tcp_output.c 91.9 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 69 70
unsigned int sysctl_tcp_notsent_lowat __read_mostly = UINT_MAX;
EXPORT_SYMBOL(sysctl_tcp_notsent_lowat);

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

84
	tp->packets_out += tcp_skb_pcount(skb);
N
Nandita Dukkipati 已提交
85
	if (!prior_packets || icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
86
	    icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
87
		tcp_rearm_rto(sk);
88
	}
L
Linus Torvalds 已提交
89 90 91 92 93 94 95 96
}

/* 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:
 */
97
static inline __u32 tcp_acceptable_seq(const struct sock *sk)
L
Linus Torvalds 已提交
98
{
99
	const struct tcp_sock *tp = tcp_sk(sk);
100

101
	if (!before(tcp_wnd_end(tp), tp->snd_nxt))
L
Linus Torvalds 已提交
102 103
		return tp->snd_nxt;
	else
104
		return tcp_wnd_end(tp);
L
Linus Torvalds 已提交
105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123
}

/* 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);
124
	const struct dst_entry *dst = __sk_dst_get(sk);
L
Linus Torvalds 已提交
125 126
	int mss = tp->advmss;

127 128 129 130 131 132 133
	if (dst) {
		unsigned int metric = dst_metric_advmss(dst);

		if (metric < mss) {
			mss = metric;
			tp->advmss = mss;
		}
L
Linus Torvalds 已提交
134 135 136 137 138 139 140
	}

	return (__u16)mss;
}

/* RFC2861. Reset CWND after idle period longer RTO to "restart window".
 * This is the first part of cwnd validation mechanism. */
141
static void tcp_cwnd_restart(struct sock *sk, const struct dst_entry *dst)
L
Linus Torvalds 已提交
142
{
143
	struct tcp_sock *tp = tcp_sk(sk);
L
Linus Torvalds 已提交
144 145 146 147
	s32 delta = tcp_time_stamp - tp->lsndtime;
	u32 restart_cwnd = tcp_init_cwnd(tp, dst);
	u32 cwnd = tp->snd_cwnd;

148
	tcp_ca_event(sk, CA_EVENT_CWND_RESTART);
L
Linus Torvalds 已提交
149

150
	tp->snd_ssthresh = tcp_current_ssthresh(sk);
L
Linus Torvalds 已提交
151 152
	restart_cwnd = min(restart_cwnd, cwnd);

153
	while ((delta -= inet_csk(sk)->icsk_rto) > 0 && cwnd > restart_cwnd)
L
Linus Torvalds 已提交
154 155 156 157 158 159
		cwnd >>= 1;
	tp->snd_cwnd = max(cwnd, restart_cwnd);
	tp->snd_cwnd_stamp = tcp_time_stamp;
	tp->snd_cwnd_used = 0;
}

160
/* Congestion state accounting after a packet has been sent. */
S
Stephen Hemminger 已提交
161
static void tcp_event_data_sent(struct tcp_sock *tp,
162
				struct sock *sk)
L
Linus Torvalds 已提交
163
{
164 165
	struct inet_connection_sock *icsk = inet_csk(sk);
	const u32 now = tcp_time_stamp;
166
	const struct dst_entry *dst = __sk_dst_get(sk);
L
Linus Torvalds 已提交
167

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

	tp->lsndtime = now;

	/* If it is a reply for ato after last received
	 * packet, enter pingpong mode.
	 */
177 178 179
	if ((u32)(now - icsk->icsk_ack.lrcvtime) < icsk->icsk_ack.ato &&
	    (!dst || !dst_metric(dst, RTAX_QUICKACK)))
			icsk->icsk_ack.pingpong = 1;
L
Linus Torvalds 已提交
180 181
}

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

189 190 191 192

u32 tcp_default_init_rwnd(u32 mss)
{
	/* Initial receive window should be twice of TCP_INIT_CWND to
W
Weiping Pan 已提交
193
	 * enable proper sending of new unsent data during fast recovery
194 195 196 197 198 199 200 201 202 203
	 * (RFC 3517, Section 4, NextSeg() rule (2)). Further place a
	 * limit when mss is larger than 1460.
	 */
	u32 init_rwnd = TCP_INIT_CWND * 2;

	if (mss > 1460)
		init_rwnd = max((1460 * init_rwnd) / mss, 2U);
	return init_rwnd;
}

L
Linus Torvalds 已提交
204 205 206 207 208 209 210 211 212
/* 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,
213 214
			       int wscale_ok, __u8 *rcv_wscale,
			       __u32 init_rcv_wnd)
L
Linus Torvalds 已提交
215 216 217 218 219 220 221 222 223 224 225 226 227
{
	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
228 229 230 231 232 233
	 * 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
234
	 */
235 236 237 238 239
	if (sysctl_tcp_workaround_signed_windows)
		(*rcv_wnd) = min(space, MAX_TCP_WINDOW);
	else
		(*rcv_wnd) = space;

L
Linus Torvalds 已提交
240 241 242
	(*rcv_wscale) = 0;
	if (wscale_ok) {
		/* Set window scaling on max possible window
243
		 * See RFC1323 for an explanation of the limit to 14
L
Linus Torvalds 已提交
244 245
		 */
		space = max_t(u32, sysctl_tcp_rmem[2], sysctl_rmem_max);
246
		space = min_t(u32, space, *window_clamp);
L
Linus Torvalds 已提交
247 248 249 250 251 252
		while (space > 65535 && (*rcv_wscale) < 14) {
			space >>= 1;
			(*rcv_wscale)++;
		}
	}

253
	if (mss > (1 << *rcv_wscale)) {
254 255 256
		if (!init_rcv_wnd) /* Use default unless specified otherwise */
			init_rcv_wnd = tcp_default_init_rwnd(mss);
		*rcv_wnd = min(*rcv_wnd, init_rcv_wnd * mss);
L
Linus Torvalds 已提交
257 258 259 260 261
	}

	/* Set the clamp no higher than max representable value */
	(*window_clamp) = min(65535U << (*rcv_wscale), *window_clamp);
}
E
Eric Dumazet 已提交
262
EXPORT_SYMBOL(tcp_select_initial_window);
L
Linus Torvalds 已提交
263 264 265 266 267 268

/* 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
269
static u16 tcp_select_window(struct sock *sk)
L
Linus Torvalds 已提交
270 271 272 273 274 275
{
	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
276
	if (new_win < cur_win) {
L
Linus Torvalds 已提交
277 278 279 280 281 282 283
		/* 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.
		 */
284
		new_win = ALIGN(cur_win, 1 << tp->rx_opt.rcv_wscale);
L
Linus Torvalds 已提交
285 286 287 288 289 290 291
	}
	tp->rcv_wnd = new_win;
	tp->rcv_wup = tp->rcv_nxt;

	/* Make sure we do not exceed the maximum possible
	 * scaled window.
	 */
292
	if (!tp->rx_opt.rcv_wscale && sysctl_tcp_workaround_signed_windows)
L
Linus Torvalds 已提交
293 294 295 296 297 298 299 300 301 302 303 304 305 306
		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;
}

307
/* Packet ECN state for a SYN-ACK */
308
static inline void TCP_ECN_send_synack(const struct tcp_sock *tp, struct sk_buff *skb)
309
{
E
Eric Dumazet 已提交
310
	TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_CWR;
311
	if (!(tp->ecn_flags & TCP_ECN_OK))
E
Eric Dumazet 已提交
312
		TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_ECE;
313 314
}

315
/* Packet ECN state for a SYN.  */
316 317 318 319 320
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;
321
	if (sock_net(sk)->ipv4.sysctl_tcp_ecn == 1) {
E
Eric Dumazet 已提交
322
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ECE | TCPHDR_CWR;
323 324 325 326 327
		tp->ecn_flags = TCP_ECN_OK;
	}
}

static __inline__ void
328
TCP_ECN_make_synack(const struct request_sock *req, struct tcphdr *th)
329 330 331 332 333
{
	if (inet_rsk(req)->ecn_ok)
		th->ece = 1;
}

334 335 336
/* Set up ECN state for a packet on a ESTABLISHED socket that is about to
 * be sent.
 */
337 338 339 340 341 342 343 344 345 346
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);
347
			if (tp->ecn_flags & TCP_ECN_QUEUE_CWR) {
348 349 350 351 352 353 354 355 356 357 358 359 360
				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;
	}
}

361 362 363 364 365
/* 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)
{
366
	skb->ip_summed = CHECKSUM_PARTIAL;
367 368
	skb->csum = 0;

E
Eric Dumazet 已提交
369
	TCP_SKB_CB(skb)->tcp_flags = flags;
370 371 372 373 374 375 376
	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
377
	if (flags & (TCPHDR_SYN | TCPHDR_FIN))
378 379 380 381
		seq++;
	TCP_SKB_CB(skb)->end_seq = seq;
}

E
Eric Dumazet 已提交
382
static inline bool tcp_urg_mode(const struct tcp_sock *tp)
I
Ilpo Järvinen 已提交
383 384 385 386
{
	return tp->snd_una != tp->snd_up;
}

A
Adam Langley 已提交
387 388 389
#define OPTION_SACK_ADVERTISE	(1 << 0)
#define OPTION_TS		(1 << 1)
#define OPTION_MD5		(1 << 2)
390
#define OPTION_WSCALE		(1 << 3)
Y
Yuchung Cheng 已提交
391
#define OPTION_FAST_OPEN_COOKIE	(1 << 8)
A
Adam Langley 已提交
392 393

struct tcp_out_options {
Y
Yuchung Cheng 已提交
394 395
	u16 options;		/* bit field of OPTION_* */
	u16 mss;		/* 0 to disable */
A
Adam Langley 已提交
396 397
	u8 ws;			/* window scale, 0 to disable */
	u8 num_sack_blocks;	/* number of SACK blocks to include */
398 399
	u8 hash_size;		/* bytes in hash_location */
	__u8 *hash_location;	/* temporary pointer, overloaded */
Y
Yuchung Cheng 已提交
400 401
	__u32 tsval, tsecr;	/* need to include OPTION_TS */
	struct tcp_fastopen_cookie *fastopen_cookie;	/* Fast open cookie */
A
Adam Langley 已提交
402 403
};

404 405 406
/* Write previously computed TCP options to the packet.
 *
 * Beware: Something in the Internet is very sensitive to the ordering of
407 408 409 410 411 412 413 414 415 416
 * 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
417
static void tcp_options_write(__be32 *ptr, struct tcp_sock *tp,
418 419
			      struct tcp_out_options *opts)
{
Y
Yuchung Cheng 已提交
420
	u16 options = opts->options;	/* mungable copy */
421 422

	if (unlikely(OPTION_MD5 & options)) {
C
Christoph Paasch 已提交
423 424
		*ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
			       (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
425 426
		/* overload cookie hash location */
		opts->hash_location = (__u8 *)ptr;
A
Adam Langley 已提交
427
		ptr += 4;
S
Stephen Hemminger 已提交
428
	}
A
Adam Langley 已提交
429

430 431 432 433 434 435
	if (unlikely(opts->mss)) {
		*ptr++ = htonl((TCPOPT_MSS << 24) |
			       (TCPOLEN_MSS << 16) |
			       opts->mss);
	}

436 437
	if (likely(OPTION_TS & options)) {
		if (unlikely(OPTION_SACK_ADVERTISE & options)) {
A
Adam Langley 已提交
438 439 440 441
			*ptr++ = htonl((TCPOPT_SACK_PERM << 24) |
				       (TCPOLEN_SACK_PERM << 16) |
				       (TCPOPT_TIMESTAMP << 8) |
				       TCPOLEN_TIMESTAMP);
442
			options &= ~OPTION_SACK_ADVERTISE;
A
Adam Langley 已提交
443 444 445 446 447 448 449 450 451 452
		} else {
			*ptr++ = htonl((TCPOPT_NOP << 24) |
				       (TCPOPT_NOP << 16) |
				       (TCPOPT_TIMESTAMP << 8) |
				       TCPOLEN_TIMESTAMP);
		}
		*ptr++ = htonl(opts->tsval);
		*ptr++ = htonl(opts->tsecr);
	}

