tcp_output.c 106.1 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

53 54
/* Default TSQ limit of four TSO segments */
int sysctl_tcp_limit_output_bytes __read_mostly = 262144;
E
Eric Dumazet 已提交
55

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
/* By default, RFC2861 behavior.  */
63
int sysctl_tcp_slow_start_after_idle __read_mostly = 1;
64

E
Eric Dumazet 已提交
65 66
static bool tcp_write_xmit(struct sock *sk, unsigned int mss_now, int nonagle,
			   int push_one, gfp_t gfp);
67

68
/* Account for new data that has been sent to the network. */
69
static void tcp_event_new_data_sent(struct sock *sk, const struct sk_buff *skb)
L
Linus Torvalds 已提交
70
{
N
Nandita Dukkipati 已提交
71
	struct inet_connection_sock *icsk = inet_csk(sk);
72
	struct tcp_sock *tp = tcp_sk(sk);
73
	unsigned int prior_packets = tp->packets_out;
74

75
	tcp_advance_send_head(sk, skb);
L
Linus Torvalds 已提交
76
	tp->snd_nxt = TCP_SKB_CB(skb)->end_seq;
77

78
	tp->packets_out += tcp_skb_pcount(skb);
N
Nandita Dukkipati 已提交
79
	if (!prior_packets || icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
80
	    icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
81
		tcp_rearm_rto(sk);
82
	}
83

84 85
	NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPORIGDATASENT,
		      tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
86 87 88 89 90 91 92 93
}

/* 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:
 */
94
static inline __u32 tcp_acceptable_seq(const struct sock *sk)
L
Linus Torvalds 已提交
95
{
96
	const struct tcp_sock *tp = tcp_sk(sk);
97

98
	if (!before(tcp_wnd_end(tp), tp->snd_nxt))
L
Linus Torvalds 已提交
99 100
		return tp->snd_nxt;
	else
101
		return tcp_wnd_end(tp);
L
Linus Torvalds 已提交
102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120
}

/* 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);
121
	const struct dst_entry *dst = __sk_dst_get(sk);
L
Linus Torvalds 已提交
122 123
	int mss = tp->advmss;

124 125 126 127 128 129 130
	if (dst) {
		unsigned int metric = dst_metric_advmss(dst);

		if (metric < mss) {
			mss = metric;
			tp->advmss = mss;
		}
L
Linus Torvalds 已提交
131 132 133 134 135 136
	}

	return (__u16)mss;
}

/* RFC2861. Reset CWND after idle period longer RTO to "restart window".
137 138 139
 * This is the first part of cwnd validation mechanism.
 */
void tcp_cwnd_restart(struct sock *sk, s32 delta)
L
Linus Torvalds 已提交
140
{
141
	struct tcp_sock *tp = tcp_sk(sk);
142
	u32 restart_cwnd = tcp_init_cwnd(tp, __sk_dst_get(sk));
L
Linus Torvalds 已提交
143 144
	u32 cwnd = tp->snd_cwnd;

145
	tcp_ca_event(sk, CA_EVENT_CWND_RESTART);
L
Linus Torvalds 已提交
146

147
	tp->snd_ssthresh = tcp_current_ssthresh(sk);
L
Linus Torvalds 已提交
148 149
	restart_cwnd = min(restart_cwnd, cwnd);

150
	while ((delta -= inet_csk(sk)->icsk_rto) > 0 && cwnd > restart_cwnd)
L
Linus Torvalds 已提交
151 152 153 154 155 156
		cwnd >>= 1;
	tp->snd_cwnd = max(cwnd, restart_cwnd);
	tp->snd_cwnd_stamp = tcp_time_stamp;
	tp->snd_cwnd_used = 0;
}

157
/* Congestion state accounting after a packet has been sent. */
S
Stephen Hemminger 已提交
158
static void tcp_event_data_sent(struct tcp_sock *tp,
159
				struct sock *sk)
L
Linus Torvalds 已提交
160
{
161 162
	struct inet_connection_sock *icsk = inet_csk(sk);
	const u32 now = tcp_time_stamp;
L
Linus Torvalds 已提交
163

