tcp_output.c 96.9 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
{
N
Nandita Dukkipati 已提交
77
	struct inet_connection_sock *icsk = inet_csk(sk);
78
	struct tcp_sock *tp = tcp_sk(sk);
79
	unsigned int prior_packets = tp->packets_out;
80

81
	tcp_advance_send_head(sk, skb);
L
Linus Torvalds 已提交
82
	tp->snd_nxt = TCP_SKB_CB(skb)->end_seq;
83

L
Lucas De Marchi 已提交
84
	/* Don't override Nagle indefinitely with F-RTO */
85 86
	if (tp->frto_counter == 2)
		tp->frto_counter = 3;
87 88

	tp->packets_out += tcp_skb_pcount(skb);
N
Nandita Dukkipati 已提交
89 90
	if (!prior_packets || icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
	    icsk->icsk_pending == ICSK_TIME_LOSS_PROBE)
91
		tcp_rearm_rto(sk);
L
Linus Torvalds 已提交
92 93 94 95 96 97 98 99
}

/* 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:
 */
100
static inline __u32 tcp_acceptable_seq(const struct sock *sk)
L
Linus Torvalds 已提交
101
{
102
	const struct tcp_sock *tp = tcp_sk(sk);
103

104
	if (!before(tcp_wnd_end(tp), tp->snd_nxt))
L
Linus Torvalds 已提交
105 106
		return tp->snd_nxt;
	else
107
		return tcp_wnd_end(tp);
L
Linus Torvalds 已提交
108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126
}

/* 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);
127
	const struct dst_entry *dst = __sk_dst_get(sk);
L
Linus Torvalds 已提交
128 129
	int mss = tp->advmss;

130 131 132 133 134 135 136
	if (dst) {
		unsigned int metric = dst_metric_advmss(dst);

		if (metric < mss) {
			mss = metric;
			tp->advmss = mss;
		}
L
Linus Torvalds 已提交
137 138 139 140 141 142 143
	}

	return (__u16)mss;
}

/* RFC2861. Reset CWND after idle period longer RTO to "restart window".
 * This is the first part of cwnd validation mechanism. */
144
static void tcp_cwnd_restart(struct sock *sk, const struct dst_entry *dst)
L
Linus Torvalds 已提交
145
{
146
	struct tcp_sock *tp = tcp_sk(sk);
L
Linus Torvalds 已提交
147 148 149 150
	s32 delta = tcp_time_stamp - tp->lsndtime;
	u32 restart_cwnd = tcp_init_cwnd(tp, dst);
	u32 cwnd = tp->snd_cwnd;

151
	tcp_ca_event(sk, CA_EVENT_CWND_RESTART);
L
Linus Torvalds 已提交
152

153
	tp->snd_ssthresh = tcp_current_ssthresh(sk);
L
Linus Torvalds 已提交
154 155
	restart_cwnd = min(restart_cwnd, cwnd);

156
	while ((delta -= inet_csk(sk)->icsk_rto) > 0 && cwnd > restart_cwnd)
L
Linus Torvalds 已提交
157 158 159 160 161 162
		cwnd >>= 1;
	tp->snd_cwnd = max(cwnd, restart_cwnd);
	tp->snd_cwnd_stamp = tcp_time_stamp;
	tp->snd_cwnd_used = 0;
}

163
/* Congestion state accounting after a packet has been sent. */
S
Stephen Hemminger 已提交
164
static void tcp_event_data_sent(struct tcp_sock *tp,
165
				struct sock *sk)
L
Linus Torvalds 已提交
166
{
167 168
	struct inet_connection_sock *icsk = inet_csk(sk);
	const u32 now = tcp_time_stamp;
L
Linus Torvalds 已提交
169

170 171
	if (sysctl_tcp_slow_start_after_idle &&
	    (!tp->packets_out && (s32)(now - tp->lsndtime) > icsk->icsk_rto))
172
		tcp_cwnd_restart(sk, __sk_dst_get(sk));
L
Linus Torvalds 已提交
173 174 175 176 177 178

	tp->lsndtime = now;

	/* If it is a reply for ato after last received
	 * packet, enter pingpong mode.
	 */
179 180
	if ((u32)(now - icsk->icsk_ack.lrcvtime) < icsk->icsk_ack.ato)
		icsk->icsk_ack.pingpong = 1;
L
Linus Torvalds 已提交
181 182
}

183
/* Account for an ACK we sent. */
S
Stephen Hemminger 已提交
184
static inline void tcp_event_ack_sent(struct sock *sk, unsigned int pkts)
L
Linus Torvalds 已提交
185
{
186 187
	tcp_dec_quickack_mode(sk, pkts);
	inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
L
Linus Torvalds 已提交
188 189 190 191 192 193 194 195 196 197 198
}

/* 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,
199 200
			       int wscale_ok, __u8 *rcv_wscale,
			       __u32 init_rcv_wnd)
L
Linus Torvalds 已提交
201 202 203 204 205 206 207 208 209 210 211 212 213
{
	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
214 215 216 217 218 219
	 * 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
220
	 */
221 222 223 224 225
	if (sysctl_tcp_workaround_signed_windows)
		(*rcv_wnd) = min(space, MAX_TCP_WINDOW);
	else
		(*rcv_wnd) = space;

L
Linus Torvalds 已提交
226 227 228
	(*rcv_wscale) = 0;
	if (wscale_ok) {
		/* Set window scaling on max possible window
229
		 * See RFC1323 for an explanation of the limit to 14
L
Linus Torvalds 已提交
230 231
		 */
		space = max_t(u32, sysctl_tcp_rmem[2], sysctl_rmem_max);
232
		space = min_t(u32, space, *window_clamp);
L
Linus Torvalds 已提交
233 234 235 236 237 238
		while (space > 65535 && (*rcv_wscale) < 14) {
			space >>= 1;
			(*rcv_wscale)++;
		}
	}

239 240 241 242
	/* 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.
	 */
243
	if (mss > (1 << *rcv_wscale)) {
244 245 246 247
		int init_cwnd = TCP_DEFAULT_INIT_RCVWND;
		if (mss > 1460)
			init_cwnd =
			max_t(u32, (1460 * TCP_DEFAULT_INIT_RCVWND) / mss, 2);
248 249 250
		/* when initializing use the value from init_rcv_wnd
		 * rather than the default from above
		 */
251 252 253 254
		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
255 256 257 258 259
	}

	/* Set the clamp no higher than max representable value */
	(*window_clamp) = min(65535U << (*rcv_wscale), *window_clamp);
}
E
Eric Dumazet 已提交
260
EXPORT_SYMBOL(tcp_select_initial_window);
L
Linus Torvalds 已提交
261 262 263 264 265 266

/* 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
267
static u16 tcp_select_window(struct sock *sk)
L
Linus Torvalds 已提交
268 269 270 271 272 273
{
	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
274
	if (new_win < cur_win) {
L
Linus Torvalds 已提交
275 276 277 278 279 280 281
		/* 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.
		 */
282
		new_win = ALIGN(cur_win, 1 << tp->rx_opt.rcv_wscale);
L
Linus Torvalds 已提交
283 284 285 286 287 288 289
	}
	tp->rcv_wnd = new_win;
	tp->rcv_wup = tp->rcv_nxt;

	/* Make sure we do not exceed the maximum possible
	 * scaled window.
	 */
290
	if (!tp->rx_opt.rcv_wscale && sysctl_tcp_workaround_signed_windows)
L
Linus Torvalds 已提交
291 292 293 294 295 296 297 298 299 300 301 302 303 304
		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;
}

305
/* Packet ECN state for a SYN-ACK */
306
static inline void TCP_ECN_send_synack(const struct tcp_sock *tp, struct sk_buff *skb)
307
{
E
Eric Dumazet 已提交
308
	TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_CWR;
309
	if (!(tp->ecn_flags & TCP_ECN_OK))
E
Eric Dumazet 已提交
310
		TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_ECE;
311 312
}

313
/* Packet ECN state for a SYN.  */
314 315 316 317 318
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;
319
	if (sock_net(sk)->ipv4.sysctl_tcp_ecn == 1) {
E
Eric Dumazet 已提交
320
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ECE | TCPHDR_CWR;
321 322 323 324 325
		tp->ecn_flags = TCP_ECN_OK;
	}
}

static __inline__ void
326
TCP_ECN_make_synack(const struct request_sock *req, struct tcphdr *th)
327 328 329 330 331
{
	if (inet_rsk(req)->ecn_ok)
		th->ece = 1;
}

332 333 334
/* Set up ECN state for a packet on a ESTABLISHED socket that is about to
 * be sent.
 */
335 336 337 338 339 340 341 342 343 344
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);
345
			if (tp->ecn_flags & TCP_ECN_QUEUE_CWR) {
346 347 348 349 350 351 352 353 354 355 356 357 358
				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;
	}
}

359 360 361 362 363
/* 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)
{
364
	skb->ip_summed = CHECKSUM_PARTIAL;
365 366
	skb->csum = 0;

E
Eric Dumazet 已提交
367
	TCP_SKB_CB(skb)->tcp_flags = flags;
368 369 370 371 372 373 374
	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
375
	if (flags & (TCPHDR_SYN | TCPHDR_FIN))
376 377 378 379
		seq++;
	TCP_SKB_CB(skb)->end_seq = seq;
}

E
Eric Dumazet 已提交
380
static inline bool tcp_urg_mode(const struct tcp_sock *tp)
I
Ilpo Järvinen 已提交
381 382 383 384
{
	return tp->snd_una != tp->snd_up;
}

A
Adam Langley 已提交
385 386 387
#define OPTION_SACK_ADVERTISE	(1 << 0)
#define OPTION_TS		(1 << 1)
#define OPTION_MD5		(1 << 2)
388
#define OPTION_WSCALE		(1 << 3)
389
#define OPTION_COOKIE_EXTENSION	(1 << 4)
Y
Yuchung Cheng 已提交
390
#define OPTION_FAST_OPEN_COOKIE	(1 << 8)
A
Adam Langley 已提交
391 392

struct tcp_out_options {
Y
Yuchung Cheng 已提交
393 394
	u16 options;		/* bit field of OPTION_* */
	u16 mss;		/* 0 to disable */
A
Adam Langley 已提交
395 396
	u8 ws;			/* window scale, 0 to disable */
	u8 num_sack_blocks;	/* number of SACK blocks to include */
397 398
	u8 hash_size;		/* bytes in hash_location */
	__u8 *hash_location;	/* temporary pointer, overloaded */
Y
Yuchung Cheng 已提交
399 400
	__u32 tsval, tsecr;	/* need to include OPTION_TS */
	struct tcp_fastopen_cookie *fastopen_cookie;	/* Fast open cookie */
A
Adam Langley 已提交
401 402
};

403 404 405 406
/* The sysctl int routines are generic, so check consistency here.
 */
static u8 tcp_cookie_size_check(u8 desired)
{
407 408 409
	int cookie_size;

	if (desired > 0)
410 411
		/* previously specified */
		return desired;
412 413 414

	cookie_size = ACCESS_ONCE(sysctl_tcp_cookie_size);
	if (cookie_size <= 0)
415 416
		/* no default specified */
		return 0;
417 418

	if (cookie_size <= TCP_COOKIE_MIN)
419 420
		/* value too small, specify minimum */
		return TCP_COOKIE_MIN;
421 422

	if (cookie_size >= TCP_COOKIE_MAX)
423 424
		/* value too large, specify maximum */
		return TCP_COOKIE_MAX;
425 426

	if (cookie_size & 1)
427
		/* 8-bit multiple, illegal, fix it */
428 429 430
		cookie_size++;

	return (u8)cookie_size;
431 432
}

433 434 435
/* Write previously computed TCP options to the packet.
 *
 * Beware: Something in the Internet is very sensitive to the ordering of
436 437 438 439 440 441 442 443 444 445
 * 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
446
static void tcp_options_write(__be32 *ptr, struct tcp_sock *tp,
447 448
			      struct tcp_out_options *opts)
{
Y
Yuchung Cheng 已提交
449
	u16 options = opts->options;	/* mungable copy */
450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473

	/* 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
474
		ptr += 4;
S
Stephen Hemminger 已提交
475
	}
A
Adam Langley 已提交
476

477 478 479 480 481 482
	if (unlikely(opts->mss)) {
		*ptr++ = htonl((TCPOPT_MSS << 24) |
			       (TCPOLEN_MSS << 16) |
			       opts->mss);
	}

