
* )
|
| 299 |
|
|
*/
|
| 300 |
|
|
sk->partial_timer.expires = jiffies+30*HZ;
|
| 301 |
|
|
sk->partial_timer.function = (void (*)(unsigned long)) tcp_send_partial;
|
| 302 |
|
|
sk->partial_timer.data = (unsigned long) sk;
|
| 303 |
|
|
add_timer(&sk->partial_timer);
|
| 304 |
|
|
restore_flags(flags);
|
| 305 |
|
|
if (tmp)
|
| 306 |
|
|
tcp_send_skb(sk, tmp);
|
| 307 |
|
|
}
|
| 308 |
|
|
|
| 309 |
|
|
/*
|
| 310 |
|
|
* This routine takes stuff off of the write queue,
|
| 311 |
|
|
* and puts it in the xmit queue. This happens as incoming acks
|
| 312 |
|
|
* open up the remote window for us.
|
| 313 |
|
|
*/
|
| 314 |
|
|
|
| 315 |
|
|
void tcp_write_xmit(struct sock *sk)
|
| 316 |
|
|
{
|
| 317 |
|
|
struct sk_buff *skb;
|
| 318 |
|
|
|
| 319 |
|
|
/*
|
| 320 |
|
|
* The bytes will have to remain here. In time closedown will
|
| 321 |
|
|
* empty the write queue and all will be happy
|
| 322 |
|
|
*/
|
| 323 |
|
|
|
| 324 |
|
|
if(sk->zapped)
|
| 325 |
|
|
return;
|
| 326 |
|
|
|
| 327 |
|
|
/*
|
| 328 |
|
|
* Anything on the transmit queue that fits the window can
|
| 329 |
|
|
* be added providing we are not
|
| 330 |
|
|
*
|
| 331 |
|
|
* a) retransmitting (Nagle's rule)
|
| 332 |
|
|
* b) exceeding our congestion window.
|
| 333 |
|
|
*/
|
| 334 |
|
|
|
| 335 |
|
|
while((skb = skb_peek(&sk->write_queue)) != NULL &&
|
| 336 |
|
|
!after(skb->end_seq, sk->window_seq) &&
|
| 337 |
|
|
(sk->retransmits == 0 ||
|
| 338 |
|
|
sk->ip_xmit_timeout != TIME_WRITE ||
|
| 339 |
|
|
!after(skb->end_seq, sk->rcv_ack_seq))
|
| 340 |
|
|
&& sk->packets_out < sk->cong_window)
|
| 341 |
|
|
{
|
| 342 |
|
|
IS_SKB(skb);
|
| 343 |
|
|
skb_unlink(skb);
|
| 344 |
|
|
|
| 345 |
|
|
/*
|
| 346 |
|
|
* See if we really need to send the whole packet.
|
| 347 |
|
|
*/
|
| 348 |
|
|
|
| 349 |
|
|
if (before(skb->end_seq, sk->rcv_ack_seq +1)) {
|
| 350 |
|
|
/*
|
| 351 |
|
|
* This is acked data. We can discard it.
|
| 352 |
|
|
* This implies the packet was sent out
|
| 353 |
|
|
* of the write queue by a zero window probe.
|
| 354 |
|
|
*/
|
| 355 |
|
|
|
| 356 |
|
|
sk->retransmits = 0;
|
| 357 |
|
|
kfree_skb(skb, FREE_WRITE);
|
| 358 |
|
|
if (!sk->dead)
|
| 359 |
|
|
sk->write_space(sk);
|
| 360 |
|
|
} else {
|
| 361 |
|
|
struct tcphdr *th;
|
| 362 |
|
|
struct iphdr *iph;
|
| 363 |
|
|
int size;
|
| 364 |
|
|
|
| 365 |
|
|
iph = skb->ip_hdr;
|
| 366 |
|
|
th = (struct tcphdr *)(((char *)iph) +(iph->ihl << 2));
|
| 367 |
|
|
|
| 368 |
|
|
/* See if we need to shrink the leading packet on
|
| 369 |
|
|
* the retransmit queue. Strictly speaking, we
|
| 370 |
|
|
* should never need to do this, but some buggy TCP
|
| 371 |
|
|
* implementations get confused if you send them
|
| 372 |
|
|
* a packet that contains both old and new data. (Feh!)
|
| 373 |
|
|
* Soooo, we have this uglyness here.
|
| 374 |
|
|
*/
|
| 375 |
|
|
if (after(sk->rcv_ack_seq,skb->seq+th->syn))
|
| 376 |
|
|
tcp_shrink_skb(sk,skb,sk->rcv_ack_seq);
|
| 377 |
|
|
|
| 378 |
|
|
size = skb->len - (((unsigned char *) th) - skb->data);
|
| 379 |
|
|
#ifndef CONFIG_NO_PATH_MTU_DISCOVERY
|
| 380 |
|
|
if (size > sk->mtu - sizeof(struct iphdr))
|
| 381 |
|
|
{
|
| 382 |
|
|
iph->frag_off &= ~htons(IP_DF);
|
| 383 |
|
|
ip_send_check(iph);
|
| 384 |
|
|
}
|
| 385 |
|
|
#endif
|
| 386 |
|
|
|
| 387 |
|
|
/*
|
| 388 |
|
|
* put in the ack seq and window at this point rather than earlier,
|
| 389 |
|
|
* in order to keep them monotonic. We really want to avoid taking
|
| 390 |
|
|
* back window allocations. That's legal, but RFC1122 says it's frowned on.
|
| 391 |
|
|
* Ack and window will in general have changed since this packet was put
|
| 392 |
|
|
* on the write queue.
|
| 393 |
|
|
*/
|
| 394 |
|
|
th->ack_seq = htonl(sk->acked_seq);
|
| 395 |
|
|
th->window = htons(tcp_select_window(sk));
|
| 396 |
|
|
|
| 397 |
|
|
tcp_send_check(th, sk->saddr, sk->daddr, size, skb);
|
| 398 |
|
|
|
| 399 |
|
|
sk->sent_seq = skb->end_seq;
|
| 400 |
|
|
|
| 401 |
|
|
/*
|
| 402 |
|
|
* IP manages our queue for some crazy reason
|
| 403 |
|
|
*/
|
| 404 |
|
|
#ifndef NO_DAVEM_FIX
|
| 405 |
|
|
sk->prot->queue_xmit(sk, skb->dev, skb, 0);
|
| 406 |
|
|
#else
|
| 407 |
|
|
sk->prot->queue_xmit(sk, skb->dev, skb, skb->free);
|
| 408 |
|
|
#endif
|
| 409 |
|
|
|
| 410 |
|
|
clear_delayed_acks(sk);
|
| 411 |
|
|
|
| 412 |
|
|
tcp_reset_xmit_timer(sk, TIME_WRITE, sk->rto);
|
| 413 |
|
|
}
|
| 414 |
|
|
}
|
| 415 |
|
|
}
|
| 416 |
|
|
|
| 417 |
|
|
|
| 418 |
|
|
/*
|
| 419 |
|
|
* A socket has timed out on its send queue and wants to do a
|
| 420 |
|
|
* little retransmitting. Currently this means TCP.
|
| 421 |
|
|
*/
|
| 422 |
|
|
|
| 423 |
|
|
void tcp_do_retransmit(struct sock *sk, int all)
|
| 424 |
|
|
{
|
| 425 |
|
|
struct sk_buff * skb;
|
| 426 |
|
|
struct proto *prot;
|
| 427 |
|
|
struct device *dev;
|
| 428 |
|
|
struct rtable *rt;
|
| 429 |
|
|
|
| 430 |
|
|
prot = sk->prot;
|
| 431 |
|
|
if (!all) {
|
| 432 |
|
|
/*
|
| 433 |
|
|
* If we are just retransmitting one packet reset
|
| 434 |
|
|
* to the start of the queue.