164 165 166
	if (tcp_packets_in_flight(tp) == 0)
		tcp_ca_event(sk, CA_EVENT_TX_START);

L
Linus Torvalds 已提交
167 168 169 170 171
	tp->lsndtime = now;

	/* If it is a reply for ato after last received
	 * packet, enter pingpong mode.
	 */
172 173
	if ((u32)(now - icsk->icsk_ack.lrcvtime) < icsk->icsk_ack.ato)
		icsk->icsk_ack.pingpong = 1;
L
Linus Torvalds 已提交
174 175
}

176
/* Account for an ACK we sent. */
S
Stephen Hemminger 已提交
177
static inline void tcp_event_ack_sent(struct sock *sk, unsigned int pkts)
L
Linus Torvalds 已提交
178
{
179 180
	tcp_dec_quickack_mode(sk, pkts);
	inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
L
Linus Torvalds 已提交
181 182
}

183 184 185 186

u32 tcp_default_init_rwnd(u32 mss)
{
	/* Initial receive window should be twice of TCP_INIT_CWND to
W
Weiping Pan 已提交
187
	 * enable proper sending of new unsent data during fast recovery
188 189 190 191 192 193 194 195 196 197
	 * (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
198 199 200 201 202 203 204 205 206
/* 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,
207 208
			       int wscale_ok, __u8 *rcv_wscale,
			       __u32 init_rcv_wnd)
L
Linus Torvalds 已提交
209 210 211 212 213 214 215 216 217 218 219 220 221
{
	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
222 223 224 225 226 227
	 * 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
228
	 */
229 230 231 232 233
	if (sysctl_tcp_workaround_signed_windows)
		(*rcv_wnd) = min(space, MAX_TCP_WINDOW);
	else
		(*rcv_wnd) = space;

L
Linus Torvalds 已提交
234 235 236
	(*rcv_wscale) = 0;
	if (wscale_ok) {
		/* Set window scaling on max possible window
237
		 * See RFC1323 for an explanation of the limit to 14
L
Linus Torvalds 已提交
238
		 */
239 240
		space = max_t(u32, space, sysctl_tcp_rmem[2]);
		space = max_t(u32, space, sysctl_rmem_max);
241
		space = min_t(u32, space, *window_clamp);
L
Linus Torvalds 已提交
242 243 244 245 246 247
		while (space > 65535 && (*rcv_wscale) < 14) {
			space >>= 1;
			(*rcv_wscale)++;
		}
	}

248
	if (mss > (1 << *rcv_wscale)) {
249 250 251
		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
252 253 254 255 256
	}

	/* Set the clamp no higher than max representable value */
	(*window_clamp) = min(65535U << (*rcv_wscale), *window_clamp);
}
E
Eric Dumazet 已提交
257
EXPORT_SYMBOL(tcp_select_initial_window);
L
Linus Torvalds 已提交
258 259 260 261 262 263

/* 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
264
static u16 tcp_select_window(struct sock *sk)
L
Linus Torvalds 已提交
265 266
{
	struct tcp_sock *tp = tcp_sk(sk);
267
	u32 old_win = tp->rcv_wnd;
L
Linus Torvalds 已提交
268 269 270 271
	u32 cur_win = tcp_receive_window(tp);
	u32 new_win = __tcp_select_window(sk);

	/* Never shrink the offered window */
S
Stephen Hemminger 已提交
272
	if (new_win < cur_win) {
L
Linus Torvalds 已提交
273 274 275 276 277 278 279
		/* Danger Will Robinson!
		 * Don't update rcv_wup/rcv_wnd here or else
		 * we will not be able to advertise a zero
		 * window in time.  --DaveM
		 *
		 * Relax Will Robinson.
		 */
280 281 282
		if (new_win == 0)
			NET_INC_STATS(sock_net(sk),
				      LINUX_MIB_TCPWANTZEROWINDOWADV);
283
		new_win = ALIGN(cur_win, 1 << tp->rx_opt.rcv_wscale);
L
Linus Torvalds 已提交
284 285 286 287 288 289 290
	}
	tp->rcv_wnd = new_win;
	tp->rcv_wup = tp->rcv_nxt;