483 484
	if (likely(OPTION_TS & options)) {
		if (unlikely(OPTION_SACK_ADVERTISE & options)) {
A
Adam Langley 已提交
485 486 487 488
			*ptr++ = htonl((TCPOPT_SACK_PERM << 24) |
				       (TCPOLEN_SACK_PERM << 16) |
				       (TCPOPT_TIMESTAMP << 8) |
				       TCPOLEN_TIMESTAMP);
489
			options &= ~OPTION_SACK_ADVERTISE;
A
Adam Langley 已提交
490 491 492 493 494 495 496 497 498 499
		} else {
			*ptr++ = htonl((TCPOPT_NOP << 24) |
				       (TCPOPT_NOP << 16) |
				       (TCPOPT_TIMESTAMP << 8) |
				       TCPOLEN_TIMESTAMP);
		}
		*ptr++ = htonl(opts->tsval);
		*ptr++ = htonl(opts->tsecr);
	}

500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538
	/* 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
539 540 541 542 543 544
		*ptr++ = htonl((TCPOPT_NOP << 24) |
			       (TCPOPT_NOP << 16) |
			       (TCPOPT_SACK_PERM << 8) |
			       TCPOLEN_SACK_PERM);
	}

545
	if (unlikely(OPTION_WSCALE & options)) {
A
Adam Langley 已提交
546 547 548 549 550 551 552 553 554
		*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
555 556 557 558 559
		int this_sack;

		*ptr++ = htonl((TCPOPT_NOP  << 24) |
			       (TCPOPT_NOP  << 16) |
			       (TCPOPT_SACK <<  8) |
A
Adam Langley 已提交
560
			       (TCPOLEN_SACK_BASE + (opts->num_sack_blocks *
S
Stephen Hemminger 已提交
561
						     TCPOLEN_SACK_PERBLOCK)));
S
Stephen Hemminger 已提交
562

A
Adam Langley 已提交
563 564
		for (this_sack = 0; this_sack < opts->num_sack_blocks;
		     ++this_sack) {
S
Stephen Hemminger 已提交
565 566 567
			*ptr++ = htonl(sp[this_sack].start_seq);
			*ptr++ = htonl(sp[this_sack].end_seq);
		}
S
Stephen Hemminger 已提交
568

569
		tp->rx_opt.dsack = 0;
S
Stephen Hemminger 已提交
570
	}
Y
Yuchung Cheng 已提交
571 572 573 574 575 576 577 578 579 580 581 582 583 584 585

	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
586 587
}

588 589 590
/* Compute TCP options for SYN packets. This is not the final
 * network wire format yet.
 */
591
static unsigned int tcp_syn_options(struct sock *sk, struct sk_buff *skb,
A
Adam Langley 已提交
592
				struct tcp_out_options *opts,
593 594
				struct tcp_md5sig_key **md5)
{
A
Adam Langley 已提交
595
	struct tcp_sock *tp = tcp_sk(sk);
596
	struct tcp_cookie_values *cvp = tp->cookie_values;
597
	unsigned int remaining = MAX_TCP_OPTION_SPACE;
598 599 600
	u8 cookie_size = (!tp->rx_opt.cookie_out_never && cvp != NULL) ?
			 tcp_cookie_size_check(cvp->cookie_desired) :
			 0;
601
	struct tcp_fastopen_request *fastopen = tp->fastopen_req;
A
Adam Langley 已提交
602

603
#ifdef CONFIG_TCP_MD5SIG
A
Adam Langley 已提交
604 605 606
	*md5 = tp->af_specific->md5_lookup(sk, sk);
	if (*md5) {
		opts->options |= OPTION_MD5;
607
		remaining -= TCPOLEN_MD5SIG_ALIGNED;
608
	}
A
Adam Langley 已提交
609 610
#else
	*md5 = NULL;
611
#endif
A
Adam Langley 已提交
612 613 614 615 616 617 618 619 620 621 622

	/* 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);
623
	remaining -= TCPOLEN_MSS_ALIGNED;
A
Adam Langley 已提交
624

625
	if (likely(sysctl_tcp_timestamps && *md5 == NULL)) {
A
Adam Langley 已提交
626
		opts->options |= OPTION_TS;
627
		opts->tsval = TCP_SKB_CB(skb)->when + tp->tsoffset;
A
Adam Langley 已提交
628
		opts->tsecr = tp->rx_opt.ts_recent;
629
		remaining -= TCPOLEN_TSTAMP_ALIGNED;
A
Adam Langley 已提交
630
	}
631
	if (likely(sysctl_tcp_window_scaling)) {
A
Adam Langley 已提交
632
		opts->ws = tp->rx_opt.rcv_wscale;
633
		opts->options |= OPTION_WSCALE;
634
		remaining -= TCPOLEN_WSCALE_ALIGNED;
A
Adam Langley 已提交
635
	}
636
	if (likely(sysctl_tcp_sack)) {
A
Adam Langley 已提交
637
		opts->options |= OPTION_SACK_ADVERTISE;
638
		if (unlikely(!(OPTION_TS & opts->options)))
639
			remaining -= TCPOLEN_SACKPERM_ALIGNED;
A
Adam Langley 已提交
640 641
	}

642 643 644 645 646 647 648 649 650 651
	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;
		}
	}
652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698
	/* 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
699 700
}

701
/* Set up TCP options for SYN-ACKs. */
702
static unsigned int tcp_synack_options(struct sock *sk,
A
Adam Langley 已提交
703
				   struct request_sock *req,
704
				   unsigned int mss, struct sk_buff *skb,
A
Adam Langley 已提交
705
				   struct tcp_out_options *opts,
706
				   struct tcp_md5sig_key **md5,
707 708
				   struct tcp_extend_values *xvp,
				   struct tcp_fastopen_cookie *foc)
709
{
A
Adam Langley 已提交
710
	struct inet_request_sock *ireq = inet_rsk(req);
711
	unsigned int remaining = MAX_TCP_OPTION_SPACE;
712 713 714
	u8 cookie_plus = (xvp != NULL && !xvp->cookie_out_never) ?
			 xvp->cookie_plus :
			 0;
A
Adam Langley 已提交
715

716
#ifdef CONFIG_TCP_MD5SIG
A
Adam Langley 已提交
717 718 719
	*md5 = tcp_rsk(req)->af_specific->md5_lookup(sk, req);
	if (*md5) {
		opts->options |= OPTION_MD5;
720 721 722 723 724 725 726
		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
727
		ireq->tstamp_ok &= !ireq->sack_ok;
728
	}
A
Adam Langley 已提交
729 730
#else
	*md5 = NULL;
731
#endif
A
Adam Langley 已提交
732

733
	/* We always send an MSS option. */
A
Adam Langley 已提交
734
	opts->mss = mss;
735
	remaining -= TCPOLEN_MSS_ALIGNED;
A
Adam Langley 已提交
736 737 738

	if (likely(ireq->wscale_ok)) {
		opts->ws = ireq->rcv_wscale;
739
		opts->options |= OPTION_WSCALE;
740
		remaining -= TCPOLEN_WSCALE_ALIGNED;
A
Adam Langley 已提交
741
	}
E
Eric Dumazet 已提交
742
	if (likely(ireq->tstamp_ok)) {
A
Adam Langley 已提交
743 744 745
		opts->options |= OPTION_TS;
		opts->tsval = TCP_SKB_CB(skb)->when;
		opts->tsecr = req->ts_recent;
746
		remaining -= TCPOLEN_TSTAMP_ALIGNED;
A
Adam Langley 已提交
747 748 749
	}
	if (likely(ireq->sack_ok)) {
		opts->options |= OPTION_SACK_ADVERTISE;
E
Eric Dumazet 已提交
750
		if (unlikely(!ireq->tstamp_ok))
751
			remaining -= TCPOLEN_SACKPERM_ALIGNED;
A
Adam Langley 已提交
752
	}
753 754 755 756 757 758 759 760 761
	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;
		}
	}
762 763 764 765
	/* 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
766
	    ireq->tstamp_ok &&
767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784
	    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
785 786
}

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

#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;
811
		opts->tsval = tcb ? tcb->when + tp->tsoffset : 0;
A
Adam Langley 已提交
812 813 814 815
		opts->tsecr = tp->rx_opt.ts_recent;
		size += TCPOLEN_TSTAMP_ALIGNED;
	}

816 817
	eff_sacks = tp->rx_opt.num_sacks + tp->rx_opt.dsack;
	if (unlikely(eff_sacks)) {
818
		const unsigned int remaining = MAX_TCP_OPTION_SPACE - size;
A
Adam Langley 已提交
819
		opts->num_sack_blocks =
820
			min_t(unsigned int, eff_sacks,
A
Adam Langley 已提交
821 822 823 824 825 826 827
			      (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
828
}
L
Linus Torvalds 已提交
829

E
Eric Dumazet 已提交
830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850

/* 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);

851 852 853 854 855 856 857
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
858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884
/*
 * 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)) {
885
			tcp_tsq_handler(sk);
E
Eric Dumazet 已提交
886 887
		} else {
			/* defer the work to tcp_release_cb() */
888
			set_bit(TCP_TSQ_DEFERRED, &tp->tsq_flags);
E
Eric Dumazet 已提交
889 890 891 892 893 894 895 896
		}
		bh_unlock_sock(sk);

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

897 898
#define TCP_DEFERRED_ALL ((1UL << TCP_TSQ_DEFERRED) |		\
			  (1UL << TCP_WRITE_TIMER_DEFERRED) |	\
899 900
			  (1UL << TCP_DELACK_TIMER_DEFERRED) |	\
			  (1UL << TCP_MTU_REDUCED_DEFERRED))
E
Eric Dumazet 已提交
901 902 903 904 905 906 907 908 909 910
/**
 * 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);
911
	unsigned long flags, nflags;
E
Eric Dumazet 已提交
912

913 914 915 916 917 918 919 920 921 922 923
	/* 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);

924
	if (flags & (1UL << TCP_WRITE_TIMER_DEFERRED)) {
925
		tcp_write_timer_handler(sk);
926 927 928
		__sock_put(sk);
	}
	if (flags & (1UL << TCP_DELACK_TIMER_DEFERRED)) {
929
		tcp_delack_timer_handler(sk);
930 931 932
		__sock_put(sk);
	}
	if (flags & (1UL << TCP_MTU_REDUCED_DEFERRED)) {
933
		sk->sk_prot->mtu_reduced(sk);
934 935
		__sock_put(sk);
	}
E
Eric Dumazet 已提交
936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957
}
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.
 */
958
static void tcp_wfree(struct sk_buff *skb)
E
Eric Dumazet 已提交
959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983
{
	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
984 985 986 987 988 989 990 991 992 993 994
/* 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.
 */
995 996
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
997
{
998 999 1000 1001
	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
1002
	struct tcp_out_options opts;
1003
	unsigned int tcp_options_size, tcp_header_size;
1004
	struct tcp_md5sig_key *md5;
1005 1006 1007 1008 1009 1010 1011 1012
	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.
	 */
1013
	if (icsk->icsk_ca_ops->flags & TCP_CONG_RTT_STAMP)
1014 1015 1016 1017 1018 1019 1020 1021 1022 1023
		__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
1024

1025 1026 1027
	inet = inet_sk(sk);
	tp = tcp_sk(sk);
	tcb = TCP_SKB_CB(skb);
A
Adam Langley 已提交
1028
	memset(&opts, 0, sizeof(opts));
L
Linus Torvalds 已提交
1029

E
Eric Dumazet 已提交
1030
	if (unlikely(tcb->tcp_flags & TCPHDR_SYN))
A
Adam Langley 已提交
1031 1032 1033 1034 1035
		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);
1036

1037
	if (tcp_packets_in_flight(tp) == 0) {
1038
		tcp_ca_event(sk, CA_EVENT_TX_START);
1039 1040 1041
		skb->ooo_okay = 1;
	} else
		skb->ooo_okay = 0;
1042

1043 1044
	skb_push(skb, tcp_header_size);
	skb_reset_transport_header(skb);
E
Eric Dumazet 已提交
1045 1046 1047 1048 1049 1050