|
| 435 |
|
|
*/
|
| 436 |
|
|
sk->send_next = sk->send_head;
|
| 437 |
|
|
sk->packets_out = 0;
|
| 438 |
|
|
}
|
| 439 |
|
|
skb = sk->send_next;
|
| 440 |
|
|
|
| 441 |
|
|
while (skb != NULL)
|
| 442 |
|
|
{
|
| 443 |
|
|
struct tcphdr *th;
|
| 444 |
|
|
struct iphdr *iph;
|
| 445 |
|
|
int size;
|
| 446 |
|
|
unsigned long flags;
|
| 447 |
|
|
|
| 448 |
|
|
dev = skb->dev;
|
| 449 |
|
|
IS_SKB(skb);
|
| 450 |
|
|
skb->when = jiffies;
|
| 451 |
|
|
|
| 452 |
|
|
/* dl1bke 960201 - @%$$! Hope this cures strange race conditions */
|
| 453 |
|
|
/* with AX.25 mode VC. (esp. DAMA) */
|
| 454 |
|
|
/* if the buffer is locked we should not retransmit */
|
| 455 |
|
|
/* anyway, so we don't need all the fuss to prepare */
|
| 456 |
|
|
/* the buffer in this case. */
|
| 457 |
|
|
/* (the skb_pull() changes skb->data while we may */
|
| 458 |
|
|
/* actually try to send the data. Ouch. A side */
|
| 459 |
|
|
/* effect is that we'll send some unnecessary data, */
|
| 460 |
|
|
/* but the alternative is disastrous... */
|
| 461 |
|
|
|
| 462 |
|
|
save_flags(flags);
|
| 463 |
|
|
cli();
|
| 464 |
|
|
|
| 465 |
|
|
if (skb_device_locked(skb)) {
|
| 466 |
|
|
restore_flags(flags);
|
| 467 |
|
|
break;
|
| 468 |
|
|
}
|
| 469 |
|
|
|
| 470 |
|
|
/* Unlink from any chain */
|
| 471 |
|
|
skb_unlink(skb);
|
| 472 |
|
|
|
| 473 |
|
|
restore_flags(flags);
|
| 474 |
|
|
|
| 475 |
|
|
/*
|
| 476 |
|
|
* Discard the surplus MAC header
|
| 477 |
|
|
*/
|
| 478 |
|
|
|
| 479 |
|
|
skb_pull(skb,((unsigned char *)skb->ip_hdr)-skb->data);
|
| 480 |
|
|
|
| 481 |
|
|
/*
|
| 482 |
|
|
* In general it's OK just to use the old packet. However we
|
| 483 |
|
|
* need to use the current ack and window fields. Urg and
|
| 484 |
|
|
* urg_ptr could possibly stand to be updated as well, but we
|
| 485 |
|
|
* don't keep the necessary data. That shouldn't be a problem,
|
| 486 |
|
|
* if the other end is doing the right thing. Since we're
|
| 487 |
|
|
* changing the packet, we have to issue a new IP identifier.
|
| 488 |
|
|
*/
|
| 489 |
|
|
|
| 490 |
|
|
iph = (struct iphdr *)skb->data;
|
| 491 |
|
|
th = (struct tcphdr *)(((char *)iph) + (iph->ihl << 2));
|
| 492 |
|
|
|
| 493 |
|
|
/* See if we need to shrink the leading packet on
|
| 494 |
|
|
* the retransmit queue. Strictly speaking, we
|
| 495 |
|
|
* should never need to do this, but some buggy TCP
|
| 496 |
|
|
* implementations get confused if you send them
|
| 497 |
|
|
* a packet that contains both old and new data. (Feh!)
|
| 498 |
|
|
* Soooo, we have this uglyness here.
|
| 499 |
|
|
*
|
| 500 |
|
|
* Is the && test needed here? If so, then it implies that
|
| 501 |
|
|
* we might be retransmitting an acked packet. This code is
|
| 502 |
|
|
* needed here to talk to Solaris 2.6 stack.
|
| 503 |
|
|
*/
|
| 504 |
|
|
if (after(sk->rcv_ack_seq,skb->seq+th->syn) && before(sk->rcv_ack_seq, skb->end_seq))
|
| 505 |
|
|
tcp_shrink_skb(sk,skb,sk->rcv_ack_seq);
|
| 506 |
|
|
|
| 507 |
|
|
size = ntohs(iph->tot_len) - (iph->ihl<<2);
|
| 508 |
|
|
|
| 509 |
|
|
/*
|
| 510 |
|
|
* Note: We ought to check for window limits here but
|
| 511 |
|
|
* currently this is done (less efficiently) elsewhere.
|
| 512 |
|
|
*/
|
| 513 |
|
|
|
| 514 |
|
|
/*
|
| 515 |
|
|
* Put a MAC header back on (may cause ARPing)
|
| 516 |
|
|
*/
|
| 517 |
|
|
|
| 518 |
|
|
{
|
| 519 |
|
|
/* ANK: UGLY, but the bug, that was here, should be fixed.
|
| 520 |
|
|
*/
|
| 521 |
|
|
struct options * opt = (struct options*)skb->proto_priv;
|
| 522 |
|
|
rt = ip_check_route(&sk->ip_route_cache, opt->srr?opt->faddr:iph->daddr,
|
| 523 |
|
|
skb->localroute, sk->bound_device);
|
| 524 |
|
|
}
|
| 525 |
|
|
|
| 526 |
|
|
iph->id = htons(ip_id_count++);
|
| 527 |
|
|
#ifndef CONFIG_NO_PATH_MTU_DISCOVERY
|
| 528 |
|
|
if (rt && ntohs(iph->tot_len) > rt->rt_mtu)
|
| 529 |
|
|
iph->frag_off &= ~htons(IP_DF);
|
| 530 |
|
|
#endif
|
| 531 |
|
|
ip_send_check(iph);
|
| 532 |
|
|
|
| 533 |
|
|
if (rt==NULL) /* Deep poo */
|
| 534 |
|
|
{
|
| 535 |
|
|
if(skb->sk)
|
| 536 |
|
|
{
|
| 537 |
|
|
skb->sk->err_soft=ENETUNREACH;
|
| 538 |
|
|
skb->sk->error_report(skb->sk);
|
| 539 |
|
|
}
|
| 540 |
|
|
/* Can't transmit this packet, no reason
|
| 541 |
|
|
* to transmit the later ones, even if
|
| 542 |
|
|
* the congestion window allows.
|
| 543 |
|
|
*/
|
| 544 |
|
|
break;
|
| 545 |
|
|
}
|
| 546 |
|
|
else
|
| 547 |
|
|
{
|
| 548 |
|
|
dev=rt->rt_dev;
|
| 549 |
|
|
if (skb->dev != dev && skb->link3 == 0
|
| 550 |
|
|
&& !skb_queue_empty(&sk->write_queue)) {
|
| 551 |
|
|
/* THIS IS UGLY. DON'T SHOW THIS TO YOUR MOTHER. --erics
|
| 552 |
|
|
* Route shifted devices.
|
| 553 |
|
|
* If this is the last packet in the
|
| 554 |
|
|
* retransmit queue, then we should walk
|
| 555 |
|
|
* the chain of packets in the write_queue
|
| 556 |
|
|
* that have the same device and
|
| 557 |
|
|
* fix routing on these packets as well.
|
| 558 |
|
|
* If we fail to do this, then every packet
|
| 559 |
|
|
* in the transmit queue will incurr a
|
| 560 |
|
|
* retransmit with the backed off retransmit
|
| 561 |
|
|
* timeout. This is very bad.
|
| 562 |
|
|
*/
|
| 563 |
|
|
struct sk_buff *skb2 = sk->write_queue.next;
|
| 564 |
|
|
while (skb2 && skb2->dev == skb->dev) {
|
| 565 |
|
|
skb2->raddr=rt->rt_gateway;
|
| 566 |
|
|
if (sysctl_ip_dynaddr & 4 || (sk->state == TCP_SYN_SENT && sysctl_ip_dynaddr & 3))
|
| 567 |
|
|
ip_rewrite_addrs (sk, skb2, dev);
|
| 568 |
|
|
skb_pull(skb2,((unsigned char *)skb2->ip_hdr)-skb2->data);
|
| 569 |
|
|
skb2->dev = dev;
|
| 570 |
|
|
skb2->arp=1;
|
| 571 |
|
|
if (rt->rt_hh)
|
| 572 |
|
|
{
|
| 573 |
|
|
memcpy(skb_push(skb2,dev->hard_header_len),rt->rt_hh->hh_data,dev->hard_header_len);
|
| 574 |
|
|
if (!rt->rt_hh->hh_uptodate)
|
| 575 |
|
|
{
|
| 576 |
|
|
skb2->arp = 0;
|
| 577 |
|
|
#if RT_CACHE_DEBUG >= 2
|
| 578 |
|
|
printk("tcp_do_retransmit(1): hh miss %08x via %08x\n", iph->daddr, rt->rt_gateway);
|
| 579 |
|
|
#endif
|
| 580 |
|
|
}
|
| 581 |
|
|
}
|
| 582 |
|
|
else if (dev->hard_header)
|
| 583 |
|
|
{
|
| 584 |
|
|
if(dev->hard_header(skb2, dev, ETH_P_IP, NULL, NULL, skb2->len)<0)
|
| 585 |
|
|
skb2->arp=0;
|
| 586 |
|
|
}
|
| 587 |
|
|
|
| 588 |
|
|
skb2 = skb2->next;
|
| 589 |
|
|
}
|
| 590 |
|
|
}
|
| 591 |
|
|
skb->raddr=rt->rt_gateway;
|
| 592 |
|
|
if ((skb->dev !=dev || skb->dev->pa_addr != skb->ip_hdr->saddr) && (sysctl_ip_dynaddr & 4 || (sk->state == TCP_SYN_SENT && sysctl_ip_dynaddr & 3)))
|
| 593 |
|
|
ip_rewrite_addrs(sk, skb, dev);
|
| 594 |
|
|
skb->dev=dev;
|
| 595 |
|
|
skb->arp=1;
|
| 596 |
|
|
#ifdef CONFIG_FIREWALL
|
| 597 |
|
|
if (call_out_firewall(PF_INET, skb->dev, iph, NULL) < FW_ACCEPT) {
|
| 598 |
|
|
/* The firewall wants us to dump the packet.
