tcp_output.c 93.3 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7
/*
 * INET		An implementation of the TCP/IP protocol suite for the LINUX
 *		operating system.  INET is implemented using the  BSD Socket
 *		interface as the means of communication with the user level.
 *
 *		Implementation of the Transmission Control Protocol(TCP).
 *
8
 * Authors:	Ross Biro
L
Linus Torvalds 已提交
9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36
 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
 *		Mark Evans, <evansmp@uhura.aston.ac.uk>
 *		Corey Minyard <wf-rch!minyard@relay.EU.net>
 *		Florian La Roche, <flla@stud.uni-sb.de>
 *		Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
 *		Linus Torvalds, <torvalds@cs.helsinki.fi>
 *		Alan Cox, <gw4pts@gw4pts.ampr.org>
 *		Matthew Dillon, <dillon@apollo.west.oic.com>
 *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
 *		Jorge Cwik, <jorge@laser.satlink.net>
 */

/*
 * Changes:	Pedro Roque	:	Retransmit queue handled by TCP.
 *				:	Fragmentation on mtu decrease
 *				:	Segment collapse on retransmit
 *				:	AF independence
 *
 *		Linus Torvalds	:	send_delayed_ack
 *		David S. Miller	:	Charge memory using the right skb
 *					during syn/ack processing.
 *		David S. Miller :	Output engine completely rewritten.
 *		Andrea Arcangeli:	SYNACK carry ts_recent in tsecr.
 *		Cacophonix Gaul :	draft-minshall-nagle-01
 *		J Hadi Salim	:	ECN support
 *
 */

37 38
#define pr_fmt(fmt) "TCP: " fmt

L
Linus Torvalds 已提交
39 40 41
#include <net/tcp.h>

#include <linux/compiler.h>
42
#include <linux/gfp.h>
L
Linus Torvalds 已提交
43 44 45
#include <linux/module.h>

/* People can turn this off for buggy TCP's found in printers etc. */
46
int sysctl_tcp_retrans_collapse __read_mostly = 1;
L
Linus Torvalds 已提交
47

48
/* People can turn this on to work with those rare, broken TCPs that
49 50
 * interpret the window field as a signed quantity.
 */
51
int sysctl_tcp_workaround_signed_windows __read_mostly = 0;
52

E
Eric Dumazet 已提交
53 54 55
/* Default TSQ limit of two TSO segments */
int sysctl_tcp_limit_output_bytes __read_mostly = 131072;

L
Linus Torvalds 已提交
56 57 58 59
/* This limits the percentage of the congestion window which we
 * will allow a single TSO frame to consume.  Building TSO frames
 * which are too large can cause TCP streams to be bursty.
 */
60
int sysctl_tcp_tso_win_divisor __read_mostly = 3;
L
Linus Torvalds 已提交
61

62
int sysctl_tcp_mtu_probing __read_mostly = 0;
63
int sysctl_tcp_base_mss __read_mostly = TCP_BASE_MSS;
J
John Heffner 已提交
64

65
/* By default, RFC2861 behavior.  */
66
int sysctl_tcp_slow_start_after_idle __read_mostly = 1;
67

68
int sysctl_tcp_cookie_size __read_mostly = 0; /* TCP_COOKIE_MAX */
69
EXPORT_SYMBOL_GPL(sysctl_tcp_cookie_size);
70

E
Eric Dumazet 已提交
71 72
static bool tcp_write_xmit(struct sock *sk, unsigned int mss_now, int nonagle,
			   int push_one, gfp_t gfp);
73

74
/* Account for new data that has been sent to the network. */
75
static void tcp_event_new_data_sent(struct sock *sk, const struct sk_buff *skb)
L
Linus Torvalds 已提交
76
{
77
	struct tcp_sock *tp = tcp_sk(sk);
78
	unsigned int prior_packets = tp->packets_out;
79

80
	tcp_advance_send_head(sk, skb);
L
Linus Torvalds 已提交
81
	tp->snd_nxt = TCP_SKB_CB(skb)->end_seq;
82

L
Lucas De Marchi 已提交
83
	/* Don't override Nagle indefinitely with F-RTO */
84 85
	if (tp->frto_counter == 2)
		tp->frto_counter = 3;
86 87

	tp->packets_out += tcp_skb_pcount(skb);
88 89
	if (!prior_packets || tp->early_retrans_delayed)
		tcp_rearm_rto(sk);
L
Linus Torvalds 已提交
90 91 92 93 94 95 96 97
}

/* SND.NXT, if window was not shrunk.
 * If window has been shrunk, what should we make? It is not clear at all.
 * Using SND.UNA we will fail to open window, SND.NXT is out of window. :-(
 * Anything in between SND.UNA...SND.UNA+SND.WND also can be already
 * invalid. OK, let's make this for now:
 */
98
static inline __u32 tcp_acceptable_seq(const struct sock *sk)
L
Linus Torvalds 已提交
99
{
100
	const struct tcp_sock *tp = tcp_sk(sk);
101

102
	if (!before(tcp_wnd_end(tp), tp->snd_nxt))
L
Linus Torvalds 已提交
103 104
		return tp->snd_nxt;
	else
105
		return tcp_wnd_end(tp);
L
Linus Torvalds 已提交
106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124
}

/* Calculate mss to advertise in SYN segment.
 * RFC1122, RFC1063, draft-ietf-tcpimpl-pmtud-01 state that:
 *
 * 1. It is independent of path mtu.
 * 2. Ideally, it is maximal possible segment size i.e. 65535-40.
 * 3. For IPv4 it is reasonable to calculate it from maximal MTU of
 *    attached devices, because some buggy hosts are confused by
 *    large MSS.
 * 4. We do not make 3, we advertise MSS, calculated from first
 *    hop device mtu, but allow to raise it to ip_rt_min_advmss.
 *    This may be overridden via information stored in routing table.
 * 5. Value 65535 for MSS is valid in IPv6 and means "as large as possible,
 *    probably even Jumbo".
 */
static __u16 tcp_advertise_mss(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
125
	const struct dst_entry *dst = __sk_dst_get(sk);
L
Linus Torvalds 已提交
126 127
	int mss = tp->advmss;

128 129 130 131 132 133 134
	if (dst) {
		unsigned int metric = dst_metric_advmss(dst);

		if (metric < mss) {
			mss = metric;
			tp->advmss = mss;
		}
L
Linus Torvalds 已提交
135 136 137 138 139 140 141
	}

	return (__u16)mss;
}

/* RFC2861. Reset CWND after idle period longer RTO to "restart window".
 * This is the first part of cwnd validation mechanism. */
142
static void tcp_cwnd_restart(struct sock *sk, const struct dst_entry *dst)
L
Linus Torvalds 已提交
143
{
144
	struct tcp_sock *tp = tcp_sk(sk);
L
Linus Torvalds 已提交
145 146 147 148
	s32 delta = tcp_time_stamp - tp->lsndtime;
	u32 restart_cwnd = tcp_init_cwnd(tp, dst);
	u32 cwnd = tp->snd_cwnd;

149
	tcp_ca_event(sk, CA_EVENT_CWND_RESTART);
L
Linus Torvalds 已提交
150

151
	tp->snd_ssthresh = tcp_current_ssthresh(sk);
L
Linus Torvalds 已提交
152 153
	restart_cwnd = min(restart_cwnd, cwnd);

154
	while ((delta -= inet_csk(sk)->icsk_rto) > 0 && cwnd > restart_cwnd)
L
Linus Torvalds 已提交
155 156 157 158 159 160
		cwnd >>= 1;
	tp->snd_cwnd = max(cwnd, restart_cwnd);
	tp->snd_cwnd_stamp = tcp_time_stamp;
	tp->snd_cwnd_used = 0;
}

161
/* Congestion state accounting after a packet has been sent. */
S
Stephen Hemminger 已提交
162
static void tcp_event_data_sent(struct tcp_sock *tp,
163
				struct sock *sk)
L
Linus Torvalds 已提交
164
{
165 166
	struct inet_connection_sock *icsk = inet_csk(sk);
	const u32 now = tcp_time_stamp;
L
Linus Torvalds 已提交
167

168 169
	if (sysctl_tcp_slow_start_after_idle &&
	    (!tp->packets_out && (s32)(now - tp->lsndtime) > icsk->icsk_rto))
170
		tcp_cwnd_restart(sk, __sk_dst_get(sk));
L
Linus Torvalds 已提交
171 172 173 174 175 176

	tp->lsndtime = now;

	/* If it is a reply for ato after last received
	 * packet, enter pingpong mode.
	 */
177 178
	if ((u32)(now - icsk->icsk_ack.lrcvtime) < icsk->icsk_ack.ato)
		icsk->icsk_ack.pingpong = 1;
L
Linus Torvalds 已提交
179 180
}

181
/* Account for an ACK we sent. */
S
Stephen Hemminger 已提交
182
static inline void tcp_event_ack_sent(struct sock *sk, unsigned int pkts)
L
Linus Torvalds 已提交
183
{
184 185
	tcp_dec_quickack_mode(sk, pkts);
	inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
L
Linus Torvalds 已提交
186 187 188 189 190 191 192 193 194 195 196
}

/* Determine a window scaling and initial window to offer.
 * Based on the assumption that the given amount of space
 * will be offered. Store the results in the tp structure.
 * NOTE: for smooth operation initial space offering should
 * be a multiple of mss if possible. We assume here that mss >= 1.
 * This MUST be enforced by all callers.
 */
void tcp_select_initial_window(int __space, __u32 mss,
			       __u32 *rcv_wnd, __u32 *window_clamp,
197 198
			       int wscale_ok, __u8 *rcv_wscale,
			       __u32 init_rcv_wnd)
L
Linus Torvalds 已提交
199 200 201 202 203 204 205 206 207 208 209 210 211
{
	unsigned int space = (__space < 0 ? 0 : __space);

	/* If no clamp set the clamp to the max possible scaled window */
	if (*window_clamp == 0)
		(*window_clamp) = (65535 << 14);
	space = min(*window_clamp, space);

	/* Quantize space offering to a multiple of mss if possible. */
	if (space > mss)
		space = (space / mss) * mss;

	/* NOTE: offering an initial window larger than 32767
212 213 214 215 216 217
	 * will break some buggy TCP stacks. If the admin tells us
	 * it is likely we could be speaking with such a buggy stack
	 * we will truncate our initial window offering to 32K-1
	 * unless the remote has sent us a window scaling option,
	 * which we interpret as a sign the remote TCP is not
	 * misinterpreting the window field as a signed quantity.
L
Linus Torvalds 已提交
218
	 */
219 220 221 222 223
	if (sysctl_tcp_workaround_signed_windows)
		(*rcv_wnd) = min(space, MAX_TCP_WINDOW);
	else
		(*rcv_wnd) = space;

L
Linus Torvalds 已提交
224 225 226
	(*rcv_wscale) = 0;
	if (wscale_ok) {
		/* Set window scaling on max possible window
227
		 * See RFC1323 for an explanation of the limit to 14
L
Linus Torvalds 已提交
228 229
		 */
		space = max_t(u32, sysctl_tcp_rmem[2], sysctl_rmem_max);
230
		space = min_t(u32, space, *window_clamp);
L
Linus Torvalds 已提交
231 232 233 234 235 236
		while (space > 65535 && (*rcv_wscale) < 14) {
			space >>= 1;
			(*rcv_wscale)++;
		}
	}

237 238 239 240
	/* Set initial window to a value enough for senders starting with
	 * initial congestion window of TCP_DEFAULT_INIT_RCVWND. Place
	 * a limit on the initial window when mss is larger than 1460.
	 */
241
	if (mss > (1 << *rcv_wscale)) {
242 243 244 245
		int init_cwnd = TCP_DEFAULT_INIT_RCVWND;
		if (mss > 1460)
			init_cwnd =
			max_t(u32, (1460 * TCP_DEFAULT_INIT_RCVWND) / mss, 2);
246 247 248
		/* when initializing use the value from init_rcv_wnd
		 * rather than the default from above
		 */
249 250 251 252
		if (init_rcv_wnd)
			*rcv_wnd = min(*rcv_wnd, init_rcv_wnd * mss);
		else
			*rcv_wnd = min(*rcv_wnd, init_cwnd * mss);
L
Linus Torvalds 已提交
253 254 255 256 257
	}

	/* Set the clamp no higher than max representable value */
	(*window_clamp) = min(65535U << (*rcv_wscale), *window_clamp);
}
E
Eric Dumazet 已提交
258
EXPORT_SYMBOL(tcp_select_initial_window);
L
Linus Torvalds 已提交
259 260 261 262 263 264

/* Chose a new window to advertise, update state in tcp_sock for the
 * socket, and return result with RFC1323 scaling applied.  The return
 * value can be stuffed directly into th->window for an outgoing
 * frame.
 */
S
Stephen Hemminger 已提交
265
static u16 tcp_select_window(struct sock *sk)
L
Linus Torvalds 已提交
266 267 268 269 270 271
{
	struct tcp_sock *tp = tcp_sk(sk);
	u32 cur_win = tcp_receive_window(tp);
	u32 new_win = __tcp_select_window(sk);

	/* Never shrink the offered window */
S
Stephen Hemminger 已提交
272
	if (new_win < cur_win) {
L
Linus Torvalds 已提交
273 274 275 276 277 278 279
		/* Danger Will Robinson!
		 * Don't update rcv_wup/rcv_wnd here or else
		 * we will not be able to advertise a zero
		 * window in time.  --DaveM
		 *
		 * Relax Will Robinson.
		 */
280
		new_win = ALIGN(cur_win, 1 << tp->rx_opt.rcv_wscale);
L
Linus Torvalds 已提交
281 282 283 284 285 286 287
	}
	tp->rcv_wnd = new_win;
	tp->rcv_wup = tp->rcv_nxt;

	/* Make sure we do not exceed the maximum possible
	 * scaled window.
	 */
288
	if (!tp->rx_opt.rcv_wscale && sysctl_tcp_workaround_signed_windows)
L
Linus Torvalds 已提交
289 290 291 292 293 294 295 296 297 298 299 300 301 302
		new_win = min(new_win, MAX_TCP_WINDOW);
	else
		new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale));

	/* RFC1323 scaling applied */
	new_win >>= tp->rx_opt.rcv_wscale;

	/* If we advertise zero window, disable fast path. */
	if (new_win == 0)
		tp->pred_flags = 0;

	return new_win;
}

303
/* Packet ECN state for a SYN-ACK */
304
static inline void TCP_ECN_send_synack(const struct tcp_sock *tp, struct sk_buff *skb)
305
{
E
Eric Dumazet 已提交
306
	TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_CWR;
307
	if (!(tp->ecn_flags & TCP_ECN_OK))
E
Eric Dumazet 已提交
308
		TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_ECE;
309 310
}

311
/* Packet ECN state for a SYN.  */
312 313 314 315 316
static inline void TCP_ECN_send_syn(struct sock *sk, struct sk_buff *skb)
{
	struct tcp_sock *tp = tcp_sk(sk);

	tp->ecn_flags = 0;
317
	if (sock_net(sk)->ipv4.sysctl_tcp_ecn == 1) {
E
Eric Dumazet 已提交
318
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ECE | TCPHDR_CWR;
319 320 321 322 323
		tp->ecn_flags = TCP_ECN_OK;
	}
}

static __inline__ void
324
TCP_ECN_make_synack(const struct request_sock *req, struct tcphdr *th)
325 326 327 328 329
{
	if (inet_rsk(req)->ecn_ok)
		th->ece = 1;
}

330 331 332
/* Set up ECN state for a packet on a ESTABLISHED socket that is about to
 * be sent.
 */
333 334 335 336 337 338 339 340 341 342
static inline void TCP_ECN_send(struct sock *sk, struct sk_buff *skb,
				int tcp_header_len)
{
	struct tcp_sock *tp = tcp_sk(sk);

	if (tp->ecn_flags & TCP_ECN_OK) {
		/* Not-retransmitted data segment: set ECT and inject CWR. */
		if (skb->len != tcp_header_len &&
		    !before(TCP_SKB_CB(skb)->seq, tp->snd_nxt)) {
			INET_ECN_xmit(sk);
343
			if (tp->ecn_flags & TCP_ECN_QUEUE_CWR) {
344 345 346 347 348 349 350 351 352 353 354 355 356
				tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR;
				tcp_hdr(skb)->cwr = 1;
				skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
			}
		} else {
			/* ACK or retransmitted segment: clear ECT|CE */
			INET_ECN_dontxmit(sk);
		}
		if (tp->ecn_flags & TCP_ECN_DEMAND_CWR)
			tcp_hdr(skb)->ece = 1;
	}
}

357 358 359 360 361
/* Constructs common control bits of non-data skb. If SYN/FIN is present,
 * auto increment end seqno.
 */
static void tcp_init_nondata_skb(struct sk_buff *skb, u32 seq, u8 flags)
{
362
	skb->ip_summed = CHECKSUM_PARTIAL;
363 364
	skb->csum = 0;

E
Eric Dumazet 已提交
365
	TCP_SKB_CB(skb)->tcp_flags = flags;
366 367 368 369 370 371 372
	TCP_SKB_CB(skb)->sacked = 0;

	skb_shinfo(skb)->gso_segs = 1;
	skb_shinfo(skb)->gso_size = 0;
	skb_shinfo(skb)->gso_type = 0;

	TCP_SKB_CB(skb)->seq = seq;
C
Changli Gao 已提交
373
	if (flags & (TCPHDR_SYN | TCPHDR_FIN))
374 375 376 377
		seq++;
	TCP_SKB_CB(skb)->end_seq = seq;
}

E
Eric Dumazet 已提交
378
static inline bool tcp_urg_mode(const struct tcp_sock *tp)
I
Ilpo Järvinen 已提交
379 380 381 382
{
	return tp->snd_una != tp->snd_up;
}

A
Adam Langley 已提交
383 384 385
#define OPTION_SACK_ADVERTISE	(1 << 0)
#define OPTION_TS		(1 << 1)
#define OPTION_MD5		(1 << 2)
386
#define OPTION_WSCALE		(1 << 3)
387
#define OPTION_COOKIE_EXTENSION	(1 << 4)
Y
Yuchung Cheng 已提交
388
#define OPTION_FAST_OPEN_COOKIE	(1 << 8)
A
Adam Langley 已提交
389 390

struct tcp_out_options {
Y
Yuchung Cheng 已提交
391 392
	u16 options;		/* bit field of OPTION_* */
	u16 mss;		/* 0 to disable */
A
Adam Langley 已提交
393 394
	u8 ws;			/* window scale, 0 to disable */
	u8 num_sack_blocks;	/* number of SACK blocks to include */
395 396
	u8 hash_size;		/* bytes in hash_location */
	__u8 *hash_location;	/* temporary pointer, overloaded */
Y
Yuchung Cheng 已提交
397 398
	__u32 tsval, tsecr;	/* need to include OPTION_TS */
	struct tcp_fastopen_cookie *fastopen_cookie;	/* Fast open cookie */
A
Adam Langley 已提交
399 400
};