	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);
1051 1052

	/* Build TCP header and checksum it. */
1053
	th = tcp_hdr(skb);
E
Eric Dumazet 已提交
1054 1055
	th->source		= inet->inet_sport;
	th->dest		= inet->inet_dport;
1056 1057
	th->seq			= htonl(tcb->seq);
	th->ack_seq		= htonl(tp->rcv_nxt);
1058
	*(((__be16 *)th) + 6)	= htons(((tcp_header_size >> 2) << 12) |
E
Eric Dumazet 已提交
1059
					tcb->tcp_flags);
1060

E
Eric Dumazet 已提交
1061
	if (unlikely(tcb->tcp_flags & TCPHDR_SYN)) {
1062 1063 1064
		/* RFC1323: The window in SYN & SYN/ACK segments
		 * is never scaled.
		 */
1065
		th->window	= htons(min(tp->rcv_wnd, 65535U));
1066 1067 1068 1069 1070
	} else {
		th->window	= htons(tcp_select_window(sk));
	}
	th->check		= 0;
	th->urg_ptr		= 0;
L
Linus Torvalds 已提交
1071

I
Ilpo Järvinen 已提交
1072
	/* The urg_mode check is necessary during a below snd_una win probe */
1073 1074 1075 1076 1077
	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)) {
1078
			th->urg_ptr = htons(0xFFFF);
1079 1080
			th->urg = 1;
		}
1081
	}
L
Linus Torvalds 已提交
1082

1083
	tcp_options_write((__be32 *)(th + 1), tp, &opts);
E
Eric Dumazet 已提交
1084
	if (likely((tcb->tcp_flags & TCPHDR_SYN) == 0))
1085
		TCP_ECN_send(sk, skb, tcp_header_size);
L
Linus Torvalds 已提交
1086

1087 1088 1089
#ifdef CONFIG_TCP_MD5SIG
	/* Calculate the MD5 hash, as we have all we need now */
	if (md5) {
E
Eric Dumazet 已提交
1090
		sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1091
		tp->af_specific->calc_md5_hash(opts.hash_location,
1092
					       md5, sk, NULL, skb);
1093 1094 1095
	}
#endif

1096
	icsk->icsk_af_ops->send_check(sk, skb);
L
Linus Torvalds 已提交
1097

E
Eric Dumazet 已提交
1098
	if (likely(tcb->tcp_flags & TCPHDR_ACK))
1099
		tcp_event_ack_sent(sk, tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
1100

1101
	if (skb->len != tcp_header_size)
1102
		tcp_event_data_sent(tp, sk);
L
Linus Torvalds 已提交
1103

1104
	if (after(tcb->end_seq, tp->snd_nxt) || tcb->seq == tcb->end_seq)
1105 1106
		TCP_ADD_STATS(sock_net(sk), TCP_MIB_OUTSEGS,
			      tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
1107

1108
	err = icsk->icsk_af_ops->queue_xmit(skb, &inet->cork.fl);
1109
	if (likely(err <= 0))
1110 1111
		return err;

1112
	tcp_enter_cwr(sk, 1);
1113

1114
	return net_xmit_eval(err);
L
Linus Torvalds 已提交
1115 1116
}

1117
/* This routine just queues the buffer for sending.
L
Linus Torvalds 已提交
1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128
 *
 * 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);
1129
	tcp_add_write_queue_tail(sk, skb);
1130 1131
	sk->sk_wmem_queued += skb->truesize;
	sk_mem_charge(sk, skb->truesize);
L
Linus Torvalds 已提交
1132 1133
}

1134
/* Initialize TSO segments for a packet. */
1135
static void tcp_set_skb_tso_segs(const struct sock *sk, struct sk_buff *skb,
1136
				 unsigned int mss_now)
1137
{
1138 1139
	if (skb->len <= mss_now || !sk_can_gso(sk) ||
	    skb->ip_summed == CHECKSUM_NONE) {
1140 1141 1142
		/* Avoid the costly divide in the normal
		 * non-TSO case.
		 */
1143 1144
		skb_shinfo(skb)->gso_segs = 1;
		skb_shinfo(skb)->gso_size = 0;
1145
		skb_shinfo(skb)->gso_type = 0;
1146
	} else {
I
Ilpo Järvinen 已提交
1147
		skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss_now);
1148
		skb_shinfo(skb)->gso_size = mss_now;
1149
		skb_shinfo(skb)->gso_type = sk->sk_gso_type;
L
Linus Torvalds 已提交
1150 1151 1152
	}
}

I
Ilpo Järvinen 已提交
1153
/* When a modification to fackets out becomes necessary, we need to check
1154
 * skb is counted to fackets_out or not.
I
Ilpo Järvinen 已提交
1155
 */
1156
static void tcp_adjust_fackets_out(struct sock *sk, const struct sk_buff *skb,
I
Ilpo Järvinen 已提交
1157 1158
				   int decr)
{
1159 1160
	struct tcp_sock *tp = tcp_sk(sk);

1161
	if (!tp->sacked_out || tcp_is_reno(tp))
I
Ilpo Järvinen 已提交
1162 1163
		return;

1164
	if (after(tcp_highest_sack_seq(tp), TCP_SKB_CB(skb)->seq))
I
Ilpo Järvinen 已提交
1165 1166 1167
		tp->fackets_out -= decr;
}

1168 1169 1170
/* Pcount in the middle of the write queue got changed, we need to do various
 * tweaks to fix counters
 */
1171
static void tcp_adjust_pcount(struct sock *sk, const struct sk_buff *skb, int decr)
1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191
{
	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) &&
1192
	    (tcp_is_fack(tp) || (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)))
1193 1194 1195 1196 1197
		tp->lost_cnt_hint -= decr;

	tcp_verify_left_out(tp);
}

L
Linus Torvalds 已提交
1198 1199
/* 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
1200
 * packet to the list.  This won't be called frequently, I hope.
L
Linus Torvalds 已提交
1201 1202
 * Remember, these are still headerless SKBs at this point.
 */
1203 1204
int tcp_fragment(struct sock *sk, struct sk_buff *skb, u32 len,
		 unsigned int mss_now)
L
Linus Torvalds 已提交
1205 1206 1207
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *buff;
1208
	int nsize, old_factor;
1209
	int nlen;
1210
	u8 flags;
L
Linus Torvalds 已提交
1211

1212 1213
	if (WARN_ON(len > skb->len))
		return -EINVAL;
1214

L
Linus Torvalds 已提交
1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227
	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
1228

1229 1230
	sk->sk_wmem_queued += buff->truesize;
	sk_mem_charge(sk, buff->truesize);
1231 1232 1233
	nlen = skb->len - len - nsize;
	buff->truesize += nlen;
	skb->truesize -= nlen;
L
Linus Torvalds 已提交
1234 1235 1236 1237 1238 1239 1240

	/* 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
1241 1242 1243
	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;
1244
	TCP_SKB_CB(buff)->sacked = TCP_SKB_CB(skb)->sacked;
L
Linus Torvalds 已提交
1245

1246
	if (!skb_shinfo(skb)->nr_frags && skb->ip_summed != CHECKSUM_PARTIAL) {
L
Linus Torvalds 已提交
1247
		/* Copy and checksum data tail into the new buffer. */
1248 1249
		buff->csum = csum_partial_copy_nocheck(skb->data + len,
						       skb_put(buff, nsize),
L
Linus Torvalds 已提交
1250 1251 1252 1253 1254 1255
						       nsize, 0);

		skb_trim(skb, len);

		skb->csum = csum_block_sub(skb->csum, buff->csum, len);
	} else {
1256
		skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
1257 1258 1259 1260 1261 1262 1263 1264 1265
		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;
1266
	buff->tstamp = skb->tstamp;
L
Linus Torvalds 已提交
1267

1268 1269
	old_factor = tcp_skb_pcount(skb);

L
Linus Torvalds 已提交
1270
	/* Fix up tso_factor for both original and new SKB.  */
1271 1272
	tcp_set_skb_tso_segs(sk, skb, mss_now);
	tcp_set_skb_tso_segs(sk, buff, mss_now);
L
Linus Torvalds 已提交
1273

1274 1275 1276
	/* If this packet has been sent out already, we must
	 * adjust the various packet counters.
	 */
1277
	if (!before(tp->snd_nxt, TCP_SKB_CB(buff)->end_seq)) {
1278 1279
		int diff = old_factor - tcp_skb_pcount(skb) -
			tcp_skb_pcount(buff);
L
Linus Torvalds 已提交
1280

1281 1282
		if (diff)
			tcp_adjust_pcount(sk, skb, diff);
L
Linus Torvalds 已提交
1283 1284 1285
	}

	/* Link BUFF into the send queue. */
1286
	skb_header_release(buff);
1287
	tcp_insert_write_queue_after(skb, buff, sk);
L
Linus Torvalds 已提交
1288 1289 1290 1291 1292 1293 1294 1295

	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.
 */
1296
static void __pskb_trim_head(struct sk_buff *skb, int len)
L
Linus Torvalds 已提交
1297 1298 1299
{
	int i, k, eat;

E
Eric Dumazet 已提交
1300 1301 1302
	eat = min_t(int, len, skb_headlen(skb));
	if (eat) {
		__skb_pull(skb, eat);
1303
		skb->avail_size -= eat;
E
Eric Dumazet 已提交
1304 1305 1306 1307
		len -= eat;
		if (!len)
			return;
	}
L
Linus Torvalds 已提交
1308 1309
	eat = len;
	k = 0;
1310
	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
E
Eric Dumazet 已提交
1311 1312 1313
		int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);

		if (size <= eat) {
1314
			skb_frag_unref(skb, i);
E
Eric Dumazet 已提交
1315
			eat -= size;
L
Linus Torvalds 已提交
1316 1317 1318 1319
		} 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
1320
				skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
L
Linus Torvalds 已提交
1321 1322 1323 1324 1325 1326 1327
				eat = 0;
			}
			k++;
		}
	}
	skb_shinfo(skb)->nr_frags = k;

1328
	skb_reset_tail_pointer(skb);
L
Linus Torvalds 已提交
1329 1330 1331 1332
	skb->data_len -= len;
	skb->len = skb->data_len;
}

1333
/* Remove acked data from a packet in the transmit queue. */
L
Linus Torvalds 已提交
1334 1335
int tcp_trim_head(struct sock *sk, struct sk_buff *skb, u32 len)
{
1336
	if (skb_unclone(skb, GFP_ATOMIC))
L
Linus Torvalds 已提交
1337 1338
		return -ENOMEM;

E
Eric Dumazet 已提交
1339
	__pskb_trim_head(skb, len);
L
Linus Torvalds 已提交
1340 1341

	TCP_SKB_CB(skb)->seq += len;
1342
	skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
1343 1344 1345

	skb->truesize	     -= len;
	sk->sk_wmem_queued   -= len;
1346
	sk_mem_uncharge(sk, len);
L
Linus Torvalds 已提交
1347 1348
	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);

1349
	/* Any change of skb->len requires recalculation of tso factor. */
L
Linus Torvalds 已提交
1350
	if (tcp_skb_pcount(skb) > 1)
1351
		tcp_set_skb_tso_segs(sk, skb, tcp_skb_mss(skb));
L
Linus Torvalds 已提交
1352 1353 1354 1355

	return 0;
}

1356 1357
/* Calculate MSS not accounting any TCP options.  */
static inline int __tcp_mtu_to_mss(struct sock *sk, int pmtu)
J
John Heffner 已提交
1358
{
1359 1360
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1361 1362 1363 1364 1365 1366 1367
	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);

1368 1369 1370 1371 1372 1373 1374 1375
	/* 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
1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388
	/* 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;
	return mss_now;
}

1389 1390 1391 1392 1393 1394 1395 1396
/* Calculate MSS. Not accounting for SACKs here.  */
int tcp_mtu_to_mss(struct sock *sk, int pmtu)
{
	/* Subtract TCP options size, not including SACKs */
	return __tcp_mtu_to_mss(sk, pmtu) -
	       (tcp_sk(sk)->tcp_header_len - sizeof(struct tcphdr));
}

J
John Heffner 已提交
1397
/* Inverse of above */
1398
int tcp_mss_to_mtu(struct sock *sk, int mss)
J
John Heffner 已提交
1399
{
1400 1401
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1402 1403 1404 1405 1406 1407 1408
	int mtu;

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

1409 1410 1411 1412 1413 1414 1415
	/* 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
1416 1417 1418
	return mtu;
}

1419
/* MTU probing init per socket */
J
John Heffner 已提交
1420 1421 1422 1423 1424 1425 1426
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) +
1427
			       icsk->icsk_af_ops->net_header_len;
J
John Heffner 已提交
1428 1429 1430
	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
1431
EXPORT_SYMBOL(tcp_mtup_init);
J
John Heffner 已提交
1432

L
Linus Torvalds 已提交
1433 1434 1435 1436 1437 1438
/* 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
1439
   It is minimum of user_mss and mss received with SYN.
L
Linus Torvalds 已提交
1440 1441
   It also does not include TCP options.