453
	if (unlikely(OPTION_SACK_ADVERTISE & options)) {
A
Adam Langley 已提交
454 455 456 457 458 459
		*ptr++ = htonl((TCPOPT_NOP << 24) |
			       (TCPOPT_NOP << 16) |
			       (TCPOPT_SACK_PERM << 8) |
			       TCPOLEN_SACK_PERM);
	}

460
	if (unlikely(OPTION_WSCALE & options)) {
A
Adam Langley 已提交
461 462 463 464 465 466 467 468 469
		*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
470 471 472 473 474
		int this_sack;

		*ptr++ = htonl((TCPOPT_NOP  << 24) |
			       (TCPOPT_NOP  << 16) |
			       (TCPOPT_SACK <<  8) |
A
Adam Langley 已提交
475
			       (TCPOLEN_SACK_BASE + (opts->num_sack_blocks *
S
Stephen Hemminger 已提交
476
						     TCPOLEN_SACK_PERBLOCK)));
S
Stephen Hemminger 已提交
477

A
Adam Langley 已提交
478 479
		for (this_sack = 0; this_sack < opts->num_sack_blocks;
		     ++this_sack) {
S
Stephen Hemminger 已提交
480 481 482
			*ptr++ = htonl(sp[this_sack].start_seq);
			*ptr++ = htonl(sp[this_sack].end_seq);
		}
S
Stephen Hemminger 已提交
483

484
		tp->rx_opt.dsack = 0;
S
Stephen Hemminger 已提交
485
	}
Y
Yuchung Cheng 已提交
486 487 488 489 490 491 492 493 494 495 496 497 498 499 500

	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
501 502
}

503 504 505
/* Compute TCP options for SYN packets. This is not the final
 * network wire format yet.
 */
506
static unsigned int tcp_syn_options(struct sock *sk, struct sk_buff *skb,
A
Adam Langley 已提交
507
				struct tcp_out_options *opts,
508 509
				struct tcp_md5sig_key **md5)
{
A
Adam Langley 已提交
510
	struct tcp_sock *tp = tcp_sk(sk);
511
	unsigned int remaining = MAX_TCP_OPTION_SPACE;
512
	struct tcp_fastopen_request *fastopen = tp->fastopen_req;
A
Adam Langley 已提交
513

514
#ifdef CONFIG_TCP_MD5SIG
A
Adam Langley 已提交
515 516 517
	*md5 = tp->af_specific->md5_lookup(sk, sk);
	if (*md5) {
		opts->options |= OPTION_MD5;
518
		remaining -= TCPOLEN_MD5SIG_ALIGNED;
519
	}
A
Adam Langley 已提交
520 521
#else
	*md5 = NULL;
522
#endif
A
Adam Langley 已提交
523 524 525 526 527 528 529 530 531 532 533

	/* 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);
534
	remaining -= TCPOLEN_MSS_ALIGNED;
A
Adam Langley 已提交
535

536
	if (likely(sysctl_tcp_timestamps && *md5 == NULL)) {
A
Adam Langley 已提交
537
		opts->options |= OPTION_TS;
538
		opts->tsval = TCP_SKB_CB(skb)->when + tp->tsoffset;
A
Adam Langley 已提交
539
		opts->tsecr = tp->rx_opt.ts_recent;
540
		remaining -= TCPOLEN_TSTAMP_ALIGNED;
A
Adam Langley 已提交
541
	}
542
	if (likely(sysctl_tcp_window_scaling)) {
A
Adam Langley 已提交
543
		opts->ws = tp->rx_opt.rcv_wscale;
544
		opts->options |= OPTION_WSCALE;
545
		remaining -= TCPOLEN_WSCALE_ALIGNED;
A
Adam Langley 已提交
546
	}
547
	if (likely(sysctl_tcp_sack)) {
A
Adam Langley 已提交
548
		opts->options |= OPTION_SACK_ADVERTISE;
549
		if (unlikely(!(OPTION_TS & opts->options)))
550
			remaining -= TCPOLEN_SACKPERM_ALIGNED;
A
Adam Langley 已提交
551 552
	}

553 554 555 556 557 558 559 560 561 562
	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;
		}
	}
563 564

	return MAX_TCP_OPTION_SPACE - remaining;
S
Stephen Hemminger 已提交
565 566
}

567
/* Set up TCP options for SYN-ACKs. */
568
static unsigned int tcp_synack_options(struct sock *sk,
A
Adam Langley 已提交
569
				   struct request_sock *req,
570
				   unsigned int mss, struct sk_buff *skb,
A
Adam Langley 已提交
571
				   struct tcp_out_options *opts,
572
				   struct tcp_md5sig_key **md5,
573
				   struct tcp_fastopen_cookie *foc)
574
{
A
Adam Langley 已提交
575
	struct inet_request_sock *ireq = inet_rsk(req);
576
	unsigned int remaining = MAX_TCP_OPTION_SPACE;
A
Adam Langley 已提交
577

578
#ifdef CONFIG_TCP_MD5SIG
A
Adam Langley 已提交
579 580 581
	*md5 = tcp_rsk(req)->af_specific->md5_lookup(sk, req);
	if (*md5) {
		opts->options |= OPTION_MD5;
582 583 584 585 586 587 588
		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
589
		ireq->tstamp_ok &= !ireq->sack_ok;
590
	}
A
Adam Langley 已提交
591 592
#else
	*md5 = NULL;
593
#endif
A
Adam Langley 已提交
594

595
	/* We always send an MSS option. */
A
Adam Langley 已提交
596
	opts->mss = mss;
597
	remaining -= TCPOLEN_MSS_ALIGNED;
A
Adam Langley 已提交
598 599 600

	if (likely(ireq->wscale_ok)) {
		opts->ws = ireq->rcv_wscale;
601
		opts->options |= OPTION_WSCALE;
602
		remaining -= TCPOLEN_WSCALE_ALIGNED;
A
Adam Langley 已提交
603
	}
E
Eric Dumazet 已提交
604
	if (likely(ireq->tstamp_ok)) {
A
Adam Langley 已提交
605 606 607
		opts->options |= OPTION_TS;
		opts->tsval = TCP_SKB_CB(skb)->when;
		opts->tsecr = req->ts_recent;
608
		remaining -= TCPOLEN_TSTAMP_ALIGNED;
A
Adam Langley 已提交
609 610 611
	}
	if (likely(ireq->sack_ok)) {
		opts->options |= OPTION_SACK_ADVERTISE;
E
Eric Dumazet 已提交
612
		if (unlikely(!ireq->tstamp_ok))
613
			remaining -= TCPOLEN_SACKPERM_ALIGNED;
A
Adam Langley 已提交
614
	}
615 616 617 618 619 620 621 622 623
	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;
		}
	}
C
Christoph Paasch 已提交
624

625
	return MAX_TCP_OPTION_SPACE - remaining;
A
Adam Langley 已提交
626 627
}

628 629 630
/* Compute TCP options for ESTABLISHED sockets. This is not the
 * final wire format yet.
 */
631
static unsigned int tcp_established_options(struct sock *sk, struct sk_buff *skb,
A
Adam Langley 已提交
632
					struct tcp_out_options *opts,
633 634
					struct tcp_md5sig_key **md5)
{
A
Adam Langley 已提交
635 636
	struct tcp_skb_cb *tcb = skb ? TCP_SKB_CB(skb) : NULL;
	struct tcp_sock *tp = tcp_sk(sk);
637
	unsigned int size = 0;
638
	unsigned int eff_sacks;
A
Adam Langley 已提交
639 640 641 642 643 644 645 646 647 648 649 650 651

#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;
652
		opts->tsval = tcb ? tcb->when + tp->tsoffset : 0;
A
Adam Langley 已提交
653 654 655 656
		opts->tsecr = tp->rx_opt.ts_recent;
		size += TCPOLEN_TSTAMP_ALIGNED;
	}

657 658
	eff_sacks = tp->rx_opt.num_sacks + tp->rx_opt.dsack;
	if (unlikely(eff_sacks)) {
659
		const unsigned int remaining = MAX_TCP_OPTION_SPACE - size;
A
Adam Langley 已提交
660
		opts->num_sack_blocks =
661
			min_t(unsigned int, eff_sacks,
A
Adam Langley 已提交
662 663 664 665 666 667 668
			      (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
669
}
L
Linus Torvalds 已提交
670

E
Eric Dumazet 已提交
671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691

/* 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);

692 693 694 695 696 697 698
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
699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725
/*
 * 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)) {
726
			tcp_tsq_handler(sk);
E
Eric Dumazet 已提交
727 728
		} else {
			/* defer the work to tcp_release_cb() */
729
			set_bit(TCP_TSQ_DEFERRED, &tp->tsq_flags);
E
Eric Dumazet 已提交
730 731 732 733 734 735 736 737
		}
		bh_unlock_sock(sk);

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

738 739
#define TCP_DEFERRED_ALL ((1UL << TCP_TSQ_DEFERRED) |		\
			  (1UL << TCP_WRITE_TIMER_DEFERRED) |	\
740 741
			  (1UL << TCP_DELACK_TIMER_DEFERRED) |	\
			  (1UL << TCP_MTU_REDUCED_DEFERRED))
E
Eric Dumazet 已提交
742 743 744 745 746 747 748 749 750 751
/**
 * 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);
752
	unsigned long flags, nflags;
E
Eric Dumazet 已提交
753

754 755 756 757 758 759 760 761 762 763 764
	/* 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);

765
	if (flags & (1UL << TCP_WRITE_TIMER_DEFERRED)) {
766
		tcp_write_timer_handler(sk);
767 768 769
		__sock_put(sk);
	}
	if (flags & (1UL << TCP_DELACK_TIMER_DEFERRED)) {
770
		tcp_delack_timer_handler(sk);
771 772 773
		__sock_put(sk);
	}
	if (flags & (1UL << TCP_MTU_REDUCED_DEFERRED)) {
774
		sk->sk_prot->mtu_reduced(sk);
775 776
		__sock_put(sk);
	}
E
Eric Dumazet 已提交
777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798
}
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.
 */
E
Eric Dumazet 已提交
799
void tcp_wfree(struct sk_buff *skb)
E
Eric Dumazet 已提交
800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824
{
	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
825 826 827 828 829 830 831 832 833 834 835
/* 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.
 */
836 837
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
838
{
839 840 841 842
	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
843
	struct tcp_out_options opts;
844
	unsigned int tcp_options_size, tcp_header_size;
845
	struct tcp_md5sig_key *md5;
846 847 848 849 850 851 852 853
	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.
	 */
854
	if (icsk->icsk_ca_ops->flags & TCP_CONG_RTT_STAMP)
855 856 857
		__net_timestamp(skb);

	if (likely(clone_it)) {
858 859 860 861 862 863 864
		const struct sk_buff *fclone = skb + 1;

		if (unlikely(skb->fclone == SKB_FCLONE_ORIG &&
			     fclone->fclone == SKB_FCLONE_CLONE))
			NET_INC_STATS_BH(sock_net(sk),
					 LINUX_MIB_TCPSPURIOUS_RTX_HOSTQUEUES);

865 866 867 868 869 870 871
		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
872

873 874 875
	inet = inet_sk(sk);
	tp = tcp_sk(sk);
	tcb = TCP_SKB_CB(skb);
A
Adam Langley 已提交
876
	memset(&opts, 0, sizeof(opts));
L
Linus Torvalds 已提交
877

E
Eric Dumazet 已提交
878
	if (unlikely(tcb->tcp_flags & TCPHDR_SYN))
A
Adam Langley 已提交
879 880 881 882 883
		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);
884