401 402 403 404
/* The sysctl int routines are generic, so check consistency here.
 */
static u8 tcp_cookie_size_check(u8 desired)
{
405 406 407
	int cookie_size;

	if (desired > 0)
408 409
		/* previously specified */
		return desired;
410 411 412

	cookie_size = ACCESS_ONCE(sysctl_tcp_cookie_size);
	if (cookie_size <= 0)
413 414
		/* no default specified */
		return 0;
415 416

	if (cookie_size <= TCP_COOKIE_MIN)
417 418
		/* value too small, specify minimum */
		return TCP_COOKIE_MIN;
419 420

	if (cookie_size >= TCP_COOKIE_MAX)
421 422
		/* value too large, specify maximum */
		return TCP_COOKIE_MAX;
423 424

	if (cookie_size & 1)
425
		/* 8-bit multiple, illegal, fix it */
426 427 428
		cookie_size++;

	return (u8)cookie_size;
429 430
}

431 432 433
/* Write previously computed TCP options to the packet.
 *
 * Beware: Something in the Internet is very sensitive to the ordering of
434 435 436 437 438 439 440 441 442 443
 * TCP options, we learned this through the hard way, so be careful here.
 * Luckily we can at least blame others for their non-compliance but from
 * inter-operatibility perspective it seems that we're somewhat stuck with
 * the ordering which we have been using if we want to keep working with
 * those broken things (not that it currently hurts anybody as there isn't
 * particular reason why the ordering would need to be changed).
 *
 * At least SACK_PERM as the first option is known to lead to a disaster
 * (but it may well be that other scenarios fail similarly).
 */
A
Adam Langley 已提交
444
static void tcp_options_write(__be32 *ptr, struct tcp_sock *tp,
445 446
			      struct tcp_out_options *opts)
{
Y
Yuchung Cheng 已提交
447
	u16 options = opts->options;	/* mungable copy */
448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471

	/* Having both authentication and cookies for security is redundant,
	 * and there's certainly not enough room.  Instead, the cookie-less
	 * extension variant is proposed.
	 *
	 * Consider the pessimal case with authentication.  The options
	 * could look like:
	 *   COOKIE|MD5(20) + MSS(4) + SACK|TS(12) + WSCALE(4) == 40
	 */
	if (unlikely(OPTION_MD5 & options)) {
		if (unlikely(OPTION_COOKIE_EXTENSION & options)) {
			*ptr++ = htonl((TCPOPT_COOKIE << 24) |
				       (TCPOLEN_COOKIE_BASE << 16) |
				       (TCPOPT_MD5SIG << 8) |
				       TCPOLEN_MD5SIG);
		} else {
			*ptr++ = htonl((TCPOPT_NOP << 24) |
				       (TCPOPT_NOP << 16) |
				       (TCPOPT_MD5SIG << 8) |
				       TCPOLEN_MD5SIG);
		}
		options &= ~OPTION_COOKIE_EXTENSION;
		/* overload cookie hash location */
		opts->hash_location = (__u8 *)ptr;
A
Adam Langley 已提交
472
		ptr += 4;
S
Stephen Hemminger 已提交
473
	}
A
Adam Langley 已提交
474

475 476 477 478 479 480
	if (unlikely(opts->mss)) {
		*ptr++ = htonl((TCPOPT_MSS << 24) |
			       (TCPOLEN_MSS << 16) |
			       opts->mss);
	}

481 482
	if (likely(OPTION_TS & options)) {
		if (unlikely(OPTION_SACK_ADVERTISE & options)) {
A
Adam Langley 已提交
483 484 485 486
			*ptr++ = htonl((TCPOPT_SACK_PERM << 24) |
				       (TCPOLEN_SACK_PERM << 16) |
				       (TCPOPT_TIMESTAMP << 8) |
				       TCPOLEN_TIMESTAMP);
487
			options &= ~OPTION_SACK_ADVERTISE;
A
Adam Langley 已提交
488 489 490 491 492 493 494 495 496 497
		} else {
			*ptr++ = htonl((TCPOPT_NOP << 24) |
				       (TCPOPT_NOP << 16) |
				       (TCPOPT_TIMESTAMP << 8) |
				       TCPOLEN_TIMESTAMP);
		}
		*ptr++ = htonl(opts->tsval);
		*ptr++ = htonl(opts->tsecr);
	}

498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536
	/* Specification requires after timestamp, so do it now.
	 *
	 * Consider the pessimal case without authentication.  The options
	 * could look like:
	 *   MSS(4) + SACK|TS(12) + COOKIE(20) + WSCALE(4) == 40
	 */
	if (unlikely(OPTION_COOKIE_EXTENSION & options)) {
		__u8 *cookie_copy = opts->hash_location;
		u8 cookie_size = opts->hash_size;

		/* 8-bit multiple handled in tcp_cookie_size_check() above,
		 * and elsewhere.
		 */
		if (0x2 & cookie_size) {
			__u8 *p = (__u8 *)ptr;

			/* 16-bit multiple */
			*p++ = TCPOPT_COOKIE;
			*p++ = TCPOLEN_COOKIE_BASE + cookie_size;
			*p++ = *cookie_copy++;
			*p++ = *cookie_copy++;
			ptr++;
			cookie_size -= 2;
		} else {
			/* 32-bit multiple */
			*ptr++ = htonl(((TCPOPT_NOP << 24) |
					(TCPOPT_NOP << 16) |
					(TCPOPT_COOKIE << 8) |
					TCPOLEN_COOKIE_BASE) +
				       cookie_size);
		}

		if (cookie_size > 0) {
			memcpy(ptr, cookie_copy, cookie_size);
			ptr += (cookie_size / 4);
		}
	}

	if (unlikely(OPTION_SACK_ADVERTISE & options)) {
A
Adam Langley 已提交
537 538 539 540 541 542
		*ptr++ = htonl((TCPOPT_NOP << 24) |
			       (TCPOPT_NOP << 16) |
			       (TCPOPT_SACK_PERM << 8) |
			       TCPOLEN_SACK_PERM);
	}

543
	if (unlikely(OPTION_WSCALE & options)) {
A
Adam Langley 已提交
544 545 546 547 548 549 550 551 552
		*ptr++ = htonl((TCPOPT_NOP << 24) |
			       (TCPOPT_WINDOW << 16) |
			       (TCPOLEN_WINDOW << 8) |
			       opts->ws);
	}

	if (unlikely(opts->num_sack_blocks)) {
		struct tcp_sack_block *sp = tp->rx_opt.dsack ?
			tp->duplicate_sack : tp->selective_acks;
S
Stephen Hemminger 已提交
553 554 555 556 557
		int this_sack;

		*ptr++ = htonl((TCPOPT_NOP  << 24) |
			       (TCPOPT_NOP  << 16) |
			       (TCPOPT_SACK <<  8) |
A
Adam Langley 已提交
558
			       (TCPOLEN_SACK_BASE + (opts->num_sack_blocks *
S
Stephen Hemminger 已提交
559
						     TCPOLEN_SACK_PERBLOCK)));
S
Stephen Hemminger 已提交
560

A
Adam Langley 已提交
561 562
		for (this_sack = 0; this_sack < opts->num_sack_blocks;
		     ++this_sack) {
S
Stephen Hemminger 已提交
563 564 565
			*ptr++ = htonl(sp[this_sack].start_seq);
			*ptr++ = htonl(sp[this_sack].end_seq);
		}
S
Stephen Hemminger 已提交
566

567
		tp->rx_opt.dsack = 0;
S
Stephen Hemminger 已提交
568
	}
Y
Yuchung Cheng 已提交
569 570 571 572 573 574 575 576 577 578 579 580 581 582 583

	if (unlikely(OPTION_FAST_OPEN_COOKIE & options)) {
		struct tcp_fastopen_cookie *foc = opts->fastopen_cookie;

		*ptr++ = htonl((TCPOPT_EXP << 24) |
			       ((TCPOLEN_EXP_FASTOPEN_BASE + foc->len) << 16) |
			       TCPOPT_FASTOPEN_MAGIC);

		memcpy(ptr, foc->val, foc->len);
		if ((foc->len & 3) == 2) {
			u8 *align = ((u8 *)ptr) + foc->len;
			align[0] = align[1] = TCPOPT_NOP;
		}
		ptr += (foc->len + 3) >> 2;
	}
A
Adam Langley 已提交
584 585
}

586 587 588
/* Compute TCP options for SYN packets. This is not the final
 * network wire format yet.
 */
589
static unsigned int tcp_syn_options(struct sock *sk, struct sk_buff *skb,
A
Adam Langley 已提交
590
				struct tcp_out_options *opts,
591 592
				struct tcp_md5sig_key **md5)
{
A
Adam Langley 已提交
593
	struct tcp_sock *tp = tcp_sk(sk);
594
	struct tcp_cookie_values *cvp = tp->cookie_values;
595
	unsigned int remaining = MAX_TCP_OPTION_SPACE;
596 597 598
	u8 cookie_size = (!tp->rx_opt.cookie_out_never && cvp != NULL) ?
			 tcp_cookie_size_check(cvp->cookie_desired) :
			 0;
599
	struct tcp_fastopen_request *fastopen = tp->fastopen_req;
A
Adam Langley 已提交
600

601
#ifdef CONFIG_TCP_MD5SIG
A
Adam Langley 已提交
602 603 604
	*md5 = tp->af_specific->md5_lookup(sk, sk);
	if (*md5) {
		opts->options |= OPTION_MD5;
605
		remaining -= TCPOLEN_MD5SIG_ALIGNED;
606
	}
A
Adam Langley 已提交
607 608
#else
	*md5 = NULL;
609
#endif
A
Adam Langley 已提交
610 611 612 613 614 615 616 617 618 619 620

	/* We always get an MSS option.  The option bytes which will be seen in
	 * normal data packets should timestamps be used, must be in the MSS
	 * advertised.  But we subtract them from tp->mss_cache so that
	 * calculations in tcp_sendmsg are simpler etc.  So account for this
	 * fact here if necessary.  If we don't do this correctly, as a
	 * receiver we won't recognize data packets as being full sized when we
	 * should, and thus we won't abide by the delayed ACK rules correctly.
	 * SACKs don't matter, we never delay an ACK when we have any of those
	 * going out.  */
	opts->mss = tcp_advertise_mss(sk);
621
	remaining -= TCPOLEN_MSS_ALIGNED;
A
Adam Langley 已提交
622

623
	if (likely(sysctl_tcp_timestamps && *md5 == NULL)) {
A
Adam Langley 已提交
624 625 626
		opts->options |= OPTION_TS;
		opts->tsval = TCP_SKB_CB(skb)->when;
		opts->tsecr = tp->rx_opt.ts_recent;
627
		remaining -= TCPOLEN_TSTAMP_ALIGNED;
A
Adam Langley 已提交
628
	}
629
	if (likely(sysctl_tcp_window_scaling)) {
A
Adam Langley 已提交
630
		opts->ws = tp->rx_opt.rcv_wscale;
631
		opts->options |= OPTION_WSCALE;
632
		remaining -= TCPOLEN_WSCALE_ALIGNED;
A
Adam Langley 已提交
633
	}
634
	if (likely(sysctl_tcp_sack)) {
A
Adam Langley 已提交
635
		opts->options |= OPTION_SACK_ADVERTISE;
636
		if (unlikely(!(OPTION_TS & opts->options)))
637
			remaining -= TCPOLEN_SACKPERM_ALIGNED;
A
Adam Langley 已提交
638 639
	}

640 641 642 643 644 645 646 647 648 649
	if (fastopen && fastopen->cookie.len >= 0) {
		u32 need = TCPOLEN_EXP_FASTOPEN_BASE + fastopen->cookie.len;
		need = (need + 3) & ~3U;  /* Align to 32 bits */
		if (remaining >= need) {
			opts->options |= OPTION_FAST_OPEN_COOKIE;
			opts->fastopen_cookie = &fastopen->cookie;
			remaining -= need;
			tp->syn_fastopen = 1;
		}
	}
650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696
	/* Note that timestamps are required by the specification.
	 *
	 * Odd numbers of bytes are prohibited by the specification, ensuring
	 * that the cookie is 16-bit aligned, and the resulting cookie pair is
	 * 32-bit aligned.
	 */
	if (*md5 == NULL &&
	    (OPTION_TS & opts->options) &&
	    cookie_size > 0) {
		int need = TCPOLEN_COOKIE_BASE + cookie_size;

		if (0x2 & need) {
			/* 32-bit multiple */
			need += 2; /* NOPs */

			if (need > remaining) {
				/* try shrinking cookie to fit */
				cookie_size -= 2;
				need -= 4;
			}
		}
		while (need > remaining && TCP_COOKIE_MIN <= cookie_size) {
			cookie_size -= 4;
			need -= 4;
		}
		if (TCP_COOKIE_MIN <= cookie_size) {
			opts->options |= OPTION_COOKIE_EXTENSION;
			opts->hash_location = (__u8 *)&cvp->cookie_pair[0];
			opts->hash_size = cookie_size;

			/* Remember for future incarnations. */
			cvp->cookie_desired = cookie_size;

			if (cvp->cookie_desired != cvp->cookie_pair_size) {
				/* Currently use random bytes as a nonce,
				 * assuming these are completely unpredictable
				 * by hostile users of the same system.
				 */
				get_random_bytes(&cvp->cookie_pair[0],
						 cookie_size);
				cvp->cookie_pair_size = cookie_size;
			}

			remaining -= need;
		}
	}
	return MAX_TCP_OPTION_SPACE - remaining;
S
Stephen Hemminger 已提交
697 698
}

699
/* Set up TCP options for SYN-ACKs. */
700
static unsigned int tcp_synack_options(struct sock *sk,
A
Adam Langley 已提交
701
				   struct request_sock *req,
702
				   unsigned int mss, struct sk_buff *skb,
A
Adam Langley 已提交
703
				   struct tcp_out_options *opts,
704
				   struct tcp_md5sig_key **md5,
705 706
				   struct tcp_extend_values *xvp,
				   struct tcp_fastopen_cookie *foc)
707
{
A
Adam Langley 已提交
708
	struct inet_request_sock *ireq = inet_rsk(req);
709
	unsigned int remaining = MAX_TCP_OPTION_SPACE;
710 711 712
	u8 cookie_plus = (xvp != NULL && !xvp->cookie_out_never) ?
			 xvp->cookie_plus :
			 0;
A
Adam Langley 已提交
713

714
#ifdef CONFIG_TCP_MD5SIG
A
Adam Langley 已提交
715 716 717
	*md5 = tcp_rsk(req)->af_specific->md5_lookup(sk, req);
	if (*md5) {
		opts->options |= OPTION_MD5;
718 719 720 721 722 723 724
		remaining -= TCPOLEN_MD5SIG_ALIGNED;

		/* We can't fit any SACK blocks in a packet with MD5 + TS
		 * options. There was discussion about disabling SACK
		 * rather than TS in order to fit in better with old,
		 * buggy kernels, but that was deemed to be unnecessary.
		 */
E
Eric Dumazet 已提交
725
		ireq->tstamp_ok &= !ireq->sack_ok;
726
	}
A
Adam Langley 已提交
727 728
#else
	*md5 = NULL;
729
#endif
A
Adam Langley 已提交
730

731
	/* We always send an MSS option. */
A
Adam Langley 已提交
732
	opts->mss = mss;
733
	remaining -= TCPOLEN_MSS_ALIGNED;
A
Adam Langley 已提交
734 735 736

	if (likely(ireq->wscale_ok)) {
		opts->ws = ireq->rcv_wscale;
737
		opts->options |= OPTION_WSCALE;
738
		remaining -= TCPOLEN_WSCALE_ALIGNED;
A
Adam Langley 已提交
739
	}
E
Eric Dumazet 已提交
740
	if (likely(ireq->tstamp_ok)) {
A
Adam Langley 已提交
741 742 743
		opts->options |= OPTION_TS;
		opts->tsval = TCP_SKB_CB(skb)->when;
		opts->tsecr = req->ts_recent;
744
		remaining -= TCPOLEN_TSTAMP_ALIGNED;
A
Adam Langley 已提交
745 746 747
	}
	if (likely(ireq->sack_ok)) {
		opts->options |= OPTION_SACK_ADVERTISE;
E
Eric Dumazet 已提交
748
		if (unlikely(!ireq->tstamp_ok))
749
			remaining -= TCPOLEN_SACKPERM_ALIGNED;
A
Adam Langley 已提交
750
	}
751 752 753 754 755 756 757 758 759
	if (foc != NULL) {
		u32 need = TCPOLEN_EXP_FASTOPEN_BASE + foc->len;
		need = (need + 3) & ~3U;  /* Align to 32 bits */
		if (remaining >= need) {
			opts->options |= OPTION_FAST_OPEN_COOKIE;
			opts->fastopen_cookie = foc;
			remaining -= need;
		}
	}
760 761 762 763
	/* Similar rationale to tcp_syn_options() applies here, too.
	 * If the <SYN> options fit, the same options should fit now!
	 */
	if (*md5 == NULL &&
E
Eric Dumazet 已提交
764
	    ireq->tstamp_ok &&
765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782
	    cookie_plus > TCPOLEN_COOKIE_BASE) {
		int need = cookie_plus; /* has TCPOLEN_COOKIE_BASE */

		if (0x2 & need) {
			/* 32-bit multiple */
			need += 2; /* NOPs */
		}
		if (need <= remaining) {
			opts->options |= OPTION_COOKIE_EXTENSION;
			opts->hash_size = cookie_plus - TCPOLEN_COOKIE_BASE;
			remaining -= need;
		} else {
			/* There's no error return, so flag it. */
			xvp->cookie_out_never = 1; /* true */
			opts->hash_size = 0;
		}
	}
	return MAX_TCP_OPTION_SPACE - remaining;
A
Adam Langley 已提交
783 784
}

785 786 787
/* Compute TCP options for ESTABLISHED sockets. This is not the
 * final wire format yet.
 */
788
static unsigned int tcp_established_options(struct sock *sk, struct sk_buff *skb,
A
Adam Langley 已提交
789
					struct tcp_out_options *opts,
790 791
					struct tcp_md5sig_key **md5)
{
A
Adam Langley 已提交
792 793
	struct tcp_skb_cb *tcb = skb ? TCP_SKB_CB(skb) : NULL;
	struct tcp_sock *tp = tcp_sk(sk);
794
	unsigned int size = 0;
795
	unsigned int eff_sacks;
A
Adam Langley 已提交
796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813

#ifdef CONFIG_TCP_MD5SIG
	*md5 = tp->af_specific->md5_lookup(sk, sk);
	if (unlikely(*md5)) {
		opts->options |= OPTION_MD5;
		size += TCPOLEN_MD5SIG_ALIGNED;
	}
#else
	*md5 = NULL;
#endif

	if (likely(tp->rx_opt.tstamp_ok)) {
		opts->options |= OPTION_TS;
		opts->tsval = tcb ? tcb->when : 0;
		opts->tsecr = tp->rx_opt.ts_recent;
		size += TCPOLEN_TSTAMP_ALIGNED;
	}