1442
   inet_csk(sk)->icsk_pmtu_cookie is last pmtu, seen by this function.
L
Linus Torvalds 已提交
1443 1444 1445 1446 1447 1448 1449 1450 1451

   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.

1452 1453
   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
1454 1455 1456 1457
 */
unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu)
{
	struct tcp_sock *tp = tcp_sk(sk);
1458
	struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1459
	int mss_now;
L
Linus Torvalds 已提交
1460

J
John Heffner 已提交
1461 1462
	if (icsk->icsk_mtup.search_high > pmtu)
		icsk->icsk_mtup.search_high = pmtu;
L
Linus Torvalds 已提交
1463

J
John Heffner 已提交
1464
	mss_now = tcp_mtu_to_mss(sk, pmtu);
1465
	mss_now = tcp_bound_to_half_wnd(tp, mss_now);
L
Linus Torvalds 已提交
1466 1467

	/* And store cached results */
1468
	icsk->icsk_pmtu_cookie = pmtu;
J
John Heffner 已提交
1469 1470
	if (icsk->icsk_mtup.enabled)
		mss_now = min(mss_now, tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low));
1471
	tp->mss_cache = mss_now;
L
Linus Torvalds 已提交
1472 1473 1474

	return mss_now;
}
E
Eric Dumazet 已提交
1475
EXPORT_SYMBOL(tcp_sync_mss);
L
Linus Torvalds 已提交
1476 1477 1478 1479

/* Compute the current effective MSS, taking SACKs and IP options,
 * and even PMTU discovery events into account.
 */
I
Ilpo Järvinen 已提交
1480
unsigned int tcp_current_mss(struct sock *sk)
L
Linus Torvalds 已提交
1481
{
1482 1483
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct dst_entry *dst = __sk_dst_get(sk);
1484
	u32 mss_now;
1485
	unsigned int header_len;
A
Adam Langley 已提交
1486 1487
	struct tcp_out_options opts;
	struct tcp_md5sig_key *md5;
1488 1489 1490

	mss_now = tp->mss_cache;

L
Linus Torvalds 已提交
1491 1492
	if (dst) {
		u32 mtu = dst_mtu(dst);
1493
		if (mtu != inet_csk(sk)->icsk_pmtu_cookie)
L
Linus Torvalds 已提交
1494 1495 1496
			mss_now = tcp_sync_mss(sk, mtu);
	}

A
Adam Langley 已提交
1497 1498 1499 1500 1501 1502 1503 1504 1505 1506
	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;
	}
1507

L
Linus Torvalds 已提交
1508 1509 1510
	return mss_now;
}

1511
/* Congestion window validation. (RFC2861) */
1512
static void tcp_cwnd_validate(struct sock *sk)
1513
{
1514
	struct tcp_sock *tp = tcp_sk(sk);
1515

1516
	if (tp->packets_out >= tp->snd_cwnd) {
1517 1518 1519 1520 1521 1522 1523 1524
		/* 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;

1525 1526
		if (sysctl_tcp_slow_start_after_idle &&
		    (s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= inet_csk(sk)->icsk_rto)
1527 1528 1529 1530
			tcp_cwnd_application_limited(sk);
	}
}

1531 1532 1533 1534
/* 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).
1535 1536 1537 1538 1539 1540 1541
 *
 * 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.
1542
 */
1543
static unsigned int tcp_mss_split_point(const struct sock *sk, const struct sk_buff *skb,
1544
					unsigned int mss_now, unsigned int max_segs)
1545
{
1546
	const struct tcp_sock *tp = tcp_sk(sk);
1547
	u32 needed, window, max_len;
1548

1549
	window = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
1550
	max_len = mss_now * max_segs;
1551

1552 1553
	if (likely(max_len <= window && skb != tcp_write_queue_tail(sk)))
		return max_len;
1554

1555 1556
	needed = min(skb->len, window);

1557 1558
	if (max_len <= needed)
		return max_len;
1559 1560

	return needed - needed % mss_now;
1561 1562 1563 1564 1565
}

/* 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.
 */
1566 1567
static inline unsigned int tcp_cwnd_test(const struct tcp_sock *tp,
					 const struct sk_buff *skb)
1568 1569 1570 1571
{
	u32 in_flight, cwnd;

	/* Don't be strict about the congestion window for the final FIN.  */
E
Eric Dumazet 已提交
1572 1573
	if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) &&
	    tcp_skb_pcount(skb) == 1)
1574 1575 1576 1577 1578 1579 1580 1581 1582 1583
		return 1;

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

	return 0;
}

1584
/* Initialize TSO state of a skb.
1585
 * This must be invoked the first time we consider transmitting
1586 1587
 * SKB onto the wire.
 */
1588
static int tcp_init_tso_segs(const struct sock *sk, struct sk_buff *skb,
1589
			     unsigned int mss_now)
1590 1591 1592
{
	int tso_segs = tcp_skb_pcount(skb);

1593
	if (!tso_segs || (tso_segs > 1 && tcp_skb_mss(skb) != mss_now)) {
1594
		tcp_set_skb_tso_segs(sk, skb, mss_now);
1595 1596 1597 1598 1599
		tso_segs = tcp_skb_pcount(skb);
	}
	return tso_segs;
}

1600
/* Minshall's variant of the Nagle send check. */
E
Eric Dumazet 已提交
1601
static inline bool tcp_minshall_check(const struct tcp_sock *tp)
1602
{
1603
	return after(tp->snd_sml, tp->snd_una) &&
1604 1605 1606
		!after(tp->snd_sml, tp->snd_nxt);
}

E
Eric Dumazet 已提交
1607
/* Return false, if packet can be sent now without violation Nagle's rules:
1608 1609
 * 1. It is full sized.
 * 2. Or it contains FIN. (already checked by caller)
F
Feng King 已提交
1610
 * 3. Or TCP_CORK is not set, and TCP_NODELAY is set.
1611 1612 1613
 * 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
1614
static inline bool tcp_nagle_check(const struct tcp_sock *tp,
1615
				  const struct sk_buff *skb,
1616
				  unsigned int mss_now, int nonagle)
1617
{
E
Eric Dumazet 已提交
1618
	return skb->len < mss_now &&
1619
		((nonagle & TCP_NAGLE_CORK) ||
E
Eric Dumazet 已提交
1620
		 (!nonagle && tp->packets_out && tcp_minshall_check(tp)));
1621 1622
}

E
Eric Dumazet 已提交
1623
/* Return true if the Nagle test allows this packet to be
1624 1625
 * sent now.
 */
E
Eric Dumazet 已提交
1626 1627
static inline bool tcp_nagle_test(const struct tcp_sock *tp, const struct sk_buff *skb,
				  unsigned int cur_mss, int nonagle)
1628 1629 1630 1631 1632 1633 1634 1635
{
	/* 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
1636
		return true;
1637

1638 1639 1640
	/* 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
1641
	if (tcp_urg_mode(tp) || (tp->frto_counter == 2) ||
E
Eric Dumazet 已提交
1642
	    (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
E
Eric Dumazet 已提交
1643
		return true;
1644 1645

	if (!tcp_nagle_check(tp, skb, cur_mss, nonagle))
E
Eric Dumazet 已提交
1646
		return true;
1647

E
Eric Dumazet 已提交
1648
	return false;
1649 1650 1651
}

/* Does at least the first segment of SKB fit into the send window? */
E
Eric Dumazet 已提交
1652 1653 1654
static bool tcp_snd_wnd_test(const struct tcp_sock *tp,
			     const struct sk_buff *skb,
			     unsigned int cur_mss)
1655 1656 1657 1658 1659 1660
{
	u32 end_seq = TCP_SKB_CB(skb)->end_seq;

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

1661
	return !after(end_seq, tcp_wnd_end(tp));
1662 1663
}

1664
/* This checks if the data bearing packet SKB (usually tcp_send_head(sk))
1665 1666 1667
 * should be put on the wire right now.  If so, it returns the number of
 * packets allowed by the congestion window.
 */
1668
static unsigned int tcp_snd_test(const struct sock *sk, struct sk_buff *skb,
1669 1670
				 unsigned int cur_mss, int nonagle)
{
1671
	const struct tcp_sock *tp = tcp_sk(sk);
1672 1673
	unsigned int cwnd_quota;

1674
	tcp_init_tso_segs(sk, skb, cur_mss);
1675 1676 1677 1678 1679

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

	cwnd_quota = tcp_cwnd_test(tp, skb);
1680
	if (cwnd_quota && !tcp_snd_wnd_test(tp, skb, cur_mss))
1681 1682 1683 1684 1685
		cwnd_quota = 0;

	return cwnd_quota;
}

1686
/* Test if sending is allowed right now. */
E
Eric Dumazet 已提交
1687
bool tcp_may_send_now(struct sock *sk)
1688
{
1689
	const struct tcp_sock *tp = tcp_sk(sk);
1690
	struct sk_buff *skb = tcp_send_head(sk);
1691

E
Eric Dumazet 已提交
1692
	return skb &&
I
Ilpo Järvinen 已提交
1693
		tcp_snd_test(sk, skb, tcp_current_mss(sk),
1694
			     (tcp_skb_is_last(sk, skb) ?
E
Eric Dumazet 已提交
1695
			      tp->nonagle : TCP_NAGLE_PUSH));
1696 1697 1698 1699 1700 1701 1702 1703 1704
}

/* 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).
 */
1705
static int tso_fragment(struct sock *sk, struct sk_buff *skb, unsigned int len,
1706
			unsigned int mss_now, gfp_t gfp)
1707 1708 1709
{
	struct sk_buff *buff;
	int nlen = skb->len - len;
1710
	u8 flags;
1711 1712

	/* All of a TSO frame must be composed of paged data.  */
1713 1714
	if (skb->len != skb->data_len)
		return tcp_fragment(sk, skb, len, mss_now);
1715

1716
	buff = sk_stream_alloc_skb(sk, 0, gfp);
1717 1718 1719
	if (unlikely(buff == NULL))
		return -ENOMEM;

1720 1721
	sk->sk_wmem_queued += buff->truesize;
	sk_mem_charge(sk, buff->truesize);
1722
	buff->truesize += nlen;
1723 1724 1725 1726 1727 1728 1729 1730
	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
1731 1732 1733
	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;
1734 1735 1736 1737

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

1738
	buff->ip_summed = skb->ip_summed = CHECKSUM_PARTIAL;
1739 1740 1741
	skb_split(skb, buff, len);

	/* Fix up tso_factor for both original and new SKB.  */
1742 1743
	tcp_set_skb_tso_segs(sk, skb, mss_now);
	tcp_set_skb_tso_segs(sk, buff, mss_now);
1744 1745 1746

	/* Link BUFF into the send queue. */
	skb_header_release(buff);
1747
	tcp_insert_write_queue_after(skb, buff, sk);
1748 1749 1750 1751 1752 1753 1754 1755 1756

	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
1757
static bool tcp_tso_should_defer(struct sock *sk, struct sk_buff *skb)
1758
{
1759
	struct tcp_sock *tp = tcp_sk(sk);
1760
	const struct inet_connection_sock *icsk = inet_csk(sk);
1761
	u32 send_win, cong_win, limit, in_flight;
1762
	int win_divisor;
1763

E
Eric Dumazet 已提交
1764
	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
J
John Heffner 已提交
1765
		goto send_now;
1766

1767
	if (icsk->icsk_ca_state != TCP_CA_Open)
J
John Heffner 已提交
1768 1769 1770
		goto send_now;

	/* Defer for less than two clock ticks. */
I
Ilpo Järvinen 已提交
1771
	if (tp->tso_deferred &&
1772
	    (((u32)jiffies << 1) >> 1) - (tp->tso_deferred >> 1) > 1)
J
John Heffner 已提交
1773
		goto send_now;
1774

1775 1776
	in_flight = tcp_packets_in_flight(tp);