E
Eric Dumazet 已提交
885
	if (tcp_packets_in_flight(tp) == 0)
886
		tcp_ca_event(sk, CA_EVENT_TX_START);
E
Eric Dumazet 已提交
887 888 889 890 891

	/* if no packet is in qdisc/device queue, then allow XPS to select
	 * another queue.
	 */
	skb->ooo_okay = sk_wmem_alloc_get(sk) == 0;
892

893 894
	skb_push(skb, tcp_header_size);
	skb_reset_transport_header(skb);
E
Eric Dumazet 已提交
895 896 897

	skb_orphan(skb);
	skb->sk = sk;
E
Eric Dumazet 已提交
898
	skb->destructor = tcp_wfree;
E
Eric Dumazet 已提交
899
	atomic_add(skb->truesize, &sk->sk_wmem_alloc);
900 901

	/* Build TCP header and checksum it. */
902
	th = tcp_hdr(skb);
E
Eric Dumazet 已提交
903 904
	th->source		= inet->inet_sport;
	th->dest		= inet->inet_dport;
905 906
	th->seq			= htonl(tcb->seq);
	th->ack_seq		= htonl(tp->rcv_nxt);
907
	*(((__be16 *)th) + 6)	= htons(((tcp_header_size >> 2) << 12) |
E
Eric Dumazet 已提交
908
					tcb->tcp_flags);
909

E
Eric Dumazet 已提交
910
	if (unlikely(tcb->tcp_flags & TCPHDR_SYN)) {
911 912 913
		/* RFC1323: The window in SYN & SYN/ACK segments
		 * is never scaled.
		 */
914
		th->window	= htons(min(tp->rcv_wnd, 65535U));
915 916 917 918 919
	} else {
		th->window	= htons(tcp_select_window(sk));
	}
	th->check		= 0;
	th->urg_ptr		= 0;
L
Linus Torvalds 已提交
920

I
Ilpo Järvinen 已提交
921
	/* The urg_mode check is necessary during a below snd_una win probe */
922 923 924 925 926
	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)) {
927
			th->urg_ptr = htons(0xFFFF);
928 929
			th->urg = 1;
		}
930
	}
L
Linus Torvalds 已提交
931

932
	tcp_options_write((__be32 *)(th + 1), tp, &opts);
E
Eric Dumazet 已提交
933
	if (likely((tcb->tcp_flags & TCPHDR_SYN) == 0))
934
		TCP_ECN_send(sk, skb, tcp_header_size);
L
Linus Torvalds 已提交
935

936 937 938
#ifdef CONFIG_TCP_MD5SIG
	/* Calculate the MD5 hash, as we have all we need now */
	if (md5) {
E
Eric Dumazet 已提交
939
		sk_nocaps_add(sk, NETIF_F_GSO_MASK);
940
		tp->af_specific->calc_md5_hash(opts.hash_location,
941
					       md5, sk, NULL, skb);
942 943 944
	}
#endif

945
	icsk->icsk_af_ops->send_check(sk, skb);
L
Linus Torvalds 已提交
946

E
Eric Dumazet 已提交
947
	if (likely(tcb->tcp_flags & TCPHDR_ACK))
948
		tcp_event_ack_sent(sk, tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
949

950
	if (skb->len != tcp_header_size)
951
		tcp_event_data_sent(tp, sk);
L
Linus Torvalds 已提交
952

953
	if (after(tcb->end_seq, tp->snd_nxt) || tcb->seq == tcb->end_seq)
954 955
		TCP_ADD_STATS(sock_net(sk), TCP_MIB_OUTSEGS,
			      tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
956

957
	err = icsk->icsk_af_ops->queue_xmit(skb, &inet->cork.fl);
958
	if (likely(err <= 0))
959 960
		return err;

961
	tcp_enter_cwr(sk, 1);
962

963
	return net_xmit_eval(err);
L
Linus Torvalds 已提交
964 965
}

966
/* This routine just queues the buffer for sending.
L
Linus Torvalds 已提交
967 968 969 970 971 972 973 974 975 976 977
 *
 * 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);
978
	tcp_add_write_queue_tail(sk, skb);
979 980
	sk->sk_wmem_queued += skb->truesize;
	sk_mem_charge(sk, skb->truesize);
L
Linus Torvalds 已提交
981 982
}

983
/* Initialize TSO segments for a packet. */
984
static void tcp_set_skb_tso_segs(const struct sock *sk, struct sk_buff *skb,
985
				 unsigned int mss_now)
986
{
987 988
	if (skb->len <= mss_now || !sk_can_gso(sk) ||
	    skb->ip_summed == CHECKSUM_NONE) {
989 990 991
		/* Avoid the costly divide in the normal
		 * non-TSO case.
		 */
992 993
		skb_shinfo(skb)->gso_segs = 1;
		skb_shinfo(skb)->gso_size = 0;
994
		skb_shinfo(skb)->gso_type = 0;
995
	} else {
I
Ilpo Järvinen 已提交
996
		skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss_now);
997
		skb_shinfo(skb)->gso_size = mss_now;
998
		skb_shinfo(skb)->gso_type = sk->sk_gso_type;
L
Linus Torvalds 已提交
999 1000 1001
	}
}

I
Ilpo Järvinen 已提交
1002
/* When a modification to fackets out becomes necessary, we need to check
1003
 * skb is counted to fackets_out or not.
I
Ilpo Järvinen 已提交
1004
 */
1005
static void tcp_adjust_fackets_out(struct sock *sk, const struct sk_buff *skb,
I
Ilpo Järvinen 已提交
1006 1007
				   int decr)
{
1008 1009
	struct tcp_sock *tp = tcp_sk(sk);

1010
	if (!tp->sacked_out || tcp_is_reno(tp))
I
Ilpo Järvinen 已提交
1011 1012
		return;

1013
	if (after(tcp_highest_sack_seq(tp), TCP_SKB_CB(skb)->seq))
I
Ilpo Järvinen 已提交
1014 1015 1016
		tp->fackets_out -= decr;
}

1017 1018 1019
/* Pcount in the middle of the write queue got changed, we need to do various
 * tweaks to fix counters
 */
1020
static void tcp_adjust_pcount(struct sock *sk, const struct sk_buff *skb, int decr)
1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040
{
	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) &&
1041
	    (tcp_is_fack(tp) || (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)))
1042 1043 1044 1045 1046
		tp->lost_cnt_hint -= decr;

	tcp_verify_left_out(tp);
}

L
Linus Torvalds 已提交
1047 1048
/* 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
1049
 * packet to the list.  This won't be called frequently, I hope.
L
Linus Torvalds 已提交
1050 1051
 * Remember, these are still headerless SKBs at this point.
 */
1052 1053
int tcp_fragment(struct sock *sk, struct sk_buff *skb, u32 len,
		 unsigned int mss_now)
L
Linus Torvalds 已提交
1054 1055 1056
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *buff;
1057
	int nsize, old_factor;
1058
	int nlen;
1059
	u8 flags;
L
Linus Torvalds 已提交
1060

1061 1062
	if (WARN_ON(len > skb->len))
		return -EINVAL;
1063

L
Linus Torvalds 已提交
1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076
	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
1077

1078 1079
	sk->sk_wmem_queued += buff->truesize;
	sk_mem_charge(sk, buff->truesize);
1080 1081 1082
	nlen = skb->len - len - nsize;
	buff->truesize += nlen;
	skb->truesize -= nlen;
L
Linus Torvalds 已提交
1083 1084 1085 1086 1087 1088 1089

	/* 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
1090 1091 1092
	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;
1093
	TCP_SKB_CB(buff)->sacked = TCP_SKB_CB(skb)->sacked;
L
Linus Torvalds 已提交
1094

1095
	if (!skb_shinfo(skb)->nr_frags && skb->ip_summed != CHECKSUM_PARTIAL) {
L
Linus Torvalds 已提交
1096
		/* Copy and checksum data tail into the new buffer. */
1097 1098
		buff->csum = csum_partial_copy_nocheck(skb->data + len,
						       skb_put(buff, nsize),
L
Linus Torvalds 已提交
1099 1100 1101 1102 1103 1104
						       nsize, 0);

		skb_trim(skb, len);

		skb->csum = csum_block_sub(skb->csum, buff->csum, len);
	} else {
1105
		skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
1106 1107 1108 1109 1110 1111 1112 1113 1114
		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;
1115
	buff->tstamp = skb->tstamp;
L
Linus Torvalds 已提交
1116

1117 1118
	old_factor = tcp_skb_pcount(skb);

L
Linus Torvalds 已提交
1119
	/* Fix up tso_factor for both original and new SKB.  */
1120 1121
	tcp_set_skb_tso_segs(sk, skb, mss_now);
	tcp_set_skb_tso_segs(sk, buff, mss_now);
L
Linus Torvalds 已提交
1122

1123 1124 1125
	/* If this packet has been sent out already, we must
	 * adjust the various packet counters.
	 */
1126
	if (!before(tp->snd_nxt, TCP_SKB_CB(buff)->end_seq)) {
1127 1128
		int diff = old_factor - tcp_skb_pcount(skb) -
			tcp_skb_pcount(buff);
L
Linus Torvalds 已提交
1129

1130 1131
		if (diff)
			tcp_adjust_pcount(sk, skb, diff);
L
Linus Torvalds 已提交
1132 1133 1134
	}

	/* Link BUFF into the send queue. */
1135
	skb_header_release(buff);
1136
	tcp_insert_write_queue_after(skb, buff, sk);
L
Linus Torvalds 已提交
1137 1138 1139 1140 1141 1142 1143 1144

	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.
 */
1145
static void __pskb_trim_head(struct sk_buff *skb, int len)
L
Linus Torvalds 已提交
1146 1147 1148
{
	int i, k, eat;

E
Eric Dumazet 已提交
1149 1150 1151 1152 1153 1154 1155
	eat = min_t(int, len, skb_headlen(skb));
	if (eat) {
		__skb_pull(skb, eat);
		len -= eat;
		if (!len)
			return;
	}
L
Linus Torvalds 已提交
1156 1157
	eat = len;
	k = 0;
1158
	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
E
Eric Dumazet 已提交
1159 1160 1161
		int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);

		if (size <= eat) {
1162
			skb_frag_unref(skb, i);
E
Eric Dumazet 已提交
1163
			eat -= size;
L
Linus Torvalds 已提交
1164 1165 1166 1167
		} 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
1168
				skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
L
Linus Torvalds 已提交
1169 1170 1171 1172 1173 1174 1175
				eat = 0;
			}
			k++;
		}
	}
	skb_shinfo(skb)->nr_frags = k;

1176
	skb_reset_tail_pointer(skb);
L
Linus Torvalds 已提交
1177 1178 1179 1180
	skb->data_len -= len;
	skb->len = skb->data_len;
}

1181
/* Remove acked data from a packet in the transmit queue. */
L
Linus Torvalds 已提交
1182 1183
int tcp_trim_head(struct sock *sk, struct sk_buff *skb, u32 len)
{
1184
	if (skb_unclone(skb, GFP_ATOMIC))
L
Linus Torvalds 已提交
1185 1186
		return -ENOMEM;

E
Eric Dumazet 已提交
1187
	__pskb_trim_head(skb, len);
L
Linus Torvalds 已提交
1188 1189

	TCP_SKB_CB(skb)->seq += len;
1190
	skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
1191 1192 1193

	skb->truesize	     -= len;
	sk->sk_wmem_queued   -= len;
1194
	sk_mem_uncharge(sk, len);
L
Linus Torvalds 已提交
1195 1196
	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);

1197
	/* Any change of skb->len requires recalculation of tso factor. */
L
Linus Torvalds 已提交
1198
	if (tcp_skb_pcount(skb) > 1)
1199
		tcp_set_skb_tso_segs(sk, skb, tcp_skb_mss(skb));
L
Linus Torvalds 已提交
1200 1201 1202 1203

	return 0;
}

1204 1205
/* 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
1206
{
1207 1208
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1209 1210 1211 1212 1213 1214 1215
	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);

1216 1217 1218 1219 1220 1221 1222 1223
	/* 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
1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236
	/* 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;
}

1237 1238 1239 1240 1241 1242 1243 1244
/* 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
1245
/* Inverse of above */
1246
int tcp_mss_to_mtu(struct sock *sk, int mss)
J
John Heffner 已提交
1247
{
1248 1249
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1250 1251 1252 1253 1254 1255 1256
	int mtu;

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

1257 1258 1259 1260 1261 1262 1263
	/* 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
1264 1265 1266
	return mtu;
}

1267
/* MTU probing init per socket */
J
John Heffner 已提交
1268 1269 1270 1271 1272 1273 1274
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) +
1275
			       icsk->icsk_af_ops->net_header_len;
J
John Heffner 已提交
1276 1277 1278
	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
1279
EXPORT_SYMBOL(tcp_mtup_init);
J
John Heffner 已提交
1280

L
Linus Torvalds 已提交
1281 1282 1283 1284 1285 1286
/* 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
1287
   It is minimum of user_mss and mss received with SYN.
L
Linus Torvalds 已提交
1288 1289
   It also does not include TCP options.