|
| 599 |
|
|
* We have to check this here, because
|
| 600 |
|
|
* the drop in ip_queue_xmit only catches the
|
| 601 |
|
|
* first time we send it. We must drop on
|
| 602 |
|
|
* every resend as well.
|
| 603 |
|
|
*/
|
| 604 |
|
|
break;
|
| 605 |
|
|
}
|
| 606 |
|
|
#endif
|
| 607 |
|
|
if (rt->rt_hh)
|
| 608 |
|
|
{
|
| 609 |
|
|
memcpy(skb_push(skb,dev->hard_header_len),rt->rt_hh->hh_data,dev->hard_header_len);
|
| 610 |
|
|
if (!rt->rt_hh->hh_uptodate)
|
| 611 |
|
|
{
|
| 612 |
|
|
skb->arp = 0;
|
| 613 |
|
|
#if RT_CACHE_DEBUG >= 2
|
| 614 |
|
|
printk("tcp_do_retransmit(2): hh miss %08x via %08x\n", iph->daddr, rt->rt_gateway);
|
| 615 |
|
|
#endif
|
| 616 |
|
|
}
|
| 617 |
|
|
}
|
| 618 |
|
|
else if (dev->hard_header)
|
| 619 |
|
|
{
|
| 620 |
|
|
if(dev->hard_header(skb, dev, ETH_P_IP, NULL, NULL, skb->len)<0)
|
| 621 |
|
|
skb->arp=0;
|
| 622 |
|
|
}
|
| 623 |
|
|
|
| 624 |
|
|
/*
|
| 625 |
|
|
* This is not the right way to handle this. We have to
|
| 626 |
|
|
* issue an up to date window and ack report with this
|
| 627 |
|
|
* retransmit to keep the odd buggy tcp that relies on
|
| 628 |
|
|
* the fact BSD does this happy.
|
| 629 |
|
|
* We don't however need to recalculate the entire
|
| 630 |
|
|
* checksum, so someone wanting a small problem to play
|
| 631 |
|
|
* with might like to implement RFC1141/RFC1624 and speed
|
| 632 |
|
|
* this up by avoiding a full checksum.
|
| 633 |
|
|
*/
|
| 634 |
|
|
|
| 635 |
|
|
th->ack_seq = htonl(sk->acked_seq);
|
| 636 |
|
|
clear_delayed_acks(sk);
|
| 637 |
|
|
th->window = ntohs(tcp_select_window(sk));
|
| 638 |
|
|
tcp_send_check(th, sk->saddr, sk->daddr, size, skb);
|
| 639 |
|
|
|
| 640 |
|
|
/*
|
| 641 |
|
|
* If the interface is (still) up and running, kick it.
|
| 642 |
|
|
*/
|
| 643 |
|
|
|
| 644 |
|
|
if (dev->flags & IFF_UP)
|
| 645 |
|
|
{
|
| 646 |
|
|
/*
|
| 647 |
|
|
* If the packet is still being sent by the device/protocol
|
| 648 |
|
|
* below then don't retransmit. This is both needed, and good -
|
| 649 |
|
|
* especially with connected mode AX.25 where it stops resends
|
| 650 |
|
|
* occurring of an as yet unsent anyway frame!
|
| 651 |
|
|
* We still add up the counts as the round trip time wants
|
| 652 |
|
|
* adjusting.
|
| 653 |
|
|
*/
|
| 654 |
|
|
if (!skb_device_locked(skb))
|
| 655 |
|
|
{
|
| 656 |
|
|
/* Now queue it */
|
| 657 |
|
|
ip_statistics.IpOutRequests++;
|
| 658 |
|
|
dev_queue_xmit(skb, dev, sk->priority);
|
| 659 |
|
|
sk->packets_out++;
|
| 660 |
|
|
} else {
|
| 661 |
|
|
/* This shouldn't happen as we skip out above if the buffer is locked */
|
| 662 |
|
|
printk(KERN_WARNING "tcp_do_retransmit: sk_buff (%p) became locked\n", skb);
|
| 663 |
|
|
}
|
| 664 |
|
|
}
|
| 665 |
|
|
}
|
| 666 |
|
|
|
| 667 |
|
|
/*
|
| 668 |
|
|
* Count retransmissions
|
| 669 |
|
|
*/
|
| 670 |
|
|
|
| 671 |
|
|
sk->prot->retransmits++;
|
| 672 |
|
|
tcp_statistics.TcpRetransSegs++;
|
| 673 |
|
|
|
| 674 |
|
|
/*
|
| 675 |
|
|
* Record the high sequence number to help avoid doing
|
| 676 |
|
|
* to much fast retransmission.
|
| 677 |
|
|
*/
|
| 678 |
|
|
if (sk->retransmits)
|
| 679 |
|
|
sk->high_seq = sk->sent_seq;
|
| 680 |
|
|
|
| 681 |
|
|
/*
|
| 682 |
|
|
* Advance the send_next pointer so we don't keep
|
| 683 |
|
|
* retransmitting the same stuff every time we get an ACK.
|
| 684 |
|
|
*/
|
| 685 |
|
|
sk->send_next = skb->link3;
|
| 686 |
|
|
|
| 687 |
|
|
/*
|
| 688 |
|
|
* Only one retransmit requested.
|
| 689 |
|
|
*/
|
| 690 |
|
|
|
| 691 |
|
|
if (!all)
|
| 692 |
|
|
break;
|
| 693 |
|
|
|
| 694 |
|
|
/*
|
| 695 |
|
|
* This should cut it off before we send too many packets.
|
| 696 |
|
|
*/
|
| 697 |
|
|
|
| 698 |
|
|
if (sk->packets_out >= sk->cong_window)
|
| 699 |
|
|
break;
|
| 700 |
|
|
|
| 701 |
|
|
skb = skb->link3;
|
| 702 |
|
|
}
|
| 703 |
|
|
}
|
| 704 |
|
|
|
| 705 |
|
|
/*
|
| 706 |
|
|
* This routine will send an RST to the other tcp.
|
| 707 |
|
|
*/
|
| 708 |
|
|
|
| 709 |
|
|
void tcp_send_reset(unsigned long saddr, unsigned long daddr, struct tcphdr *th,
|
| 710 |
|
|
struct proto *prot, struct options *opt, struct device *dev, int tos, int ttl)
|
| 711 |
|
|
{
|
| 712 |
|
|
struct sk_buff *buff;
|
| 713 |
|
|
struct tcphdr *t1;
|
| 714 |
|
|
int tmp;
|
| 715 |
|
|
struct device *ndev=NULL;
|
| 716 |
|
|
|
| 717 |
|
|
/*
|
| 718 |
|
|
* Cannot reset a reset (Think about it).
|
| 719 |
|
|
*/
|
| 720 |
|
|
|
| 721 |
|
|
if(th->rst)
|
| 722 |
|
|
return;
|
| 723 |
|
|
|
| 724 |
|
|
/*
|
| 725 |
|
|
* We need to grab some memory, and put together an RST,
|
| 726 |
|
|
* and then put it into the queue to be sent.
|
| 727 |
|
|
*/
|
| 728 |
|
|
|
| 729 |
|
|
buff = alloc_skb(MAX_RESET_SIZE, GFP_ATOMIC);
|
| 730 |
|
|
if (buff == NULL)
|
| 731 |
|
|
return;
|
| 732 |
|
|
|
| 733 |
|
|
buff->sk = NULL;
|
| 734 |
|
|
buff->dev = dev;
|
| 735 |
|
|
buff->localroute = 0;
|
| 736 |
|
|
buff->csum = 0;
|
| 737 |
|
|
|
| 738 |
|
|
/*
|
| 739 |
|
|
* Put in the IP header and routing stuff.
|
| 740 |
|
|
*/
|
| 741 |
|
|
|
| 742 |
|
|
tmp = prot->build_header(buff, saddr, daddr, &ndev, IPPROTO_TCP, opt,
|
| 743 |
|
|
sizeof(struct tcphdr),tos,ttl,NULL);
|
| 744 |
|
|
if (tmp < 0)
|
| 745 |
|
|
{
|
| 746 |
|
|
buff->free = 1;
|
| 747 |
|
|
sock_wfree(NULL, buff);
|
| 748 |
|
|
return;
|
| 749 |
|
|
}
|
| 750 |
|
|
|
| 751 |
|
|
t1 =(struct tcphdr *)skb_put(buff,sizeof(struct tcphdr));
|
| 752 |
|
|
memset(t1, 0, sizeof(*t1));
|
| 753 |
|
|
|
| 754 |
|
|
/*
|
| 755 |
|
|
* Swap the send and the receive.