1777
	BUG_ON(tcp_skb_pcount(skb) <= 1 || (tp->snd_cwnd <= in_flight));
1778

1779
	send_win = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
1780 1781 1782 1783 1784 1785

	/* 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);

1786
	/* If a full-sized TSO skb can be sent, do it. */
1787 1788
	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
1789
		goto send_now;
1790

1791 1792 1793 1794
	/* 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;

1795 1796
	win_divisor = ACCESS_ONCE(sysctl_tcp_tso_win_divisor);
	if (win_divisor) {
1797 1798 1799 1800 1801
		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.
		 */
1802
		chunk /= win_divisor;
1803
		if (limit >= chunk)
J
John Heffner 已提交
1804
			goto send_now;
1805 1806 1807 1808 1809 1810
	} 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.
		 */
1811
		if (limit > tcp_max_tso_deferred_mss(tp) * tp->mss_cache)
J
John Heffner 已提交
1812
			goto send_now;
1813 1814 1815
	}

	/* Ok, it looks like it is advisable to defer.  */
1816
	tp->tso_deferred = 1 | (jiffies << 1);
J
John Heffner 已提交
1817

E
Eric Dumazet 已提交
1818
	return true;
J
John Heffner 已提交
1819 1820 1821

send_now:
	tp->tso_deferred = 0;
E
Eric Dumazet 已提交
1822
	return false;
1823 1824
}

J
John Heffner 已提交
1825
/* Create a new MTU probe if we are ready.
1826 1827 1828 1829
 * 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
1830 1831
 * Returns 0 if we should wait to probe (no cwnd available),
 *         1 if a probe was sent,
1832 1833
 *         -1 otherwise
 */
J
John Heffner 已提交
1834 1835 1836 1837 1838 1839 1840
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;
1841
	int size_needed;
J
John Heffner 已提交
1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852
	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 ||
1853
	    tp->rx_opt.num_sacks || tp->rx_opt.dsack)
J
John Heffner 已提交
1854 1855 1856
		return -1;

	/* Very simple search strategy: just double the MSS. */
I
Ilpo Järvinen 已提交
1857
	mss_now = tcp_current_mss(sk);
1858
	probe_size = 2 * tp->mss_cache;
1859
	size_needed = probe_size + (tp->reordering + 1) * tp->mss_cache;
J
John Heffner 已提交
1860 1861 1862 1863 1864 1865
	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? */
1866
	if (tp->write_seq - tp->snd_nxt < size_needed)
J
John Heffner 已提交
1867 1868
		return -1;

1869 1870
	if (tp->snd_wnd < size_needed)
		return -1;
1871
	if (after(tp->snd_nxt + size_needed, tcp_wnd_end(tp)))
1872
		return 0;
J
John Heffner 已提交
1873

1874 1875 1876
	/* 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
1877 1878 1879 1880 1881 1882 1883 1884
			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;
1885 1886
	sk->sk_wmem_queued += nskb->truesize;
	sk_mem_charge(sk, nskb->truesize);
J
John Heffner 已提交
1887

1888
	skb = tcp_send_head(sk);
J
John Heffner 已提交
1889 1890 1891

	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
1892
	TCP_SKB_CB(nskb)->tcp_flags = TCPHDR_ACK;
J
John Heffner 已提交
1893 1894
	TCP_SKB_CB(nskb)->sacked = 0;
	nskb->csum = 0;
1895
	nskb->ip_summed = skb->ip_summed;
J
John Heffner 已提交
1896

1897 1898
	tcp_insert_write_queue_before(nskb, skb, sk);

J
John Heffner 已提交
1899
	len = 0;
1900
	tcp_for_write_queue_from_safe(skb, next, sk) {
J
John Heffner 已提交
1901 1902 1903 1904 1905
		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,
1906 1907
							    skb_put(nskb, copy),
							    copy, nskb->csum);
J
John Heffner 已提交
1908 1909 1910 1911

		if (skb->len <= copy) {
			/* We've eaten all the data from this skb.
			 * Throw it away. */
E
Eric Dumazet 已提交
1912
			TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;
1913
			tcp_unlink_write_queue(skb, sk);
1914
			sk_wmem_free_skb(sk, skb);
J
John Heffner 已提交
1915
		} else {
E
Eric Dumazet 已提交
1916
			TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags &
C
Changli Gao 已提交
1917
						   ~(TCPHDR_FIN|TCPHDR_PSH);
J
John Heffner 已提交
1918 1919
			if (!skb_shinfo(skb)->nr_frags) {
				skb_pull(skb, copy);
1920
				if (skb->ip_summed != CHECKSUM_PARTIAL)
1921 1922
					skb->csum = csum_partial(skb->data,
								 skb->len, 0);
J
John Heffner 已提交
1923 1924 1925 1926 1927 1928 1929 1930
			} else {
				__pskb_trim_head(skb, copy);
				tcp_set_skb_tso_segs(sk, skb, mss_now);
			}
			TCP_SKB_CB(skb)->seq += copy;
		}

		len += copy;
1931 1932 1933

		if (len >= probe_size)
			break;
J
John Heffner 已提交
1934 1935 1936 1937 1938 1939 1940 1941
	}
	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
1942
		 * effectively two packets. */
J
John Heffner 已提交
1943
		tp->snd_cwnd--;
1944
		tcp_event_new_data_sent(sk, nskb);
J
John Heffner 已提交
1945 1946

		icsk->icsk_mtup.probe_size = tcp_mss_to_mtu(sk, nskb->len);
1947 1948
		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
1949 1950 1951 1952 1953 1954 1955

		return 1;
	}

	return -1;
}

L
Linus Torvalds 已提交
1956 1957 1958 1959
/* This routine writes packets to the network.  It advances the
 * send_head.  This happens as incoming acks open up the remote
 * window for us.
 *
1960 1961 1962 1963
 * 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.
 *
N
Nandita Dukkipati 已提交
1964 1965 1966
 * Send at most one packet when push_one > 0. Temporarily ignore
 * cwnd limit to force at most one packet out when push_one == 2.

E
Eric Dumazet 已提交
1967 1968
 * 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
1969
 */
E
Eric Dumazet 已提交
1970 1971
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
1972 1973
{
	struct tcp_sock *tp = tcp_sk(sk);
1974
	struct sk_buff *skb;
1975 1976
	unsigned int tso_segs, sent_pkts;
	int cwnd_quota;
J
John Heffner 已提交
1977
	int result;
L
Linus Torvalds 已提交
1978

1979
	sent_pkts = 0;
J
John Heffner 已提交
1980

1981 1982 1983 1984
	if (!push_one) {
		/* Do MTU probing. */
		result = tcp_mtu_probe(sk);
		if (!result) {
E
Eric Dumazet 已提交
1985
			return false;
1986 1987 1988
		} else if (result > 0) {
			sent_pkts = 1;
		}
J
John Heffner 已提交
1989 1990
	}

1991
	while ((skb = tcp_send_head(sk))) {
1992 1993
		unsigned int limit;

E
Eric Dumazet 已提交
1994

1995
		tso_segs = tcp_init_tso_segs(sk, skb, mss_now);
1996
		BUG_ON(!tso_segs);
1997

1998 1999 2000
		if (unlikely(tp->repair) && tp->repair_queue == TCP_SEND_QUEUE)
			goto repair; /* Skip network transmission */

2001
		cwnd_quota = tcp_cwnd_test(tp, skb);
N
Nandita Dukkipati 已提交
2002 2003 2004 2005 2006 2007 2008
		if (!cwnd_quota) {
			if (push_one == 2)
				/* Force out a loss probe pkt. */
				cwnd_quota = 1;
			else
				break;
		}
2009 2010 2011 2012

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

2013 2014 2015 2016 2017 2018
		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 {
2019
			if (!push_one && tcp_tso_should_defer(sk, skb))
2020 2021
				break;
		}
2022

E
Eric Dumazet 已提交
2023 2024 2025 2026 2027 2028 2029
		/* 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;
		}
2030
		limit = mss_now;
2031
		if (tso_segs > 1 && !tcp_urg_mode(tp))
2032
			limit = tcp_mss_split_point(sk, skb, mss_now,
2033 2034 2035
						    min_t(unsigned int,
							  cwnd_quota,
							  sk->sk_gso_max_segs));
L
Linus Torvalds 已提交
2036

2037
		if (skb->len > limit &&
2038
		    unlikely(tso_fragment(sk, skb, limit, mss_now, gfp)))
2039 2040
			break;

2041
		TCP_SKB_CB(skb)->when = tcp_time_stamp;
2042

2043
		if (unlikely(tcp_transmit_skb(sk, skb, 1, gfp)))
2044
			break;
L
Linus Torvalds 已提交
2045

2046
repair:
2047 2048 2049
		/* Advance the send_head.  This one is sent out.
		 * This call will increment packets_out.
		 */
2050
		tcp_event_new_data_sent(sk, skb);
L
Linus Torvalds 已提交
2051

2052
		tcp_minshall_update(tp, mss_now, skb);
2053
		sent_pkts += tcp_skb_pcount(skb);
2054 2055 2056

		if (push_one)
			break;
2057
	}
L
Linus Torvalds 已提交
2058

2059
	if (likely(sent_pkts)) {
2060 2061
		if (tcp_in_cwnd_reduction(sk))
			tp->prr_out += sent_pkts;
N
Nandita Dukkipati 已提交
2062 2063 2064 2065

		/* Send one loss probe per tail loss episode. */
		if (push_one != 2)
			tcp_schedule_loss_probe(sk);
2066
		tcp_cwnd_validate(sk);
E
Eric Dumazet 已提交
2067
		return false;
L
Linus Torvalds 已提交
2068
	}
N
Nandita Dukkipati 已提交
2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134
	return (push_one == 2) || (!tp->packets_out && tcp_send_head(sk));
}

bool tcp_schedule_loss_probe(struct sock *sk)
{
	struct inet_connection_sock *icsk = inet_csk(sk);
	struct tcp_sock *tp = tcp_sk(sk);
	u32 timeout, tlp_time_stamp, rto_time_stamp;
	u32 rtt = tp->srtt >> 3;

	if (WARN_ON(icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS))
		return false;
	/* No consecutive loss probes. */
	if (WARN_ON(icsk->icsk_pending == ICSK_TIME_LOSS_PROBE)) {
		tcp_rearm_rto(sk);
		return false;
	}
	/* Don't do any loss probe on a Fast Open connection before 3WHS
	 * finishes.
	 */
	if (sk->sk_state == TCP_SYN_RECV)
		return false;

	/* TLP is only scheduled when next timer event is RTO. */
	if (icsk->icsk_pending != ICSK_TIME_RETRANS)
		return false;

	/* Schedule a loss probe in 2*RTT for SACK capable connections
	 * in Open state, that are either limited by cwnd or application.
	 */
	if (sysctl_tcp_early_retrans < 3 || !rtt || !tp->packets_out ||
	    !tcp_is_sack(tp) || inet_csk(sk)->icsk_ca_state != TCP_CA_Open)
		return false;

	if ((tp->snd_cwnd > tcp_packets_in_flight(tp)) &&
	     tcp_send_head(sk))
		return false;

	/* Probe timeout is at least 1.5*rtt + TCP_DELACK_MAX to account
	 * for delayed ack when there's one outstanding packet.
	 */
	timeout = rtt << 1;
	if (tp->packets_out == 1)
		timeout = max_t(u32, timeout,
				(rtt + (rtt >> 1) + TCP_DELACK_MAX));
	timeout = max_t(u32, timeout, msecs_to_jiffies(10));

	/* If RTO is shorter, just schedule TLP in its place. */
	tlp_time_stamp = tcp_time_stamp + timeout;
	rto_time_stamp = (u32)inet_csk(sk)->icsk_timeout;
	if ((s32)(tlp_time_stamp - rto_time_stamp) > 0) {
		s32 delta = rto_time_stamp - tcp_time_stamp;
		if (delta > 0)
			timeout = delta;
	}

	inet_csk_reset_xmit_timer(sk, ICSK_TIME_LOSS_PROBE, timeout,
				  TCP_RTO_MAX);
	return true;
}