	/* Make sure we do not exceed the maximum possible
	 * scaled window.
	 */
291
	if (!tp->rx_opt.rcv_wscale && sysctl_tcp_workaround_signed_windows)
L
Linus Torvalds 已提交
292 293 294 295 296 297 298 299
		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. */
300
	if (new_win == 0) {
L
Linus Torvalds 已提交
301
		tp->pred_flags = 0;
302 303 304 305 306 307
		if (old_win)
			NET_INC_STATS(sock_net(sk),
				      LINUX_MIB_TCPTOZEROWINDOWADV);
	} else if (old_win == 0) {
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPFROMZEROWINDOWADV);
	}
L
Linus Torvalds 已提交
308 309 310 311

	return new_win;
}

312
/* Packet ECN state for a SYN-ACK */
313
static void tcp_ecn_send_synack(struct sock *sk, struct sk_buff *skb)
314
{
315 316
	const struct tcp_sock *tp = tcp_sk(sk);

E
Eric Dumazet 已提交
317
	TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_CWR;
318
	if (!(tp->ecn_flags & TCP_ECN_OK))
E
Eric Dumazet 已提交
319
		TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_ECE;
320 321
	else if (tcp_ca_needs_ecn(sk))
		INET_ECN_xmit(sk);
322 323
}

324
/* Packet ECN state for a SYN.  */
325
static void tcp_ecn_send_syn(struct sock *sk, struct sk_buff *skb)
326 327
{
	struct tcp_sock *tp = tcp_sk(sk);
328 329 330 331 332 333 334 335 336
	bool use_ecn = sock_net(sk)->ipv4.sysctl_tcp_ecn == 1 ||
		       tcp_ca_needs_ecn(sk);

	if (!use_ecn) {
		const struct dst_entry *dst = __sk_dst_get(sk);

		if (dst && dst_feature(dst, RTAX_FEATURE_ECN))
			use_ecn = true;
	}
337 338

	tp->ecn_flags = 0;
339 340

	if (use_ecn) {
E
Eric Dumazet 已提交
341
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ECE | TCPHDR_CWR;
342
		tp->ecn_flags = TCP_ECN_OK;
343 344
		if (tcp_ca_needs_ecn(sk))
			INET_ECN_xmit(sk);
345 346 347
	}
}

348 349 350 351 352 353 354 355 356
static void tcp_ecn_clear_syn(struct sock *sk, struct sk_buff *skb)
{
	if (sock_net(sk)->ipv4.sysctl_tcp_ecn_fallback)
		/* tp->ecn_flags are cleared at a later point in time when
		 * SYN ACK is ultimatively being received.
		 */
		TCP_SKB_CB(skb)->tcp_flags &= ~(TCPHDR_ECE | TCPHDR_CWR);
}

357
static void
358
tcp_ecn_make_synack(const struct request_sock *req, struct tcphdr *th)
359
{
360
	if (inet_rsk(req)->ecn_ok)
361 362 363
		th->ece = 1;
}

364 365 366
/* Set up ECN state for a packet on a ESTABLISHED socket that is about to
 * be sent.
 */
367
static void tcp_ecn_send(struct sock *sk, struct sk_buff *skb,
368
			 struct tcphdr *th, int tcp_header_len)
369 370 371 372 373 374 375 376
{
	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);
377
			if (tp->ecn_flags & TCP_ECN_QUEUE_CWR) {
378
				tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR;
379
				th->cwr = 1;
380 381
				skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
			}
382
		} else if (!tcp_ca_needs_ecn(sk)) {
383 384 385 386
			/* ACK or retransmitted segment: clear ECT|CE */
			INET_ECN_dontxmit(sk);
		}
		if (tp->ecn_flags & TCP_ECN_DEMAND_CWR)
387
			th->ece = 1;
388 389 390
	}
}