1290
   inet_csk(sk)->icsk_pmtu_cookie is last pmtu, seen by this function.
L
Linus Torvalds 已提交
1291 1292 1293 1294 1295 1296 1297 1298 1299

   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.

1300 1301
   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
1302 1303 1304 1305
 */
unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu)
{
	struct tcp_sock *tp = tcp_sk(sk);
1306
	struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1307
	int mss_now;
L
Linus Torvalds 已提交
1308

J
John Heffner 已提交
1309 1310
	if (icsk->icsk_mtup.search_high > pmtu)
		icsk->icsk_mtup.search_high = pmtu;
L
Linus Torvalds 已提交
1311

J
John Heffner 已提交
1312
	mss_now = tcp_mtu_to_mss(sk, pmtu);
1313
	mss_now = tcp_bound_to_half_wnd(tp, mss_now);
L
Linus Torvalds 已提交
1314 1315

	/* And store cached results */
1316
	icsk->icsk_pmtu_cookie = pmtu;
J
John Heffner 已提交
1317 1318
	if (icsk->icsk_mtup.enabled)
		mss_now = min(mss_now, tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low));
1319
	tp->mss_cache = mss_now;
L
Linus Torvalds 已提交
1320 1321 1322

	return mss_now;
}
E
Eric Dumazet 已提交
1323
EXPORT_SYMBOL(tcp_sync_mss);
L
Linus Torvalds 已提交
1324 1325 1326 1327

/* Compute the current effective MSS, taking SACKs and IP options,
 * and even PMTU discovery events into account.
 */
I
Ilpo Järvinen 已提交
1328
unsigned int tcp_current_mss(struct sock *sk)
L
Linus Torvalds 已提交
1329
{
1330 1331
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct dst_entry *dst = __sk_dst_get(sk);
1332
	u32 mss_now;
1333
	unsigned int header_len;
A
Adam Langley 已提交
1334 1335
	struct tcp_out_options opts;
	struct tcp_md5sig_key *md5;
1336 1337 1338

	mss_now = tp->mss_cache;

L
Linus Torvalds 已提交
1339 1340
	if (dst) {
		u32 mtu = dst_mtu(dst);
1341
		if (mtu != inet_csk(sk)->icsk_pmtu_cookie)
L
Linus Torvalds 已提交
1342 1343 1344
			mss_now = tcp_sync_mss(sk, mtu);
	}

A
Adam Langley 已提交
1345 1346 1347 1348 1349 1350 1351 1352 1353 1354
	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;
	}
1355

L
Linus Torvalds 已提交
1356 1357 1358
	return mss_now;
}

1359
/* Congestion window validation. (RFC2861) */
1360
static void tcp_cwnd_validate(struct sock *sk)
1361
{
1362
	struct tcp_sock *tp = tcp_sk(sk);
1363

1364
	if (tp->packets_out >= tp->snd_cwnd) {
1365 1366 1367 1368 1369 1370 1371 1372
		/* 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;

1373 1374
		if (sysctl_tcp_slow_start_after_idle &&
		    (s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= inet_csk(sk)->icsk_rto)
1375 1376 1377 1378
			tcp_cwnd_application_limited(sk);
	}
}

1379 1380 1381 1382
/* 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).
1383 1384 1385 1386 1387 1388 1389
 *
 * 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.
1390
 */
1391
static unsigned int tcp_mss_split_point(const struct sock *sk, const struct sk_buff *skb,
1392
					unsigned int mss_now, unsigned int max_segs)
1393
{
1394
	const struct tcp_sock *tp = tcp_sk(sk);
1395
	u32 needed, window, max_len;
1396

1397
	window = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
1398
	max_len = mss_now * max_segs;
1399

1400 1401
	if (likely(max_len <= window && skb != tcp_write_queue_tail(sk)))
		return max_len;
1402

1403 1404
	needed = min(skb->len, window);

1405 1406
	if (max_len <= needed)
		return max_len;
1407 1408

	return needed - needed % mss_now;
1409 1410 1411 1412 1413
}

/* 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.
 */
1414 1415
static inline unsigned int tcp_cwnd_test(const struct tcp_sock *tp,
					 const struct sk_buff *skb)
1416 1417 1418 1419
{
	u32 in_flight, cwnd;

	/* Don't be strict about the congestion window for the final FIN.  */
E
Eric Dumazet 已提交
1420 1421
	if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) &&
	    tcp_skb_pcount(skb) == 1)
1422 1423 1424 1425 1426 1427 1428 1429 1430 1431
		return 1;

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

	return 0;
}

1432
/* Initialize TSO state of a skb.
1433
 * This must be invoked the first time we consider transmitting
1434 1435
 * SKB onto the wire.
 */
1436
static int tcp_init_tso_segs(const struct sock *sk, struct sk_buff *skb,
1437
			     unsigned int mss_now)
1438 1439 1440
{
	int tso_segs = tcp_skb_pcount(skb);

1441
	if (!tso_segs || (tso_segs > 1 && tcp_skb_mss(skb) != mss_now)) {
1442
		tcp_set_skb_tso_segs(sk, skb, mss_now);
1443 1444 1445 1446 1447
		tso_segs = tcp_skb_pcount(skb);
	}
	return tso_segs;
}

1448
/* Minshall's variant of the Nagle send check. */
E
Eric Dumazet 已提交
1449
static inline bool tcp_minshall_check(const struct tcp_sock *tp)
1450
{
1451
	return after(tp->snd_sml, tp->snd_una) &&
1452 1453 1454
		!after(tp->snd_sml, tp->snd_nxt);
}

E
Eric Dumazet 已提交
1455
/* Return false, if packet can be sent now without violation Nagle's rules:
1456 1457
 * 1. It is full sized.
 * 2. Or it contains FIN. (already checked by caller)
F
Feng King 已提交
1458
 * 3. Or TCP_CORK is not set, and TCP_NODELAY is set.
1459 1460 1461
 * 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
1462
static inline bool tcp_nagle_check(const struct tcp_sock *tp,
1463
				  const struct sk_buff *skb,
1464
				  unsigned int mss_now, int nonagle)
1465
{
E
Eric Dumazet 已提交
1466
	return skb->len < mss_now &&
1467
		((nonagle & TCP_NAGLE_CORK) ||
E
Eric Dumazet 已提交
1468
		 (!nonagle && tp->packets_out && tcp_minshall_check(tp)));
1469 1470
}

E
Eric Dumazet 已提交
1471
/* Return true if the Nagle test allows this packet to be
1472 1473
 * sent now.
 */
E
Eric Dumazet 已提交
1474 1475
static inline bool tcp_nagle_test(const struct tcp_sock *tp, const struct sk_buff *skb,
				  unsigned int cur_mss, int nonagle)
1476 1477 1478 1479 1480 1481 1482 1483
{
	/* 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
1484
		return true;
1485

Y
Yuchung Cheng 已提交
1486 1487
	/* Don't use the nagle rule for urgent data (or for the final FIN). */
	if (tcp_urg_mode(tp) || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
E
Eric Dumazet 已提交
1488
		return true;
1489 1490

	if (!tcp_nagle_check(tp, skb, cur_mss, nonagle))
E
Eric Dumazet 已提交
1491
		return true;
1492

E
Eric Dumazet 已提交
1493
	return false;
1494 1495 1496
}

/* Does at least the first segment of SKB fit into the send window? */
E
Eric Dumazet 已提交
1497 1498 1499
static bool tcp_snd_wnd_test(const struct tcp_sock *tp,
			     const struct sk_buff *skb,
			     unsigned int cur_mss)
1500 1501 1502 1503 1504 1505
{
	u32 end_seq = TCP_SKB_CB(skb)->end_seq;

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

1506
	return !after(end_seq, tcp_wnd_end(tp));
1507 1508
}

1509
/* This checks if the data bearing packet SKB (usually tcp_send_head(sk))
1510 1511 1512
 * should be put on the wire right now.  If so, it returns the number of
 * packets allowed by the congestion window.
 */
1513
static unsigned int tcp_snd_test(const struct sock *sk, struct sk_buff *skb,
1514 1515
				 unsigned int cur_mss, int nonagle)
{
1516
	const struct tcp_sock *tp = tcp_sk(sk);
1517 1518
	unsigned int cwnd_quota;

1519
	tcp_init_tso_segs(sk, skb, cur_mss);
1520 1521 1522 1523 1524

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

	cwnd_quota = tcp_cwnd_test(tp, skb);
1525
	if (cwnd_quota && !tcp_snd_wnd_test(tp, skb, cur_mss))
1526 1527 1528 1529 1530
		cwnd_quota = 0;

	return cwnd_quota;
}

1531
/* Test if sending is allowed right now. */
E
Eric Dumazet 已提交
1532
bool tcp_may_send_now(struct sock *sk)
1533
{
1534
	const struct tcp_sock *tp = tcp_sk(sk);
1535
	struct sk_buff *skb = tcp_send_head(sk);
1536

E
Eric Dumazet 已提交
1537
	return skb &&
I
Ilpo Järvinen 已提交
1538
		tcp_snd_test(sk, skb, tcp_current_mss(sk),
1539
			     (tcp_skb_is_last(sk, skb) ?
E
Eric Dumazet 已提交
1540
			      tp->nonagle : TCP_NAGLE_PUSH));
1541 1542 1543 1544 1545 1546 1547 1548 1549
}

/* 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).
 */
1550
static int tso_fragment(struct sock *sk, struct sk_buff *skb, unsigned int len,
1551
			unsigned int mss_now, gfp_t gfp)
1552 1553 1554
{
	struct sk_buff *buff;
	int nlen = skb->len - len;
1555
	u8 flags;
1556 1557

	/* All of a TSO frame must be composed of paged data.  */
1558 1559
	if (skb->len != skb->data_len)
		return tcp_fragment(sk, skb, len, mss_now);
1560

1561
	buff = sk_stream_alloc_skb(sk, 0, gfp);
1562 1563 1564
	if (unlikely(buff == NULL))
		return -ENOMEM;

1565 1566
	sk->sk_wmem_queued += buff->truesize;
	sk_mem_charge(sk, buff->truesize);
1567
	buff->truesize += nlen;
1568 1569 1570 1571 1572 1573 1574 1575
	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
1576 1577 1578
	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;
1579 1580 1581 1582

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

1583
	buff->ip_summed = skb->ip_summed = CHECKSUM_PARTIAL;
1584 1585 1586
	skb_split(skb, buff, len);