/* When probe timeout (PTO) fires, send a new segment if one exists, else
 * retransmit the last segment.
 */
void tcp_send_loss_probe(struct sock *sk)
{
N
Nandita Dukkipati 已提交
2135
	struct tcp_sock *tp = tcp_sk(sk);
N
Nandita Dukkipati 已提交
2136 2137 2138 2139 2140 2141 2142 2143 2144 2145
	struct sk_buff *skb;
	int pcount;
	int mss = tcp_current_mss(sk);
	int err = -1;

	if (tcp_send_head(sk) != NULL) {
		err = tcp_write_xmit(sk, mss, TCP_NAGLE_OFF, 2, GFP_ATOMIC);
		goto rearm_timer;
	}

N
Nandita Dukkipati 已提交
2146 2147 2148 2149
	/* At most one outstanding TLP retransmission. */
	if (tp->tlp_high_seq)
		goto rearm_timer;

N
Nandita Dukkipati 已提交
2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171
	/* Retransmit last segment. */
	skb = tcp_write_queue_tail(sk);
	if (WARN_ON(!skb))
		goto rearm_timer;

	pcount = tcp_skb_pcount(skb);
	if (WARN_ON(!pcount))
		goto rearm_timer;

	if ((pcount > 1) && (skb->len > (pcount - 1) * mss)) {
		if (unlikely(tcp_fragment(sk, skb, (pcount - 1) * mss, mss)))
			goto rearm_timer;
		skb = tcp_write_queue_tail(sk);
	}

	if (WARN_ON(!skb || !tcp_skb_pcount(skb)))
		goto rearm_timer;

	/* Probe with zero data doesn't trigger fast recovery. */
	if (skb->len > 0)
		err = __tcp_retransmit_skb(sk, skb);

N
Nandita Dukkipati 已提交
2172 2173 2174 2175
	/* Record snd_nxt for loss detection. */
	if (likely(!err))
		tp->tlp_high_seq = tp->snd_nxt;

N
Nandita Dukkipati 已提交
2176 2177 2178 2179 2180 2181 2182 2183 2184
rearm_timer:
	inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
				  inet_csk(sk)->icsk_rto,
				  TCP_RTO_MAX);

	if (likely(!err))
		NET_INC_STATS_BH(sock_net(sk),
				 LINUX_MIB_TCPLOSSPROBES);
	return;
L
Linus Torvalds 已提交
2185 2186
}

2187 2188 2189 2190
/* 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.
 */
2191 2192
void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
			       int nonagle)
2193
{
2194 2195 2196 2197 2198 2199 2200
	/* 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;

2201 2202
	if (tcp_write_xmit(sk, cur_mss, nonagle, 0,
			   sk_gfp_atomic(sk, GFP_ATOMIC)))
2203
		tcp_check_probe_timer(sk);
2204 2205
}

2206 2207 2208 2209 2210
/* 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)
{
2211
	struct sk_buff *skb = tcp_send_head(sk);
2212 2213 2214

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

2215
	tcp_write_xmit(sk, mss_now, TCP_NAGLE_PUSH, 1, sk->sk_allocation);
2216 2217
}

L
Linus Torvalds 已提交
2218 2219
/* This function returns the amount that we can raise the
 * usable window based on the following constraints
2220
 *
L
Linus Torvalds 已提交
2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238
 * 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.
2239
 *
L
Linus Torvalds 已提交
2240 2241
 * Of course, if the sender implements sender side SWS prevention
 * then this will not be a problem.
2242
 *
L
Linus Torvalds 已提交
2243
 * BSD seems to make the following compromise:
2244
 *
L
Linus Torvalds 已提交
2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271
 *	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)
{
2272
	struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
2273
	struct tcp_sock *tp = tcp_sk(sk);
S
Stephen Hemminger 已提交
2274
	/* MSS for the peer's data.  Previous versions used mss_clamp
L
Linus Torvalds 已提交
2275 2276 2277 2278 2279
	 * 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
	 */
2280
	int mss = icsk->icsk_ack.rcv_mss;
L
Linus Torvalds 已提交
2281 2282 2283 2284 2285
	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)
2286
		mss = full_space;
L
Linus Torvalds 已提交
2287

2288
	if (free_space < (full_space >> 1)) {
2289
		icsk->icsk_ack.quick = 0;
L
Linus Torvalds 已提交
2290

2291
		if (sk_under_memory_pressure(sk))
2292 2293
			tp->rcv_ssthresh = min(tp->rcv_ssthresh,
					       4U * tp->advmss);
L
Linus Torvalds 已提交
2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325

		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)
2326
			window = (free_space / mss) * mss;
2327
		else if (mss == full_space &&
2328
			 free_space > window + (full_space >> 1))
2329
			window = free_space;
L
Linus Torvalds 已提交
2330 2331 2332 2333 2334
	}

	return window;
}

2335 2336
/* 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
2337 2338
{
	struct tcp_sock *tp = tcp_sk(sk);
2339
	struct sk_buff *next_skb = tcp_write_queue_next(sk, skb);
2340
	int skb_size, next_skb_size;
L
Linus Torvalds 已提交
2341

2342 2343
	skb_size = skb->len;
	next_skb_size = next_skb->len;
L
Linus Torvalds 已提交
2344

2345
	BUG_ON(tcp_skb_pcount(skb) != 1 || tcp_skb_pcount(next_skb) != 1);
2346

2347
	tcp_highest_sack_combine(sk, next_skb, skb);
L
Linus Torvalds 已提交
2348

2349
	tcp_unlink_write_queue(next_skb, sk);
L
Linus Torvalds 已提交
2350

2351 2352
	skb_copy_from_linear_data(next_skb, skb_put(skb, next_skb_size),
				  next_skb_size);
L
Linus Torvalds 已提交
2353

2354 2355
	if (next_skb->ip_summed == CHECKSUM_PARTIAL)
		skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
2356

2357 2358
	if (skb->ip_summed != CHECKSUM_PARTIAL)
		skb->csum = csum_block_add(skb->csum, next_skb->csum, skb_size);
L
Linus Torvalds 已提交
2359

2360 2361
	/* 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
2362

2363
	/* Merge over control information. This moves PSH/FIN etc. over */
E
Eric Dumazet 已提交
2364
	TCP_SKB_CB(skb)->tcp_flags |= TCP_SKB_CB(next_skb)->tcp_flags;
2365 2366 2367 2368 2369 2370 2371

	/* 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 */
2372 2373 2374
	tcp_clear_retrans_hints_partial(tp);
	if (next_skb == tp->retransmit_skb_hint)
		tp->retransmit_skb_hint = skb;
2375

2376 2377
	tcp_adjust_pcount(sk, next_skb, tcp_skb_pcount(next_skb));

2378
	sk_wmem_free_skb(sk, next_skb);
L
Linus Torvalds 已提交
2379 2380
}

2381
/* Check if coalescing SKBs is legal. */
E
Eric Dumazet 已提交
2382
static bool tcp_can_collapse(const struct sock *sk, const struct sk_buff *skb)
2383 2384
{
	if (tcp_skb_pcount(skb) > 1)
E
Eric Dumazet 已提交
2385
		return false;
2386 2387
	/* TODO: SACK collapsing could be used to remove this condition */
	if (skb_shinfo(skb)->nr_frags != 0)
E
Eric Dumazet 已提交
2388
		return false;
2389
	if (skb_cloned(skb))
E
Eric Dumazet 已提交
2390
		return false;
2391
	if (skb == tcp_send_head(sk))
E
Eric Dumazet 已提交
2392
		return false;
2393 2394
	/* 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
2395
		return false;
2396

E
Eric Dumazet 已提交
2397
	return true;
2398 2399
}

2400 2401 2402
/* Collapse packets in the retransmit queue to make to create
 * less packets on the wire. This is only done on retransmission.
 */
2403 2404 2405 2406 2407
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
2408
	bool first = true;
2409 2410 2411

	if (!sysctl_tcp_retrans_collapse)
		return;
E
Eric Dumazet 已提交
2412
	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)
2413 2414 2415 2416 2417 2418 2419 2420 2421
		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
2422
			first = false;
2423 2424 2425 2426 2427 2428 2429 2430
			continue;
		}

		if (space < 0)
			break;
		/* Punt if not enough space exists in the first SKB for
		 * the data in the second
		 */
2431
		if (skb->len > skb_availroom(to))
2432 2433 2434 2435 2436 2437 2438 2439 2440
			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
2441 2442 2443 2444
/* 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.
 */
2445
int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
2446 2447
{
	struct tcp_sock *tp = tcp_sk(sk);
J
John Heffner 已提交
2448
	struct inet_connection_sock *icsk = inet_csk(sk);
2449
	unsigned int cur_mss;
L
Linus Torvalds 已提交
2450

J
John Heffner 已提交
2451 2452 2453 2454 2455
	/* Inconslusive MTU probe */
	if (icsk->icsk_mtup.probe_size) {
		icsk->icsk_mtup.probe_size = 0;
	}

L
Linus Torvalds 已提交
2456
	/* Do not sent more than we queued. 1/4 is reserved for possible
S
Stephen Hemminger 已提交
2457
	 * copying overhead: fragmentation, tunneling, mangling etc.
L
Linus Torvalds 已提交
2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469
	 */
	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;
	}

2470 2471 2472
	if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk))
		return -EHOSTUNREACH; /* Routing failure or similar. */

I
Ilpo Järvinen 已提交
2473
	cur_mss = tcp_current_mss(sk);
2474

L
Linus Torvalds 已提交
2475 2476 2477 2478 2479
	/* 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.
	 */
2480 2481
	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
2482 2483 2484
		return -EAGAIN;

	if (skb->len > cur_mss) {
2485
		if (tcp_fragment(sk, skb, cur_mss, cur_mss))
L
Linus Torvalds 已提交
2486
			return -ENOMEM; /* We'll try again later. */
2487
	} else {
2488 2489 2490 2491 2492 2493
		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
2494 2495
	}

2496
	tcp_retrans_try_collapse(sk, skb, cur_mss);
L
Linus Torvalds 已提交
2497 2498 2499 2500 2501

	/* 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
2502
	if (skb->len > 0 &&
E
Eric Dumazet 已提交
2503
	    (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) &&
S
Stephen Hemminger 已提交
2504
	    tp->snd_una == (TCP_SKB_CB(skb)->end_seq - 1)) {
L
Linus Torvalds 已提交
2505
		if (!pskb_trim(skb, 0)) {
2506 2507
			/* 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
2508
					     TCP_SKB_CB(skb)->tcp_flags);
L
Linus Torvalds 已提交
2509 2510 2511 2512 2513 2514 2515 2516 2517
			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
2518 2519 2520 2521
	/* 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);
2522 2523
		return nskb ? tcp_transmit_skb(sk, nskb, 0, GFP_ATOMIC) :
			      -ENOBUFS;
E
Eric Dumazet 已提交
2524
	} else {
2525
		return tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
E
Eric Dumazet 已提交
2526
	}
2527 2528 2529 2530 2531 2532
}

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
2533 2534 2535

	if (err == 0) {
		/* Update global TCP statistics. */
P
Pavel Emelyanov 已提交
2536
		TCP_INC_STATS(sock_net(sk), TCP_MIB_RETRANSSEGS);
L
Linus Torvalds 已提交
2537 2538 2539 2540

		tp->total_retrans++;

#if FASTRETRANS_DEBUG > 0
2541
		if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) {
2542
			net_dbg_ratelimited("retrans_out leaked\n");
L
Linus Torvalds 已提交
2543 2544
		}
#endif
2545 2546
		if (!tp->retrans_out)
			tp->lost_retrans_low = tp->snd_nxt;
L
Linus Torvalds 已提交
2547 2548 2549 2550 2551 2552 2553
		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
2554
		tp->undo_retrans += tcp_skb_pcount(skb);
L
Linus Torvalds 已提交
2555 2556 2557 2558 2559 2560 2561 2562 2563

		/* 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;
}

2564 2565 2566
/* Check if we forward retransmits are possible in the current
 * window/congestion state.
 */
E
Eric Dumazet 已提交
2567
static bool tcp_can_forward_retransmit(struct sock *sk)
2568 2569
{
	const struct inet_connection_sock *icsk = inet_csk(sk);
2570
	const struct tcp_sock *tp = tcp_sk(sk);
2571 2572 2573

	/* Forward retransmissions are possible only during Recovery. */
	if (icsk->icsk_ca_state != TCP_CA_Recovery)
E
Eric Dumazet 已提交
2574
		return false;
2575 2576 2577