814 815
	eff_sacks = tp->rx_opt.num_sacks + tp->rx_opt.dsack;
	if (unlikely(eff_sacks)) {
816
		const unsigned int remaining = MAX_TCP_OPTION_SPACE - size;
A
Adam Langley 已提交
817
		opts->num_sack_blocks =
818
			min_t(unsigned int, eff_sacks,
A
Adam Langley 已提交
819 820 821 822 823 824 825
			      (remaining - TCPOLEN_SACK_BASE_ALIGNED) /
			      TCPOLEN_SACK_PERBLOCK);
		size += TCPOLEN_SACK_BASE_ALIGNED +
			opts->num_sack_blocks * TCPOLEN_SACK_PERBLOCK;
	}

	return size;
S
Stephen Hemminger 已提交
826
}
L
Linus Torvalds 已提交
827

E
Eric Dumazet 已提交
828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848

/* TCP SMALL QUEUES (TSQ)
 *
 * TSQ goal is to keep small amount of skbs per tcp flow in tx queues (qdisc+dev)
 * to reduce RTT and bufferbloat.
 * We do this using a special skb destructor (tcp_wfree).
 *
 * Its important tcp_wfree() can be replaced by sock_wfree() in the event skb
 * needs to be reallocated in a driver.
 * The invariant being skb->truesize substracted from sk->sk_wmem_alloc
 *
 * Since transmit from skb destructor is forbidden, we use a tasklet
 * to process all sockets that eventually need to send more skbs.
 * We use one tasklet per cpu, with its own queue of sockets.
 */
struct tsq_tasklet {
	struct tasklet_struct	tasklet;
	struct list_head	head; /* queue of tcp sockets */
};
static DEFINE_PER_CPU(struct tsq_tasklet, tsq_tasklet);

849 850 851 852 853 854 855
static void tcp_tsq_handler(struct sock *sk)
{
	if ((1 << sk->sk_state) &
	    (TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_CLOSING |
	     TCPF_CLOSE_WAIT  | TCPF_LAST_ACK))
		tcp_write_xmit(sk, tcp_current_mss(sk), 0, 0, GFP_ATOMIC);
}
E
Eric Dumazet 已提交
856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882
/*
 * One tasklest per cpu tries to send more skbs.
 * We run in tasklet context but need to disable irqs when
 * transfering tsq->head because tcp_wfree() might
 * interrupt us (non NAPI drivers)
 */
static void tcp_tasklet_func(unsigned long data)
{
	struct tsq_tasklet *tsq = (struct tsq_tasklet *)data;
	LIST_HEAD(list);
	unsigned long flags;
	struct list_head *q, *n;
	struct tcp_sock *tp;
	struct sock *sk;

	local_irq_save(flags);
	list_splice_init(&tsq->head, &list);
	local_irq_restore(flags);

	list_for_each_safe(q, n, &list) {
		tp = list_entry(q, struct tcp_sock, tsq_node);
		list_del(&tp->tsq_node);

		sk = (struct sock *)tp;
		bh_lock_sock(sk);

		if (!sock_owned_by_user(sk)) {
883
			tcp_tsq_handler(sk);
E
Eric Dumazet 已提交
884 885
		} else {
			/* defer the work to tcp_release_cb() */
886
			set_bit(TCP_TSQ_DEFERRED, &tp->tsq_flags);
E
Eric Dumazet 已提交
887 888 889 890 891 892 893 894
		}
		bh_unlock_sock(sk);

		clear_bit(TSQ_QUEUED, &tp->tsq_flags);
		sk_free(sk);
	}
}

895 896
#define TCP_DEFERRED_ALL ((1UL << TCP_TSQ_DEFERRED) |		\
			  (1UL << TCP_WRITE_TIMER_DEFERRED) |	\
897 898
			  (1UL << TCP_DELACK_TIMER_DEFERRED) |	\
			  (1UL << TCP_MTU_REDUCED_DEFERRED))
E
Eric Dumazet 已提交
899 900 901 902 903 904 905 906 907 908
/**
 * tcp_release_cb - tcp release_sock() callback
 * @sk: socket
 *
 * called from release_sock() to perform protocol dependent
 * actions before socket release.
 */
void tcp_release_cb(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
909
	unsigned long flags, nflags;
E
Eric Dumazet 已提交
910

911 912 913 914 915 916 917 918 919 920 921
	/* perform an atomic operation only if at least one flag is set */
	do {
		flags = tp->tsq_flags;
		if (!(flags & TCP_DEFERRED_ALL))
			return;
		nflags = flags & ~TCP_DEFERRED_ALL;
	} while (cmpxchg(&tp->tsq_flags, flags, nflags) != flags);

	if (flags & (1UL << TCP_TSQ_DEFERRED))
		tcp_tsq_handler(sk);

922
	if (flags & (1UL << TCP_WRITE_TIMER_DEFERRED)) {
923
		tcp_write_timer_handler(sk);
924 925 926
		__sock_put(sk);
	}
	if (flags & (1UL << TCP_DELACK_TIMER_DEFERRED)) {
927
		tcp_delack_timer_handler(sk);
928 929 930
		__sock_put(sk);
	}
	if (flags & (1UL << TCP_MTU_REDUCED_DEFERRED)) {
931
		sk->sk_prot->mtu_reduced(sk);
932 933
		__sock_put(sk);
	}
E
Eric Dumazet 已提交
934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955
}
EXPORT_SYMBOL(tcp_release_cb);

void __init tcp_tasklet_init(void)
{
	int i;

	for_each_possible_cpu(i) {
		struct tsq_tasklet *tsq = &per_cpu(tsq_tasklet, i);

		INIT_LIST_HEAD(&tsq->head);
		tasklet_init(&tsq->tasklet,
			     tcp_tasklet_func,
			     (unsigned long)tsq);
	}
}

/*
 * Write buffer destructor automatically called from kfree_skb.
 * We cant xmit new skbs from this context, as we might already
 * hold qdisc lock.
 */
956
static void tcp_wfree(struct sk_buff *skb)
E
Eric Dumazet 已提交
957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981
{
	struct sock *sk = skb->sk;
	struct tcp_sock *tp = tcp_sk(sk);

	if (test_and_clear_bit(TSQ_THROTTLED, &tp->tsq_flags) &&
	    !test_and_set_bit(TSQ_QUEUED, &tp->tsq_flags)) {
		unsigned long flags;
		struct tsq_tasklet *tsq;

		/* Keep a ref on socket.
		 * This last ref will be released in tcp_tasklet_func()
		 */
		atomic_sub(skb->truesize - 1, &sk->sk_wmem_alloc);

		/* queue this socket to tasklet queue */
		local_irq_save(flags);
		tsq = &__get_cpu_var(tsq_tasklet);
		list_add(&tp->tsq_node, &tsq->head);
		tasklet_schedule(&tsq->tasklet);
		local_irq_restore(flags);
	} else {
		sock_wfree(skb);
	}
}

L
Linus Torvalds 已提交
982 983 984 985 986 987 988 989 990 991 992
/* This routine actually transmits TCP packets queued in by
 * tcp_do_sendmsg().  This is used by both the initial
 * transmission and possible later retransmissions.
 * All SKB's seen here are completely headerless.  It is our
 * job to build the TCP header, and pass the packet down to
 * IP so it can do the same plus pass the packet off to the
 * device.
 *
 * We are working here with either a clone of the original
 * SKB, or a fresh unique copy made by the retransmit engine.
 */
993 994
static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it,
			    gfp_t gfp_mask)
L
Linus Torvalds 已提交
995
{
996 997 998 999
	const struct inet_connection_sock *icsk = inet_csk(sk);
	struct inet_sock *inet;
	struct tcp_sock *tp;
	struct tcp_skb_cb *tcb;
A
Adam Langley 已提交
1000
	struct tcp_out_options opts;
1001
	unsigned int tcp_options_size, tcp_header_size;
1002
	struct tcp_md5sig_key *md5;
1003 1004 1005 1006 1007 1008 1009 1010
	struct tcphdr *th;
	int err;

	BUG_ON(!skb || !tcp_skb_pcount(skb));

	/* If congestion control is doing timestamping, we must
	 * take such a timestamp before we potentially clone/copy.
	 */
1011
	if (icsk->icsk_ca_ops->flags & TCP_CONG_RTT_STAMP)
1012 1013 1014 1015 1016 1017 1018 1019 1020 1021
		__net_timestamp(skb);

	if (likely(clone_it)) {
		if (unlikely(skb_cloned(skb)))
			skb = pskb_copy(skb, gfp_mask);
		else
			skb = skb_clone(skb, gfp_mask);
		if (unlikely(!skb))
			return -ENOBUFS;
	}
L
Linus Torvalds 已提交
1022

1023 1024 1025
	inet = inet_sk(sk);
	tp = tcp_sk(sk);
	tcb = TCP_SKB_CB(skb);
A
Adam Langley 已提交
1026
	memset(&opts, 0, sizeof(opts));
L
Linus Torvalds 已提交
1027

E
Eric Dumazet 已提交
1028
	if (unlikely(tcb->tcp_flags & TCPHDR_SYN))
A
Adam Langley 已提交
1029 1030 1031 1032 1033
		tcp_options_size = tcp_syn_options(sk, skb, &opts, &md5);
	else
		tcp_options_size = tcp_established_options(sk, skb, &opts,
							   &md5);
	tcp_header_size = tcp_options_size + sizeof(struct tcphdr);
1034

1035
	if (tcp_packets_in_flight(tp) == 0) {
1036
		tcp_ca_event(sk, CA_EVENT_TX_START);
1037 1038 1039
		skb->ooo_okay = 1;
	} else
		skb->ooo_okay = 0;
1040

1041 1042
	skb_push(skb, tcp_header_size);
	skb_reset_transport_header(skb);
E
Eric Dumazet 已提交
1043 1044 1045 1046 1047 1048

	skb_orphan(skb);
	skb->sk = sk;
	skb->destructor = (sysctl_tcp_limit_output_bytes > 0) ?
			  tcp_wfree : sock_wfree;
	atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1049 1050

	/* Build TCP header and checksum it. */
1051
	th = tcp_hdr(skb);
E
Eric Dumazet 已提交
1052 1053
	th->source		= inet->inet_sport;
	th->dest		= inet->inet_dport;
1054 1055
	th->seq			= htonl(tcb->seq);
	th->ack_seq		= htonl(tp->rcv_nxt);
1056
	*(((__be16 *)th) + 6)	= htons(((tcp_header_size >> 2) << 12) |
E
Eric Dumazet 已提交
1057
					tcb->tcp_flags);
1058

E
Eric Dumazet 已提交
1059
	if (unlikely(tcb->tcp_flags & TCPHDR_SYN)) {
1060 1061 1062
		/* RFC1323: The window in SYN & SYN/ACK segments
		 * is never scaled.
		 */
1063
		th->window	= htons(min(tp->rcv_wnd, 65535U));
1064 1065 1066 1067 1068
	} else {
		th->window	= htons(tcp_select_window(sk));
	}
	th->check		= 0;
	th->urg_ptr		= 0;
L
Linus Torvalds 已提交
1069

I
Ilpo Järvinen 已提交
1070
	/* The urg_mode check is necessary during a below snd_una win probe */
1071 1072 1073 1074 1075
	if (unlikely(tcp_urg_mode(tp) && before(tcb->seq, tp->snd_up))) {
		if (before(tp->snd_up, tcb->seq + 0x10000)) {
			th->urg_ptr = htons(tp->snd_up - tcb->seq);
			th->urg = 1;
		} else if (after(tcb->seq + 0xFFFF, tp->snd_nxt)) {
1076
			th->urg_ptr = htons(0xFFFF);
1077 1078
			th->urg = 1;
		}
1079
	}
L
Linus Torvalds 已提交
1080

1081
	tcp_options_write((__be32 *)(th + 1), tp, &opts);
E
Eric Dumazet 已提交
1082
	if (likely((tcb->tcp_flags & TCPHDR_SYN) == 0))
1083
		TCP_ECN_send(sk, skb, tcp_header_size);
L
Linus Torvalds 已提交
1084

1085 1086 1087
#ifdef CONFIG_TCP_MD5SIG
	/* Calculate the MD5 hash, as we have all we need now */
	if (md5) {
E
Eric Dumazet 已提交
1088
		sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1089
		tp->af_specific->calc_md5_hash(opts.hash_location,
1090
					       md5, sk, NULL, skb);
1091 1092 1093
	}
#endif

1094
	icsk->icsk_af_ops->send_check(sk, skb);
L
Linus Torvalds 已提交
1095

E
Eric Dumazet 已提交
1096
	if (likely(tcb->tcp_flags & TCPHDR_ACK))
1097
		tcp_event_ack_sent(sk, tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
1098

1099
	if (skb->len != tcp_header_size)
1100
		tcp_event_data_sent(tp, sk);
L
Linus Torvalds 已提交
1101

1102
	if (after(tcb->end_seq, tp->snd_nxt) || tcb->seq == tcb->end_seq)
1103 1104
		TCP_ADD_STATS(sock_net(sk), TCP_MIB_OUTSEGS,
			      tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
1105

1106
	err = icsk->icsk_af_ops->queue_xmit(skb, &inet->cork.fl);
1107
	if (likely(err <= 0))
1108 1109
		return err;

1110
	tcp_enter_cwr(sk, 1);
1111

1112
	return net_xmit_eval(err);
L
Linus Torvalds 已提交
1113 1114
}

1115
/* This routine just queues the buffer for sending.
L
Linus Torvalds 已提交
1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126
 *
 * NOTE: probe0 timer is not checked, do not forget tcp_push_pending_frames,
 * otherwise socket can stall.
 */
static void tcp_queue_skb(struct sock *sk, struct sk_buff *skb)
{
	struct tcp_sock *tp = tcp_sk(sk);

	/* Advance write_seq and place onto the write_queue. */
	tp->write_seq = TCP_SKB_CB(skb)->end_seq;
	skb_header_release(skb);
1127
	tcp_add_write_queue_tail(sk, skb);
1128 1129
	sk->sk_wmem_queued += skb->truesize;
	sk_mem_charge(sk, skb->truesize);
L
Linus Torvalds 已提交
1130 1131
}

1132
/* Initialize TSO segments for a packet. */
1133
static void tcp_set_skb_tso_segs(const struct sock *sk, struct sk_buff *skb,
1134
				 unsigned int mss_now)
1135
{
1136 1137
	if (skb->len <= mss_now || !sk_can_gso(sk) ||
	    skb->ip_summed == CHECKSUM_NONE) {
1138 1139 1140
		/* Avoid the costly divide in the normal
		 * non-TSO case.
		 */
1141 1142 1143
		skb_shinfo(skb)->gso_segs = 1;
		skb_shinfo(skb)->gso_size = 0;
		skb_shinfo(skb)->gso_type = 0;
1144
	} else {
I
Ilpo Järvinen 已提交
1145
		skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss_now);
1146
		skb_shinfo(skb)->gso_size = mss_now;
1147
		skb_shinfo(skb)->gso_type = sk->sk_gso_type;
L
Linus Torvalds 已提交
1148 1149 1150
	}
}

I
Ilpo Järvinen 已提交
1151
/* When a modification to fackets out becomes necessary, we need to check
1152
 * skb is counted to fackets_out or not.
I
Ilpo Järvinen 已提交
1153
 */
1154
static void tcp_adjust_fackets_out(struct sock *sk, const struct sk_buff *skb,
I
Ilpo Järvinen 已提交
1155 1156
				   int decr)
{
1157 1158
	struct tcp_sock *tp = tcp_sk(sk);

1159
	if (!tp->sacked_out || tcp_is_reno(tp))
I
Ilpo Järvinen 已提交
1160 1161
		return;

1162
	if (after(tcp_highest_sack_seq(tp), TCP_SKB_CB(skb)->seq))
I
Ilpo Järvinen 已提交
1163 1164 1165
		tp->fackets_out -= decr;
}

1166 1167 1168
/* Pcount in the middle of the write queue got changed, we need to do various
 * tweaks to fix counters
 */
1169
static void tcp_adjust_pcount(struct sock *sk, const struct sk_buff *skb, int decr)
1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189
{
	struct tcp_sock *tp = tcp_sk(sk);

	tp->packets_out -= decr;

	if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)
		tp->sacked_out -= decr;
	if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS)
		tp->retrans_out -= decr;
	if (TCP_SKB_CB(skb)->sacked & TCPCB_LOST)
		tp->lost_out -= decr;

	/* Reno case is special. Sigh... */
	if (tcp_is_reno(tp) && decr > 0)
		tp->sacked_out -= min_t(u32, tp->sacked_out, decr);

	tcp_adjust_fackets_out(sk, skb, decr);

	if (tp->lost_skb_hint &&
	    before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(tp->lost_skb_hint)->seq) &&
1190
	    (tcp_is_fack(tp) || (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)))
1191 1192 1193 1194 1195
		tp->lost_cnt_hint -= decr;

	tcp_verify_left_out(tp);
}

L
Linus Torvalds 已提交
1196 1197
/* Function to create two new TCP segments.  Shrinks the given segment
 * to the specified size and appends a new segment with the rest of the
1198
 * packet to the list.  This won't be called frequently, I hope.
L
Linus Torvalds 已提交
1199 1200
 * Remember, these are still headerless SKBs at this point.
 */
1201 1202
int tcp_fragment(struct sock *sk, struct sk_buff *skb, u32 len,
		 unsigned int mss_now)
L
Linus Torvalds 已提交
1203 1204 1205
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *buff;
1206
	int nsize, old_factor;
1207
	int nlen;
1208
	u8 flags;
L
Linus Torvalds 已提交
1209

1210 1211
	if (WARN_ON(len > skb->len))
		return -EINVAL;
1212

L
Linus Torvalds 已提交
1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225
	nsize = skb_headlen(skb) - len;
	if (nsize < 0)
		nsize = 0;

	if (skb_cloned(skb) &&
	    skb_is_nonlinear(skb) &&
	    pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
		return -ENOMEM;

	/* Get a new skb... force flag on. */
	buff = sk_stream_alloc_skb(sk, nsize, GFP_ATOMIC);
	if (buff == NULL)
		return -ENOMEM; /* We'll just try again later. */
H
Herbert Xu 已提交
1226

1227 1228
	sk->sk_wmem_queued += buff->truesize;
	sk_mem_charge(sk, buff->truesize);
1229 1230 1231
	nlen = skb->len - len - nsize;
	buff->truesize += nlen;
	skb->truesize -= nlen;
L
Linus Torvalds 已提交
1232 1233 1234 1235 1236 1237 1238

	/* Correct the sequence numbers. */
	TCP_SKB_CB(buff)->seq = TCP_SKB_CB(skb)->seq + len;
	TCP_SKB_CB(buff)->end_seq = TCP_SKB_CB(skb)->end_seq;
	TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(buff)->seq;

	/* PSH and FIN should only be set in the second packet. */
E
Eric Dumazet 已提交
1239 1240 1241
	flags = TCP_SKB_CB(skb)->tcp_flags;
	TCP_SKB_CB(skb)->tcp_flags = flags & ~(TCPHDR_FIN | TCPHDR_PSH);
	TCP_SKB_CB(buff)->tcp_flags = flags;
1242
	TCP_SKB_CB(buff)->sacked = TCP_SKB_CB(skb)->sacked;
L
Linus Torvalds 已提交
1243