	/* Fix up tso_factor for both original and new SKB.  */
1587 1588
	tcp_set_skb_tso_segs(sk, skb, mss_now);
	tcp_set_skb_tso_segs(sk, buff, mss_now);
1589 1590 1591

	/* Link BUFF into the send queue. */
	skb_header_release(buff);
1592
	tcp_insert_write_queue_after(skb, buff, sk);
1593 1594 1595 1596 1597 1598 1599 1600 1601

	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
1602
static bool tcp_tso_should_defer(struct sock *sk, struct sk_buff *skb)
1603
{
1604
	struct tcp_sock *tp = tcp_sk(sk);
1605
	const struct inet_connection_sock *icsk = inet_csk(sk);
1606
	u32 send_win, cong_win, limit, in_flight;
1607
	int win_divisor;
1608

E
Eric Dumazet 已提交
1609
	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
J
John Heffner 已提交
1610
		goto send_now;
1611

1612
	if (icsk->icsk_ca_state != TCP_CA_Open)
J
John Heffner 已提交
1613 1614 1615
		goto send_now;

	/* Defer for less than two clock ticks. */
I
Ilpo Järvinen 已提交
1616
	if (tp->tso_deferred &&
1617
	    (((u32)jiffies << 1) >> 1) - (tp->tso_deferred >> 1) > 1)
J
John Heffner 已提交
1618
		goto send_now;
1619

1620 1621
	in_flight = tcp_packets_in_flight(tp);

1622
	BUG_ON(tcp_skb_pcount(skb) <= 1 || (tp->snd_cwnd <= in_flight));
1623

1624
	send_win = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
1625 1626 1627 1628 1629 1630

	/* 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);

1631
	/* If a full-sized TSO skb can be sent, do it. */
1632
	if (limit >= min_t(unsigned int, sk->sk_gso_max_size,
E
Eric Dumazet 已提交
1633
			   tp->xmit_size_goal_segs * tp->mss_cache))
J
John Heffner 已提交
1634
		goto send_now;
1635

1636 1637 1638 1639
	/* 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;

1640 1641
	win_divisor = ACCESS_ONCE(sysctl_tcp_tso_win_divisor);
	if (win_divisor) {
1642 1643 1644 1645 1646
		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.
		 */
1647
		chunk /= win_divisor;
1648
		if (limit >= chunk)
J
John Heffner 已提交
1649
			goto send_now;
1650 1651 1652 1653 1654 1655
	} 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.
		 */
1656
		if (limit > tcp_max_tso_deferred_mss(tp) * tp->mss_cache)
J
John Heffner 已提交
1657
			goto send_now;
1658 1659
	}

E
Eric Dumazet 已提交
1660 1661 1662 1663 1664
	/* Ok, it looks like it is advisable to defer.
	 * Do not rearm the timer if already set to not break TCP ACK clocking.
	 */
	if (!tp->tso_deferred)
		tp->tso_deferred = 1 | (jiffies << 1);
J
John Heffner 已提交
1665

E
Eric Dumazet 已提交
1666
	return true;
J
John Heffner 已提交
1667 1668 1669

send_now:
	tp->tso_deferred = 0;
E
Eric Dumazet 已提交
1670
	return false;
1671 1672
}

J
John Heffner 已提交
1673
/* Create a new MTU probe if we are ready.
1674 1675 1676 1677
 * 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
1678 1679
 * Returns 0 if we should wait to probe (no cwnd available),
 *         1 if a probe was sent,
1680 1681
 *         -1 otherwise
 */
J
John Heffner 已提交
1682 1683 1684 1685 1686 1687 1688
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;
1689
	int size_needed;
J
John Heffner 已提交
1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700
	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 ||
1701
	    tp->rx_opt.num_sacks || tp->rx_opt.dsack)
J
John Heffner 已提交
1702 1703 1704
		return -1;

	/* Very simple search strategy: just double the MSS. */
I
Ilpo Järvinen 已提交
1705
	mss_now = tcp_current_mss(sk);
1706
	probe_size = 2 * tp->mss_cache;
1707
	size_needed = probe_size + (tp->reordering + 1) * tp->mss_cache;
J
John Heffner 已提交
1708 1709 1710 1711 1712 1713
	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? */
1714
	if (tp->write_seq - tp->snd_nxt < size_needed)
J
John Heffner 已提交
1715 1716
		return -1;

1717 1718
	if (tp->snd_wnd < size_needed)
		return -1;
1719
	if (after(tp->snd_nxt + size_needed, tcp_wnd_end(tp)))
1720
		return 0;
J
John Heffner 已提交
1721

1722 1723 1724
	/* 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
1725 1726 1727 1728 1729 1730 1731 1732
			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;
1733 1734
	sk->sk_wmem_queued += nskb->truesize;
	sk_mem_charge(sk, nskb->truesize);
J
John Heffner 已提交
1735

1736
	skb = tcp_send_head(sk);
J
John Heffner 已提交
1737 1738 1739

	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
1740
	TCP_SKB_CB(nskb)->tcp_flags = TCPHDR_ACK;
J
John Heffner 已提交
1741 1742
	TCP_SKB_CB(nskb)->sacked = 0;
	nskb->csum = 0;
1743
	nskb->ip_summed = skb->ip_summed;
J
John Heffner 已提交
1744

1745 1746
	tcp_insert_write_queue_before(nskb, skb, sk);

J
John Heffner 已提交
1747
	len = 0;
1748
	tcp_for_write_queue_from_safe(skb, next, sk) {
J
John Heffner 已提交
1749 1750 1751 1752 1753
		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,
1754 1755
							    skb_put(nskb, copy),
							    copy, nskb->csum);
J
John Heffner 已提交
1756 1757 1758 1759

		if (skb->len <= copy) {
			/* We've eaten all the data from this skb.
			 * Throw it away. */
E
Eric Dumazet 已提交
1760
			TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;
1761
			tcp_unlink_write_queue(skb, sk);
1762
			sk_wmem_free_skb(sk, skb);
J
John Heffner 已提交
1763
		} else {
E
Eric Dumazet 已提交
1764
			TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags &
C
Changli Gao 已提交
1765
						   ~(TCPHDR_FIN|TCPHDR_PSH);
J
John Heffner 已提交
1766 1767
			if (!skb_shinfo(skb)->nr_frags) {
				skb_pull(skb, copy);
1768
				if (skb->ip_summed != CHECKSUM_PARTIAL)
1769 1770
					skb->csum = csum_partial(skb->data,
								 skb->len, 0);
J
John Heffner 已提交
1771 1772 1773 1774 1775 1776 1777 1778
			} else {
				__pskb_trim_head(skb, copy);
				tcp_set_skb_tso_segs(sk, skb, mss_now);
			}
			TCP_SKB_CB(skb)->seq += copy;
		}

		len += copy;
1779 1780 1781

		if (len >= probe_size)
			break;
J
John Heffner 已提交
1782 1783 1784 1785 1786 1787 1788 1789
	}
	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
1790
		 * effectively two packets. */
J
John Heffner 已提交
1791
		tp->snd_cwnd--;
1792
		tcp_event_new_data_sent(sk, nskb);
J
John Heffner 已提交
1793 1794

		icsk->icsk_mtup.probe_size = tcp_mss_to_mtu(sk, nskb->len);
1795 1796
		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
1797 1798 1799 1800 1801 1802 1803

		return 1;
	}

	return -1;
}

L
Linus Torvalds 已提交
1804 1805 1806 1807
/* This routine writes packets to the network.  It advances the
 * send_head.  This happens as incoming acks open up the remote
 * window for us.
 *
1808 1809 1810 1811
 * 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
1812 1813 1814
 * 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
1815 1816
 * 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
1817
 */
E
Eric Dumazet 已提交
1818 1819
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
1820 1821
{
	struct tcp_sock *tp = tcp_sk(sk);
1822
	struct sk_buff *skb;
1823 1824
	unsigned int tso_segs, sent_pkts;
	int cwnd_quota;
J
John Heffner 已提交
1825
	int result;
L
Linus Torvalds 已提交
1826

1827
	sent_pkts = 0;
J
John Heffner 已提交
1828

1829 1830 1831 1832
	if (!push_one) {
		/* Do MTU probing. */
		result = tcp_mtu_probe(sk);
		if (!result) {
E
Eric Dumazet 已提交
1833
			return false;
1834 1835 1836
		} else if (result > 0) {
			sent_pkts = 1;
		}
J
John Heffner 已提交
1837 1838
	}

1839
	while ((skb = tcp_send_head(sk))) {
1840 1841
		unsigned int limit;

1842
		tso_segs = tcp_init_tso_segs(sk, skb, mss_now);
1843
		BUG_ON(!tso_segs);
1844

1845 1846 1847
		if (unlikely(tp->repair) && tp->repair_queue == TCP_SEND_QUEUE)
			goto repair; /* Skip network transmission */

1848
		cwnd_quota = tcp_cwnd_test(tp, skb);
N
Nandita Dukkipati 已提交
1849 1850 1851 1852 1853 1854 1855
		if (!cwnd_quota) {
			if (push_one == 2)
				/* Force out a loss probe pkt. */
				cwnd_quota = 1;
			else
				break;
		}
1856 1857 1858 1859

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

1860 1861 1862 1863 1864 1865
		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 {
1866
			if (!push_one && tcp_tso_should_defer(sk, skb))
1867 1868
				break;
		}
1869

E
Eric Dumazet 已提交
1870 1871 1872 1873 1874 1875
		/* TCP Small Queues :
		 * Control number of packets in qdisc/devices to two packets / or ~1 ms.
		 * This allows for :
		 *  - better RTT estimation and ACK scheduling
		 *  - faster recovery
		 *  - high rates
E
Eric Dumazet 已提交
1876
		 */
E
Eric Dumazet 已提交
1877 1878 1879
		limit = max(skb->truesize, sk->sk_pacing_rate >> 10);

		if (atomic_read(&sk->sk_wmem_alloc) > limit) {
E
Eric Dumazet 已提交
1880 1881 1882
			set_bit(TSQ_THROTTLED, &tp->tsq_flags);
			break;
		}
E
Eric Dumazet 已提交
1883

1884
		limit = mss_now;
1885
		if (tso_segs > 1 && !tcp_urg_mode(tp))
1886
			limit = tcp_mss_split_point(sk, skb, mss_now,
1887 1888 1889
						    min_t(unsigned int,
							  cwnd_quota,
							  sk->sk_gso_max_segs));
L
Linus Torvalds 已提交
1890

1891
		if (skb->len > limit &&
1892
		    unlikely(tso_fragment(sk, skb, limit, mss_now, gfp)))
1893 1894
			break;

1895
		TCP_SKB_CB(skb)->when = tcp_time_stamp;
1896

1897
		if (unlikely(tcp_transmit_skb(sk, skb, 1, gfp)))
1898
			break;
L
Linus Torvalds 已提交
1899

1900
repair:
1901 1902 1903
		/* Advance the send_head.  This one is sent out.
		 * This call will increment packets_out.
		 */
1904
		tcp_event_new_data_sent(sk, skb);
L
Linus Torvalds 已提交
1905

1906
		tcp_minshall_update(tp, mss_now, skb);
1907
		sent_pkts += tcp_skb_pcount(skb);
1908 1909 1910

		if (push_one)
			break;
1911
	}
L
Linus Torvalds 已提交
1912

1913
	if (likely(sent_pkts)) {
1914 1915
		if (tcp_in_cwnd_reduction(sk))
			tp->prr_out += sent_pkts;
N
Nandita Dukkipati 已提交
1916 1917 1918 1919

		/* Send one loss probe per tail loss episode. */
		if (push_one != 2)
			tcp_schedule_loss_probe(sk);
1920
		tcp_cwnd_validate(sk);
E
Eric Dumazet 已提交
1921
		return false;
L
Linus Torvalds 已提交
1922
	}
N
Nandita Dukkipati 已提交
1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988
	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
1989
	struct tcp_sock *tp = tcp_sk(sk);
N
Nandita Dukkipati 已提交
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
	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
2000 2001 2002 2003
	/* At most one outstanding TLP retransmission. */
	if (tp->tlp_high_seq)
		goto rearm_timer;