	/* No forward retransmissions in Reno are possible. */
	if (tcp_is_reno(tp))
E
Eric Dumazet 已提交
2578
		return false;
2579 2580 2581 2582 2583 2584 2585 2586 2587 2588

	/* 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
2589
		return false;
2590

E
Eric Dumazet 已提交
2591
	return true;
2592 2593
}

L
Linus Torvalds 已提交
2594 2595 2596 2597 2598 2599 2600 2601 2602 2603
/* 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)
{
2604
	const struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
2605 2606
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb;
2607
	struct sk_buff *hole = NULL;
2608
	u32 last_lost;
2609
	int mib_idx;
2610
	int fwd_rexmitting = 0;
2611

2612 2613 2614
	if (!tp->packets_out)
		return;

2615 2616 2617
	if (!tp->lost_out)
		tp->retransmit_high = tp->snd_una;

2618
	if (tp->retransmit_skb_hint) {
2619
		skb = tp->retransmit_skb_hint;
2620 2621 2622 2623
		last_lost = TCP_SKB_CB(skb)->end_seq;
		if (after(last_lost, tp->retransmit_high))
			last_lost = tp->retransmit_high;
	} else {
2624
		skb = tcp_write_queue_head(sk);
2625 2626
		last_lost = tp->snd_una;
	}
L
Linus Torvalds 已提交
2627

2628 2629
	tcp_for_write_queue_from(skb, sk) {
		__u8 sacked = TCP_SKB_CB(skb)->sacked;
L
Linus Torvalds 已提交
2630

2631 2632 2633
		if (skb == tcp_send_head(sk))
			break;
		/* we could do better than to assign each time */
2634 2635
		if (hole == NULL)
			tp->retransmit_skb_hint = skb;
2636 2637 2638 2639 2640 2641 2642 2643 2644 2645

		/* 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
2646

2647 2648 2649 2650 2651
		if (fwd_rexmitting) {
begin_fwd:
			if (!before(TCP_SKB_CB(skb)->seq, tcp_highest_sack_seq(tp)))
				break;
			mib_idx = LINUX_MIB_TCPFORWARDRETRANS;
2652

2653
		} else if (!before(TCP_SKB_CB(skb)->seq, tp->retransmit_high)) {
2654
			tp->retransmit_high = last_lost;
2655 2656 2657 2658 2659 2660 2661 2662 2663
			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
2664

2665
		} else if (!(sacked & TCPCB_LOST)) {
2666
			if (hole == NULL && !(sacked & (TCPCB_SACKED_RETRANS|TCPCB_SACKED_ACKED)))
2667 2668
				hole = skb;
			continue;
L
Linus Torvalds 已提交
2669

2670
		} else {
2671
			last_lost = TCP_SKB_CB(skb)->end_seq;
2672 2673 2674 2675 2676
			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
2677

2678
		if (sacked & (TCPCB_SACKED_ACKED|TCPCB_SACKED_RETRANS))
L
Linus Torvalds 已提交
2679 2680
			continue;

2681 2682
		if (tcp_retransmit_skb(sk, skb)) {
			NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPRETRANSFAIL);
2683
			return;
2684
		}
2685
		NET_INC_STATS_BH(sock_net(sk), mib_idx);
L
Linus Torvalds 已提交
2686

2687
		if (tcp_in_cwnd_reduction(sk))
2688 2689
			tp->prr_out += tcp_skb_pcount(skb);

2690
		if (skb == tcp_write_queue_head(sk))
2691 2692 2693
			inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
						  inet_csk(sk)->icsk_rto,
						  TCP_RTO_MAX);
L
Linus Torvalds 已提交
2694 2695 2696 2697 2698 2699 2700 2701
	}
}

/* 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)
{
2702
	struct tcp_sock *tp = tcp_sk(sk);
2703
	struct sk_buff *skb = tcp_write_queue_tail(sk);
L
Linus Torvalds 已提交
2704
	int mss_now;
2705

L
Linus Torvalds 已提交
2706 2707 2708 2709
	/* 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
2710
	mss_now = tcp_current_mss(sk);
L
Linus Torvalds 已提交
2711

2712
	if (tcp_send_head(sk) != NULL) {
E
Eric Dumazet 已提交
2713
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_FIN;
L
Linus Torvalds 已提交
2714 2715 2716 2717 2718
		TCP_SKB_CB(skb)->end_seq++;
		tp->write_seq++;
	} else {
		/* Socket is locked, keep trying until memory is available. */
		for (;;) {
2719 2720
			skb = alloc_skb_fclone(MAX_TCP_HEADER,
					       sk->sk_allocation);
L
Linus Torvalds 已提交
2721 2722 2723 2724 2725 2726 2727 2728
			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(). */
2729
		tcp_init_nondata_skb(skb, tp->write_seq,
C
Changli Gao 已提交
2730
				     TCPHDR_ACK | TCPHDR_FIN);
L
Linus Torvalds 已提交
2731 2732
		tcp_queue_skb(sk, skb);
	}
2733
	__tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_OFF);
L
Linus Torvalds 已提交
2734 2735 2736 2737 2738
}

/* 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
2739
 * by RFC 2525, section 2.17.  -DaveM
L
Linus Torvalds 已提交
2740
 */
A
Al Viro 已提交
2741
void tcp_send_active_reset(struct sock *sk, gfp_t priority)
L
Linus Torvalds 已提交
2742 2743 2744 2745 2746 2747
{
	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
2748
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTFAILED);
L
Linus Torvalds 已提交
2749 2750 2751 2752 2753
		return;
	}

	/* Reserve space for headers and prepare control bits. */
	skb_reserve(skb, MAX_TCP_HEADER);
2754
	tcp_init_nondata_skb(skb, tcp_acceptable_seq(sk),
C
Changli Gao 已提交
2755
			     TCPHDR_ACK | TCPHDR_RST);
L
Linus Torvalds 已提交
2756 2757
	/* Send it off. */
	TCP_SKB_CB(skb)->when = tcp_time_stamp;
2758
	if (tcp_transmit_skb(sk, skb, 0, priority))
P
Pavel Emelyanov 已提交
2759
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTFAILED);
2760

P
Pavel Emelyanov 已提交
2761
	TCP_INC_STATS(sock_net(sk), TCP_MIB_OUTRSTS);
L
Linus Torvalds 已提交
2762 2763
}

2764 2765
/* 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
2766 2767 2768 2769 2770 2771
 * 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)
{
2772
	struct sk_buff *skb;
L
Linus Torvalds 已提交
2773

2774
	skb = tcp_write_queue_head(sk);
E
Eric Dumazet 已提交
2775
	if (skb == NULL || !(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
2776
		pr_debug("%s: wrong queue state\n", __func__);
L
Linus Torvalds 已提交
2777 2778
		return -EFAULT;
	}
E
Eric Dumazet 已提交
2779
	if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)) {
L
Linus Torvalds 已提交
2780 2781 2782 2783
		if (skb_cloned(skb)) {
			struct sk_buff *nskb = skb_copy(skb, GFP_ATOMIC);
			if (nskb == NULL)
				return -ENOMEM;
2784
			tcp_unlink_write_queue(skb, sk);
L
Linus Torvalds 已提交
2785
			skb_header_release(nskb);
2786
			__tcp_add_write_queue_head(sk, nskb);
2787 2788 2789
			sk_wmem_free_skb(sk, skb);
			sk->sk_wmem_queued += nskb->truesize;
			sk_mem_charge(sk, nskb->truesize);
L
Linus Torvalds 已提交
2790 2791 2792
			skb = nskb;
		}

E
Eric Dumazet 已提交
2793
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ACK;
L
Linus Torvalds 已提交
2794 2795 2796
		TCP_ECN_send_synack(tcp_sk(sk), skb);
	}
	TCP_SKB_CB(skb)->when = tcp_time_stamp;
2797
	return tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
L
Linus Torvalds 已提交
2798 2799
}

2800 2801 2802 2803 2804 2805 2806 2807 2808 2809
/**
 * 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.
 */
2810
struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
2811
				struct request_sock *req,
2812 2813
				struct request_values *rvp,
				struct tcp_fastopen_cookie *foc)
L
Linus Torvalds 已提交
2814
{
2815
	struct tcp_out_options opts;
2816
	struct tcp_extend_values *xvp = tcp_xv(rvp);
2817
	struct inet_request_sock *ireq = inet_rsk(req);
L
Linus Torvalds 已提交
2818
	struct tcp_sock *tp = tcp_sk(sk);
E
Eric Dumazet 已提交
2819
	const struct tcp_cookie_values *cvp = tp->cookie_values;
L
Linus Torvalds 已提交
2820 2821
	struct tcphdr *th;
	struct sk_buff *skb;
2822
	struct tcp_md5sig_key *md5;
2823
	int tcp_header_size;
2824
	int mss;
E
Eric Dumazet 已提交
2825
	int s_data_desired = 0;
L
Linus Torvalds 已提交
2826

E
Eric Dumazet 已提交
2827 2828
	if (cvp != NULL && cvp->s_data_constant && cvp->s_data_desired)
		s_data_desired = cvp->s_data_desired;
2829 2830
	skb = alloc_skb(MAX_TCP_HEADER + 15 + s_data_desired,
			sk_gfp_atomic(sk, GFP_ATOMIC));
2831 2832
	if (unlikely(!skb)) {
		dst_release(dst);
L
Linus Torvalds 已提交
2833
		return NULL;
2834
	}
L
Linus Torvalds 已提交
2835 2836 2837
	/* Reserve space for headers. */
	skb_reserve(skb, MAX_TCP_HEADER);

2838
	skb_dst_set(skb, dst);
L
Linus Torvalds 已提交
2839

2840
	mss = dst_metric_advmss(dst);
2841 2842 2843
	if (tp->rx_opt.user_mss && tp->rx_opt.user_mss < mss)
		mss = tp->rx_opt.user_mss;

A
Adam Langley 已提交
2844 2845 2846 2847
	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);
2848 2849 2850 2851 2852 2853

		/* 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
2854 2855
		/* tcp_full_space because it is guaranteed to be the first packet */
		tcp_select_initial_window(tcp_full_space(sk),
2856
			mss - (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0),
A
Adam Langley 已提交
2857 2858 2859
			&req->rcv_wnd,
			&req->window_clamp,
			ireq->wscale_ok,
2860 2861
			&rcv_wscale,
			dst_metric(dst, RTAX_INITRWND));
A
Adam Langley 已提交
2862 2863 2864 2865
		ireq->rcv_wscale = rcv_wscale;
	}

	memset(&opts, 0, sizeof(opts));
2866 2867 2868 2869 2870
#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
2871
	TCP_SKB_CB(skb)->when = tcp_time_stamp;
2872
	tcp_header_size = tcp_synack_options(sk, req, mss,
2873
					     skb, &opts, &md5, xvp, foc)
2874
			+ sizeof(*th);
2875

2876 2877
	skb_push(skb, tcp_header_size);
	skb_reset_transport_header(skb);
L
Linus Torvalds 已提交
2878

2879
	th = tcp_hdr(skb);
L
Linus Torvalds 已提交
2880 2881 2882 2883
	memset(th, 0, sizeof(struct tcphdr));
	th->syn = 1;
	th->ack = 1;
	TCP_ECN_make_synack(req, th);
2884
	th->source = ireq->loc_port;
2885
	th->dest = ireq->rmt_port;
2886 2887 2888 2889
	/* 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
2890
			     TCPHDR_SYN | TCPHDR_ACK);
2891 2892

	if (OPTION_COOKIE_EXTENSION & opts.options) {
E
Eric Dumazet 已提交
2893 2894
		if (s_data_desired) {
			u8 *buf = skb_put(skb, s_data_desired);
2895 2896

			/* copy data directly from the listening socket. */
E
Eric Dumazet 已提交
2897 2898
			memcpy(buf, cvp->s_data_payload, s_data_desired);
			TCP_SKB_CB(skb)->end_seq += s_data_desired;
2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914
		}

		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 */
2915
			*tail-- ^= (((__force u32)th->dest << 16) | (__force u32)th->source);
D
David S. Miller 已提交
2916
			*tail-- ^= (u32)(unsigned long)cvp; /* per sockopt */
2917 2918 2919 2920 2921 2922 2923 2924 2925

			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
2926
	th->seq = htonl(TCP_SKB_CB(skb)->seq);
2927 2928
	/* 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
2929 2930