1244
	if (!skb_shinfo(skb)->nr_frags && skb->ip_summed != CHECKSUM_PARTIAL) {
L
Linus Torvalds 已提交
1245
		/* Copy and checksum data tail into the new buffer. */
1246 1247
		buff->csum = csum_partial_copy_nocheck(skb->data + len,
						       skb_put(buff, nsize),
L
Linus Torvalds 已提交
1248 1249 1250 1251 1252 1253
						       nsize, 0);

		skb_trim(skb, len);

		skb->csum = csum_block_sub(skb->csum, buff->csum, len);
	} else {
1254
		skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
1255 1256 1257 1258 1259 1260 1261 1262 1263
		skb_split(skb, buff, len);
	}

	buff->ip_summed = skb->ip_summed;

	/* Looks stupid, but our code really uses when of
	 * skbs, which it never sent before. --ANK
	 */
	TCP_SKB_CB(buff)->when = TCP_SKB_CB(skb)->when;
1264
	buff->tstamp = skb->tstamp;
L
Linus Torvalds 已提交
1265

1266 1267
	old_factor = tcp_skb_pcount(skb);

L
Linus Torvalds 已提交
1268
	/* Fix up tso_factor for both original and new SKB.  */
1269 1270
	tcp_set_skb_tso_segs(sk, skb, mss_now);
	tcp_set_skb_tso_segs(sk, buff, mss_now);
L
Linus Torvalds 已提交
1271

1272 1273 1274
	/* If this packet has been sent out already, we must
	 * adjust the various packet counters.
	 */
1275
	if (!before(tp->snd_nxt, TCP_SKB_CB(buff)->end_seq)) {
1276 1277
		int diff = old_factor - tcp_skb_pcount(skb) -
			tcp_skb_pcount(buff);
L
Linus Torvalds 已提交
1278

1279 1280
		if (diff)
			tcp_adjust_pcount(sk, skb, diff);
L
Linus Torvalds 已提交
1281 1282 1283
	}

	/* Link BUFF into the send queue. */
1284
	skb_header_release(buff);
1285
	tcp_insert_write_queue_after(skb, buff, sk);
L
Linus Torvalds 已提交
1286 1287 1288 1289 1290 1291 1292 1293

	return 0;
}

/* This is similar to __pskb_pull_head() (it will go to core/skbuff.c
 * eventually). The difference is that pulled data not copied, but
 * immediately discarded.
 */
1294
static void __pskb_trim_head(struct sk_buff *skb, int len)
L
Linus Torvalds 已提交
1295 1296 1297
{
	int i, k, eat;

E
Eric Dumazet 已提交
1298 1299 1300
	eat = min_t(int, len, skb_headlen(skb));
	if (eat) {
		__skb_pull(skb, eat);
1301
		skb->avail_size -= eat;
E
Eric Dumazet 已提交
1302 1303 1304 1305
		len -= eat;
		if (!len)
			return;
	}
L
Linus Torvalds 已提交
1306 1307
	eat = len;
	k = 0;
1308
	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
E
Eric Dumazet 已提交
1309 1310 1311
		int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);

		if (size <= eat) {
1312
			skb_frag_unref(skb, i);
E
Eric Dumazet 已提交
1313
			eat -= size;
L
Linus Torvalds 已提交
1314 1315 1316 1317
		} else {
			skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
			if (eat) {
				skb_shinfo(skb)->frags[k].page_offset += eat;
E
Eric Dumazet 已提交
1318
				skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
L
Linus Torvalds 已提交
1319 1320 1321 1322 1323 1324 1325
				eat = 0;
			}
			k++;
		}
	}
	skb_shinfo(skb)->nr_frags = k;

1326
	skb_reset_tail_pointer(skb);
L
Linus Torvalds 已提交
1327 1328 1329 1330
	skb->data_len -= len;
	skb->len = skb->data_len;
}

1331
/* Remove acked data from a packet in the transmit queue. */
L
Linus Torvalds 已提交
1332 1333
int tcp_trim_head(struct sock *sk, struct sk_buff *skb, u32 len)
{
1334
	if (skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
L
Linus Torvalds 已提交
1335 1336
		return -ENOMEM;

E
Eric Dumazet 已提交
1337
	__pskb_trim_head(skb, len);
L
Linus Torvalds 已提交
1338 1339

	TCP_SKB_CB(skb)->seq += len;
1340
	skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
1341 1342 1343

	skb->truesize	     -= len;
	sk->sk_wmem_queued   -= len;
1344
	sk_mem_uncharge(sk, len);
L
Linus Torvalds 已提交
1345 1346
	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);

1347
	/* Any change of skb->len requires recalculation of tso factor. */
L
Linus Torvalds 已提交
1348
	if (tcp_skb_pcount(skb) > 1)
1349
		tcp_set_skb_tso_segs(sk, skb, tcp_skb_mss(skb));
L
Linus Torvalds 已提交
1350 1351 1352 1353

	return 0;
}

1354
/* Calculate MSS. Not accounting for SACKs here.  */
1355
int tcp_mtu_to_mss(struct sock *sk, int pmtu)
J
John Heffner 已提交
1356
{
1357 1358
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1359 1360 1361 1362 1363 1364 1365
	int mss_now;

	/* Calculate base mss without TCP options:
	   It is MMS_S - sizeof(tcphdr) of rfc1122
	 */
	mss_now = pmtu - icsk->icsk_af_ops->net_header_len - sizeof(struct tcphdr);

1366 1367 1368 1369 1370 1371 1372 1373
	/* IPv6 adds a frag_hdr in case RTAX_FEATURE_ALLFRAG is set */
	if (icsk->icsk_af_ops->net_frag_header_len) {
		const struct dst_entry *dst = __sk_dst_get(sk);

		if (dst && dst_allfrag(dst))
			mss_now -= icsk->icsk_af_ops->net_frag_header_len;
	}

J
John Heffner 已提交
1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391
	/* Clamp it (mss_clamp does not include tcp options) */
	if (mss_now > tp->rx_opt.mss_clamp)
		mss_now = tp->rx_opt.mss_clamp;

	/* Now subtract optional transport overhead */
	mss_now -= icsk->icsk_ext_hdr_len;

	/* Then reserve room for full set of TCP options and 8 bytes of data */
	if (mss_now < 48)
		mss_now = 48;

	/* Now subtract TCP options size, not including SACKs */
	mss_now -= tp->tcp_header_len - sizeof(struct tcphdr);

	return mss_now;
}

/* Inverse of above */
1392
int tcp_mss_to_mtu(struct sock *sk, int mss)
J
John Heffner 已提交
1393
{
1394 1395
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1396 1397 1398 1399 1400 1401 1402
	int mtu;

	mtu = mss +
	      tp->tcp_header_len +
	      icsk->icsk_ext_hdr_len +
	      icsk->icsk_af_ops->net_header_len;

1403 1404 1405 1406 1407 1408 1409
	/* IPv6 adds a frag_hdr in case RTAX_FEATURE_ALLFRAG is set */
	if (icsk->icsk_af_ops->net_frag_header_len) {
		const struct dst_entry *dst = __sk_dst_get(sk);

		if (dst && dst_allfrag(dst))
			mtu += icsk->icsk_af_ops->net_frag_header_len;
	}
J
John Heffner 已提交
1410 1411 1412
	return mtu;
}

1413
/* MTU probing init per socket */
J
John Heffner 已提交
1414 1415 1416 1417 1418 1419 1420
void tcp_mtup_init(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct inet_connection_sock *icsk = inet_csk(sk);

	icsk->icsk_mtup.enabled = sysctl_tcp_mtu_probing > 1;
	icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp + sizeof(struct tcphdr) +
1421
			       icsk->icsk_af_ops->net_header_len;
J
John Heffner 已提交
1422 1423 1424
	icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, sysctl_tcp_base_mss);
	icsk->icsk_mtup.probe_size = 0;
}
E
Eric Dumazet 已提交
1425
EXPORT_SYMBOL(tcp_mtup_init);
J
John Heffner 已提交
1426

L
Linus Torvalds 已提交
1427 1428 1429 1430 1431 1432
/* This function synchronize snd mss to current pmtu/exthdr set.

   tp->rx_opt.user_mss is mss set by user by TCP_MAXSEG. It does NOT counts
   for TCP options, but includes only bare TCP header.

   tp->rx_opt.mss_clamp is mss negotiated at connection setup.
S
Stephen Hemminger 已提交
1433
   It is minimum of user_mss and mss received with SYN.
L
Linus Torvalds 已提交
1434 1435
   It also does not include TCP options.

1436
   inet_csk(sk)->icsk_pmtu_cookie is last pmtu, seen by this function.
L
Linus Torvalds 已提交
1437 1438 1439 1440 1441 1442 1443 1444 1445

   tp->mss_cache is current effective sending mss, including
   all tcp options except for SACKs. It is evaluated,
   taking into account current pmtu, but never exceeds
   tp->rx_opt.mss_clamp.

   NOTE1. rfc1122 clearly states that advertised MSS
   DOES NOT include either tcp or ip options.

1446 1447
   NOTE2. inet_csk(sk)->icsk_pmtu_cookie and tp->mss_cache
   are READ ONLY outside this function.		--ANK (980731)
L
Linus Torvalds 已提交
1448 1449 1450 1451
 */
unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu)
{
	struct tcp_sock *tp = tcp_sk(sk);
1452
	struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1453
	int mss_now;
L
Linus Torvalds 已提交
1454

J
John Heffner 已提交
1455 1456
	if (icsk->icsk_mtup.search_high > pmtu)
		icsk->icsk_mtup.search_high = pmtu;
L
Linus Torvalds 已提交
1457

J
John Heffner 已提交
1458
	mss_now = tcp_mtu_to_mss(sk, pmtu);
1459
	mss_now = tcp_bound_to_half_wnd(tp, mss_now);
L
Linus Torvalds 已提交
1460 1461

	/* And store cached results */
1462
	icsk->icsk_pmtu_cookie = pmtu;
J
John Heffner 已提交
1463 1464
	if (icsk->icsk_mtup.enabled)
		mss_now = min(mss_now, tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low));
1465
	tp->mss_cache = mss_now;
L
Linus Torvalds 已提交
1466 1467 1468

	return mss_now;
}
E
Eric Dumazet 已提交
1469
EXPORT_SYMBOL(tcp_sync_mss);
L
Linus Torvalds 已提交
1470 1471 1472 1473

/* Compute the current effective MSS, taking SACKs and IP options,
 * and even PMTU discovery events into account.
 */
I
Ilpo Järvinen 已提交
1474
unsigned int tcp_current_mss(struct sock *sk)
L
Linus Torvalds 已提交
1475
{
1476 1477
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct dst_entry *dst = __sk_dst_get(sk);
1478
	u32 mss_now;
1479
	unsigned int header_len;
A
Adam Langley 已提交
1480 1481
	struct tcp_out_options opts;
	struct tcp_md5sig_key *md5;
1482 1483 1484

	mss_now = tp->mss_cache;

L
Linus Torvalds 已提交
1485 1486
	if (dst) {
		u32 mtu = dst_mtu(dst);
1487
		if (mtu != inet_csk(sk)->icsk_pmtu_cookie)
L
Linus Torvalds 已提交
1488 1489 1490
			mss_now = tcp_sync_mss(sk, mtu);
	}

A
Adam Langley 已提交
1491 1492 1493 1494 1495 1496 1497 1498 1499 1500
	header_len = tcp_established_options(sk, NULL, &opts, &md5) +
		     sizeof(struct tcphdr);
	/* The mss_cache is sized based on tp->tcp_header_len, which assumes
	 * some common options. If this is an odd packet (because we have SACK
	 * blocks etc) then our calculated header_len will be different, and
	 * we have to adjust mss_now correspondingly */
	if (header_len != tp->tcp_header_len) {
		int delta = (int) header_len - tp->tcp_header_len;
		mss_now -= delta;
	}
1501

L
Linus Torvalds 已提交
1502 1503 1504
	return mss_now;
}

1505
/* Congestion window validation. (RFC2861) */
1506
static void tcp_cwnd_validate(struct sock *sk)
1507
{
1508
	struct tcp_sock *tp = tcp_sk(sk);
1509

1510
	if (tp->packets_out >= tp->snd_cwnd) {
1511 1512 1513 1514 1515 1516 1517 1518
		/* Network is feed fully. */
		tp->snd_cwnd_used = 0;
		tp->snd_cwnd_stamp = tcp_time_stamp;
	} else {
		/* Network starves. */
		if (tp->packets_out > tp->snd_cwnd_used)
			tp->snd_cwnd_used = tp->packets_out;

1519 1520
		if (sysctl_tcp_slow_start_after_idle &&
		    (s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= inet_csk(sk)->icsk_rto)
1521 1522 1523 1524
			tcp_cwnd_application_limited(sk);
	}
}

1525 1526 1527 1528
/* Returns the portion of skb which can be sent right away without
 * introducing MSS oddities to segment boundaries. In rare cases where
 * mss_now != mss_cache, we will request caller to create a small skb
 * per input skb which could be mostly avoided here (if desired).
1529 1530 1531 1532 1533 1534 1535
 *
 * We explicitly want to create a request for splitting write queue tail
 * to a small skb for Nagle purposes while avoiding unnecessary modulos,
 * thus all the complexity (cwnd_len is always MSS multiple which we
 * return whenever allowed by the other factors). Basically we need the
 * modulo only when the receiver window alone is the limiting factor or
 * when we would be allowed to send the split-due-to-Nagle skb fully.
1536
 */
1537
static unsigned int tcp_mss_split_point(const struct sock *sk, const struct sk_buff *skb,
1538
					unsigned int mss_now, unsigned int max_segs)
1539
{
1540
	const struct tcp_sock *tp = tcp_sk(sk);
1541
	u32 needed, window, max_len;
1542

1543
	window = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
1544
	max_len = mss_now * max_segs;
1545

1546 1547
	if (likely(max_len <= window && skb != tcp_write_queue_tail(sk)))
		return max_len;
1548

1549 1550
	needed = min(skb->len, window);

1551 1552
	if (max_len <= needed)
		return max_len;
1553 1554

	return needed - needed % mss_now;
1555 1556 1557 1558 1559
}

/* Can at least one segment of SKB be sent right now, according to the
 * congestion window rules?  If so, return how many segments are allowed.
 */
1560 1561
static inline unsigned int tcp_cwnd_test(const struct tcp_sock *tp,
					 const struct sk_buff *skb)
1562 1563 1564 1565
{
	u32 in_flight, cwnd;

	/* Don't be strict about the congestion window for the final FIN.  */
E
Eric Dumazet 已提交
1566 1567
	if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) &&
	    tcp_skb_pcount(skb) == 1)
1568 1569 1570 1571 1572 1573 1574 1575 1576 1577
		return 1;

	in_flight = tcp_packets_in_flight(tp);
	cwnd = tp->snd_cwnd;
	if (in_flight < cwnd)
		return (cwnd - in_flight);

	return 0;
}

1578
/* Initialize TSO state of a skb.
1579
 * This must be invoked the first time we consider transmitting
1580 1581
 * SKB onto the wire.
 */
1582
static int tcp_init_tso_segs(const struct sock *sk, struct sk_buff *skb,
1583
			     unsigned int mss_now)
1584 1585 1586
{
	int tso_segs = tcp_skb_pcount(skb);

1587
	if (!tso_segs || (tso_segs > 1 && tcp_skb_mss(skb) != mss_now)) {
1588
		tcp_set_skb_tso_segs(sk, skb, mss_now);
1589 1590 1591 1592 1593
		tso_segs = tcp_skb_pcount(skb);
	}
	return tso_segs;
}

1594
/* Minshall's variant of the Nagle send check. */
E
Eric Dumazet 已提交
1595
static inline bool tcp_minshall_check(const struct tcp_sock *tp)
1596
{
1597
	return after(tp->snd_sml, tp->snd_una) &&
1598 1599 1600
		!after(tp->snd_sml, tp->snd_nxt);
}

E
Eric Dumazet 已提交
1601
/* Return false, if packet can be sent now without violation Nagle's rules:
1602 1603
 * 1. It is full sized.
 * 2. Or it contains FIN. (already checked by caller)
F
Feng King 已提交
1604
 * 3. Or TCP_CORK is not set, and TCP_NODELAY is set.
1605 1606 1607
 * 4. Or TCP_CORK is not set, and all sent packets are ACKed.
 *    With Minshall's modification: all sent small packets are ACKed.
 */
E
Eric Dumazet 已提交
1608
static inline bool tcp_nagle_check(const struct tcp_sock *tp,
1609
				  const struct sk_buff *skb,
1610
				  unsigned int mss_now, int nonagle)
1611
{
E
Eric Dumazet 已提交
1612
	return skb->len < mss_now &&
1613
		((nonagle & TCP_NAGLE_CORK) ||
E
Eric Dumazet 已提交
1614
		 (!nonagle && tp->packets_out && tcp_minshall_check(tp)));
1615 1616
}

E
Eric Dumazet 已提交
1617
/* Return true if the Nagle test allows this packet to be
1618 1619
 * sent now.
 */
E
Eric Dumazet 已提交
1620 1621
static inline bool tcp_nagle_test(const struct tcp_sock *tp, const struct sk_buff *skb,
				  unsigned int cur_mss, int nonagle)
1622 1623 1624 1625 1626 1627 1628 1629
{
	/* Nagle rule does not apply to frames, which sit in the middle of the
	 * write_queue (they have no chances to get new data).
	 *
	 * This is implemented in the callers, where they modify the 'nonagle'
	 * argument based upon the location of SKB in the send queue.
	 */
	if (nonagle & TCP_NAGLE_PUSH)
E
Eric Dumazet 已提交
1630
		return true;
1631

1632 1633 1634
	/* Don't use the nagle rule for urgent data (or for the final FIN).
	 * Nagle can be ignored during F-RTO too (see RFC4138).
	 */
I
Ilpo Järvinen 已提交
1635
	if (tcp_urg_mode(tp) || (tp->frto_counter == 2) ||
E
Eric Dumazet 已提交
1636
	    (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
E
Eric Dumazet 已提交
1637
		return true;
1638 1639

	if (!tcp_nagle_check(tp, skb, cur_mss, nonagle))
E
Eric Dumazet 已提交
1640
		return true;
1641

E
Eric Dumazet 已提交
1642
	return false;
1643 1644 1645
}

/* Does at least the first segment of SKB fit into the send window? */
E
Eric Dumazet 已提交
1646 1647 1648
static bool tcp_snd_wnd_test(const struct tcp_sock *tp,
			     const struct sk_buff *skb,
			     unsigned int cur_mss)
1649 1650 1651 1652 1653 1654
{
	u32 end_seq = TCP_SKB_CB(skb)->end_seq;

	if (skb->len > cur_mss)
		end_seq = TCP_SKB_CB(skb)->seq + cur_mss;