|
| 756 |
|
|
*/
|
| 757 |
|
|
|
| 758 |
|
|
t1->dest = th->source;
|
| 759 |
|
|
t1->source = th->dest;
|
| 760 |
|
|
t1->doff = sizeof(*t1)/4;
|
| 761 |
|
|
t1->rst = 1;
|
| 762 |
|
|
|
| 763 |
|
|
if(th->ack)
|
| 764 |
|
|
{
|
| 765 |
|
|
t1->seq = th->ack_seq;
|
| 766 |
|
|
}
|
| 767 |
|
|
else
|
| 768 |
|
|
{
|
| 769 |
|
|
t1->ack = 1;
|
| 770 |
|
|
if(!th->syn)
|
| 771 |
|
|
t1->ack_seq = th->seq;
|
| 772 |
|
|
else
|
| 773 |
|
|
t1->ack_seq = htonl(ntohl(th->seq)+1);
|
| 774 |
|
|
}
|
| 775 |
|
|
|
| 776 |
|
|
tcp_send_check(t1, saddr, daddr, sizeof(*t1), buff);
|
| 777 |
|
|
prot->queue_xmit(NULL, ndev, buff, 1);
|
| 778 |
|
|
tcp_statistics.TcpOutSegs++;
|
| 779 |
|
|
}
|
| 780 |
|
|
|
| 781 |
|
|
#ifdef CONFIG_RST_COOKIES
|
| 782 |
|
|
/*
|
| 783 |
|
|
* This routine will send a bad SYNACK to the remote tcp
|
| 784 |
|
|
* containing a secure sequence number.
|
| 785 |
|
|
* This should evoke a reset with a cookie, so we can verify
|
| 786 |
|
|
* the existence of the remote machine.
|
| 787 |
|
|
*/
|
| 788 |
|
|
|
| 789 |
|
|
void tcp_send_synack_probe(unsigned long saddr, unsigned long daddr, struct tcphdr *th,
|
| 790 |
|
|
struct proto *prot, struct options *opt, struct device *dev, int tos, int ttl)
|
| 791 |
|
|
{
|
| 792 |
|
|
struct sk_buff *buff;
|
| 793 |
|
|
struct tcphdr *t1;
|
| 794 |
|
|
int tmp;
|
| 795 |
|
|
struct device *ndev=NULL;
|
| 796 |
|
|
|
| 797 |
|
|
/*
|
| 798 |
|
|
* We need to grab some memory, and put together a SYNACK,
|
| 799 |
|
|
* and then put it into the queue to be sent.
|
| 800 |
|
|
*/
|
| 801 |
|
|
|
| 802 |
|
|
buff = alloc_skb(MAX_SYN_SIZE, GFP_ATOMIC);
|
| 803 |
|
|
if (buff == NULL)
|
| 804 |
|
|
return;
|
| 805 |
|
|
|
| 806 |
|
|
buff->sk = NULL;
|
| 807 |
|
|
buff->dev = dev;
|
| 808 |
|
|
buff->localroute = 0;
|
| 809 |
|
|
buff->csum = 0;
|
| 810 |
|
|
|
| 811 |
|
|
/*
|
| 812 |
|
|
* Put in the IP header and routing stuff.
|
| 813 |
|
|
*/
|
| 814 |
|
|
|
| 815 |
|
|
tmp = prot->build_header(buff, saddr, daddr, &ndev, IPPROTO_TCP, opt,
|
| 816 |
|
|
sizeof(struct tcphdr),tos,ttl,NULL);
|
| 817 |
|
|
if (tmp < 0)
|
| 818 |
|
|
{
|
| 819 |
|
|
buff->free = 1;
|
| 820 |
|
|
sock_wfree(NULL, buff);
|
| 821 |
|
|
return;
|
| 822 |
|
|
}
|
| 823 |
|
|
|
| 824 |
|
|
t1 = (struct tcphdr *)skb_put(buff,sizeof(struct tcphdr));
|
| 825 |
|
|
|
| 826 |
|
|
memcpy(t1, th, sizeof(*t1));
|
| 827 |
|
|
/*
|
| 828 |
|
|
* Swap the send and the receive.
|
| 829 |
|
|
*/
|
| 830 |
|
|
t1->dest = th->source;
|
| 831 |
|
|
t1->source = th->dest;
|
| 832 |
|
|
t1->ack_seq = t1->seq = htonl(secure_tcp_probe_number(daddr,saddr,
|
| 833 |
|
|
ntohs(th->source),ntohs(th->dest),ntohl(th->seq),0));
|
| 834 |
|
|
t1->window = htons(1024); /* make up a window here. */
|
| 835 |
|
|
t1->syn = 1;
|
| 836 |
|
|
t1->ack = 1;
|
| 837 |
|
|
t1->urg = 0;
|
| 838 |
|
|
t1->rst = 0;
|
| 839 |
|
|
t1->psh = 0;
|
| 840 |
|
|
t1->fin = 0; /* In case someone sent us a SYN|FIN frame! */
|
| 841 |
|
|
t1->doff = sizeof(*t1)/4;
|
| 842 |
|
|
t1->res1 = 0; /* RFC requires this, we upset ECN without it */
|
| 843 |
|
|
t1->res2 = 0;
|
| 844 |
|
|
|
| 845 |
|
|
tcp_send_check(t1, saddr, daddr, sizeof(*t1), buff);
|
| 846 |
|
|
prot->queue_xmit(NULL, ndev, buff, 1);
|
| 847 |
|
|
tcp_statistics.TcpOutSegs++;
|
| 848 |
|
|
}
|
| 849 |
|
|
#endif
|
| 850 |
|
|
|
| 851 |
|
|
/*
|
| 852 |
|
|
* Send a fin.
|
| 853 |
|
|
*/
|
| 854 |
|
|
|
| 855 |
|
|
void tcp_send_fin(struct sock *sk)
|
| 856 |
|
|
{
|
| 857 |
|
|
struct proto *prot =(struct proto *)sk->prot;
|
| 858 |
|
|
struct tcphdr *th =(struct tcphdr *)&sk->dummy_th;
|
| 859 |
|
|
struct tcphdr *t1;
|
| 860 |
|
|
struct sk_buff *buff;
|
| 861 |
|
|
struct device *dev=sk->bound_device;
|
| 862 |
|
|
int tmp;
|
| 863 |
|
|
|
| 864 |
|
|
buff = sock_wmalloc(sk, MAX_RESET_SIZE,1 , GFP_KERNEL);
|
| 865 |
|
|
|
| 866 |
|
|
if (buff == NULL)
|
| 867 |
|
|
{
|
| 868 |
|
|
/* This is a disaster if it occurs */
|
| 869 |
|
|
printk(KERN_CRIT "tcp_send_fin: Impossible malloc failure");
|
| 870 |
|
|
return;
|
| 871 |
|
|
}
|
| 872 |
|
|
|
| 873 |
|
|
/*
|
| 874 |
|
|
* Administrivia
|
| 875 |
|
|
*/
|
| 876 |
|
|
|
| 877 |
|
|
buff->sk = sk;
|
| 878 |
|
|
buff->localroute = sk->localroute;
|
| 879 |
|
|
buff->csum = 0;
|
| 880 |
|
|
|
| 881 |
|
|
/*
|
| 882 |
|
|
* Put in the IP header and routing stuff.
|
| 883 |
|
|
*/
|
| 884 |
|
|
|
| 885 |
|
|
tmp = prot->build_header(buff,sk->saddr, sk->daddr, &dev,
|
| 886 |
|
|
IPPROTO_TCP, sk->opt,
|
| 887 |
|
|
sizeof(struct tcphdr),sk->ip_tos,sk->ip_ttl,&sk->ip_route_cache);
|
| 888 |
|
|
if (tmp < 0)
|
| 889 |
|
|
{
|
| 890 |
|
|
/* Oh oh. We couldn't route the packet, and we can't afford
|
| 891 |
|
|
* to drop it from the queue, since we will fail to retransmit
|
| 892 |
|
|
* then, and we never try to initiate a close again.
|
| 893 |
|
|
* Drop it onto the loopback device. The worst thing that
|
| 894 |
|
|
* happens is that the send gets droped when it comes out the
|
| 895 |
|
|
* the other side. If we get lucky it might even get forward
|
| 896 |
|
|
* to its real destination.
|
| 897 |
|
|
* WARNING: there are a few subtle points here.