391 392 393 394 395
/* 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)
{
396
	skb->ip_summed = CHECKSUM_PARTIAL;
397 398
	skb->csum = 0;

E
Eric Dumazet 已提交
399
	TCP_SKB_CB(skb)->tcp_flags = flags;
400 401
	TCP_SKB_CB(skb)->sacked = 0;

402
	tcp_skb_pcount_set(skb, 1);
403 404

	TCP_SKB_CB(skb)->seq = seq;
C
Changli Gao 已提交
405
	if (flags & (TCPHDR_SYN | TCPHDR_FIN))
406 407 408 409
		seq++;
	TCP_SKB_CB(skb)->end_seq = seq;
}

E
Eric Dumazet 已提交
410
static inline bool tcp_urg_mode(const struct tcp_sock *tp)
I
Ilpo Järvinen 已提交
411 412 413 414
{
	return tp->snd_una != tp->snd_up;
}

A
Adam Langley 已提交
415 416 417
#define OPTION_SACK_ADVERTISE	(1 << 0)
#define OPTION_TS		(1 << 1)
#define OPTION_MD5		(1 << 2)
418
#define OPTION_WSCALE		(1 << 3)
Y
Yuchung Cheng 已提交
419
#define OPTION_FAST_OPEN_COOKIE	(1 << 8)
A
Adam Langley 已提交
420 421

struct tcp_out_options {
Y
Yuchung Cheng 已提交
422 423
	u16 options;		/* bit field of OPTION_* */
	u16 mss;		/* 0 to disable */
A
Adam Langley 已提交
424 425
	u8 ws;			/* window scale, 0 to disable */
	u8 num_sack_blocks;	/* number of SACK blocks to include */
426 427
	u8 hash_size;		/* bytes in hash_location */
	__u8 *hash_location;	/* temporary pointer, overloaded */
Y
Yuchung Cheng 已提交
428 429
	__u32 tsval, tsecr;	/* need to include OPTION_TS */
	struct tcp_fastopen_cookie *fastopen_cookie;	/* Fast open cookie */
A
Adam Langley 已提交
430 431
};

432 433 434
/* Write previously computed TCP options to the packet.
 *
 * Beware: Something in the Internet is very sensitive to the ordering of
435 436
 * 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
S
stephen hemminger 已提交
437
 * inter-operability perspective it seems that we're somewhat stuck with
438 439 440 441 442 443 444
 * 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
445
static void tcp_options_write(__be32 *ptr, struct tcp_sock *tp,
446 447
			      struct tcp_out_options *opts)
{
Y
Yuchung Cheng 已提交
448
	u16 options = opts->options;	/* mungable copy */
449 450

	if (unlikely(OPTION_MD5 & options)) {
C
Christoph Paasch 已提交
451 452
		*ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
			       (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
453 454
		/* overload cookie hash location */
		opts->hash_location = (__u8 *)ptr;
A
Adam Langley 已提交
455
		ptr += 4;
S
Stephen Hemminger 已提交
456
	}
A
Adam Langley 已提交
457

458 459 460 461 462 463
	if (unlikely(opts->mss)) {
		*ptr++ = htonl((TCPOPT_MSS << 24) |
			       (TCPOLEN_MSS << 16) |
			       opts->mss);
	}

464 465
	if (likely(OPTION_TS & options)) {
		if (unlikely(OPTION_SACK_ADVERTISE & options)) {
A
Adam Langley 已提交
466 467 468 469
			*ptr++ = htonl((TCPOPT_SACK_PERM << 24) |
				       (TCPOLEN_SACK_PERM << 16) |
				       (TCPOPT_TIMESTAMP << 8) |
				       TCPOLEN_TIMESTAMP);
470
			options &= ~OPTION_SACK_ADVERTISE;
A
Adam Langley 已提交
471 472 473 474 475 476 477 478 479 480
		} else {
			*ptr++ = htonl((TCPOPT_NOP << 24) |
				       (TCPOPT_NOP << 16) |
				       (TCPOPT_TIMESTAMP << 8) |
				       TCPOLEN_TIMESTAMP);
		}
		*ptr++ = htonl(opts->tsval);
		*ptr++ = htonl(opts->tsecr);
	}