1655
	return !after(end_seq, tcp_wnd_end(tp));
1656 1657
}

1658
/* This checks if the data bearing packet SKB (usually tcp_send_head(sk))
1659 1660 1661
 * should be put on the wire right now.  If so, it returns the number of
 * packets allowed by the congestion window.
 */
1662
static unsigned int tcp_snd_test(const struct sock *sk, struct sk_buff *skb,
1663 1664
				 unsigned int cur_mss, int nonagle)
{
1665
	const struct tcp_sock *tp = tcp_sk(sk);
1666 1667
	unsigned int cwnd_quota;

1668
	tcp_init_tso_segs(sk, skb, cur_mss);
1669 1670 1671 1672 1673

	if (!tcp_nagle_test(tp, skb, cur_mss, nonagle))
		return 0;

	cwnd_quota = tcp_cwnd_test(tp, skb);
1674
	if (cwnd_quota && !tcp_snd_wnd_test(tp, skb, cur_mss))
1675 1676 1677 1678 1679
		cwnd_quota = 0;

	return cwnd_quota;
}

1680
/* Test if sending is allowed right now. */
E
Eric Dumazet 已提交
1681
bool tcp_may_send_now(struct sock *sk)
1682
{
1683
	const struct tcp_sock *tp = tcp_sk(sk);
1684
	struct sk_buff *skb = tcp_send_head(sk);
1685

E
Eric Dumazet 已提交
1686
	return skb &&
I
Ilpo Järvinen 已提交
1687
		tcp_snd_test(sk, skb, tcp_current_mss(sk),
1688
			     (tcp_skb_is_last(sk, skb) ?
E
Eric Dumazet 已提交
1689
			      tp->nonagle : TCP_NAGLE_PUSH));
1690 1691 1692 1693 1694 1695 1696 1697 1698
}

/* Trim TSO SKB to LEN bytes, put the remaining data into a new packet
 * which is put after SKB on the list.  It is very much like
 * tcp_fragment() except that it may make several kinds of assumptions
 * in order to speed up the splitting operation.  In particular, we
 * know that all the data is in scatter-gather pages, and that the
 * packet has never been sent out before (and thus is not cloned).
 */
1699
static int tso_fragment(struct sock *sk, struct sk_buff *skb, unsigned int len,
1700
			unsigned int mss_now, gfp_t gfp)
1701 1702 1703
{
	struct sk_buff *buff;
	int nlen = skb->len - len;
1704
	u8 flags;
1705 1706

	/* All of a TSO frame must be composed of paged data.  */
1707 1708
	if (skb->len != skb->data_len)
		return tcp_fragment(sk, skb, len, mss_now);
1709

1710
	buff = sk_stream_alloc_skb(sk, 0, gfp);
1711 1712 1713
	if (unlikely(buff == NULL))
		return -ENOMEM;

1714 1715
	sk->sk_wmem_queued += buff->truesize;
	sk_mem_charge(sk, buff->truesize);
1716
	buff->truesize += nlen;
1717 1718 1719 1720 1721 1722 1723 1724
	skb->truesize -= nlen;

	/* Correct the sequence numbers. */
	TCP_SKB_CB(buff)->seq = TCP_SKB_CB(skb)->seq + len;
	TCP_SKB_CB(buff)->end_seq = TCP_SKB_CB(skb)->end_seq;
	TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(buff)->seq;

	/* PSH and FIN should only be set in the second packet. */
E
Eric Dumazet 已提交
1725 1726 1727
	flags = TCP_SKB_CB(skb)->tcp_flags;
	TCP_SKB_CB(skb)->tcp_flags = flags & ~(TCPHDR_FIN | TCPHDR_PSH);
	TCP_SKB_CB(buff)->tcp_flags = flags;
1728 1729 1730 1731

	/* This packet was never sent out yet, so no SACK bits. */
	TCP_SKB_CB(buff)->sacked = 0;

1732
	buff->ip_summed = skb->ip_summed = CHECKSUM_PARTIAL;
1733 1734 1735
	skb_split(skb, buff, len);

	/* Fix up tso_factor for both original and new SKB.  */
1736 1737
	tcp_set_skb_tso_segs(sk, skb, mss_now);
	tcp_set_skb_tso_segs(sk, buff, mss_now);
1738 1739 1740

	/* Link BUFF into the send queue. */
	skb_header_release(buff);
1741
	tcp_insert_write_queue_after(skb, buff, sk);
1742 1743 1744 1745 1746 1747 1748 1749 1750

	return 0;
}

/* Try to defer sending, if possible, in order to minimize the amount
 * of TSO splitting we do.  View it as a kind of TSO Nagle test.
 *
 * This algorithm is from John Heffner.
 */
E
Eric Dumazet 已提交
1751
static bool tcp_tso_should_defer(struct sock *sk, struct sk_buff *skb)
1752
{
1753
	struct tcp_sock *tp = tcp_sk(sk);
1754
	const struct inet_connection_sock *icsk = inet_csk(sk);
1755
	u32 send_win, cong_win, limit, in_flight;
1756
	int win_divisor;
1757

E
Eric Dumazet 已提交
1758
	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
J
John Heffner 已提交
1759
		goto send_now;
1760

1761
	if (icsk->icsk_ca_state != TCP_CA_Open)
J
John Heffner 已提交
1762 1763 1764
		goto send_now;

	/* Defer for less than two clock ticks. */
I
Ilpo Järvinen 已提交
1765
	if (tp->tso_deferred &&
1766
	    (((u32)jiffies << 1) >> 1) - (tp->tso_deferred >> 1) > 1)
J
John Heffner 已提交
1767
		goto send_now;
1768

1769 1770
	in_flight = tcp_packets_in_flight(tp);

1771
	BUG_ON(tcp_skb_pcount(skb) <= 1 || (tp->snd_cwnd <= in_flight));
1772

1773
	send_win = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
1774 1775 1776 1777 1778 1779

	/* From in_flight test above, we know that cwnd > in_flight.  */
	cong_win = (tp->snd_cwnd - in_flight) * tp->mss_cache;

	limit = min(send_win, cong_win);

1780
	/* If a full-sized TSO skb can be sent, do it. */
1781 1782
	if (limit >= min_t(unsigned int, sk->sk_gso_max_size,
			   sk->sk_gso_max_segs * tp->mss_cache))
J
John Heffner 已提交
1783
		goto send_now;
1784

1785 1786 1787 1788
	/* Middle in queue won't get any more data, full sendable already? */
	if ((skb != tcp_write_queue_tail(sk)) && (limit >= skb->len))
		goto send_now;

1789 1790
	win_divisor = ACCESS_ONCE(sysctl_tcp_tso_win_divisor);
	if (win_divisor) {
1791 1792 1793 1794 1795
		u32 chunk = min(tp->snd_wnd, tp->snd_cwnd * tp->mss_cache);

		/* If at least some fraction of a window is available,
		 * just use it.
		 */
1796
		chunk /= win_divisor;
1797
		if (limit >= chunk)
J
John Heffner 已提交
1798
			goto send_now;
1799 1800 1801 1802 1803 1804
	} else {
		/* Different approach, try not to defer past a single
		 * ACK.  Receiver should ACK every other full sized
		 * frame, so if we have space for more than 3 frames
		 * then send now.
		 */
1805
		if (limit > tcp_max_tso_deferred_mss(tp) * tp->mss_cache)
J
John Heffner 已提交
1806
			goto send_now;
1807 1808 1809
	}

	/* Ok, it looks like it is advisable to defer.  */
1810
	tp->tso_deferred = 1 | (jiffies << 1);
J
John Heffner 已提交
1811

E
Eric Dumazet 已提交
1812
	return true;
J
John Heffner 已提交
1813 1814 1815

send_now:
	tp->tso_deferred = 0;
E
Eric Dumazet 已提交
1816
	return false;
1817 1818
}

J
John Heffner 已提交
1819
/* Create a new MTU probe if we are ready.
1820 1821 1822 1823
 * MTU probe is regularly attempting to increase the path MTU by
 * deliberately sending larger packets.  This discovers routing
 * changes resulting in larger path MTUs.
 *
J
John Heffner 已提交
1824 1825
 * Returns 0 if we should wait to probe (no cwnd available),
 *         1 if a probe was sent,
1826 1827
 *         -1 otherwise
 */
J
John Heffner 已提交
1828 1829 1830 1831 1832 1833 1834
static int tcp_mtu_probe(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct inet_connection_sock *icsk = inet_csk(sk);
	struct sk_buff *skb, *nskb, *next;
	int len;
	int probe_size;
1835
	int size_needed;
J
John Heffner 已提交
1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846
	int copy;
	int mss_now;

	/* Not currently probing/verifying,
	 * not in recovery,
	 * have enough cwnd, and
	 * not SACKing (the variable headers throw things off) */
	if (!icsk->icsk_mtup.enabled ||
	    icsk->icsk_mtup.probe_size ||
	    inet_csk(sk)->icsk_ca_state != TCP_CA_Open ||
	    tp->snd_cwnd < 11 ||
1847
	    tp->rx_opt.num_sacks || tp->rx_opt.dsack)
J
John Heffner 已提交
1848 1849 1850
		return -1;

	/* Very simple search strategy: just double the MSS. */
I
Ilpo Järvinen 已提交
1851
	mss_now = tcp_current_mss(sk);
1852
	probe_size = 2 * tp->mss_cache;
1853
	size_needed = probe_size + (tp->reordering + 1) * tp->mss_cache;
J
John Heffner 已提交
1854 1855 1856 1857 1858 1859
	if (probe_size > tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_high)) {
		/* TODO: set timer for probe_converge_event */
		return -1;
	}

	/* Have enough data in the send queue to probe? */
1860
	if (tp->write_seq - tp->snd_nxt < size_needed)
J
John Heffner 已提交
1861 1862
		return -1;

1863 1864
	if (tp->snd_wnd < size_needed)
		return -1;
1865
	if (after(tp->snd_nxt + size_needed, tcp_wnd_end(tp)))
1866
		return 0;
J
John Heffner 已提交
1867

1868 1869 1870
	/* Do we need to wait to drain cwnd? With none in flight, don't stall */
	if (tcp_packets_in_flight(tp) + 2 > tp->snd_cwnd) {
		if (!tcp_packets_in_flight(tp))
J
John Heffner 已提交
1871 1872 1873 1874 1875 1876 1877 1878
			return -1;
		else
			return 0;
	}

	/* We're allowed to probe.  Build it now. */
	if ((nskb = sk_stream_alloc_skb(sk, probe_size, GFP_ATOMIC)) == NULL)
		return -1;
1879 1880
	sk->sk_wmem_queued += nskb->truesize;
	sk_mem_charge(sk, nskb->truesize);
J
John Heffner 已提交
1881

1882
	skb = tcp_send_head(sk);
J
John Heffner 已提交
1883 1884 1885

	TCP_SKB_CB(nskb)->seq = TCP_SKB_CB(skb)->seq;
	TCP_SKB_CB(nskb)->end_seq = TCP_SKB_CB(skb)->seq + probe_size;
E
Eric Dumazet 已提交
1886
	TCP_SKB_CB(nskb)->tcp_flags = TCPHDR_ACK;
J
John Heffner 已提交
1887 1888
	TCP_SKB_CB(nskb)->sacked = 0;
	nskb->csum = 0;
1889
	nskb->ip_summed = skb->ip_summed;
J
John Heffner 已提交
1890

1891 1892
	tcp_insert_write_queue_before(nskb, skb, sk);

J
John Heffner 已提交
1893
	len = 0;
1894
	tcp_for_write_queue_from_safe(skb, next, sk) {
J
John Heffner 已提交
1895 1896 1897 1898 1899
		copy = min_t(int, skb->len, probe_size - len);
		if (nskb->ip_summed)
			skb_copy_bits(skb, 0, skb_put(nskb, copy), copy);
		else
			nskb->csum = skb_copy_and_csum_bits(skb, 0,
1900 1901
							    skb_put(nskb, copy),
							    copy, nskb->csum);
J
John Heffner 已提交
1902 1903 1904 1905

		if (skb->len <= copy) {
			/* We've eaten all the data from this skb.
			 * Throw it away. */
E
Eric Dumazet 已提交
1906
			TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;
1907
			tcp_unlink_write_queue(skb, sk);
1908
			sk_wmem_free_skb(sk, skb);
J
John Heffner 已提交
1909
		} else {
E
Eric Dumazet 已提交
1910
			TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags &
C
Changli Gao 已提交
1911
						   ~(TCPHDR_FIN|TCPHDR_PSH);
J
John Heffner 已提交
1912 1913
			if (!skb_shinfo(skb)->nr_frags) {
				skb_pull(skb, copy);
1914
				if (skb->ip_summed != CHECKSUM_PARTIAL)
1915 1916
					skb->csum = csum_partial(skb->data,
								 skb->len, 0);
J
John Heffner 已提交
1917 1918 1919 1920 1921 1922 1923 1924
			} else {
				__pskb_trim_head(skb, copy);
				tcp_set_skb_tso_segs(sk, skb, mss_now);
			}
			TCP_SKB_CB(skb)->seq += copy;
		}

		len += copy;
1925 1926 1927

		if (len >= probe_size)
			break;
J
John Heffner 已提交
1928 1929 1930 1931 1932 1933 1934 1935
	}
	tcp_init_tso_segs(sk, nskb, nskb->len);

	/* We're ready to send.  If this fails, the probe will
	 * be resegmented into mss-sized pieces by tcp_write_xmit(). */
	TCP_SKB_CB(nskb)->when = tcp_time_stamp;
	if (!tcp_transmit_skb(sk, nskb, 1, GFP_ATOMIC)) {
		/* Decrement cwnd here because we are sending
1936
		 * effectively two packets. */
J
John Heffner 已提交
1937
		tp->snd_cwnd--;
1938
		tcp_event_new_data_sent(sk, nskb);
J
John Heffner 已提交
1939 1940

		icsk->icsk_mtup.probe_size = tcp_mss_to_mtu(sk, nskb->len);
1941 1942
		tp->mtu_probe.probe_seq_start = TCP_SKB_CB(nskb)->seq;
		tp->mtu_probe.probe_seq_end = TCP_SKB_CB(nskb)->end_seq;
J
John Heffner 已提交
1943 1944 1945 1946 1947 1948 1949

		return 1;
	}

	return -1;
}

L
Linus Torvalds 已提交
1950 1951 1952 1953
/* This routine writes packets to the network.  It advances the
 * send_head.  This happens as incoming acks open up the remote
 * window for us.
 *
1954 1955 1956 1957
 * LARGESEND note: !tcp_urg_mode is overkill, only frames between
 * snd_up-64k-mss .. snd_up cannot be large. However, taking into
 * account rare use of URG, this is not a big flaw.
 *
E
Eric Dumazet 已提交
1958 1959
 * Returns true, if no segments are in flight and we have queued segments,
 * but cannot send anything now because of SWS or another problem.
L
Linus Torvalds 已提交
1960
 */
E
Eric Dumazet 已提交
1961 1962
static bool tcp_write_xmit(struct sock *sk, unsigned int mss_now, int nonagle,
			   int push_one, gfp_t gfp)
L
Linus Torvalds 已提交
1963 1964
{
	struct tcp_sock *tp = tcp_sk(sk);
1965
	struct sk_buff *skb;
1966 1967
	unsigned int tso_segs, sent_pkts;
	int cwnd_quota;
J
John Heffner 已提交
1968
	int result;
L
Linus Torvalds 已提交
1969

1970
	sent_pkts = 0;
J
John Heffner 已提交
1971

1972 1973 1974 1975
	if (!push_one) {
		/* Do MTU probing. */
		result = tcp_mtu_probe(sk);
		if (!result) {
E
Eric Dumazet 已提交
1976
			return false;
1977 1978 1979
		} else if (result > 0) {
			sent_pkts = 1;
		}
J
John Heffner 已提交
1980 1981
	}

1982
	while ((skb = tcp_send_head(sk))) {
1983 1984
		unsigned int limit;

E
Eric Dumazet 已提交
1985

1986
		tso_segs = tcp_init_tso_segs(sk, skb, mss_now);
1987
		BUG_ON(!tso_segs);
1988

1989 1990 1991
		if (unlikely(tp->repair) && tp->repair_queue == TCP_SEND_QUEUE)
			goto repair; /* Skip network transmission */

1992 1993 1994 1995 1996 1997 1998
		cwnd_quota = tcp_cwnd_test(tp, skb);
		if (!cwnd_quota)
			break;

		if (unlikely(!tcp_snd_wnd_test(tp, skb, mss_now)))
			break;

1999 2000 2001 2002 2003 2004
		if (tso_segs == 1) {
			if (unlikely(!tcp_nagle_test(tp, skb, mss_now,
						     (tcp_skb_is_last(sk, skb) ?
						      nonagle : TCP_NAGLE_PUSH))))
				break;
		} else {
2005
			if (!push_one && tcp_tso_should_defer(sk, skb))
2006 2007
				break;
		}
2008

E
Eric Dumazet 已提交
2009 2010 2011 2012 2013 2014 2015
		/* TSQ : sk_wmem_alloc accounts skb truesize,
		 * including skb overhead. But thats OK.
		 */
		if (atomic_read(&sk->sk_wmem_alloc) >= sysctl_tcp_limit_output_bytes) {
			set_bit(TSQ_THROTTLED, &tp->tsq_flags);
			break;
		}
2016
		limit = mss_now;
2017
		if (tso_segs > 1 && !tcp_urg_mode(tp))
2018
			limit = tcp_mss_split_point(sk, skb, mss_now,
2019 2020 2021
						    min_t(unsigned int,
							  cwnd_quota,
							  sk->sk_gso_max_segs));
L
Linus Torvalds 已提交
2022

2023
		if (skb->len > limit &&
2024
		    unlikely(tso_fragment(sk, skb, limit, mss_now, gfp)))
2025 2026
			break;

2027
		TCP_SKB_CB(skb)->when = tcp_time_stamp;
2028

2029
		if (unlikely(tcp_transmit_skb(sk, skb, 1, gfp)))
2030
			break;
L
Linus Torvalds 已提交
2031

2032
repair:
2033 2034 2035
		/* Advance the send_head.  This one is sent out.
		 * This call will increment packets_out.
		 */
2036
		tcp_event_new_data_sent(sk, skb);
L
Linus Torvalds 已提交
2037

2038
		tcp_minshall_update(tp, mss_now, skb);
2039
		sent_pkts += tcp_skb_pcount(skb);
2040 2041 2042

		if (push_one)
			break;
2043
	}
L
Linus Torvalds 已提交
2044

2045
	if (likely(sent_pkts)) {
2046 2047
		if (tcp_in_cwnd_reduction(sk))
			tp->prr_out += sent_pkts;
2048
		tcp_cwnd_validate(sk);
E
Eric Dumazet 已提交
2049
		return false;
L
Linus Torvalds 已提交
2050
	}
2051
	return !tp->packets_out && tcp_send_head(sk);
L
Linus Torvalds 已提交
2052 2053
}