|
| 898 |
|
|
* 1) We assume that if we build the header using the
|
| 899 |
|
|
* loopback we can not fail. The only way this can happen
|
| 900 |
|
|
* right now is if someone marks the loopback as
|
| 901 |
|
|
* a gateway. This should never happen. Be careful
|
| 902 |
|
|
* not to change that without taking this case into account.
|
| 903 |
|
|
* 2) If we fail to queue up the FIN packet here we get
|
| 904 |
|
|
* bitten later when we receive a simultaneous FIN.
|
| 905 |
|
|
* See the comments in tcp_fin().
|
| 906 |
|
|
*/
|
| 907 |
|
|
dev = &loopback_dev;
|
| 908 |
|
|
tmp = prot->build_header(buff,sk->saddr, sk->daddr, &dev,
|
| 909 |
|
|
IPPROTO_TCP, sk->opt,
|
| 910 |
|
|
sizeof(struct tcphdr),sk->ip_tos,sk->ip_ttl,&sk->ip_route_cache);
|
| 911 |
|
|
if (tmp < 0) {
|
| 912 |
|
|
printk(KERN_CRIT "tcp_send_fin: Impossible loopback failure");
|
| 913 |
|
|
return;
|
| 914 |
|
|
}
|
| 915 |
|
|
}
|
| 916 |
|
|
|
| 917 |
|
|
clear_delayed_acks(sk);
|
| 918 |
|
|
|
| 919 |
|
|
/*
|
| 920 |
|
|
* We ought to check if the end of the queue is a buffer and
|
| 921 |
|
|
* if so simply add the fin to that buffer, not send it ahead.
|
| 922 |
|
|
*/
|
| 923 |
|
|
|
| 924 |
|
|
t1 =(struct tcphdr *)skb_put(buff,sizeof(struct tcphdr));
|
| 925 |
|
|
buff->dev = dev;
|
| 926 |
|
|
memcpy(t1, th, sizeof(*t1));
|
| 927 |
|
|
buff->seq = sk->write_seq;
|
| 928 |
|
|
sk->write_seq++;
|
| 929 |
|
|
buff->end_seq = sk->write_seq;
|
| 930 |
|
|
t1->seq = htonl(buff->seq);
|
| 931 |
|
|
t1->ack_seq = htonl(sk->acked_seq);
|
| 932 |
|
|
t1->window = htons(tcp_select_window(sk));
|
| 933 |
|
|
t1->fin = 1;
|
| 934 |
|
|
tcp_send_check(t1, sk->saddr, sk->daddr, sizeof(*t1), buff);
|
| 935 |
|
|
|
| 936 |
|
|
/*
|
| 937 |
|
|
* If there is data in the write queue, the fin must be appended to
|
| 938 |
|
|
* the write queue.
|
| 939 |
|
|
*/
|
| 940 |
|
|
|
| 941 |
|
|
if (skb_peek(&sk->write_queue) != NULL)
|
| 942 |
|
|
{
|
| 943 |
|
|
buff->free = 0;
|
| 944 |
|
|
if (buff->next != NULL)
|
| 945 |
|
|
{
|
| 946 |
|
|
printk(KERN_ERR "tcp_send_fin: next != NULL\n");
|
| 947 |
|
|
skb_unlink(buff);
|
| 948 |
|
|
}
|
| 949 |
|
|
skb_queue_tail(&sk->write_queue, buff);
|
| 950 |
|
|
}
|
| 951 |
|
|
else
|
| 952 |
|
|
{
|
| 953 |
|
|
sk->sent_seq = sk->write_seq;
|
| 954 |
|
|
sk->prot->queue_xmit(sk, dev, buff, 0);
|
| 955 |
|
|
tcp_reset_xmit_timer(sk, TIME_WRITE, sk->rto);
|
| 956 |
|
|
}
|
| 957 |
|
|
}
|
| 958 |
|
|
|
| 959 |
|
|
|
| 960 |
|
|
void tcp_send_synack(struct sock * newsk, struct sock * sk, struct sk_buff * skb, int destroy)
|
| 961 |
|
|
{
|
| 962 |
|
|
struct tcphdr *t1;
|
| 963 |
|
|
unsigned char *ptr;
|
| 964 |
|
|
struct sk_buff * buff;
|
| 965 |
|
|
struct device *ndev=newsk->bound_device;
|
| 966 |
|
|
int tmp;
|
| 967 |
|
|
|
| 968 |
|
|
buff = sock_wmalloc(newsk, MAX_SYN_SIZE, 1, GFP_ATOMIC);
|
| 969 |
|
|
if (buff == NULL)
|
| 970 |
|
|
{
|
| 971 |
|
|
sk->err = ENOMEM;
|
| 972 |
|
|
destroy_sock(newsk);
|
| 973 |
|
|
kfree_skb(skb, FREE_READ);
|
| 974 |
|
|
tcp_statistics.TcpAttemptFails++;
|
| 975 |
|
|
return;
|
| 976 |
|
|
}
|
| 977 |
|
|
|
| 978 |
|
|
buff->sk = newsk;
|
| 979 |
|
|
buff->localroute = newsk->localroute;
|
| 980 |
|
|
|
| 981 |
|
|
/*
|
| 982 |
|
|
* Put in the IP header and routing stuff.
|
| 983 |
|
|
*/
|
| 984 |
|
|
|
| 985 |
|
|
tmp = sk->prot->build_header(buff, newsk->saddr, newsk->daddr, &ndev,
|
| 986 |
|
|
IPPROTO_TCP, newsk->opt, MAX_SYN_SIZE,sk->ip_tos,sk->ip_ttl,&newsk->ip_route_cache);
|
| 987 |
|
|
|
| 988 |
|
|
/*
|
| 989 |
|
|
* Something went wrong.
|
| 990 |
|
|
*/
|
| 991 |
|
|
|
| 992 |
|
|
if (tmp < 0)
|
| 993 |
|
|
{
|
| 994 |
|
|
sk->err = tmp;
|
| 995 |
|
|
buff->free = 1;
|
| 996 |
|
|
kfree_skb(buff,FREE_WRITE);
|
| 997 |
|
|
destroy_sock(newsk);
|
| 998 |
|
|
skb->sk = sk;
|
| 999 |
|
|
kfree_skb(skb, FREE_READ);
|
| 1000 |
|
|
tcp_statistics.TcpAttemptFails++;
|
| 1001 |
|
|
return;
|
| 1002 |
|
|
}
|
| 1003 |
|
|
|
| 1004 |
|
|
t1 =(struct tcphdr *)skb_put(buff,sizeof(struct tcphdr));
|
| 1005 |
|
|
|
| 1006 |
|
|
memcpy(t1, skb->h.th, sizeof(*t1));
|
| 1007 |
|
|
buff->seq = newsk->write_seq++;
|
| 1008 |
|
|
buff->end_seq = newsk->write_seq;
|
| 1009 |
|
|
/*
|
| 1010 |
|
|
* Swap the send and the receive.
|
| 1011 |
|
|
*/
|
| 1012 |
|
|
t1->dest = skb->h.th->source;
|
| 1013 |
|
|
t1->source = newsk->dummy_th.source;
|
| 1014 |
|
|
t1->seq = ntohl(buff->seq);
|
| 1015 |
|
|
newsk->sent_seq = newsk->write_seq;
|
| 1016 |
|
|
t1->window = ntohs(tcp_select_window(newsk));
|
| 1017 |
|
|
t1->syn = 1;
|
| 1018 |
|
|
t1->ack = 1;
|
| 1019 |
|
|
t1->urg = 0;
|
| 1020 |
|
|
t1->rst = 0;
|
| 1021 |
|
|
t1->psh = 0;
|
| 1022 |
|
|
t1->ack_seq = htonl(newsk->acked_seq);
|
| 1023 |
|
|
t1->doff = sizeof(*t1)/4+1;
|
| 1024 |
|
|
t1->res1 = 0;
|
| 1025 |
|
|
t1->res2 = 0;
|
| 1026 |
|
|
ptr = skb_put(buff,4);
|
| 1027 |
|
|
ptr[0] = 2;
|
| 1028 |
|
|
ptr[1] = 4;
|
| 1029 |
|
|
ptr[2] = ((newsk->mtu) >> 8) & 0xff;
|
| 1030 |
|
|
ptr[3] =(newsk->mtu) & 0xff;
|
| 1031 |
|
|
buff->csum = csum_partial(ptr, 4, 0);
|
| 1032 |
|
|
#ifdef CONFIG_SYN_COOKIES
|
| 1033 |
|
|
/* Don't save buff on the newsk chain if we are going to destroy
|
| 1034 |
|
|
* newsk anyway in a second, it just delays getting rid of newsk.