N
Nandita Dukkipati 已提交
2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025
	/* 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
2026 2027 2028 2029
	/* Record snd_nxt for loss detection. */
	if (likely(!err))
		tp->tlp_high_seq = tp->snd_nxt;

N
Nandita Dukkipati 已提交
2030 2031 2032 2033 2034 2035 2036 2037 2038
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
2039 2040
}

2041 2042 2043 2044
/* 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.
 */
2045 2046
void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
			       int nonagle)
2047
{
2048 2049 2050 2051 2052 2053 2054
	/* 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;

2055 2056
	if (tcp_write_xmit(sk, cur_mss, nonagle, 0,
			   sk_gfp_atomic(sk, GFP_ATOMIC)))
2057
		tcp_check_probe_timer(sk);
2058 2059
}

2060 2061 2062 2063 2064
/* 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)
{
2065
	struct sk_buff *skb = tcp_send_head(sk);
2066 2067 2068

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

2069
	tcp_write_xmit(sk, mss_now, TCP_NAGLE_PUSH, 1, sk->sk_allocation);
2070 2071
}

L
Linus Torvalds 已提交
2072 2073
/* This function returns the amount that we can raise the
 * usable window based on the following constraints
2074
 *
L
Linus Torvalds 已提交
2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092
 * 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.
2093
 *
L
Linus Torvalds 已提交
2094 2095
 * Of course, if the sender implements sender side SWS prevention
 * then this will not be a problem.
2096
 *
L
Linus Torvalds 已提交
2097
 * BSD seems to make the following compromise:
2098
 *
L
Linus Torvalds 已提交
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
 *	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)
{
2126
	struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
2127
	struct tcp_sock *tp = tcp_sk(sk);
S
Stephen Hemminger 已提交
2128
	/* MSS for the peer's data.  Previous versions used mss_clamp
L
Linus Torvalds 已提交
2129 2130 2131 2132 2133
	 * 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
	 */
2134
	int mss = icsk->icsk_ack.rcv_mss;
L
Linus Torvalds 已提交
2135 2136 2137 2138 2139
	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)
2140
		mss = full_space;
L
Linus Torvalds 已提交
2141

2142
	if (free_space < (full_space >> 1)) {
2143
		icsk->icsk_ack.quick = 0;
L
Linus Torvalds 已提交
2144

2145
		if (sk_under_memory_pressure(sk))
2146 2147
			tp->rcv_ssthresh = min(tp->rcv_ssthresh,
					       4U * tp->advmss);
L
Linus Torvalds 已提交
2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179

		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)
2180
			window = (free_space / mss) * mss;
2181
		else if (mss == full_space &&
2182
			 free_space > window + (full_space >> 1))
2183
			window = free_space;
L
Linus Torvalds 已提交
2184 2185 2186 2187 2188
	}

	return window;
}

2189 2190
/* 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
2191 2192
{
	struct tcp_sock *tp = tcp_sk(sk);
2193
	struct sk_buff *next_skb = tcp_write_queue_next(sk, skb);
2194
	int skb_size, next_skb_size;
L
Linus Torvalds 已提交
2195

2196 2197
	skb_size = skb->len;
	next_skb_size = next_skb->len;
L
Linus Torvalds 已提交
2198

2199
	BUG_ON(tcp_skb_pcount(skb) != 1 || tcp_skb_pcount(next_skb) != 1);
2200

2201
	tcp_highest_sack_combine(sk, next_skb, skb);
L
Linus Torvalds 已提交
2202

2203
	tcp_unlink_write_queue(next_skb, sk);
L
Linus Torvalds 已提交
2204

2205 2206
	skb_copy_from_linear_data(next_skb, skb_put(skb, next_skb_size),
				  next_skb_size);
L
Linus Torvalds 已提交
2207

2208 2209
	if (next_skb->ip_summed == CHECKSUM_PARTIAL)
		skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
2210

2211 2212
	if (skb->ip_summed != CHECKSUM_PARTIAL)
		skb->csum = csum_block_add(skb->csum, next_skb->csum, skb_size);
L
Linus Torvalds 已提交
2213

2214 2215
	/* 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
2216

2217
	/* Merge over control information. This moves PSH/FIN etc. over */
E
Eric Dumazet 已提交
2218
	TCP_SKB_CB(skb)->tcp_flags |= TCP_SKB_CB(next_skb)->tcp_flags;
2219 2220 2221 2222 2223 2224 2225

	/* 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 */
2226 2227 2228
	tcp_clear_retrans_hints_partial(tp);
	if (next_skb == tp->retransmit_skb_hint)
		tp->retransmit_skb_hint = skb;
2229

2230 2231
	tcp_adjust_pcount(sk, next_skb, tcp_skb_pcount(next_skb));

2232
	sk_wmem_free_skb(sk, next_skb);
L
Linus Torvalds 已提交
2233 2234
}

2235
/* Check if coalescing SKBs is legal. */
E
Eric Dumazet 已提交
2236
static bool tcp_can_collapse(const struct sock *sk, const struct sk_buff *skb)
2237 2238
{
	if (tcp_skb_pcount(skb) > 1)
E
Eric Dumazet 已提交
2239
		return false;
2240 2241
	/* TODO: SACK collapsing could be used to remove this condition */
	if (skb_shinfo(skb)->nr_frags != 0)
E
Eric Dumazet 已提交
2242
		return false;
2243
	if (skb_cloned(skb))
E
Eric Dumazet 已提交
2244
		return false;
2245
	if (skb == tcp_send_head(sk))
E
Eric Dumazet 已提交
2246
		return false;
2247 2248
	/* 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
2249
		return false;
2250

E
Eric Dumazet 已提交
2251
	return true;
2252 2253
}

2254 2255 2256
/* Collapse packets in the retransmit queue to make to create
 * less packets on the wire. This is only done on retransmission.
 */
2257 2258 2259 2260 2261
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
2262
	bool first = true;
2263 2264 2265

	if (!sysctl_tcp_retrans_collapse)
		return;
E
Eric Dumazet 已提交
2266
	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)
2267 2268 2269 2270 2271 2272 2273 2274 2275
		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
2276
			first = false;
2277 2278 2279 2280 2281 2282 2283 2284
			continue;
		}

		if (space < 0)
			break;
		/* Punt if not enough space exists in the first SKB for
		 * the data in the second
		 */
2285
		if (skb->len > skb_availroom(to))
2286 2287 2288 2289 2290 2291 2292 2293 2294
			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
2295 2296 2297 2298
/* 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.
 */
2299
int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
2300 2301
{
	struct tcp_sock *tp = tcp_sk(sk);
J
John Heffner 已提交
2302
	struct inet_connection_sock *icsk = inet_csk(sk);
2303
	unsigned int cur_mss;
L
Linus Torvalds 已提交
2304

J
John Heffner 已提交
2305 2306 2307 2308 2309
	/* Inconslusive MTU probe */
	if (icsk->icsk_mtup.probe_size) {
		icsk->icsk_mtup.probe_size = 0;
	}

L
Linus Torvalds 已提交
2310
	/* Do not sent more than we queued. 1/4 is reserved for possible
S
Stephen Hemminger 已提交
2311
	 * copying overhead: fragmentation, tunneling, mangling etc.
L
Linus Torvalds 已提交
2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323
	 */
	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;
	}

2324 2325 2326
	if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk))
		return -EHOSTUNREACH; /* Routing failure or similar. */

I
Ilpo Järvinen 已提交
2327
	cur_mss = tcp_current_mss(sk);
2328

L
Linus Torvalds 已提交
2329 2330 2331 2332 2333
	/* 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.
	 */
2334 2335
	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
2336 2337 2338
		return -EAGAIN;

	if (skb->len > cur_mss) {
2339
		if (tcp_fragment(sk, skb, cur_mss, cur_mss))
L
Linus Torvalds 已提交
2340
			return -ENOMEM; /* We'll try again later. */
2341
	} else {
2342 2343 2344 2345 2346 2347
		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
2348 2349
	}

2350
	tcp_retrans_try_collapse(sk, skb, cur_mss);
L
Linus Torvalds 已提交
2351 2352 2353 2354 2355

	/* 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
2356
	if (skb->len > 0 &&
E
Eric Dumazet 已提交
2357
	    (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) &&
S
Stephen Hemminger 已提交
2358
	    tp->snd_una == (TCP_SKB_CB(skb)->end_seq - 1)) {
L
Linus Torvalds 已提交
2359
		if (!pskb_trim(skb, 0)) {
2360 2361
			/* 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
2362
					     TCP_SKB_CB(skb)->tcp_flags);
L
Linus Torvalds 已提交
2363 2364 2365 2366 2367 2368 2369 2370 2371
			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;

2372 2373 2374 2375 2376 2377
	/* make sure skb->data is aligned on arches that require it
	 * and check if ack-trimming & collapsing extended the headroom
	 * beyond what csum_start can cover.
	 */
	if (unlikely((NET_IP_ALIGN && ((unsigned long)skb->data & 3)) ||
		     skb_headroom(skb) >= 0xFFFF)) {
E
Eric Dumazet 已提交
2378 2379
		struct sk_buff *nskb = __pskb_copy(skb, MAX_TCP_HEADER,
						   GFP_ATOMIC);
2380 2381
		return nskb ? tcp_transmit_skb(sk, nskb, 0, GFP_ATOMIC) :
			      -ENOBUFS;
E
Eric Dumazet 已提交
2382
	} else {
2383
		return tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
E
Eric Dumazet 已提交
2384
	}
2385 2386 2387 2388 2389 2390
}

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
2391 2392 2393

	if (err == 0) {
		/* Update global TCP statistics. */
P
Pavel Emelyanov 已提交
2394
		TCP_INC_STATS(sock_net(sk), TCP_MIB_RETRANSSEGS);
L
Linus Torvalds 已提交
2395 2396 2397 2398

		tp->total_retrans++;

#if FASTRETRANS_DEBUG > 0
2399
		if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) {
2400
			net_dbg_ratelimited("retrans_out leaked\n");
L
Linus Torvalds 已提交
2401 2402
		}
#endif
2403 2404
		if (!tp->retrans_out)
			tp->lost_retrans_low = tp->snd_nxt;
L
Linus Torvalds 已提交
2405 2406 2407 2408 2409 2410 2411
		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
2412
		tp->undo_retrans += tcp_skb_pcount(skb);
L
Linus Torvalds 已提交
2413 2414 2415 2416 2417

		/* 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;
2418 2419
	} else {
		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPRETRANSFAIL);
L
Linus Torvalds 已提交
2420 2421 2422 2423
	}
	return err;
}

2424 2425 2426
/* Check if we forward retransmits are possible in the current
 * window/congestion state.
 */
E
Eric Dumazet 已提交
2427
static bool tcp_can_forward_retransmit(struct sock *sk)
2428 2429
{
	const struct inet_connection_sock *icsk = inet_csk(sk);
2430
	const struct tcp_sock *tp = tcp_sk(sk);
2431 2432 2433

	/* Forward retransmissions are possible only during Recovery. */
	if (icsk->icsk_ca_state != TCP_CA_Recovery)
E
Eric Dumazet 已提交
2434
		return false;
2435 2436 2437

	/* No forward retransmissions in Reno are possible. */
	if (tcp_is_reno(tp))
E
Eric Dumazet 已提交
2438
		return false;
2439 2440 2441 2442 2443 2444 2445 2446 2447 2448

	/* 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
2449
		return false;
2450