2054 2055 2056 2057
/* Push out any pending frames which were held back due to
 * TCP_CORK or attempt at coalescing tiny packets.
 * The socket must be locked by the caller.
 */
2058 2059
void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
			       int nonagle)
2060
{
2061 2062 2063 2064 2065 2066 2067
	/* If we are closed, the bytes will have to remain here.
	 * In time closedown will finish, we empty the write queue and
	 * all will be happy.
	 */
	if (unlikely(sk->sk_state == TCP_CLOSE))
		return;

2068 2069
	if (tcp_write_xmit(sk, cur_mss, nonagle, 0,
			   sk_gfp_atomic(sk, GFP_ATOMIC)))
2070
		tcp_check_probe_timer(sk);
2071 2072
}

2073 2074 2075 2076 2077
/* Send _single_ skb sitting at the send head. This function requires
 * true push pending frames to setup probe timer etc.
 */
void tcp_push_one(struct sock *sk, unsigned int mss_now)
{
2078
	struct sk_buff *skb = tcp_send_head(sk);
2079 2080 2081

	BUG_ON(!skb || skb->len < mss_now);

2082
	tcp_write_xmit(sk, mss_now, TCP_NAGLE_PUSH, 1, sk->sk_allocation);
2083 2084
}

L
Linus Torvalds 已提交
2085 2086
/* This function returns the amount that we can raise the
 * usable window based on the following constraints
2087
 *
L
Linus Torvalds 已提交
2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105
 * 1. The window can never be shrunk once it is offered (RFC 793)
 * 2. We limit memory per socket
 *
 * RFC 1122:
 * "the suggested [SWS] avoidance algorithm for the receiver is to keep
 *  RECV.NEXT + RCV.WIN fixed until:
 *  RCV.BUFF - RCV.USER - RCV.WINDOW >= min(1/2 RCV.BUFF, MSS)"
 *
 * i.e. don't raise the right edge of the window until you can raise
 * it at least MSS bytes.
 *
 * Unfortunately, the recommended algorithm breaks header prediction,
 * since header prediction assumes th->window stays fixed.
 *
 * Strictly speaking, keeping th->window fixed violates the receiver
 * side SWS prevention criteria. The problem is that under this rule
 * a stream of single byte packets will cause the right side of the
 * window to always advance by a single byte.
2106
 *
L
Linus Torvalds 已提交
2107 2108
 * Of course, if the sender implements sender side SWS prevention
 * then this will not be a problem.
2109
 *
L
Linus Torvalds 已提交
2110
 * BSD seems to make the following compromise:
2111
 *
L
Linus Torvalds 已提交
2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138
 *	If the free space is less than the 1/4 of the maximum
 *	space available and the free space is less than 1/2 mss,
 *	then set the window to 0.
 *	[ Actually, bsd uses MSS and 1/4 of maximal _window_ ]
 *	Otherwise, just prevent the window from shrinking
 *	and from being larger than the largest representable value.
 *
 * This prevents incremental opening of the window in the regime
 * where TCP is limited by the speed of the reader side taking
 * data out of the TCP receive queue. It does nothing about
 * those cases where the window is constrained on the sender side
 * because the pipeline is full.
 *
 * BSD also seems to "accidentally" limit itself to windows that are a
 * multiple of MSS, at least until the free space gets quite small.
 * This would appear to be a side effect of the mbuf implementation.
 * Combining these two algorithms results in the observed behavior
 * of having a fixed window size at almost all times.
 *
 * Below we obtain similar behavior by forcing the offered window to
 * a multiple of the mss when it is feasible to do so.
 *
 * Note, we don't "adjust" for TIMESTAMP or SACK option bytes.
 * Regular options like TIMESTAMP are taken into account.
 */
u32 __tcp_select_window(struct sock *sk)
{
2139
	struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
2140
	struct tcp_sock *tp = tcp_sk(sk);
S
Stephen Hemminger 已提交
2141
	/* MSS for the peer's data.  Previous versions used mss_clamp
L
Linus Torvalds 已提交
2142 2143 2144 2145 2146
	 * here.  I don't know if the value based on our guesses
	 * of peer's MSS is better for the performance.  It's more correct
	 * but may be worse for the performance because of rcv_mss
	 * fluctuations.  --SAW  1998/11/1
	 */
2147
	int mss = icsk->icsk_ack.rcv_mss;
L
Linus Torvalds 已提交
2148 2149 2150 2151 2152
	int free_space = tcp_space(sk);
	int full_space = min_t(int, tp->window_clamp, tcp_full_space(sk));
	int window;

	if (mss > full_space)
2153
		mss = full_space;
L
Linus Torvalds 已提交
2154

2155
	if (free_space < (full_space >> 1)) {
2156
		icsk->icsk_ack.quick = 0;
L
Linus Torvalds 已提交
2157

2158
		if (sk_under_memory_pressure(sk))
2159 2160
			tp->rcv_ssthresh = min(tp->rcv_ssthresh,
					       4U * tp->advmss);
L
Linus Torvalds 已提交
2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192

		if (free_space < mss)
			return 0;
	}

	if (free_space > tp->rcv_ssthresh)
		free_space = tp->rcv_ssthresh;

	/* Don't do rounding if we are using window scaling, since the
	 * scaled window will not line up with the MSS boundary anyway.
	 */
	window = tp->rcv_wnd;
	if (tp->rx_opt.rcv_wscale) {
		window = free_space;

		/* Advertise enough space so that it won't get scaled away.
		 * Import case: prevent zero window announcement if
		 * 1<<rcv_wscale > mss.
		 */
		if (((window >> tp->rx_opt.rcv_wscale) << tp->rx_opt.rcv_wscale) != window)
			window = (((window >> tp->rx_opt.rcv_wscale) + 1)
				  << tp->rx_opt.rcv_wscale);
	} else {
		/* Get the largest window that is a nice multiple of mss.
		 * Window clamp already applied above.
		 * If our current window offering is within 1 mss of the
		 * free space we just keep it. This prevents the divide
		 * and multiply from happening most of the time.
		 * We also don't do any window rounding when the free space
		 * is too small.
		 */
		if (window <= free_space - mss || window > free_space)
2193
			window = (free_space / mss) * mss;
2194
		else if (mss == full_space &&
2195
			 free_space > window + (full_space >> 1))
2196
			window = free_space;
L
Linus Torvalds 已提交
2197 2198 2199 2200 2201
	}

	return window;
}

2202 2203
/* Collapses two adjacent SKB's during retransmission. */
static void tcp_collapse_retrans(struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
2204 2205
{
	struct tcp_sock *tp = tcp_sk(sk);
2206
	struct sk_buff *next_skb = tcp_write_queue_next(sk, skb);
2207
	int skb_size, next_skb_size;
L
Linus Torvalds 已提交
2208

2209 2210
	skb_size = skb->len;
	next_skb_size = next_skb->len;
L
Linus Torvalds 已提交
2211

2212
	BUG_ON(tcp_skb_pcount(skb) != 1 || tcp_skb_pcount(next_skb) != 1);
2213

2214
	tcp_highest_sack_combine(sk, next_skb, skb);
L
Linus Torvalds 已提交
2215

2216
	tcp_unlink_write_queue(next_skb, sk);
L
Linus Torvalds 已提交
2217

2218 2219
	skb_copy_from_linear_data(next_skb, skb_put(skb, next_skb_size),
				  next_skb_size);
L
Linus Torvalds 已提交
2220

2221 2222
	if (next_skb->ip_summed == CHECKSUM_PARTIAL)
		skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
2223

2224 2225
	if (skb->ip_summed != CHECKSUM_PARTIAL)
		skb->csum = csum_block_add(skb->csum, next_skb->csum, skb_size);
L
Linus Torvalds 已提交
2226

2227 2228
	/* Update sequence range on original skb. */
	TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(next_skb)->end_seq;
L
Linus Torvalds 已提交
2229

2230
	/* Merge over control information. This moves PSH/FIN etc. over */
E
Eric Dumazet 已提交
2231
	TCP_SKB_CB(skb)->tcp_flags |= TCP_SKB_CB(next_skb)->tcp_flags;
2232 2233 2234 2235 2236 2237 2238

	/* All done, get rid of second SKB and account for it so
	 * packet counting does not break.
	 */
	TCP_SKB_CB(skb)->sacked |= TCP_SKB_CB(next_skb)->sacked & TCPCB_EVER_RETRANS;

	/* changed transmit queue under us so clear hints */
2239 2240 2241
	tcp_clear_retrans_hints_partial(tp);
	if (next_skb == tp->retransmit_skb_hint)
		tp->retransmit_skb_hint = skb;
2242

2243 2244
	tcp_adjust_pcount(sk, next_skb, tcp_skb_pcount(next_skb));

2245
	sk_wmem_free_skb(sk, next_skb);
L
Linus Torvalds 已提交
2246 2247
}

2248
/* Check if coalescing SKBs is legal. */
E
Eric Dumazet 已提交
2249
static bool tcp_can_collapse(const struct sock *sk, const struct sk_buff *skb)
2250 2251
{
	if (tcp_skb_pcount(skb) > 1)
E
Eric Dumazet 已提交
2252
		return false;
2253 2254
	/* TODO: SACK collapsing could be used to remove this condition */
	if (skb_shinfo(skb)->nr_frags != 0)
E
Eric Dumazet 已提交
2255
		return false;
2256
	if (skb_cloned(skb))
E
Eric Dumazet 已提交
2257
		return false;
2258
	if (skb == tcp_send_head(sk))
E
Eric Dumazet 已提交
2259
		return false;
2260 2261
	/* Some heurestics for collapsing over SACK'd could be invented */
	if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)
E
Eric Dumazet 已提交
2262
		return false;
2263

E
Eric Dumazet 已提交
2264
	return true;
2265 2266
}

2267 2268 2269
/* Collapse packets in the retransmit queue to make to create
 * less packets on the wire. This is only done on retransmission.
 */
2270 2271 2272 2273 2274
static void tcp_retrans_try_collapse(struct sock *sk, struct sk_buff *to,
				     int space)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb = to, *tmp;
E
Eric Dumazet 已提交
2275
	bool first = true;
2276 2277 2278

	if (!sysctl_tcp_retrans_collapse)
		return;
E
Eric Dumazet 已提交
2279
	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)
2280 2281 2282 2283 2284 2285 2286 2287 2288
		return;

	tcp_for_write_queue_from_safe(skb, tmp, sk) {
		if (!tcp_can_collapse(sk, skb))
			break;

		space -= skb->len;

		if (first) {
E
Eric Dumazet 已提交
2289
			first = false;
2290 2291 2292 2293 2294 2295 2296 2297
			continue;
		}

		if (space < 0)
			break;
		/* Punt if not enough space exists in the first SKB for
		 * the data in the second
		 */
2298
		if (skb->len > skb_availroom(to))
2299 2300 2301 2302 2303 2304 2305 2306 2307
			break;

		if (after(TCP_SKB_CB(skb)->end_seq, tcp_wnd_end(tp)))
			break;

		tcp_collapse_retrans(sk, to);
	}
}

L
Linus Torvalds 已提交
2308 2309 2310 2311
/* This retransmits one SKB.  Policy decisions and retransmit queue
 * state updates are done by the caller.  Returns non-zero if an
 * error occurred which prevented the send.
 */
2312
int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
2313 2314
{
	struct tcp_sock *tp = tcp_sk(sk);
J
John Heffner 已提交
2315
	struct inet_connection_sock *icsk = inet_csk(sk);
2316
	unsigned int cur_mss;
L
Linus Torvalds 已提交
2317

J
John Heffner 已提交
2318 2319 2320 2321 2322
	/* Inconslusive MTU probe */
	if (icsk->icsk_mtup.probe_size) {
		icsk->icsk_mtup.probe_size = 0;
	}

L
Linus Torvalds 已提交
2323
	/* Do not sent more than we queued. 1/4 is reserved for possible
S
Stephen Hemminger 已提交
2324
	 * copying overhead: fragmentation, tunneling, mangling etc.
L
Linus Torvalds 已提交
2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336
	 */
	if (atomic_read(&sk->sk_wmem_alloc) >
	    min(sk->sk_wmem_queued + (sk->sk_wmem_queued >> 2), sk->sk_sndbuf))
		return -EAGAIN;

	if (before(TCP_SKB_CB(skb)->seq, tp->snd_una)) {
		if (before(TCP_SKB_CB(skb)->end_seq, tp->snd_una))
			BUG();
		if (tcp_trim_head(sk, skb, tp->snd_una - TCP_SKB_CB(skb)->seq))
			return -ENOMEM;
	}

2337 2338 2339
	if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk))
		return -EHOSTUNREACH; /* Routing failure or similar. */

I
Ilpo Järvinen 已提交
2340
	cur_mss = tcp_current_mss(sk);
2341

L
Linus Torvalds 已提交
2342 2343 2344 2345 2346
	/* If receiver has shrunk his window, and skb is out of
	 * new window, do not retransmit it. The exception is the
	 * case, when window is shrunk to zero. In this case
	 * our retransmit serves as a zero window probe.
	 */
2347 2348
	if (!before(TCP_SKB_CB(skb)->seq, tcp_wnd_end(tp)) &&
	    TCP_SKB_CB(skb)->seq != tp->snd_una)
L
Linus Torvalds 已提交
2349 2350 2351
		return -EAGAIN;

	if (skb->len > cur_mss) {
2352
		if (tcp_fragment(sk, skb, cur_mss, cur_mss))
L
Linus Torvalds 已提交
2353
			return -ENOMEM; /* We'll try again later. */
2354
	} else {
2355 2356 2357 2358 2359 2360
		int oldpcount = tcp_skb_pcount(skb);

		if (unlikely(oldpcount > 1)) {
			tcp_init_tso_segs(sk, skb, cur_mss);
			tcp_adjust_pcount(sk, skb, oldpcount - tcp_skb_pcount(skb));
		}
L
Linus Torvalds 已提交
2361 2362
	}

2363
	tcp_retrans_try_collapse(sk, skb, cur_mss);
L
Linus Torvalds 已提交
2364 2365 2366 2367 2368

	/* Some Solaris stacks overoptimize and ignore the FIN on a
	 * retransmit when old data is attached.  So strip it off
	 * since it is cheap to do so and saves bytes on the network.
	 */
S
Stephen Hemminger 已提交
2369
	if (skb->len > 0 &&
E
Eric Dumazet 已提交
2370
	    (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) &&
S
Stephen Hemminger 已提交
2371
	    tp->snd_una == (TCP_SKB_CB(skb)->end_seq - 1)) {
L
Linus Torvalds 已提交
2372
		if (!pskb_trim(skb, 0)) {
2373 2374
			/* Reuse, even though it does some unnecessary work */
			tcp_init_nondata_skb(skb, TCP_SKB_CB(skb)->end_seq - 1,
E
Eric Dumazet 已提交
2375
					     TCP_SKB_CB(skb)->tcp_flags);
L
Linus Torvalds 已提交
2376 2377 2378 2379 2380 2381 2382 2383 2384
			skb->ip_summed = CHECKSUM_NONE;
		}
	}

	/* Make a copy, if the first transmission SKB clone we made
	 * is still in somebody's hands, else make a clone.
	 */
	TCP_SKB_CB(skb)->when = tcp_time_stamp;

E
Eric Dumazet 已提交
2385 2386 2387 2388
	/* make sure skb->data is aligned on arches that require it */
	if (unlikely(NET_IP_ALIGN && ((unsigned long)skb->data & 3))) {
		struct sk_buff *nskb = __pskb_copy(skb, MAX_TCP_HEADER,
						   GFP_ATOMIC);
2389 2390
		return nskb ? tcp_transmit_skb(sk, nskb, 0, GFP_ATOMIC) :
			      -ENOBUFS;
E
Eric Dumazet 已提交
2391
	} else {
2392
		return tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
E
Eric Dumazet 已提交
2393
	}
2394 2395 2396 2397 2398 2399
}

int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
{
	struct tcp_sock *tp = tcp_sk(sk);
	int err = __tcp_retransmit_skb(sk, skb);
L
Linus Torvalds 已提交
2400 2401 2402

	if (err == 0) {
		/* Update global TCP statistics. */
P
Pavel Emelyanov 已提交
2403
		TCP_INC_STATS(sock_net(sk), TCP_MIB_RETRANSSEGS);
L
Linus Torvalds 已提交
2404 2405 2406 2407

		tp->total_retrans++;

#if FASTRETRANS_DEBUG > 0
2408
		if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) {
2409
			net_dbg_ratelimited("retrans_out leaked\n");
L
Linus Torvalds 已提交
2410 2411
		}
#endif
2412 2413
		if (!tp->retrans_out)
			tp->lost_retrans_low = tp->snd_nxt;
L
Linus Torvalds 已提交
2414 2415 2416 2417 2418 2419 2420
		TCP_SKB_CB(skb)->sacked |= TCPCB_RETRANS;
		tp->retrans_out += tcp_skb_pcount(skb);

		/* Save stamp of the first retransmit. */
		if (!tp->retrans_stamp)
			tp->retrans_stamp = TCP_SKB_CB(skb)->when;

Y
Yuchung Cheng 已提交
2421
		tp->undo_retrans += tcp_skb_pcount(skb);
L
Linus Torvalds 已提交
2422 2423 2424 2425 2426 2427 2428 2429 2430

		/* snd_nxt is stored to detect loss of retransmitted segment,
		 * see tcp_input.c tcp_sacktag_write_queue().
		 */
		TCP_SKB_CB(skb)->ack_seq = tp->snd_nxt;
	}
	return err;
}

2431 2432 2433
/* Check if we forward retransmits are possible in the current
 * window/congestion state.
 */
E
Eric Dumazet 已提交
2434
static bool tcp_can_forward_retransmit(struct sock *sk)
2435 2436
{
	const struct inet_connection_sock *icsk = inet_csk(sk);
2437
	const struct tcp_sock *tp = tcp_sk(sk);
2438 2439 2440

	/* Forward retransmissions are possible only during Recovery. */
	if (icsk->icsk_ca_state != TCP_CA_Recovery)
E
Eric Dumazet 已提交
2441
		return false;
2442 2443 2444

	/* No forward retransmissions in Reno are possible. */
	if (tcp_is_reno(tp))
E
Eric Dumazet 已提交
2445
		return false;
2446 2447 2448 2449 2450 2451 2452 2453 2454 2455

	/* Yeah, we have to make difficult choice between forward transmission
	 * and retransmission... Both ways have their merits...
	 *
	 * For now we do not retransmit anything, while we have some new
	 * segments to send. In the other cases, follow rule 3 for
	 * NextSeg() specified in RFC3517.
	 */

	if (tcp_may_send_now(sk))
E
Eric Dumazet 已提交
2456
		return false;
2457

E
Eric Dumazet 已提交
2458
	return true;
2459 2460
}

L
Linus Torvalds 已提交
2461 2462 2463 2464 2465 2466 2467 2468 2469 2470
/* This gets called after a retransmit timeout, and the initially
 * retransmitted data is acknowledged.  It tries to continue
 * resending the rest of the retransmit queue, until either
 * we've sent it all or the congestion window limit is reached.
 * If doing SACK, the first ACK which comes back for a timeout
 * based retransmit packet might feed us FACK information again.
 * If so, we use it to avoid unnecessarily retransmissions.
 */
void tcp_xmit_retransmit_queue(struct sock *sk)
{
2471
	const struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
2472 2473
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb;
2474
	struct sk_buff *hole = NULL;
2475
	u32 last_lost;
2476
	int mib_idx;
2477
	int fwd_rexmitting = 0;
2478