|
| 1035 |
|
|
*/
|
| 1036 |
|
|
if (destroy) {
|
| 1037 |
|
|
/* BUFF was charged to NEWSK, _this_ is what we want
|
| 1038 |
|
|
* to undo so the SYN cookie can be killed now. SKB
|
| 1039 |
|
|
* is charged to SK, below we will undo that when
|
| 1040 |
|
|
* we kfree SKB.
|
| 1041 |
|
|
*/
|
| 1042 |
|
|
buff->sk = NULL;
|
| 1043 |
|
|
atomic_sub(buff->truesize, &newsk->wmem_alloc);
|
| 1044 |
|
|
}
|
| 1045 |
|
|
#endif
|
| 1046 |
|
|
tcp_send_check(t1, newsk->saddr, newsk->daddr, sizeof(*t1)+4, buff);
|
| 1047 |
|
|
if (destroy)
|
| 1048 |
|
|
newsk->prot->queue_xmit(NULL, ndev, buff, 1);
|
| 1049 |
|
|
else
|
| 1050 |
|
|
newsk->prot->queue_xmit(newsk, ndev, buff, 0);
|
| 1051 |
|
|
|
| 1052 |
|
|
|
| 1053 |
|
|
#ifdef CONFIG_SYN_COOKIES
|
| 1054 |
|
|
if (destroy) {
|
| 1055 |
|
|
/*
|
| 1056 |
|
|
* Get rid of the newsk structure if this was a cookie.
|
| 1057 |
|
|
*/
|
| 1058 |
|
|
destroy_sock(newsk);
|
| 1059 |
|
|
skb->sk = sk;
|
| 1060 |
|
|
kfree_skb(skb, FREE_READ);
|
| 1061 |
|
|
} else {
|
| 1062 |
|
|
#endif
|
| 1063 |
|
|
tcp_reset_xmit_timer(newsk, TIME_WRITE , TCP_TIMEOUT_INIT);
|
| 1064 |
|
|
skb->sk = newsk;
|
| 1065 |
|
|
|
| 1066 |
|
|
/*
|
| 1067 |
|
|
* Charge the sock_buff to newsk.
|
| 1068 |
|
|
*/
|
| 1069 |
|
|
atomic_sub(skb->truesize, &sk->rmem_alloc);
|
| 1070 |
|
|
atomic_add(skb->truesize, &newsk->rmem_alloc);
|
| 1071 |
|
|
|
| 1072 |
|
|
skb_queue_tail(&sk->receive_queue,skb);
|
| 1073 |
|
|
sk->ack_backlog++;
|
| 1074 |
|
|
#ifdef CONFIG_SYN_COOKIES
|
| 1075 |
|
|
}
|
| 1076 |
|
|
#endif
|
| 1077 |
|
|
tcp_statistics.TcpOutSegs++;
|
| 1078 |
|
|
}
|
| 1079 |
|
|
|
| 1080 |
|
|
/*
|
| 1081 |
|
|
* Set up the timers for sending a delayed ack..
|
| 1082 |
|
|
*
|
| 1083 |
|
|
* rules for delaying an ack:
|
| 1084 |
|
|
* - delay time <= 0.5 HZ
|
| 1085 |
|
|
* - must send at least every 2 full sized packets
|
| 1086 |
|
|
* - we don't have a window update to send
|
| 1087 |
|
|
*
|
| 1088 |
|
|
* additional thoughts:
|
| 1089 |
|
|
* - we should not delay sending an ACK if we have ato > 0.5 HZ.
|
| 1090 |
|
|
* My thinking about this is that in this case we will just be
|
| 1091 |
|
|
* systematically skewing the RTT calculation. (The rule about
|
| 1092 |
|
|
* sending every two full sized packets will never need to be
|
| 1093 |
|
|
* invoked, the delayed ack will be sent before the ATO timeout
|
| 1094 |
|
|
* every time. Of course, the relies on our having a good estimate
|
| 1095 |
|
|
* for packet interarrival times.)
|
| 1096 |
|
|
*/
|
| 1097 |
|
|
void tcp_send_delayed_ack(struct sock * sk, int max_timeout, unsigned long timeout)
|
| 1098 |
|
|
{
|
| 1099 |
|
|
/* Calculate new timeout */
|
| 1100 |
|
|
if (timeout > max_timeout)
|
| 1101 |
|
|
timeout = max_timeout;
|
| 1102 |
|
|
if (sk->bytes_rcv >= sk->max_unacked)
|
| 1103 |
|
|
timeout = 0;
|
| 1104 |
|
|
timeout += jiffies;
|
| 1105 |
|
|
|
| 1106 |
|
|
/* Use new timeout only if there wasn't a older one earlier */
|
| 1107 |
|
|
if (!del_timer(&sk->delack_timer) || timeout < sk->delack_timer.expires)
|
| 1108 |
|
|
sk->delack_timer.expires = timeout;
|
| 1109 |
|
|
|
| 1110 |
|
|
sk->ack_backlog++;
|
| 1111 |
|
|
add_timer(&sk->delack_timer);
|
| 1112 |
|
|
}
|
| 1113 |
|
|
|
| 1114 |
|
|
|
| 1115 |
|
|
|
| 1116 |
|
|
/*
|
| 1117 |
|
|
* This routine sends an ack and also updates the window.
|
| 1118 |
|
|
*/
|
| 1119 |
|
|
|
| 1120 |
|
|
void tcp_send_ack(struct sock *sk)
|
| 1121 |
|
|
{
|
| 1122 |
|
|
struct sk_buff *buff;
|
| 1123 |
|
|
struct tcphdr *t1;
|
| 1124 |
|
|
struct device *dev = sk->bound_device;
|
| 1125 |
|
|
int tmp;
|
| 1126 |
|
|
|
| 1127 |
|
|
if(sk->zapped)
|
| 1128 |
|
|
return; /* We have been reset, we may not send again */
|
| 1129 |
|
|
|
| 1130 |
|
|
/*
|
| 1131 |
|
|
* If we have nothing queued for transmit and the transmit timer
|
| 1132 |
|
|
* is on we are just doing an ACK timeout and need to switch
|
| 1133 |
|
|
* to a keepalive.
|
| 1134 |
|
|
*/
|
| 1135 |
|
|
|
| 1136 |
|
|
clear_delayed_acks(sk);
|
| 1137 |
|
|
|
| 1138 |
|
|
if (sk->send_head == NULL
|
| 1139 |
|
|
&& skb_queue_empty(&sk->write_queue)
|
| 1140 |
|
|
&& sk->ip_xmit_timeout == TIME_WRITE)
|
| 1141 |
|
|
{
|
| 1142 |
|
|
if (sk->keepopen)
|
| 1143 |
|
|
tcp_reset_xmit_timer(sk,TIME_KEEPOPEN,TCP_TIMEOUT_LEN);
|
| 1144 |
|
|
else
|
| 1145 |
|
|
del_timer(&sk->retransmit_timer);
|
| 1146 |
|
|
}
|
| 1147 |
|
|
|
| 1148 |
|
|
/*
|
| 1149 |
|
|
* We need to grab some memory, and put together an ack,
|
| 1150 |
|
|
* and then put it into the queue to be sent.
|
| 1151 |
|
|
*/
|
| 1152 |
|
|
|
| 1153 |
|
|
buff = sock_wmalloc(sk, MAX_ACK_SIZE, 1, GFP_ATOMIC);
|
| 1154 |
|
|
if (buff == NULL)
|
| 1155 |
|
|
{
|
| 1156 |
|
|
/*
|
| 1157 |
|
|
* Force it to send an ack. We don't have to do this
|
| 1158 |
|
|
* (ACK is unreliable) but it's much better use of
|
| 1159 |
|
|
* bandwidth on slow links to send a spare ack than
|
| 1160 |
|
|
* resend packets.
|
| 1161 |
|
|
*/
|
| 1162 |
|
|
|
| 1163 |
|
|
tcp_send_delayed_ack(sk, HZ/2, HZ/2);
|
| 1164 |
|
|
return;
|
| 1165 |
|
|
}
|
| 1166 |
|
|
|
| 1167 |
|
|
/*
|
| 1168 |
|
|
* Assemble a suitable TCP frame
|
| 1169 |
|
|
*/
|
| 1170 |
|
|
|
| 1171 |
|
|
buff->sk = sk;
|
| 1172 |
|
|
buff->localroute = sk->localroute;
|
| 1173 |
|
|
buff->csum = 0;
|
| 1174 |
|
|
|
| 1175 |
|
|
/*
|
| 1176 |
|
|
* Put in the IP header and routing stuff.