E
Eric Dumazet 已提交
2451
	return true;
2452 2453
}

L
Linus Torvalds 已提交
2454 2455 2456 2457 2458 2459 2460 2461 2462 2463
/* 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)
{
2464
	const struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
2465 2466
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb;
2467
	struct sk_buff *hole = NULL;
2468
	u32 last_lost;
2469
	int mib_idx;
2470
	int fwd_rexmitting = 0;
2471

2472 2473 2474
	if (!tp->packets_out)
		return;

2475 2476 2477
	if (!tp->lost_out)
		tp->retransmit_high = tp->snd_una;

2478
	if (tp->retransmit_skb_hint) {
2479
		skb = tp->retransmit_skb_hint;
2480 2481 2482 2483
		last_lost = TCP_SKB_CB(skb)->end_seq;
		if (after(last_lost, tp->retransmit_high))
			last_lost = tp->retransmit_high;
	} else {
2484
		skb = tcp_write_queue_head(sk);
2485 2486
		last_lost = tp->snd_una;
	}
L
Linus Torvalds 已提交
2487

2488 2489
	tcp_for_write_queue_from(skb, sk) {
		__u8 sacked = TCP_SKB_CB(skb)->sacked;
L
Linus Torvalds 已提交
2490

2491 2492 2493
		if (skb == tcp_send_head(sk))
			break;
		/* we could do better than to assign each time */
2494 2495
		if (hole == NULL)
			tp->retransmit_skb_hint = skb;
2496 2497 2498 2499 2500 2501 2502 2503 2504 2505

		/* 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
2506

2507 2508 2509 2510 2511
		if (fwd_rexmitting) {
begin_fwd:
			if (!before(TCP_SKB_CB(skb)->seq, tcp_highest_sack_seq(tp)))
				break;
			mib_idx = LINUX_MIB_TCPFORWARDRETRANS;
2512

2513
		} else if (!before(TCP_SKB_CB(skb)->seq, tp->retransmit_high)) {
2514
			tp->retransmit_high = last_lost;
2515 2516 2517 2518 2519 2520 2521 2522 2523
			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
2524

2525
		} else if (!(sacked & TCPCB_LOST)) {
2526
			if (hole == NULL && !(sacked & (TCPCB_SACKED_RETRANS|TCPCB_SACKED_ACKED)))
2527 2528
				hole = skb;
			continue;
L
Linus Torvalds 已提交
2529

2530
		} else {
2531
			last_lost = TCP_SKB_CB(skb)->end_seq;
2532 2533 2534 2535 2536
			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
2537

2538
		if (sacked & (TCPCB_SACKED_ACKED|TCPCB_SACKED_RETRANS))
L
Linus Torvalds 已提交
2539 2540
			continue;

2541
		if (tcp_retransmit_skb(sk, skb))
2542
			return;
2543

2544
		NET_INC_STATS_BH(sock_net(sk), mib_idx);
L
Linus Torvalds 已提交
2545

2546
		if (tcp_in_cwnd_reduction(sk))
2547 2548
			tp->prr_out += tcp_skb_pcount(skb);

2549
		if (skb == tcp_write_queue_head(sk))
2550 2551 2552
			inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
						  inet_csk(sk)->icsk_rto,
						  TCP_RTO_MAX);
L
Linus Torvalds 已提交
2553 2554 2555 2556 2557 2558 2559 2560
	}
}

/* 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)
{
2561
	struct tcp_sock *tp = tcp_sk(sk);
2562
	struct sk_buff *skb = tcp_write_queue_tail(sk);
L
Linus Torvalds 已提交
2563
	int mss_now;
2564

L
Linus Torvalds 已提交
2565 2566 2567 2568
	/* 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
2569
	mss_now = tcp_current_mss(sk);
L
Linus Torvalds 已提交
2570

2571
	if (tcp_send_head(sk) != NULL) {
E
Eric Dumazet 已提交
2572
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_FIN;
L
Linus Torvalds 已提交
2573 2574 2575 2576 2577
		TCP_SKB_CB(skb)->end_seq++;
		tp->write_seq++;
	} else {
		/* Socket is locked, keep trying until memory is available. */
		for (;;) {
2578 2579
			skb = alloc_skb_fclone(MAX_TCP_HEADER,
					       sk->sk_allocation);
L
Linus Torvalds 已提交
2580 2581 2582 2583 2584 2585 2586 2587
			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(). */
2588
		tcp_init_nondata_skb(skb, tp->write_seq,
C
Changli Gao 已提交
2589
				     TCPHDR_ACK | TCPHDR_FIN);
L
Linus Torvalds 已提交
2590 2591
		tcp_queue_skb(sk, skb);
	}
2592
	__tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_OFF);
L
Linus Torvalds 已提交
2593 2594 2595 2596 2597
}

/* 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
2598
 * by RFC 2525, section 2.17.  -DaveM
L
Linus Torvalds 已提交
2599
 */
A
Al Viro 已提交
2600
void tcp_send_active_reset(struct sock *sk, gfp_t priority)
L
Linus Torvalds 已提交
2601 2602 2603 2604 2605 2606
{
	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
2607
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTFAILED);
L
Linus Torvalds 已提交
2608 2609 2610 2611 2612
		return;
	}

	/* Reserve space for headers and prepare control bits. */
	skb_reserve(skb, MAX_TCP_HEADER);
2613
	tcp_init_nondata_skb(skb, tcp_acceptable_seq(sk),
C
Changli Gao 已提交
2614
			     TCPHDR_ACK | TCPHDR_RST);
L
Linus Torvalds 已提交
2615 2616
	/* Send it off. */
	TCP_SKB_CB(skb)->when = tcp_time_stamp;
2617
	if (tcp_transmit_skb(sk, skb, 0, priority))
P
Pavel Emelyanov 已提交
2618
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTFAILED);
2619

P
Pavel Emelyanov 已提交
2620
	TCP_INC_STATS(sock_net(sk), TCP_MIB_OUTRSTS);
L
Linus Torvalds 已提交
2621 2622
}

2623 2624
/* 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
2625 2626 2627 2628 2629 2630
 * 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)
{
2631
	struct sk_buff *skb;
L
Linus Torvalds 已提交
2632

2633
	skb = tcp_write_queue_head(sk);
E
Eric Dumazet 已提交
2634
	if (skb == NULL || !(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
2635
		pr_debug("%s: wrong queue state\n", __func__);
L
Linus Torvalds 已提交
2636 2637
		return -EFAULT;
	}
E
Eric Dumazet 已提交
2638
	if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)) {
L
Linus Torvalds 已提交
2639 2640 2641 2642
		if (skb_cloned(skb)) {
			struct sk_buff *nskb = skb_copy(skb, GFP_ATOMIC);
			if (nskb == NULL)
				return -ENOMEM;
2643
			tcp_unlink_write_queue(skb, sk);
L
Linus Torvalds 已提交
2644
			skb_header_release(nskb);
2645
			__tcp_add_write_queue_head(sk, nskb);
2646 2647 2648
			sk_wmem_free_skb(sk, skb);
			sk->sk_wmem_queued += nskb->truesize;
			sk_mem_charge(sk, nskb->truesize);
L
Linus Torvalds 已提交
2649 2650 2651
			skb = nskb;
		}

E
Eric Dumazet 已提交
2652
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ACK;
L
Linus Torvalds 已提交
2653 2654 2655
		TCP_ECN_send_synack(tcp_sk(sk), skb);
	}
	TCP_SKB_CB(skb)->when = tcp_time_stamp;
2656
	return tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
L
Linus Torvalds 已提交
2657 2658
}

2659 2660 2661 2662 2663 2664 2665 2666 2667
/**
 * tcp_make_synack - Prepare a SYN-ACK.
 * sk: listener socket
 * dst: dst entry attached to the SYNACK
 * req: request_sock pointer
 *
 * Allocate one skb and build a SYNACK packet.
 * @dst is consumed : Caller should not use it again.
 */
2668
struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
2669
				struct request_sock *req,
2670
				struct tcp_fastopen_cookie *foc)
L
Linus Torvalds 已提交
2671
{
2672
	struct tcp_out_options opts;
2673
	struct inet_request_sock *ireq = inet_rsk(req);
L
Linus Torvalds 已提交
2674 2675 2676
	struct tcp_sock *tp = tcp_sk(sk);
	struct tcphdr *th;
	struct sk_buff *skb;
2677
	struct tcp_md5sig_key *md5;
2678
	int tcp_header_size;
2679
	int mss;
L
Linus Torvalds 已提交
2680

2681
	skb = sock_wmalloc(sk, MAX_TCP_HEADER + 15, 1, GFP_ATOMIC);
2682 2683
	if (unlikely(!skb)) {
		dst_release(dst);
L
Linus Torvalds 已提交
2684
		return NULL;
2685
	}
L
Linus Torvalds 已提交
2686 2687 2688
	/* Reserve space for headers. */
	skb_reserve(skb, MAX_TCP_HEADER);

2689
	skb_dst_set(skb, dst);
2690
	security_skb_owned_by(skb, sk);
L
Linus Torvalds 已提交
2691

2692
	mss = dst_metric_advmss(dst);
2693 2694 2695
	if (tp->rx_opt.user_mss && tp->rx_opt.user_mss < mss)
		mss = tp->rx_opt.user_mss;

A
Adam Langley 已提交
2696 2697 2698 2699
	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);
2700 2701 2702 2703 2704 2705

		/* 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
2706 2707
		/* tcp_full_space because it is guaranteed to be the first packet */
		tcp_select_initial_window(tcp_full_space(sk),
2708
			mss - (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0),
A
Adam Langley 已提交
2709 2710 2711
			&req->rcv_wnd,
			&req->window_clamp,
			ireq->wscale_ok,
2712 2713
			&rcv_wscale,
			dst_metric(dst, RTAX_INITRWND));
A
Adam Langley 已提交
2714 2715 2716 2717
		ireq->rcv_wscale = rcv_wscale;
	}

	memset(&opts, 0, sizeof(opts));
2718 2719 2720 2721 2722
#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
2723
	TCP_SKB_CB(skb)->when = tcp_time_stamp;
C
Christoph Paasch 已提交
2724 2725
	tcp_header_size = tcp_synack_options(sk, req, mss, skb, &opts, &md5,
					     foc) + sizeof(*th);
2726

2727 2728
	skb_push(skb, tcp_header_size);
	skb_reset_transport_header(skb);
L
Linus Torvalds 已提交
2729

2730
	th = tcp_hdr(skb);
L
Linus Torvalds 已提交
2731 2732 2733 2734
	memset(th, 0, sizeof(struct tcphdr));
	th->syn = 1;
	th->ack = 1;
	TCP_ECN_make_synack(req, th);
2735
	th->source = ireq->loc_port;
2736
	th->dest = ireq->rmt_port;
2737 2738 2739 2740
	/* 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
2741
			     TCPHDR_SYN | TCPHDR_ACK);
2742

L
Linus Torvalds 已提交
2743
	th->seq = htonl(TCP_SKB_CB(skb)->seq);
2744 2745
	/* 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
2746 2747

	/* RFC1323: The window in SYN & SYN/ACK segments is never scaled. */
2748
	th->window = htons(min(req->rcv_wnd, 65535U));
2749
	tcp_options_write((__be32 *)(th + 1), tp, &opts);
L
Linus Torvalds 已提交
2750
	th->doff = (tcp_header_size >> 2);
2751
	TCP_ADD_STATS(sock_net(sk), TCP_MIB_OUTSEGS, tcp_skb_pcount(skb));
2752 2753 2754 2755

#ifdef CONFIG_TCP_MD5SIG
	/* Okay, we have all we need - do the md5 hash if needed */
	if (md5) {
2756
		tcp_rsk(req)->af_specific->calc_md5_hash(opts.hash_location,
2757
					       md5, NULL, req, skb);
2758 2759 2760
	}
#endif

L
Linus Torvalds 已提交
2761 2762
	return skb;
}
E
Eric Dumazet 已提交
2763
EXPORT_SYMBOL(tcp_make_synack);
L
Linus Torvalds 已提交
2764