2479 2480 2481
	if (!tp->packets_out)
		return;

2482 2483 2484
	if (!tp->lost_out)
		tp->retransmit_high = tp->snd_una;

2485
	if (tp->retransmit_skb_hint) {
2486
		skb = tp->retransmit_skb_hint;
2487 2488 2489 2490
		last_lost = TCP_SKB_CB(skb)->end_seq;
		if (after(last_lost, tp->retransmit_high))
			last_lost = tp->retransmit_high;
	} else {
2491
		skb = tcp_write_queue_head(sk);
2492 2493
		last_lost = tp->snd_una;
	}
L
Linus Torvalds 已提交
2494

2495 2496
	tcp_for_write_queue_from(skb, sk) {
		__u8 sacked = TCP_SKB_CB(skb)->sacked;
L
Linus Torvalds 已提交
2497

2498 2499 2500
		if (skb == tcp_send_head(sk))
			break;
		/* we could do better than to assign each time */
2501 2502
		if (hole == NULL)
			tp->retransmit_skb_hint = skb;
2503 2504 2505 2506 2507 2508 2509 2510 2511 2512

		/* Assume this retransmit will generate
		 * only one packet for congestion window
		 * calculation purposes.  This works because
		 * tcp_retransmit_skb() will chop up the
		 * packet to be MSS sized and all the
		 * packet counting works out.
		 */
		if (tcp_packets_in_flight(tp) >= tp->snd_cwnd)
			return;
L
Linus Torvalds 已提交
2513

2514 2515 2516 2517 2518
		if (fwd_rexmitting) {
begin_fwd:
			if (!before(TCP_SKB_CB(skb)->seq, tcp_highest_sack_seq(tp)))
				break;
			mib_idx = LINUX_MIB_TCPFORWARDRETRANS;
2519

2520
		} else if (!before(TCP_SKB_CB(skb)->seq, tp->retransmit_high)) {
2521
			tp->retransmit_high = last_lost;
2522 2523 2524 2525 2526 2527 2528 2529 2530
			if (!tcp_can_forward_retransmit(sk))
				break;
			/* Backtrack if necessary to non-L'ed skb */
			if (hole != NULL) {
				skb = hole;
				hole = NULL;
			}
			fwd_rexmitting = 1;
			goto begin_fwd;
L
Linus Torvalds 已提交
2531

2532
		} else if (!(sacked & TCPCB_LOST)) {
2533
			if (hole == NULL && !(sacked & (TCPCB_SACKED_RETRANS|TCPCB_SACKED_ACKED)))
2534 2535
				hole = skb;
			continue;
L
Linus Torvalds 已提交
2536

2537
		} else {
2538
			last_lost = TCP_SKB_CB(skb)->end_seq;
2539 2540 2541 2542 2543
			if (icsk->icsk_ca_state != TCP_CA_Loss)
				mib_idx = LINUX_MIB_TCPFASTRETRANS;
			else
				mib_idx = LINUX_MIB_TCPSLOWSTARTRETRANS;
		}
L
Linus Torvalds 已提交
2544

2545
		if (sacked & (TCPCB_SACKED_ACKED|TCPCB_SACKED_RETRANS))
L
Linus Torvalds 已提交
2546 2547
			continue;

2548 2549
		if (tcp_retransmit_skb(sk, skb)) {
			NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPRETRANSFAIL);
2550
			return;
2551
		}
2552
		NET_INC_STATS_BH(sock_net(sk), mib_idx);
L
Linus Torvalds 已提交
2553

2554
		if (tcp_in_cwnd_reduction(sk))
2555 2556
			tp->prr_out += tcp_skb_pcount(skb);

2557
		if (skb == tcp_write_queue_head(sk))
2558 2559 2560
			inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
						  inet_csk(sk)->icsk_rto,
						  TCP_RTO_MAX);
L
Linus Torvalds 已提交
2561 2562 2563 2564 2565 2566 2567 2568
	}
}

/* Send a fin.  The caller locks the socket for us.  This cannot be
 * allowed to fail queueing a FIN frame under any circumstances.
 */
void tcp_send_fin(struct sock *sk)
{
2569
	struct tcp_sock *tp = tcp_sk(sk);
2570
	struct sk_buff *skb = tcp_write_queue_tail(sk);
L
Linus Torvalds 已提交
2571
	int mss_now;
2572

L
Linus Torvalds 已提交
2573 2574 2575 2576
	/* Optimization, tack on the FIN if we have a queue of
	 * unsent frames.  But be careful about outgoing SACKS
	 * and IP options.
	 */
I
Ilpo Järvinen 已提交
2577
	mss_now = tcp_current_mss(sk);
L
Linus Torvalds 已提交
2578

2579
	if (tcp_send_head(sk) != NULL) {
E
Eric Dumazet 已提交
2580
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_FIN;
L
Linus Torvalds 已提交
2581 2582 2583 2584 2585
		TCP_SKB_CB(skb)->end_seq++;
		tp->write_seq++;
	} else {
		/* Socket is locked, keep trying until memory is available. */
		for (;;) {
2586 2587
			skb = alloc_skb_fclone(MAX_TCP_HEADER,
					       sk->sk_allocation);
L
Linus Torvalds 已提交
2588 2589 2590 2591 2592 2593 2594 2595
			if (skb)
				break;
			yield();
		}

		/* Reserve space for headers and prepare control bits. */
		skb_reserve(skb, MAX_TCP_HEADER);
		/* FIN eats a sequence byte, write_seq advanced by tcp_queue_skb(). */
2596
		tcp_init_nondata_skb(skb, tp->write_seq,
C
Changli Gao 已提交
2597
				     TCPHDR_ACK | TCPHDR_FIN);
L
Linus Torvalds 已提交
2598 2599
		tcp_queue_skb(sk, skb);
	}
2600
	__tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_OFF);
L
Linus Torvalds 已提交
2601 2602 2603 2604 2605
}

/* We get here when a process closes a file descriptor (either due to
 * an explicit close() or as a byproduct of exit()'ing) and there
 * was unread data in the receive queue.  This behavior is recommended
2606
 * by RFC 2525, section 2.17.  -DaveM
L
Linus Torvalds 已提交
2607
 */
A
Al Viro 已提交
2608
void tcp_send_active_reset(struct sock *sk, gfp_t priority)
L
Linus Torvalds 已提交
2609 2610 2611 2612 2613 2614
{
	struct sk_buff *skb;

	/* NOTE: No TCP options attached and we never retransmit this. */
	skb = alloc_skb(MAX_TCP_HEADER, priority);
	if (!skb) {
P
Pavel Emelyanov 已提交
2615
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTFAILED);
L
Linus Torvalds 已提交
2616 2617 2618 2619 2620
		return;
	}

	/* Reserve space for headers and prepare control bits. */
	skb_reserve(skb, MAX_TCP_HEADER);
2621
	tcp_init_nondata_skb(skb, tcp_acceptable_seq(sk),
C
Changli Gao 已提交
2622
			     TCPHDR_ACK | TCPHDR_RST);
L
Linus Torvalds 已提交
2623 2624
	/* Send it off. */
	TCP_SKB_CB(skb)->when = tcp_time_stamp;
2625
	if (tcp_transmit_skb(sk, skb, 0, priority))
P
Pavel Emelyanov 已提交
2626
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTFAILED);
2627

P
Pavel Emelyanov 已提交
2628
	TCP_INC_STATS(sock_net(sk), TCP_MIB_OUTRSTS);
L
Linus Torvalds 已提交
2629 2630
}

2631 2632
/* Send a crossed SYN-ACK during socket establishment.
 * WARNING: This routine must only be called when we have already sent
L
Linus Torvalds 已提交
2633 2634 2635 2636 2637 2638
 * a SYN packet that crossed the incoming SYN that caused this routine
 * to get called. If this assumption fails then the initial rcv_wnd
 * and rcv_wscale values will not be correct.
 */
int tcp_send_synack(struct sock *sk)
{
2639
	struct sk_buff *skb;
L
Linus Torvalds 已提交
2640

2641
	skb = tcp_write_queue_head(sk);
E
Eric Dumazet 已提交
2642
	if (skb == NULL || !(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
2643
		pr_debug("%s: wrong queue state\n", __func__);
L
Linus Torvalds 已提交
2644 2645
		return -EFAULT;
	}
E
Eric Dumazet 已提交
2646
	if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)) {
L
Linus Torvalds 已提交
2647 2648 2649 2650
		if (skb_cloned(skb)) {
			struct sk_buff *nskb = skb_copy(skb, GFP_ATOMIC);
			if (nskb == NULL)
				return -ENOMEM;
2651
			tcp_unlink_write_queue(skb, sk);
L
Linus Torvalds 已提交
2652
			skb_header_release(nskb);
2653
			__tcp_add_write_queue_head(sk, nskb);
2654 2655 2656
			sk_wmem_free_skb(sk, skb);
			sk->sk_wmem_queued += nskb->truesize;
			sk_mem_charge(sk, nskb->truesize);
L
Linus Torvalds 已提交
2657 2658 2659
			skb = nskb;
		}

E
Eric Dumazet 已提交
2660
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ACK;
L
Linus Torvalds 已提交
2661 2662 2663
		TCP_ECN_send_synack(tcp_sk(sk), skb);
	}
	TCP_SKB_CB(skb)->when = tcp_time_stamp;
2664
	return tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
L
Linus Torvalds 已提交
2665 2666
}

2667 2668 2669 2670 2671 2672 2673 2674 2675 2676
/**
 * tcp_make_synack - Prepare a SYN-ACK.
 * sk: listener socket
 * dst: dst entry attached to the SYNACK
 * req: request_sock pointer
 * rvp: request_values pointer
 *
 * Allocate one skb and build a SYNACK packet.
 * @dst is consumed : Caller should not use it again.
 */
2677
struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
2678
				struct request_sock *req,
2679 2680
				struct request_values *rvp,
				struct tcp_fastopen_cookie *foc)
L
Linus Torvalds 已提交
2681
{
2682
	struct tcp_out_options opts;
2683
	struct tcp_extend_values *xvp = tcp_xv(rvp);
2684
	struct inet_request_sock *ireq = inet_rsk(req);
L
Linus Torvalds 已提交
2685
	struct tcp_sock *tp = tcp_sk(sk);
E
Eric Dumazet 已提交
2686
	const struct tcp_cookie_values *cvp = tp->cookie_values;
L
Linus Torvalds 已提交
2687 2688
	struct tcphdr *th;
	struct sk_buff *skb;
2689
	struct tcp_md5sig_key *md5;
2690
	int tcp_header_size;
2691
	int mss;
E
Eric Dumazet 已提交
2692
	int s_data_desired = 0;
L
Linus Torvalds 已提交
2693

E
Eric Dumazet 已提交
2694 2695
	if (cvp != NULL && cvp->s_data_constant && cvp->s_data_desired)
		s_data_desired = cvp->s_data_desired;
2696 2697
	skb = alloc_skb(MAX_TCP_HEADER + 15 + s_data_desired,
			sk_gfp_atomic(sk, GFP_ATOMIC));
2698 2699
	if (unlikely(!skb)) {
		dst_release(dst);
L
Linus Torvalds 已提交
2700
		return NULL;
2701
	}
L
Linus Torvalds 已提交
2702 2703 2704
	/* Reserve space for headers. */
	skb_reserve(skb, MAX_TCP_HEADER);

2705
	skb_dst_set(skb, dst);
L
Linus Torvalds 已提交
2706

2707
	mss = dst_metric_advmss(dst);
2708 2709 2710
	if (tp->rx_opt.user_mss && tp->rx_opt.user_mss < mss)
		mss = tp->rx_opt.user_mss;

A
Adam Langley 已提交
2711 2712 2713 2714
	if (req->rcv_wnd == 0) { /* ignored for retransmitted syns */
		__u8 rcv_wscale;
		/* Set this up on the first call only */
		req->window_clamp = tp->window_clamp ? : dst_metric(dst, RTAX_WINDOW);
2715 2716 2717 2718 2719 2720

		/* limit the window selection if the user enforce a smaller rx buffer */
		if (sk->sk_userlocks & SOCK_RCVBUF_LOCK &&
		    (req->window_clamp > tcp_full_space(sk) || req->window_clamp == 0))
			req->window_clamp = tcp_full_space(sk);

A
Adam Langley 已提交
2721 2722
		/* tcp_full_space because it is guaranteed to be the first packet */
		tcp_select_initial_window(tcp_full_space(sk),
2723
			mss - (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0),
A
Adam Langley 已提交
2724 2725 2726
			&req->rcv_wnd,
			&req->window_clamp,
			ireq->wscale_ok,
2727 2728
			&rcv_wscale,
			dst_metric(dst, RTAX_INITRWND));
A
Adam Langley 已提交
2729 2730 2731 2732
		ireq->rcv_wscale = rcv_wscale;
	}

	memset(&opts, 0, sizeof(opts));
2733 2734 2735 2736 2737
#ifdef CONFIG_SYN_COOKIES
	if (unlikely(req->cookie_ts))
		TCP_SKB_CB(skb)->when = cookie_init_timestamp(req);
	else
#endif
A
Adam Langley 已提交
2738
	TCP_SKB_CB(skb)->when = tcp_time_stamp;
2739
	tcp_header_size = tcp_synack_options(sk, req, mss,
2740
					     skb, &opts, &md5, xvp, foc)
2741
			+ sizeof(*th);
2742

2743 2744
	skb_push(skb, tcp_header_size);
	skb_reset_transport_header(skb);
L
Linus Torvalds 已提交
2745

2746
	th = tcp_hdr(skb);
L
Linus Torvalds 已提交
2747 2748 2749 2750
	memset(th, 0, sizeof(struct tcphdr));
	th->syn = 1;
	th->ack = 1;
	TCP_ECN_make_synack(req, th);
2751
	th->source = ireq->loc_port;
2752
	th->dest = ireq->rmt_port;
2753 2754 2755 2756
	/* Setting of flags are superfluous here for callers (and ECE is
	 * not even correctly set)
	 */
	tcp_init_nondata_skb(skb, tcp_rsk(req)->snt_isn,
C
Changli Gao 已提交
2757
			     TCPHDR_SYN | TCPHDR_ACK);
2758 2759

	if (OPTION_COOKIE_EXTENSION & opts.options) {
E
Eric Dumazet 已提交
2760 2761
		if (s_data_desired) {
			u8 *buf = skb_put(skb, s_data_desired);
2762 2763

			/* copy data directly from the listening socket. */
E
Eric Dumazet 已提交
2764 2765
			memcpy(buf, cvp->s_data_payload, s_data_desired);
			TCP_SKB_CB(skb)->end_seq += s_data_desired;
2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781
		}

		if (opts.hash_size > 0) {
			__u32 workspace[SHA_WORKSPACE_WORDS];
			u32 *mess = &xvp->cookie_bakery[COOKIE_DIGEST_WORDS];
			u32 *tail = &mess[COOKIE_MESSAGE_WORDS-1];

			/* Secret recipe depends on the Timestamp, (future)
			 * Sequence and Acknowledgment Numbers, Initiator
			 * Cookie, and others handled by IP variant caller.
			 */
			*tail-- ^= opts.tsval;
			*tail-- ^= tcp_rsk(req)->rcv_isn + 1;
			*tail-- ^= TCP_SKB_CB(skb)->seq + 1;

			/* recommended */
2782
			*tail-- ^= (((__force u32)th->dest << 16) | (__force u32)th->source);
D
David S. Miller 已提交
2783
			*tail-- ^= (u32)(unsigned long)cvp; /* per sockopt */
2784 2785 2786 2787 2788 2789 2790 2791 2792

			sha_transform((__u32 *)&xvp->cookie_bakery[0],
				      (char *)mess,
				      &workspace[0]);
			opts.hash_location =
				(__u8 *)&xvp->cookie_bakery[0];
		}
	}

L
Linus Torvalds 已提交
2793
	th->seq = htonl(TCP_SKB_CB(skb)->seq);
2794 2795
	/* XXX data is queued and acked as is. No buffer/window check */
	th->ack_seq = htonl(tcp_rsk(req)->rcv_nxt);
L
Linus Torvalds 已提交
2796 2797

	/* RFC1323: The window in SYN & SYN/ACK segments is never scaled. */
2798
	th->window = htons(min(req->rcv_wnd, 65535U));
2799
	tcp_options_write((__be32 *)(th + 1), tp, &opts);
L
Linus Torvalds 已提交
2800
	th->doff = (tcp_header_size >> 2);
2801
	TCP_ADD_STATS(sock_net(sk), TCP_MIB_OUTSEGS, tcp_skb_pcount(skb));
2802 2803 2804 2805

#ifdef CONFIG_TCP_MD5SIG
	/* Okay, we have all we need - do the md5 hash if needed */
	if (md5) {
2806
		tcp_rsk(req)->af_specific->calc_md5_hash(opts.hash_location,
2807
					       md5, NULL, req, skb);
2808 2809 2810
	}
#endif

L
Linus Torvalds 已提交
2811 2812
	return skb;
}
E
Eric Dumazet 已提交
2813
EXPORT_SYMBOL(tcp_make_synack);
L
Linus Torvalds 已提交
2814

2815
/* Do all connect socket setups that can be done AF independent. */
P
Pavel Emelyanov 已提交
2816
void tcp_connect_init(struct sock *sk)
L
Linus Torvalds 已提交
2817
{
2818
	const struct dst_entry *dst = __sk_dst_get(sk);
L
Linus Torvalds 已提交
2819 2820 2821 2822 2823 2824 2825
	struct tcp_sock *tp = tcp_sk(sk);
	__u8 rcv_wscale;

	/* We'll fix this up when we get a response from the other end.
	 * See tcp_input.c:tcp_rcv_state_process case TCP_SYN_SENT.
	 */
	tp->tcp_header_len = sizeof(struct tcphdr) +
2826
		(sysctl_tcp_timestamps ? TCPOLEN_TSTAMP_ALIGNED : 0);
L
Linus Torvalds 已提交
2827

2828 2829 2830 2831 2832
#ifdef CONFIG_TCP_MD5SIG
	if (tp->af_specific->md5_lookup(sk, sk) != NULL)
		tp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
#endif

L
Linus Torvalds 已提交
2833 2834 2835 2836
	/* If user gave his TCP_MAXSEG, record it to clamp */
	if (tp->rx_opt.user_mss)
		tp->rx_opt.mss_clamp = tp->rx_opt.user_mss;
	tp->max_window = 0;
J
John Heffner 已提交
2837
	tcp_mtup_init(sk);
L
Linus Torvalds 已提交
2838 2839 2840 2841
	tcp_sync_mss(sk, dst_mtu(dst));

	if (!tp->window_clamp)
		tp->window_clamp = dst_metric(dst, RTAX_WINDOW);
2842
	tp->advmss = dst_metric_advmss(dst);
2843 2844 2845
	if (tp->rx_opt.user_mss && tp->rx_opt.user_mss < tp->advmss)
		tp->advmss = tp->rx_opt.user_mss;