481
	if (unlikely(OPTION_SACK_ADVERTISE & options)) {
A
Adam Langley 已提交
482 483 484 485 486 487
		*ptr++ = htonl((TCPOPT_NOP << 24) |
			       (TCPOPT_NOP << 16) |
			       (TCPOPT_SACK_PERM << 8) |
			       TCPOLEN_SACK_PERM);
	}

488
	if (unlikely(OPTION_WSCALE & options)) {
A
Adam Langley 已提交
489 490 491 492 493 494 495 496 497
		*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
498 499 500 501 502
		int this_sack;

		*ptr++ = htonl((TCPOPT_NOP  << 24) |
			       (TCPOPT_NOP  << 16) |
			       (TCPOPT_SACK <<  8) |
A
Adam Langley 已提交
503
			       (TCPOLEN_SACK_BASE + (opts->num_sack_blocks *
S
Stephen Hemminger 已提交
504
						     TCPOLEN_SACK_PERBLOCK)));
S
Stephen Hemminger 已提交
505

A
Adam Langley 已提交
506 507
		for (this_sack = 0; this_sack < opts->num_sack_blocks;
		     ++this_sack) {
S
Stephen Hemminger 已提交
508 509 510
			*ptr++ = htonl(sp[this_sack].start_seq);
			*ptr++ = htonl(sp[this_sack].end_seq);
		}
S
Stephen Hemminger 已提交
511

512
		tp->rx_opt.dsack = 0;
S
Stephen Hemminger 已提交
513
	}
Y
Yuchung Cheng 已提交
514 515 516

	if (unlikely(OPTION_FAST_OPEN_COOKIE & options)) {
		struct tcp_fastopen_cookie *foc = opts->fastopen_cookie;
517 518 519 520 521 522 523 524 525 526 527 528 529
		u8 *p = (u8 *)ptr;
		u32 len; /* Fast Open option length */

		if (foc->exp) {
			len = TCPOLEN_EXP_FASTOPEN_BASE + foc->len;
			*ptr = htonl((TCPOPT_EXP << 24) | (len << 16) |
				     TCPOPT_FASTOPEN_MAGIC);
			p += TCPOLEN_EXP_FASTOPEN_BASE;
		} else {
			len = TCPOLEN_FASTOPEN_BASE + foc->len;
			*p++ = TCPOPT_FASTOPEN;
			*p++ = len;
		}
Y
Yuchung Cheng 已提交
530

531 532 533 534
		memcpy(p, foc->val, foc->len);
		if ((len & 3) == 2) {
			p[foc->len] = TCPOPT_NOP;
			p[foc->len + 1] = TCPOPT_NOP;
Y
Yuchung Cheng 已提交
535
		}
536
		ptr += (len + 3) >> 2;
Y
Yuchung Cheng 已提交
537
	}
A
Adam Langley 已提交
538 539
}

540 541 542
/* Compute TCP options for SYN packets. This is not the final
 * network wire format yet.
 */
543
static unsigned int tcp_syn_options(struct sock *sk, struct sk_buff *skb,
A
Adam Langley 已提交
544
				struct tcp_out_options *opts,
545 546
				struct tcp_md5sig_key **md5)
{
A
Adam Langley 已提交
547
	struct tcp_sock *tp = tcp_sk(sk);
548
	unsigned int remaining = MAX_TCP_OPTION_SPACE;
549
	struct tcp_fastopen_request *fastopen = tp->fastopen_req;
A
Adam Langley 已提交
550