|
| 1177 |
|
|
*/
|
| 1178 |
|
|
|
| 1179 |
|
|
tmp = sk->prot->build_header(buff, sk->saddr, sk->daddr, &dev,
|
| 1180 |
|
|
IPPROTO_TCP, sk->opt, MAX_ACK_SIZE,sk->ip_tos,sk->ip_ttl,&sk->ip_route_cache);
|
| 1181 |
|
|
if (tmp < 0)
|
| 1182 |
|
|
{
|
| 1183 |
|
|
buff->free = 1;
|
| 1184 |
|
|
sock_wfree(sk, buff);
|
| 1185 |
|
|
return;
|
| 1186 |
|
|
}
|
| 1187 |
|
|
|
| 1188 |
|
|
#ifndef CONFIG_NO_PATH_MTU_DISCOVERY
|
| 1189 |
|
|
buff->ip_hdr->frag_off |= htons(IP_DF);
|
| 1190 |
|
|
#endif
|
| 1191 |
|
|
|
| 1192 |
|
|
t1 =(struct tcphdr *)skb_put(buff,sizeof(struct tcphdr));
|
| 1193 |
|
|
|
| 1194 |
|
|
/*
|
| 1195 |
|
|
* Fill in the packet and send it
|
| 1196 |
|
|
*/
|
| 1197 |
|
|
|
| 1198 |
|
|
memcpy(t1, &sk->dummy_th, sizeof(*t1));
|
| 1199 |
|
|
t1->seq = htonl(sk->sent_seq);
|
| 1200 |
|
|
t1->ack_seq = htonl(sk->acked_seq);
|
| 1201 |
|
|
t1->window = htons(tcp_select_window(sk));
|
| 1202 |
|
|
|
| 1203 |
|
|
tcp_send_check(t1, sk->saddr, sk->daddr, sizeof(*t1), buff);
|
| 1204 |
|
|
if (sk->debug)
|
| 1205 |
|
|
printk(KERN_ERR "\rtcp_ack: seq %x ack %x\n", sk->sent_seq, sk->acked_seq);
|
| 1206 |
|
|
sk->prot->queue_xmit(sk, dev, buff, 1);
|
| 1207 |
|
|
tcp_statistics.TcpOutSegs++;
|
| 1208 |
|
|
}
|
| 1209 |
|
|
|
| 1210 |
|
|
/*
|
| 1211 |
|
|
* This routine sends a packet with an out of date sequence
|
| 1212 |
|
|
* number. It assumes the other end will try to ack it.
|
| 1213 |
|
|
*/
|
| 1214 |
|
|
|
| 1215 |
|
|
void tcp_write_wakeup(struct sock *sk)
|
| 1216 |
|
|
{
|
| 1217 |
|
|
struct sk_buff *buff,*skb;
|
| 1218 |
|
|
struct tcphdr *t1;
|
| 1219 |
|
|
struct device *dev=sk->bound_device;
|
| 1220 |
|
|
int tmp;
|
| 1221 |
|
|
|
| 1222 |
|
|
if (sk->zapped)
|
| 1223 |
|
|
return; /* After a valid reset we can send no more */
|
| 1224 |
|
|
|
| 1225 |
|
|
/*
|
| 1226 |
|
|
* Write data can still be transmitted/retransmitted in the
|
| 1227 |
|
|
* following states. If any other state is encountered, return.
|
| 1228 |
|
|
* [listen/close will never occur here anyway]
|
| 1229 |
|
|
*/
|
| 1230 |
|
|
|
| 1231 |
|
|
if (sk->state != TCP_ESTABLISHED &&
|
| 1232 |
|
|
sk->state != TCP_CLOSE_WAIT &&
|
| 1233 |
|
|
sk->state != TCP_FIN_WAIT1 &&
|
| 1234 |
|
|
sk->state != TCP_LAST_ACK &&
|
| 1235 |
|
|
sk->state != TCP_CLOSING
|
| 1236 |
|
|
)
|
| 1237 |
|
|
{
|
| 1238 |
|
|
return;
|
| 1239 |
|
|
}
|
| 1240 |
|
|
if ( before(sk->sent_seq, sk->window_seq) &&
|
| 1241 |
|
|
(skb=skb_peek(&sk->write_queue)))
|
| 1242 |
|
|
{
|
| 1243 |
|
|
/*
|
| 1244 |
|
|
* We are probing the opening of a window
|
| 1245 |
|
|
* but the window size is != 0
|
| 1246 |
|
|
* must have been a result SWS avoidance ( sender )
|
| 1247 |
|
|
*/
|
| 1248 |
|
|
|
| 1249 |
|
|
struct iphdr *iph;
|
| 1250 |
|
|
struct tcphdr *th;
|
| 1251 |
|
|
struct tcphdr *nth;
|
| 1252 |
|
|
unsigned long win_size;
|
| 1253 |
|
|
#if 0
|
| 1254 |
|
|
unsigned long ow_size;
|
| 1255 |
|
|
#endif
|
| 1256 |
|
|
|
| 1257 |
|
|
/*
|
| 1258 |
|
|
* Recover the buffer pointers
|
| 1259 |
|
|
*/
|
| 1260 |
|
|
|
| 1261 |
|
|
iph = (struct iphdr *)skb->ip_hdr;
|
| 1262 |
|
|
th = (struct tcphdr *)(((char *)iph) +(iph->ihl << 2));
|
| 1263 |
|
|
|
| 1264 |
|
|
/*
|
| 1265 |
|
|
* How many bytes can we send ?
|
| 1266 |
|
|
*/
|
| 1267 |
|
|
|
| 1268 |
|
|
/* During window probes, don't try to send more than is
|
| 1269 |
|
|
* actually in the skb we've taken off the send queue here.
|
| 1270 |
|
|
*/
|
| 1271 |
|
|
win_size = skb->len - (((unsigned char *) th) - skb->data);
|
| 1272 |
|
|
win_size -= th->doff * 4;
|
| 1273 |
|
|
|
| 1274 |
|
|
/* Don't send more than the offered window! */
|
| 1275 |
|
|
win_size = min(win_size, sk->window_seq - sk->sent_seq);
|
| 1276 |
|
|
|
| 1277 |
|
|
/*
|
| 1278 |
|
|
* Grab the data for a temporary frame
|
| 1279 |
|
|
*/
|
| 1280 |
|
|
|
| 1281 |
|
|
buff = sock_wmalloc(sk, win_size + th->doff * 4 +
|
| 1282 |
|
|
(iph->ihl << 2) +
|
| 1283 |
|
|
sk->prot->max_header + 15,
|
| 1284 |
|
|
1, GFP_ATOMIC);
|
| 1285 |
|
|
if ( buff == NULL )
|
| 1286 |
|
|
return;
|
| 1287 |
|
|
|
| 1288 |
|
|
/*
|
| 1289 |
|
|
* If we strip the packet on the write queue we must
|
| 1290 |
|
|
* be ready to retransmit this one
|
| 1291 |
|
|
*/
|
| 1292 |
|
|
|
| 1293 |
|
|
buff->free = /*0*/1;
|
| 1294 |
|
|
|
| 1295 |
|
|
buff->sk = sk;
|
| 1296 |
|
|
buff->localroute = sk->localroute;
|
| 1297 |
|
|
|
| 1298 |
|
|
/*
|
| 1299 |
|
|
* Put headers on the new packet
|
| 1300 |
|
|
*/
|
| 1301 |
|
|
|
| 1302 |
|
|
tmp = sk->prot->build_header(buff, sk->saddr, sk->daddr, &dev,
|
| 1303 |
|
|
IPPROTO_TCP, sk->opt, buff->truesize,
|
| 1304 |
|
|
sk->ip_tos,sk->ip_ttl,&sk->ip_route_cache);
|
| 1305 |
|
|
if (tmp < 0)
|
| 1306 |
|
|
{
|
| 1307 |
|
|
sock_wfree(sk, buff);
|
| 1308 |
|
|
return;
|
| 1309 |
|
|
}
|
| 1310 |
|
|
|
| 1311 |
|
|
/*
|
| 1312 |
|
|
* Move the TCP header over
|
| 1313 |
|
|
*/
|
| 1314 |
|
|
|
| 1315 |
|
|
buff->dev = dev;
|
| 1316 |
|
|
|
| 1317 |
|
|
nth = (struct tcphdr *) skb_put(buff,sizeof(*th));
|
| 1318 |
|
|
|
| 1319 |
|
|
memcpy(nth, th, sizeof(*th));
|
| 1320 |
|
|
|
| 1321 |
|
|
/*
|
| 1322 |
|
|
* Correct the new header
|
| 1323 |
|
|
*/
|
| 1324 |
|
|
|
| 1325 |
|
|
nth->ack = 1;
|
| 1326 |
|
|
nth->ack_seq = htonl(sk->acked_seq);
|
| 1327 |
|
|
nth->window = htons(tcp_select_window(sk));
|
| 1328 |
|
|
nth->check = 0;
|
| 1329 |
|
|
|
| 1330 |
|
|
/*
|
| 1331 |
|
|
* Copy TCP options and data start to our new buffer
|
| 1332 |
|
|
*/
|
| 1333 |
|
|
|
| 1334 |
|
|
buff->csum = csum_partial_copy((void *)(th + 1), skb_put(buff,win_size),
|
| 1335 |
|
|
win_size + th->doff*4 - sizeof(*th), 0);
|
| 1336 |
|
|
|
| 1337 |
|
|
/*
|
| 1338 |
|
|
* Remember our right edge sequence number.