L
Linus Torvalds 已提交
2846 2847
	tcp_initialize_rcv_mss(sk);

2848 2849 2850 2851 2852
	/* limit the window selection if the user enforce a smaller rx buffer */
	if (sk->sk_userlocks & SOCK_RCVBUF_LOCK &&
	    (tp->window_clamp > tcp_full_space(sk) || tp->window_clamp == 0))
		tp->window_clamp = tcp_full_space(sk);

L
Linus Torvalds 已提交
2853 2854 2855 2856
	tcp_select_initial_window(tcp_full_space(sk),
				  tp->advmss - (tp->rx_opt.ts_recent_stamp ? tp->tcp_header_len - sizeof(struct tcphdr) : 0),
				  &tp->rcv_wnd,
				  &tp->window_clamp,
2857
				  sysctl_tcp_window_scaling,
2858 2859
				  &rcv_wscale,
				  dst_metric(dst, RTAX_INITRWND));
L
Linus Torvalds 已提交
2860 2861 2862 2863 2864 2865 2866

	tp->rx_opt.rcv_wscale = rcv_wscale;
	tp->rcv_ssthresh = tp->rcv_wnd;

	sk->sk_err = 0;
	sock_reset_flag(sk, SOCK_DONE);
	tp->snd_wnd = 0;
2867
	tcp_init_wl(tp, 0);
L
Linus Torvalds 已提交
2868 2869
	tp->snd_una = tp->write_seq;
	tp->snd_sml = tp->write_seq;
I
Ilpo Järvinen 已提交
2870
	tp->snd_up = tp->write_seq;
P
Pavel Emelyanov 已提交
2871
	tp->snd_nxt = tp->write_seq;
P
Pavel Emelyanov 已提交
2872 2873 2874 2875 2876

	if (likely(!tp->repair))
		tp->rcv_nxt = 0;
	tp->rcv_wup = tp->rcv_nxt;
	tp->copied_seq = tp->rcv_nxt;
L
Linus Torvalds 已提交
2877

2878 2879
	inet_csk(sk)->icsk_rto = TCP_TIMEOUT_INIT;
	inet_csk(sk)->icsk_retransmits = 0;
L
Linus Torvalds 已提交
2880 2881 2882
	tcp_clear_retrans(tp);
}

2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907
static void tcp_connect_queue_skb(struct sock *sk, struct sk_buff *skb)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);

	tcb->end_seq += skb->len;
	skb_header_release(skb);
	__tcp_add_write_queue_tail(sk, skb);
	sk->sk_wmem_queued += skb->truesize;
	sk_mem_charge(sk, skb->truesize);
	tp->write_seq = tcb->end_seq;
	tp->packets_out += tcp_skb_pcount(skb);
}

/* Build and send a SYN with data and (cached) Fast Open cookie. However,
 * queue a data-only packet after the regular SYN, such that regular SYNs
 * are retransmitted on timeouts. Also if the remote SYN-ACK acknowledges
 * only the SYN sequence, the data are retransmitted in the first ACK.
 * If cookie is not cached or other error occurs, falls back to send a
 * regular SYN with Fast Open cookie request option.
 */
static int tcp_send_syn_data(struct sock *sk, struct sk_buff *syn)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct tcp_fastopen_request *fo = tp->fastopen_req;
2908
	int syn_loss = 0, space, i, err = 0, iovlen = fo->data->msg_iovlen;
2909
	struct sk_buff *syn_data = NULL, *data;
2910 2911
	unsigned long last_syn_loss = 0;

2912
	tp->rx_opt.mss_clamp = tp->advmss;  /* If MSS is not cached */
2913 2914 2915 2916 2917 2918 2919 2920
	tcp_fastopen_cache_get(sk, &tp->rx_opt.mss_clamp, &fo->cookie,
			       &syn_loss, &last_syn_loss);
	/* Recurring FO SYN losses: revert to regular handshake temporarily */
	if (syn_loss > 1 &&
	    time_before(jiffies, last_syn_loss + (60*HZ << syn_loss))) {
		fo->cookie.len = -1;
		goto fallback;
	}
2921

2922 2923 2924
	if (sysctl_tcp_fastopen & TFO_CLIENT_NO_COOKIE)
		fo->cookie.len = -1;
	else if (fo->cookie.len <= 0)
2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966
		goto fallback;

	/* MSS for SYN-data is based on cached MSS and bounded by PMTU and
	 * user-MSS. Reserve maximum option space for middleboxes that add
	 * private TCP options. The cost is reduced data space in SYN :(
	 */
	if (tp->rx_opt.user_mss && tp->rx_opt.user_mss < tp->rx_opt.mss_clamp)
		tp->rx_opt.mss_clamp = tp->rx_opt.user_mss;
	space = tcp_mtu_to_mss(sk, inet_csk(sk)->icsk_pmtu_cookie) -
		MAX_TCP_OPTION_SPACE;

	syn_data = skb_copy_expand(syn, skb_headroom(syn), space,
				   sk->sk_allocation);
	if (syn_data == NULL)
		goto fallback;

	for (i = 0; i < iovlen && syn_data->len < space; ++i) {
		struct iovec *iov = &fo->data->msg_iov[i];
		unsigned char __user *from = iov->iov_base;
		int len = iov->iov_len;

		if (syn_data->len + len > space)
			len = space - syn_data->len;
		else if (i + 1 == iovlen)
			/* No more data pending in inet_wait_for_connect() */
			fo->data = NULL;

		if (skb_add_data(syn_data, from, len))
			goto fallback;
	}

	/* Queue a data-only packet after the regular SYN for retransmission */
	data = pskb_copy(syn_data, sk->sk_allocation);
	if (data == NULL)
		goto fallback;
	TCP_SKB_CB(data)->seq++;
	TCP_SKB_CB(data)->tcp_flags &= ~TCPHDR_SYN;
	TCP_SKB_CB(data)->tcp_flags = (TCPHDR_ACK|TCPHDR_PSH);
	tcp_connect_queue_skb(sk, data);
	fo->copied = data->len;

	if (tcp_transmit_skb(sk, syn_data, 0, sk->sk_allocation) == 0) {
2967
		tp->syn_data = (fo->copied > 0);
2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPFASTOPENACTIVE);
		goto done;
	}
	syn_data = NULL;

fallback:
	/* Send a regular SYN with Fast Open cookie request option */
	if (fo->cookie.len > 0)
		fo->cookie.len = 0;
	err = tcp_transmit_skb(sk, syn, 1, sk->sk_allocation);
	if (err)
		tp->syn_fastopen = 0;
	kfree_skb(syn_data);
done:
	fo->cookie.len = -1;  /* Exclude Fast Open option for SYN retries */
	return err;
}

2986
/* Build a SYN and send it off. */
L
Linus Torvalds 已提交
2987 2988 2989 2990
int tcp_connect(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *buff;
2991
	int err;
L
Linus Torvalds 已提交
2992 2993 2994

	tcp_connect_init(sk);

A
Andrey Vagin 已提交
2995 2996 2997 2998 2999
	if (unlikely(tp->repair)) {
		tcp_finish_connect(sk, NULL);
		return 0;
	}

3000
	buff = alloc_skb_fclone(MAX_TCP_HEADER + 15, sk->sk_allocation);
L
Linus Torvalds 已提交
3001 3002 3003 3004 3005 3006
	if (unlikely(buff == NULL))
		return -ENOBUFS;

	/* Reserve space for headers. */
	skb_reserve(buff, MAX_TCP_HEADER);

C
Changli Gao 已提交
3007
	tcp_init_nondata_skb(buff, tp->write_seq++, TCPHDR_SYN);
3008 3009
	tp->retrans_stamp = TCP_SKB_CB(buff)->when = tcp_time_stamp;
	tcp_connect_queue_skb(sk, buff);
3010
	TCP_ECN_send_syn(sk, buff);
L
Linus Torvalds 已提交
3011

3012 3013 3014
	/* Send off SYN; include data in Fast Open. */
	err = tp->fastopen_req ? tcp_send_syn_data(sk, buff) :
	      tcp_transmit_skb(sk, buff, 1, sk->sk_allocation);
3015 3016
	if (err == -ECONNREFUSED)
		return err;
3017 3018 3019 3020 3021 3022

	/* We change tp->snd_nxt after the tcp_transmit_skb() call
	 * in order to make this packet get counted in tcpOutSegs.
	 */
	tp->snd_nxt = tp->write_seq;
	tp->pushed_seq = tp->write_seq;
P
Pavel Emelyanov 已提交
3023
	TCP_INC_STATS(sock_net(sk), TCP_MIB_ACTIVEOPENS);
L
Linus Torvalds 已提交
3024 3025

	/* Timer for repeating the SYN until an answer. */
3026 3027
	inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
				  inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
L
Linus Torvalds 已提交
3028 3029
	return 0;
}
E
Eric Dumazet 已提交
3030
EXPORT_SYMBOL(tcp_connect);
L
Linus Torvalds 已提交
3031 3032 3033 3034 3035 3036 3037

/* Send out a delayed ack, the caller does the policy checking
 * to see if we should even be here.  See tcp_input.c:tcp_ack_snd_check()
 * for details.
 */
void tcp_send_delayed_ack(struct sock *sk)
{
3038 3039
	struct inet_connection_sock *icsk = inet_csk(sk);
	int ato = icsk->icsk_ack.ato;
L
Linus Torvalds 已提交
3040 3041 3042
	unsigned long timeout;

	if (ato > TCP_DELACK_MIN) {
3043
		const struct tcp_sock *tp = tcp_sk(sk);
3044
		int max_ato = HZ / 2;
L
Linus Torvalds 已提交
3045

3046 3047
		if (icsk->icsk_ack.pingpong ||
		    (icsk->icsk_ack.pending & ICSK_ACK_PUSHED))
L
Linus Torvalds 已提交
3048 3049 3050 3051 3052
			max_ato = TCP_DELACK_MAX;

		/* Slow path, intersegment interval is "high". */

		/* If some rtt estimate is known, use it to bound delayed ack.
3053
		 * Do not use inet_csk(sk)->icsk_rto here, use results of rtt measurements
L
Linus Torvalds 已提交
3054 3055 3056
		 * directly.
		 */
		if (tp->srtt) {
3057
			int rtt = max(tp->srtt >> 3, TCP_DELACK_MIN);
L
Linus Torvalds 已提交
3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069

			if (rtt < max_ato)
				max_ato = rtt;
		}

		ato = min(ato, max_ato);
	}

	/* Stay within the limit we were given */
	timeout = jiffies + ato;

	/* Use new timeout only if there wasn't a older one earlier. */
3070
	if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
L
Linus Torvalds 已提交
3071 3072 3073
		/* If delack timer was blocked or is about to expire,
		 * send ACK now.
		 */
3074 3075
		if (icsk->icsk_ack.blocked ||
		    time_before_eq(icsk->icsk_ack.timeout, jiffies + (ato >> 2))) {
L
Linus Torvalds 已提交
3076 3077 3078 3079
			tcp_send_ack(sk);
			return;
		}

3080 3081
		if (!time_before(timeout, icsk->icsk_ack.timeout))
			timeout = icsk->icsk_ack.timeout;
L
Linus Torvalds 已提交
3082
	}
3083 3084 3085
	icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
	icsk->icsk_ack.timeout = timeout;
	sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
L
Linus Torvalds 已提交
3086 3087 3088 3089 3090
}

/* This routine sends an ack and also updates the window. */
void tcp_send_ack(struct sock *sk)
{
3091
	struct sk_buff *buff;
L
Linus Torvalds 已提交
3092

3093 3094 3095
	/* If we have been reset, we may not send again. */
	if (sk->sk_state == TCP_CLOSE)
		return;
L
Linus Torvalds 已提交
3096

3097 3098 3099 3100
	/* We are not putting this on the write queue, so
	 * tcp_transmit_skb() will set the ownership to this
	 * sock.
	 */
3101
	buff = alloc_skb(MAX_TCP_HEADER, sk_gfp_atomic(sk, GFP_ATOMIC));
3102 3103 3104 3105 3106 3107
	if (buff == NULL) {
		inet_csk_schedule_ack(sk);
		inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
		inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
					  TCP_DELACK_MAX, TCP_RTO_MAX);
		return;
L
Linus Torvalds 已提交
3108
	}
3109 3110 3111

	/* Reserve space for headers and prepare control bits. */
	skb_reserve(buff, MAX_TCP_HEADER);
C
Changli Gao 已提交
3112
	tcp_init_nondata_skb(buff, tcp_acceptable_seq(sk), TCPHDR_ACK);
3113 3114 3115

	/* Send it off, this clears delayed acks for us. */
	TCP_SKB_CB(buff)->when = tcp_time_stamp;
3116
	tcp_transmit_skb(sk, buff, 0, sk_gfp_atomic(sk, GFP_ATOMIC));
L
Linus Torvalds 已提交
3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135
}

/* This routine sends a packet with an out of date sequence
 * number. It assumes the other end will try to ack it.
 *
 * Question: what should we make while urgent mode?
 * 4.4BSD forces sending single byte of data. We cannot send
 * out of window data, because we have SND.NXT==SND.MAX...
 *
 * Current solution: to send TWO zero-length segments in urgent mode:
 * one is with SEG.SEQ=SND.UNA to deliver urgent pointer, another is
 * out-of-date with SND.UNA-1 to probe window.
 */
static int tcp_xmit_probe_skb(struct sock *sk, int urgent)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb;

	/* We don't queue it, tcp_transmit_skb() sets ownership. */
3136
	skb = alloc_skb(MAX_TCP_HEADER, sk_gfp_atomic(sk, GFP_ATOMIC));
3137
	if (skb == NULL)
L
Linus Torvalds 已提交
3138 3139 3140 3141 3142 3143 3144 3145
		return -1;

	/* Reserve space for headers and set control bits. */
	skb_reserve(skb, MAX_TCP_HEADER);
	/* Use a previous sequence.  This should cause the other
	 * end to send an ack.  Don't queue or clone SKB, just
	 * send it.
	 */
C
Changli Gao 已提交
3146
	tcp_init_nondata_skb(skb, tp->snd_una - !urgent, TCPHDR_ACK);
L
Linus Torvalds 已提交
3147
	TCP_SKB_CB(skb)->when = tcp_time_stamp;
3148
	return tcp_transmit_skb(sk, skb, 0, GFP_ATOMIC);
L
Linus Torvalds 已提交
3149 3150
}

P
Pavel Emelyanov 已提交
3151 3152 3153 3154
void tcp_send_window_probe(struct sock *sk)
{
	if (sk->sk_state == TCP_ESTABLISHED) {
		tcp_sk(sk)->snd_wl1 = tcp_sk(sk)->rcv_nxt - 1;
P
Pavel Emelyanov 已提交
3155
		tcp_sk(sk)->snd_nxt = tcp_sk(sk)->write_seq;
P
Pavel Emelyanov 已提交
3156 3157 3158 3159
		tcp_xmit_probe_skb(sk, 0);
	}
}

3160
/* Initiate keepalive or window probe from timer. */
L
Linus Torvalds 已提交
3161 3162
int tcp_write_wakeup(struct sock *sk)
{
3163 3164
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb;
L
Linus Torvalds 已提交
3165

3166 3167 3168 3169 3170 3171
	if (sk->sk_state == TCP_CLOSE)
		return -1;

	if ((skb = tcp_send_head(sk)) != NULL &&
	    before(TCP_SKB_CB(skb)->seq, tcp_wnd_end(tp))) {
		int err;
I
Ilpo Järvinen 已提交
3172
		unsigned int mss = tcp_current_mss(sk);
3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184
		unsigned int seg_size = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;

		if (before(tp->pushed_seq, TCP_SKB_CB(skb)->end_seq))
			tp->pushed_seq = TCP_SKB_CB(skb)->end_seq;

		/* We are probing the opening of a window
		 * but the window size is != 0
		 * must have been a result SWS avoidance ( sender )
		 */
		if (seg_size < TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq ||
		    skb->len > mss) {
			seg_size = min(seg_size, mss);
E
Eric Dumazet 已提交
3185
			TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
3186 3187 3188 3189 3190
			if (tcp_fragment(sk, skb, seg_size, mss))
				return -1;
		} else if (!tcp_skb_pcount(skb))
			tcp_set_skb_tso_segs(sk, skb, mss);

E
Eric Dumazet 已提交
3191
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
3192 3193 3194 3195 3196 3197
		TCP_SKB_CB(skb)->when = tcp_time_stamp;
		err = tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
		if (!err)
			tcp_event_new_data_sent(sk, skb);
		return err;
	} else {
I
Ilpo Järvinen 已提交
3198
		if (between(tp->snd_up, tp->snd_una + 1, tp->snd_una + 0xFFFF))
3199 3200
			tcp_xmit_probe_skb(sk, 1);
		return tcp_xmit_probe_skb(sk, 0);
L
Linus Torvalds 已提交
3201 3202 3203 3204 3205 3206 3207 3208
	}
}

/* A window probe timeout has occurred.  If window is not closed send
 * a partial packet else a zero probe.
 */
void tcp_send_probe0(struct sock *sk)
{
3209
	struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
3210 3211 3212 3213 3214
	struct tcp_sock *tp = tcp_sk(sk);
	int err;

	err = tcp_write_wakeup(sk);

3215
	if (tp->packets_out || !tcp_send_head(sk)) {
L
Linus Torvalds 已提交
3216
		/* Cancel probe timer, if it is not required. */
3217
		icsk->icsk_probes_out = 0;
3218
		icsk->icsk_backoff = 0;
L
Linus Torvalds 已提交
3219 3220 3221 3222
		return;
	}

	if (err <= 0) {
3223 3224
		if (icsk->icsk_backoff < sysctl_tcp_retries2)
			icsk->icsk_backoff++;
3225
		icsk->icsk_probes_out++;
3226
		inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
3227 3228
					  min(icsk->icsk_rto << icsk->icsk_backoff, TCP_RTO_MAX),
					  TCP_RTO_MAX);
L
Linus Torvalds 已提交
3229 3230
	} else {
		/* If packet was not sent due to local congestion,
3231
		 * do not backoff and do not remember icsk_probes_out.
L
Linus Torvalds 已提交
3232 3233 3234 3235
		 * Let local senders to fight for local resources.
		 *
		 * Use accumulated backoff yet.
		 */
3236 3237
		if (!icsk->icsk_probes_out)
			icsk->icsk_probes_out = 1;
3238
		inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
3239
					  min(icsk->icsk_rto << icsk->icsk_backoff,
3240 3241
					      TCP_RESOURCE_PROBE_INTERVAL),
					  TCP_RTO_MAX);
L
Linus Torvalds 已提交
3242 3243
	}
}