2765
/* Do all connect socket setups that can be done AF independent. */
P
Pavel Emelyanov 已提交
2766
void tcp_connect_init(struct sock *sk)
L
Linus Torvalds 已提交
2767
{
2768
	const struct dst_entry *dst = __sk_dst_get(sk);
L
Linus Torvalds 已提交
2769 2770 2771 2772 2773 2774 2775
	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) +
2776
		(sysctl_tcp_timestamps ? TCPOLEN_TSTAMP_ALIGNED : 0);
L
Linus Torvalds 已提交
2777

2778 2779 2780 2781 2782
#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
2783 2784 2785 2786
	/* 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
2787
	tcp_mtup_init(sk);
L
Linus Torvalds 已提交
2788 2789 2790 2791
	tcp_sync_mss(sk, dst_mtu(dst));

	if (!tp->window_clamp)
		tp->window_clamp = dst_metric(dst, RTAX_WINDOW);
2792
	tp->advmss = dst_metric_advmss(dst);
2793 2794 2795
	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
2796 2797
	tcp_initialize_rcv_mss(sk);

2798 2799 2800 2801 2802
	/* 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
2803 2804 2805 2806
	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,
2807
				  sysctl_tcp_window_scaling,
2808 2809
				  &rcv_wscale,
				  dst_metric(dst, RTAX_INITRWND));
L
Linus Torvalds 已提交
2810 2811 2812 2813 2814 2815 2816

	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;
2817
	tcp_init_wl(tp, 0);
L
Linus Torvalds 已提交
2818 2819
	tp->snd_una = tp->write_seq;
	tp->snd_sml = tp->write_seq;
I
Ilpo Järvinen 已提交
2820
	tp->snd_up = tp->write_seq;
P
Pavel Emelyanov 已提交
2821
	tp->snd_nxt = tp->write_seq;
P
Pavel Emelyanov 已提交
2822 2823 2824

	if (likely(!tp->repair))
		tp->rcv_nxt = 0;
2825 2826
	else
		tp->rcv_tstamp = tcp_time_stamp;
P
Pavel Emelyanov 已提交
2827 2828
	tp->rcv_wup = tp->rcv_nxt;
	tp->copied_seq = tp->rcv_nxt;
L
Linus Torvalds 已提交
2829

2830 2831
	inet_csk(sk)->icsk_rto = TCP_TIMEOUT_INIT;
	inet_csk(sk)->icsk_retransmits = 0;
L
Linus Torvalds 已提交
2832 2833 2834
	tcp_clear_retrans(tp);
}

2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859
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;
2860
	int syn_loss = 0, space, i, err = 0, iovlen = fo->data->msg_iovlen;
2861
	struct sk_buff *syn_data = NULL, *data;
2862 2863
	unsigned long last_syn_loss = 0;

2864
	tp->rx_opt.mss_clamp = tp->advmss;  /* If MSS is not cached */
2865 2866 2867 2868 2869 2870 2871 2872
	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;
	}
2873

2874 2875 2876
	if (sysctl_tcp_fastopen & TFO_CLIENT_NO_COOKIE)
		fo->cookie.len = -1;
	else if (fo->cookie.len <= 0)
2877 2878 2879 2880 2881 2882 2883 2884
		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;
2885
	space = __tcp_mtu_to_mss(sk, inet_csk(sk)->icsk_pmtu_cookie) -
2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918
		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) {
2919
		tp->syn_data = (fo->copied > 0);
2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937
		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;
}

2938
/* Build a SYN and send it off. */
L
Linus Torvalds 已提交
2939 2940 2941 2942
int tcp_connect(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *buff;
2943
	int err;
L
Linus Torvalds 已提交
2944 2945 2946

	tcp_connect_init(sk);

A
Andrey Vagin 已提交
2947 2948 2949 2950 2951
	if (unlikely(tp->repair)) {
		tcp_finish_connect(sk, NULL);
		return 0;
	}

2952
	buff = alloc_skb_fclone(MAX_TCP_HEADER + 15, sk->sk_allocation);
L
Linus Torvalds 已提交
2953 2954 2955 2956 2957 2958
	if (unlikely(buff == NULL))
		return -ENOBUFS;

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

C
Changli Gao 已提交
2959
	tcp_init_nondata_skb(buff, tp->write_seq++, TCPHDR_SYN);
2960 2961
	tp->retrans_stamp = TCP_SKB_CB(buff)->when = tcp_time_stamp;
	tcp_connect_queue_skb(sk, buff);
2962
	TCP_ECN_send_syn(sk, buff);
L
Linus Torvalds 已提交
2963

2964 2965 2966
	/* 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);
2967 2968
	if (err == -ECONNREFUSED)
		return err;
2969 2970 2971 2972 2973 2974

	/* 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
2975
	TCP_INC_STATS(sock_net(sk), TCP_MIB_ACTIVEOPENS);
L
Linus Torvalds 已提交
2976 2977

	/* Timer for repeating the SYN until an answer. */
2978 2979
	inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
				  inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
L
Linus Torvalds 已提交
2980 2981
	return 0;
}
E
Eric Dumazet 已提交
2982
EXPORT_SYMBOL(tcp_connect);
L
Linus Torvalds 已提交
2983 2984 2985 2986 2987 2988 2989

/* 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)
{
2990 2991
	struct inet_connection_sock *icsk = inet_csk(sk);
	int ato = icsk->icsk_ack.ato;
L
Linus Torvalds 已提交
2992 2993 2994
	unsigned long timeout;

	if (ato > TCP_DELACK_MIN) {
2995
		const struct tcp_sock *tp = tcp_sk(sk);
2996
		int max_ato = HZ / 2;
L
Linus Torvalds 已提交
2997

2998 2999
		if (icsk->icsk_ack.pingpong ||
		    (icsk->icsk_ack.pending & ICSK_ACK_PUSHED))
L
Linus Torvalds 已提交
3000 3001 3002 3003 3004
			max_ato = TCP_DELACK_MAX;

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

		/* If some rtt estimate is known, use it to bound delayed ack.
3005
		 * Do not use inet_csk(sk)->icsk_rto here, use results of rtt measurements
L
Linus Torvalds 已提交
3006 3007 3008
		 * directly.
		 */
		if (tp->srtt) {
3009
			int rtt = max(tp->srtt >> 3, TCP_DELACK_MIN);
L
Linus Torvalds 已提交
3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021

			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. */
3022
	if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
L
Linus Torvalds 已提交
3023 3024 3025
		/* If delack timer was blocked or is about to expire,
		 * send ACK now.
		 */
3026 3027
		if (icsk->icsk_ack.blocked ||
		    time_before_eq(icsk->icsk_ack.timeout, jiffies + (ato >> 2))) {
L
Linus Torvalds 已提交
3028 3029 3030 3031
			tcp_send_ack(sk);
			return;
		}

3032 3033
		if (!time_before(timeout, icsk->icsk_ack.timeout))
			timeout = icsk->icsk_ack.timeout;
L
Linus Torvalds 已提交
3034
	}
3035 3036 3037
	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
3038 3039 3040 3041 3042
}

/* This routine sends an ack and also updates the window. */
void tcp_send_ack(struct sock *sk)
{
3043
	struct sk_buff *buff;
L
Linus Torvalds 已提交
3044

3045 3046 3047
	/* If we have been reset, we may not send again. */
	if (sk->sk_state == TCP_CLOSE)
		return;
L
Linus Torvalds 已提交
3048

3049 3050 3051 3052
	/* We are not putting this on the write queue, so
	 * tcp_transmit_skb() will set the ownership to this
	 * sock.
	 */
3053
	buff = alloc_skb(MAX_TCP_HEADER, sk_gfp_atomic(sk, GFP_ATOMIC));
3054 3055 3056 3057 3058 3059
	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
3060
	}
3061 3062 3063

	/* Reserve space for headers and prepare control bits. */
	skb_reserve(buff, MAX_TCP_HEADER);
C
Changli Gao 已提交
3064
	tcp_init_nondata_skb(buff, tcp_acceptable_seq(sk), TCPHDR_ACK);
3065 3066 3067

	/* Send it off, this clears delayed acks for us. */
	TCP_SKB_CB(buff)->when = tcp_time_stamp;
3068
	tcp_transmit_skb(sk, buff, 0, sk_gfp_atomic(sk, GFP_ATOMIC));
L
Linus Torvalds 已提交
3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087
}

/* 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. */
3088
	skb = alloc_skb(MAX_TCP_HEADER, sk_gfp_atomic(sk, GFP_ATOMIC));
3089
	if (skb == NULL)
L
Linus Torvalds 已提交
3090 3091 3092 3093 3094 3095 3096 3097
		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
3098
	tcp_init_nondata_skb(skb, tp->snd_una - !urgent, TCPHDR_ACK);
L
Linus Torvalds 已提交
3099
	TCP_SKB_CB(skb)->when = tcp_time_stamp;
3100
	return tcp_transmit_skb(sk, skb, 0, GFP_ATOMIC);
L
Linus Torvalds 已提交
3101 3102
}

P
Pavel Emelyanov 已提交
3103 3104 3105 3106
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
3107
		tcp_sk(sk)->snd_nxt = tcp_sk(sk)->write_seq;
P
Pavel Emelyanov 已提交
3108 3109 3110 3111
		tcp_xmit_probe_skb(sk, 0);
	}
}

3112
/* Initiate keepalive or window probe from timer. */
L
Linus Torvalds 已提交
3113 3114
int tcp_write_wakeup(struct sock *sk)
{
3115 3116
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb;
L
Linus Torvalds 已提交
3117

3118 3119 3120 3121 3122 3123
	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
3124
		unsigned int mss = tcp_current_mss(sk);
3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136
		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
3137
			TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
3138 3139 3140 3141 3142
			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
3143
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
3144 3145 3146 3147 3148 3149
		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
3150
		if (between(tp->snd_up, tp->snd_una + 1, tp->snd_una + 0xFFFF))
3151 3152
			tcp_xmit_probe_skb(sk, 1);
		return tcp_xmit_probe_skb(sk, 0);
L
Linus Torvalds 已提交
3153 3154 3155 3156 3157 3158 3159 3160
	}
}

/* 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)
{
3161
	struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
3162 3163 3164 3165 3166
	struct tcp_sock *tp = tcp_sk(sk);
	int err;

	err = tcp_write_wakeup(sk);

3167
	if (tp->packets_out || !tcp_send_head(sk)) {
L
Linus Torvalds 已提交
3168
		/* Cancel probe timer, if it is not required. */
3169
		icsk->icsk_probes_out = 0;
3170
		icsk->icsk_backoff = 0;
L
Linus Torvalds 已提交
3171 3172 3173 3174
		return;
	}

	if (err <= 0) {
3175 3176
		if (icsk->icsk_backoff < sysctl_tcp_retries2)
			icsk->icsk_backoff++;
3177
		icsk->icsk_probes_out++;
3178
		inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
3179 3180
					  min(icsk->icsk_rto << icsk->icsk_backoff, TCP_RTO_MAX),
					  TCP_RTO_MAX);
L
Linus Torvalds 已提交
3181 3182
	} else {
		/* If packet was not sent due to local congestion,
3183
		 * do not backoff and do not remember icsk_probes_out.
L
Linus Torvalds 已提交
3184 3185 3186 3187
		 * Let local senders to fight for local resources.
		 *
		 * Use accumulated backoff yet.
		 */
3188 3189
		if (!icsk->icsk_probes_out)
			icsk->icsk_probes_out = 1;
3190
		inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
3191
					  min(icsk->icsk_rto << icsk->icsk_backoff,
3192 3193
					      TCP_RESOURCE_PROBE_INTERVAL),
					  TCP_RTO_MAX);
L
Linus Torvalds 已提交
3194 3195
	}
}