	/* RFC1323: The window in SYN & SYN/ACK segments is never scaled. */
2931
	th->window = htons(min(req->rcv_wnd, 65535U));
2932
	tcp_options_write((__be32 *)(th + 1), tp, &opts);
L
Linus Torvalds 已提交
2933
	th->doff = (tcp_header_size >> 2);
2934
	TCP_ADD_STATS(sock_net(sk), TCP_MIB_OUTSEGS, tcp_skb_pcount(skb));
2935 2936 2937 2938

#ifdef CONFIG_TCP_MD5SIG
	/* Okay, we have all we need - do the md5 hash if needed */
	if (md5) {
2939
		tcp_rsk(req)->af_specific->calc_md5_hash(opts.hash_location,
2940
					       md5, NULL, req, skb);
2941 2942 2943
	}
#endif

L
Linus Torvalds 已提交
2944 2945
	return skb;
}
E
Eric Dumazet 已提交
2946
EXPORT_SYMBOL(tcp_make_synack);
L
Linus Torvalds 已提交
2947

2948
/* Do all connect socket setups that can be done AF independent. */
P
Pavel Emelyanov 已提交
2949
void tcp_connect_init(struct sock *sk)
L
Linus Torvalds 已提交
2950
{
2951
	const struct dst_entry *dst = __sk_dst_get(sk);
L
Linus Torvalds 已提交
2952 2953 2954 2955 2956 2957 2958
	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) +
2959
		(sysctl_tcp_timestamps ? TCPOLEN_TSTAMP_ALIGNED : 0);
L
Linus Torvalds 已提交
2960

2961 2962 2963 2964 2965
#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
2966 2967 2968 2969
	/* 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
2970
	tcp_mtup_init(sk);
L
Linus Torvalds 已提交
2971 2972 2973 2974
	tcp_sync_mss(sk, dst_mtu(dst));

	if (!tp->window_clamp)
		tp->window_clamp = dst_metric(dst, RTAX_WINDOW);
2975
	tp->advmss = dst_metric_advmss(dst);
2976 2977 2978
	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
2979 2980
	tcp_initialize_rcv_mss(sk);

2981 2982 2983 2984 2985
	/* 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
2986 2987 2988 2989
	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,
2990
				  sysctl_tcp_window_scaling,
2991 2992
				  &rcv_wscale,
				  dst_metric(dst, RTAX_INITRWND));
L
Linus Torvalds 已提交
2993 2994 2995 2996 2997 2998 2999

	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;
3000
	tcp_init_wl(tp, 0);
L
Linus Torvalds 已提交
3001 3002
	tp->snd_una = tp->write_seq;
	tp->snd_sml = tp->write_seq;
I
Ilpo Järvinen 已提交
3003
	tp->snd_up = tp->write_seq;
P
Pavel Emelyanov 已提交
3004
	tp->snd_nxt = tp->write_seq;
P
Pavel Emelyanov 已提交
3005 3006 3007 3008 3009

	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
3010

3011 3012
	inet_csk(sk)->icsk_rto = TCP_TIMEOUT_INIT;
	inet_csk(sk)->icsk_retransmits = 0;
L
Linus Torvalds 已提交
3013 3014 3015
	tcp_clear_retrans(tp);
}

3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040
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;
3041
	int syn_loss = 0, space, i, err = 0, iovlen = fo->data->msg_iovlen;
3042
	struct sk_buff *syn_data = NULL, *data;
3043 3044
	unsigned long last_syn_loss = 0;

3045
	tp->rx_opt.mss_clamp = tp->advmss;  /* If MSS is not cached */
3046 3047 3048 3049 3050 3051 3052 3053
	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;
	}
3054

3055 3056 3057
	if (sysctl_tcp_fastopen & TFO_CLIENT_NO_COOKIE)
		fo->cookie.len = -1;
	else if (fo->cookie.len <= 0)
3058 3059 3060 3061 3062 3063 3064 3065
		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;
3066
	space = __tcp_mtu_to_mss(sk, inet_csk(sk)->icsk_pmtu_cookie) -
3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099
		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) {
3100
		tp->syn_data = (fo->copied > 0);
3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118
		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;
}

3119
/* Build a SYN and send it off. */
L
Linus Torvalds 已提交
3120 3121 3122 3123
int tcp_connect(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *buff;
3124
	int err;
L
Linus Torvalds 已提交
3125 3126 3127

	tcp_connect_init(sk);

A
Andrey Vagin 已提交
3128 3129 3130 3131 3132
	if (unlikely(tp->repair)) {
		tcp_finish_connect(sk, NULL);
		return 0;
	}

3133
	buff = alloc_skb_fclone(MAX_TCP_HEADER + 15, sk->sk_allocation);
L
Linus Torvalds 已提交
3134 3135 3136 3137 3138 3139
	if (unlikely(buff == NULL))
		return -ENOBUFS;

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

C
Changli Gao 已提交
3140
	tcp_init_nondata_skb(buff, tp->write_seq++, TCPHDR_SYN);
3141 3142
	tp->retrans_stamp = TCP_SKB_CB(buff)->when = tcp_time_stamp;
	tcp_connect_queue_skb(sk, buff);
3143
	TCP_ECN_send_syn(sk, buff);
L
Linus Torvalds 已提交
3144

3145 3146 3147
	/* 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);
3148 3149
	if (err == -ECONNREFUSED)
		return err;
3150 3151 3152 3153 3154 3155

	/* 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
3156
	TCP_INC_STATS(sock_net(sk), TCP_MIB_ACTIVEOPENS);
L
Linus Torvalds 已提交
3157 3158

	/* Timer for repeating the SYN until an answer. */
3159 3160
	inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
				  inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
L
Linus Torvalds 已提交
3161 3162
	return 0;
}
E
Eric Dumazet 已提交
3163
EXPORT_SYMBOL(tcp_connect);
L
Linus Torvalds 已提交
3164 3165 3166 3167 3168 3169 3170

/* 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)
{
3171 3172
	struct inet_connection_sock *icsk = inet_csk(sk);
	int ato = icsk->icsk_ack.ato;
L
Linus Torvalds 已提交
3173 3174 3175
	unsigned long timeout;

	if (ato > TCP_DELACK_MIN) {
3176
		const struct tcp_sock *tp = tcp_sk(sk);
3177
		int max_ato = HZ / 2;
L
Linus Torvalds 已提交
3178

3179 3180
		if (icsk->icsk_ack.pingpong ||
		    (icsk->icsk_ack.pending & ICSK_ACK_PUSHED))
L
Linus Torvalds 已提交
3181 3182 3183 3184 3185
			max_ato = TCP_DELACK_MAX;

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

		/* If some rtt estimate is known, use it to bound delayed ack.
3186
		 * Do not use inet_csk(sk)->icsk_rto here, use results of rtt measurements
L
Linus Torvalds 已提交
3187 3188 3189
		 * directly.
		 */
		if (tp->srtt) {
3190
			int rtt = max(tp->srtt >> 3, TCP_DELACK_MIN);
L
Linus Torvalds 已提交
3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202

			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. */
3203
	if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
L
Linus Torvalds 已提交
3204 3205 3206
		/* If delack timer was blocked or is about to expire,
		 * send ACK now.
		 */
3207 3208
		if (icsk->icsk_ack.blocked ||
		    time_before_eq(icsk->icsk_ack.timeout, jiffies + (ato >> 2))) {
L
Linus Torvalds 已提交
3209 3210 3211 3212
			tcp_send_ack(sk);
			return;
		}

3213 3214
		if (!time_before(timeout, icsk->icsk_ack.timeout))
			timeout = icsk->icsk_ack.timeout;
L
Linus Torvalds 已提交
3215
	}
3216 3217 3218
	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
3219 3220 3221 3222 3223
}

/* This routine sends an ack and also updates the window. */
void tcp_send_ack(struct sock *sk)
{
3224
	struct sk_buff *buff;
L
Linus Torvalds 已提交
3225

3226 3227 3228
	/* If we have been reset, we may not send again. */
	if (sk->sk_state == TCP_CLOSE)
		return;
L
Linus Torvalds 已提交
3229

3230 3231 3232 3233
	/* We are not putting this on the write queue, so
	 * tcp_transmit_skb() will set the ownership to this
	 * sock.
	 */
3234
	buff = alloc_skb(MAX_TCP_HEADER, sk_gfp_atomic(sk, GFP_ATOMIC));
3235 3236 3237 3238 3239 3240
	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
3241
	}
3242 3243 3244

	/* Reserve space for headers and prepare control bits. */
	skb_reserve(buff, MAX_TCP_HEADER);
C
Changli Gao 已提交
3245
	tcp_init_nondata_skb(buff, tcp_acceptable_seq(sk), TCPHDR_ACK);
3246 3247 3248

	/* Send it off, this clears delayed acks for us. */
	TCP_SKB_CB(buff)->when = tcp_time_stamp;
3249
	tcp_transmit_skb(sk, buff, 0, sk_gfp_atomic(sk, GFP_ATOMIC));
L
Linus Torvalds 已提交
3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268
}

/* 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. */
3269
	skb = alloc_skb(MAX_TCP_HEADER, sk_gfp_atomic(sk, GFP_ATOMIC));
3270
	if (skb == NULL)
L
Linus Torvalds 已提交
3271 3272 3273 3274 3275 3276 3277 3278
		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
3279
	tcp_init_nondata_skb(skb, tp->snd_una - !urgent, TCPHDR_ACK);
L
Linus Torvalds 已提交
3280
	TCP_SKB_CB(skb)->when = tcp_time_stamp;
3281
	return tcp_transmit_skb(sk, skb, 0, GFP_ATOMIC);
L
Linus Torvalds 已提交
3282 3283
}

P
Pavel Emelyanov 已提交
3284 3285 3286 3287
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
3288
		tcp_sk(sk)->snd_nxt = tcp_sk(sk)->write_seq;
P
Pavel Emelyanov 已提交
3289 3290 3291 3292
		tcp_xmit_probe_skb(sk, 0);
	}
}

3293
/* Initiate keepalive or window probe from timer. */
L
Linus Torvalds 已提交
3294 3295
int tcp_write_wakeup(struct sock *sk)
{
3296 3297
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb;
L
Linus Torvalds 已提交
3298

3299 3300 3301 3302 3303 3304
	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
3305
		unsigned int mss = tcp_current_mss(sk);
3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317
		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
3318
			TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
3319 3320 3321 3322 3323
			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
3324
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
3325 3326 3327 3328 3329 3330
		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
3331
		if (between(tp->snd_up, tp->snd_una + 1, tp->snd_una + 0xFFFF))
3332 3333
			tcp_xmit_probe_skb(sk, 1);
		return tcp_xmit_probe_skb(sk, 0);
L
Linus Torvalds 已提交
3334 3335 3336 3337 3338 3339 3340 3341
	}
}

/* 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)
{
3342
	struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
3343 3344 3345 3346 3347
	struct tcp_sock *tp = tcp_sk(sk);
	int err;

	err = tcp_write_wakeup(sk);

3348
	if (tp->packets_out || !tcp_send_head(sk)) {
L
Linus Torvalds 已提交
3349
		/* Cancel probe timer, if it is not required. */
3350
		icsk->icsk_probes_out = 0;
3351
		icsk->icsk_backoff = 0;
L
Linus Torvalds 已提交
3352 3353 3354 3355
		return;
	}

	if (err <= 0) {
3356 3357
		if (icsk->icsk_backoff < sysctl_tcp_retries2)
			icsk->icsk_backoff++;
3358
		icsk->icsk_probes_out++;
3359
		inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
3360 3361
					  min(icsk->icsk_rto << icsk->icsk_backoff, TCP_RTO_MAX),
					  TCP_RTO_MAX);
L
Linus Torvalds 已提交
3362 3363
	} else {
		/* If packet was not sent due to local congestion,
3364
		 * do not backoff and do not remember icsk_probes_out.
L
Linus Torvalds 已提交
3365 3366 3367 3368
		 * Let local senders to fight for local resources.
		 *
		 * Use accumulated backoff yet.
		 */
3369 3370
		if (!icsk->icsk_probes_out)
			icsk->icsk_probes_out = 1;
3371
		inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
3372
					  min(icsk->icsk_rto << icsk->icsk_backoff,
3373 3374
					      TCP_RESOURCE_PROBE_INTERVAL),
					  TCP_RTO_MAX);
L
Linus Torvalds 已提交
3375 3376
	}
}