|
| 1339 |
|
|
*/
|
| 1340 |
|
|
|
| 1341 |
|
|
buff->end_seq = sk->sent_seq + win_size;
|
| 1342 |
|
|
sk->sent_seq = buff->end_seq; /* Hack */
|
| 1343 |
|
|
if(th->urg && ntohs(th->urg_ptr) < win_size)
|
| 1344 |
|
|
nth->urg = 0;
|
| 1345 |
|
|
|
| 1346 |
|
|
/*
|
| 1347 |
|
|
* Checksum the split buffer
|
| 1348 |
|
|
*/
|
| 1349 |
|
|
|
| 1350 |
|
|
tcp_send_check(nth, sk->saddr, sk->daddr,
|
| 1351 |
|
|
nth->doff * 4 + win_size , buff);
|
| 1352 |
|
|
}
|
| 1353 |
|
|
else
|
| 1354 |
|
|
{
|
| 1355 |
|
|
buff = sock_wmalloc(sk,MAX_ACK_SIZE,1, GFP_ATOMIC);
|
| 1356 |
|
|
if (buff == NULL)
|
| 1357 |
|
|
return;
|
| 1358 |
|
|
|
| 1359 |
|
|
buff->free = 1;
|
| 1360 |
|
|
buff->sk = sk;
|
| 1361 |
|
|
buff->localroute = sk->localroute;
|
| 1362 |
|
|
buff->csum = 0;
|
| 1363 |
|
|
|
| 1364 |
|
|
/*
|
| 1365 |
|
|
* Put in the IP header and routing stuff.
|
| 1366 |
|
|
*/
|
| 1367 |
|
|
|
| 1368 |
|
|
tmp = sk->prot->build_header(buff, sk->saddr, sk->daddr, &dev,
|
| 1369 |
|
|
IPPROTO_TCP, sk->opt, MAX_ACK_SIZE,sk->ip_tos,sk->ip_ttl,&sk->ip_route_cache);
|
| 1370 |
|
|
if (tmp < 0)
|
| 1371 |
|
|
{
|
| 1372 |
|
|
sock_wfree(sk, buff);
|
| 1373 |
|
|
return;
|
| 1374 |
|
|
}
|
| 1375 |
|
|
|
| 1376 |
|
|
t1 = (struct tcphdr *)skb_put(buff,sizeof(struct tcphdr));
|
| 1377 |
|
|
memcpy(t1,(void *) &sk->dummy_th, sizeof(*t1));
|
| 1378 |
|
|
|
| 1379 |
|
|
/*
|
| 1380 |
|
|
* Use a previous sequence.
|
| 1381 |
|
|
* This should cause the other end to send an ack.
|
| 1382 |
|
|
*/
|
| 1383 |
|
|
|
| 1384 |
|
|
t1->seq = htonl(sk->sent_seq-1);
|
| 1385 |
|
|
/* t1->fin = 0; -- We are sending a 'previous' sequence, and 0 bytes of data - thus no FIN bit */
|
| 1386 |
|
|
t1->ack_seq = htonl(sk->acked_seq);
|
| 1387 |
|
|
t1->window = htons(tcp_select_window(sk));
|
| 1388 |
|
|
tcp_send_check(t1, sk->saddr, sk->daddr, sizeof(*t1), buff);
|
| 1389 |
|
|
|
| 1390 |
|
|
}
|
| 1391 |
|
|
|
| 1392 |
|
|
/*
|
| 1393 |
|
|
* Send it.
|
| 1394 |
|
|
*/
|
| 1395 |
|
|
|
| 1396 |
|
|
sk->prot->queue_xmit(sk, dev, buff, 1);
|
| 1397 |
|
|
tcp_statistics.TcpOutSegs++;
|
| 1398 |
|
|
}
|
| 1399 |
|
|
|
| 1400 |
|
|
/*
|
| 1401 |
|
|
* A window probe timeout has occurred.
|
| 1402 |
|
|
*/
|
| 1403 |
|
|
|
| 1404 |
|
|
void tcp_send_probe0(struct sock *sk)
|
| 1405 |
|
|
{
|
| 1406 |
|
|
if (sk->zapped)
|
| 1407 |
|
|
return; /* After a valid reset we can send no more */
|
| 1408 |
|
|
|
| 1409 |
|
|
tcp_write_wakeup(sk);
|
| 1410 |
|
|
|
| 1411 |
|
|
sk->backoff++;
|
| 1412 |
|
|
sk->rto = min(sk->rto << 1, 120*HZ);
|
| 1413 |
|
|
sk->retransmits++;
|
| 1414 |
|
|
sk->prot->retransmits ++;
|
| 1415 |
|
|
tcp_reset_xmit_timer (sk, TIME_PROBE0, sk->rto);
|
| 1416 |
|
|
}
|
| 1417 |
|
|
|
| 1418 |
|
|
/*
|
| 1419 |
|
|
* Remove the portion of a packet that has already been sent.
|
| 1420 |
|
|
* Needed to deal with buggy TCP implementations that can't deal
|
| 1421 |
|
|
* with seeing a packet that contains some data that has already
|
| 1422 |
|
|
* been received.
|
| 1423 |
|
|
*
|
| 1424 |
|
|
* Note that the SYN sequence number is at the start of the packet
|
| 1425 |
|
|
* while the FIN is at the end. This means that we always clear out
|
| 1426 |
|
|
* the SYN bit, and never clear out the FIN bit.
|
| 1427 |
|
|
*/
|
| 1428 |
|
|
void tcp_shrink_skb(struct sock *sk, struct sk_buff *skb, u32 ack)
|
| 1429 |
|
|
{
|
| 1430 |
|
|
struct iphdr *iph;
|
| 1431 |
|
|
struct tcphdr *th;
|
| 1432 |
|
|
unsigned char *old, *new;
|
| 1433 |
|
|
unsigned long len;
|
| 1434 |
|
|
int diff;
|
| 1435 |
|
|
|
| 1436 |
|
|
/*
|
| 1437 |
|
|
* Recover the buffer pointers
|
| 1438 |
|
|
*/
|
| 1439 |
|
|
|
| 1440 |
|
|
iph = (struct iphdr *)skb->ip_hdr;
|
| 1441 |
|
|
th = (struct tcphdr *)(((char *)iph) +(iph->ihl << 2));
|
| 1442 |
|
|
|
| 1443 |
|
|
/* how much data are we droping from the tcp frame */
|
| 1444 |
|
|
diff = ack - (skb->seq + th->syn);
|
| 1445 |
|
|
/* how much data are we keeping in the tcp frame */
|
| 1446 |
|
|
len = (skb->end_seq - th->fin) - ack;
|
| 1447 |
|
|
|
| 1448 |
|
|
/* pointers to new start of remaining data, and old start */
|
| 1449 |
|
|
new = (unsigned char *)th + th->doff*4;
|
| 1450 |
|
|
old = new+diff;
|
| 1451 |
|
|
|
| 1452 |
|
|
/* Update our starting seq number */
|
| 1453 |
|
|
skb->seq = ack;
|
| 1454 |
|
|
th->seq = htonl(ack);
|
| 1455 |
|
|
th->syn = 0; /* Turn SYN off as it is logically at the start of the packet */
|
| 1456 |
|
|
|
| 1457 |
|
|
iph->tot_len = htons(ntohs(iph->tot_len)-diff);
|
| 1458 |
|
|
ip_send_check(iph);
|
| 1459 |
|
|
|
| 1460 |
|
|
/* Get the partial checksum for the IP options */
|
| 1461 |
|
|
if (th->doff*4 - sizeof(*th) > 0)
|
| 1462 |
|
|
skb->csum = csum_partial((void *)(th+1),
|
| 1463 |
|
|
th->doff*4-sizeof(*th),0);
|
| 1464 |
|
|
else
|
| 1465 |
|
|
skb->csum = 0;
|
| 1466 |
|
|
|
| 1467 |
|
|
/* Copy the good data down and get it's checksum */
|
| 1468 |
|
|
skb->csum = csum_partial_copy((void *)old,(void *)new,len,skb->csum);
|
| 1469 |
|
|
|
| 1470 |
|
|
/* shorten the skb */
|
| 1471 |
|
|
skb_trim(skb,skb->len-diff);
|
| 1472 |
|
|
|
| 1473 |
|
|
/* Checksum the shrunk buffer */
|
| 1474 |
|
|
tcp_send_check(th, sk->saddr, sk->daddr,
|
| 1475 |
|
|
th->doff * 4 + len , skb);
|
| 1476 |
|
|
}
|