551
#ifdef CONFIG_TCP_MD5SIG
A
Adam Langley 已提交
552 553 554
	*md5 = tp->af_specific->md5_lookup(sk, sk);
	if (*md5) {
		opts->options |= OPTION_MD5;
555
		remaining -= TCPOLEN_MD5SIG_ALIGNED;
556
	}
A
Adam Langley 已提交
557 558
#else
	*md5 = NULL;
559
#endif
A
Adam Langley 已提交
560 561 562 563 564 565 566 567 568 569 570

	/* 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);
571
	remaining -= TCPOLEN_MSS_ALIGNED;
A
Adam Langley 已提交
572

573
	if (likely(sysctl_tcp_timestamps && !*md5)) {
A
Adam Langley 已提交
574
		opts->options |= OPTION_TS;
E
Eric Dumazet 已提交
575
		opts->tsval = tcp_skb_timestamp(skb) + tp->tsoffset;
A
Adam Langley 已提交
576
		opts->tsecr = tp->rx_opt.ts_recent;
577
		remaining -= TCPOLEN_TSTAMP_ALIGNED;
A
Adam Langley 已提交
578
	}
579
	if (likely(sysctl_tcp_window_scaling)) {
A
Adam Langley 已提交
580
		opts->ws = tp->rx_opt.rcv_wscale;
581
		opts->options |= OPTION_WSCALE;
582
		remaining -= TCPOLEN_WSCALE_ALIGNED;
A
Adam Langley 已提交
583
	}
584
	if (likely(sysctl_tcp_sack)) {
A
Adam Langley 已提交
585
		opts->options |= OPTION_SACK_ADVERTISE;
586
		if (unlikely(!(OPTION_TS & opts->options)))
587
			remaining -= TCPOLEN_SACKPERM_ALIGNED;
A
Adam Langley 已提交
588 589
	}

590
	if (fastopen && fastopen->cookie.len >= 0) {
591 592 593 594
		u32 need = fastopen->cookie.len;

		need += fastopen->cookie.exp ? TCPOLEN_EXP_FASTOPEN_BASE :
					       TCPOLEN_FASTOPEN_BASE;
595 596 597 598 599 600
		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;
601
			tp->syn_fastopen_exp = fastopen->cookie.exp ? 1 : 0;
602 603
		}
	}
604 605

	return MAX_TCP_OPTION_SPACE - remaining;
S
Stephen Hemminger 已提交
606 607
}

608
/* Set up TCP options for SYN-ACKs. */
609 610 611 612 613
static unsigned int tcp_synack_options(struct request_sock *req,
				       unsigned int mss, struct sk_buff *skb,
				       struct tcp_out_options *opts,
				       const struct tcp_md5sig_key *md5,
				       struct tcp_fastopen_cookie *foc)
614
{
A
Adam Langley 已提交
615
	struct inet_request_sock *ireq = inet_rsk(req);
616
	unsigned int remaining = MAX_TCP_OPTION_SPACE;
A
Adam Langley 已提交
617

618
#ifdef CONFIG_TCP_MD5SIG
E
Eric Dumazet 已提交
619
	if (md5) {
A
Adam Langley 已提交
620
		opts->options |= OPTION_MD5;
621 622 623 624 625 626 627
		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
628
		ireq->tstamp_ok &= !ireq->sack_ok;
629 630
	}
#endif
A
Adam Langley 已提交
631

632
	/* We always send an MSS option. */
A
Adam Langley 已提交
633
	opts->mss = mss;
634
	remaining -= TCPOLEN_MSS_ALIGNED;
A
Adam Langley 已提交
635 636 637

	if (likely(ireq->wscale_ok)) {
		opts->ws = ireq->rcv_wscale;
638
		opts->options |= OPTION_WSCALE;
639
		remaining -= TCPOLEN_WSCALE_ALIGNED;
A
Adam Langley 已提交
640
	}
E
Eric Dumazet 已提交
641
	if (likely(ireq->tstamp_ok)) {
A
Adam Langley 已提交
642
		opts->options |= OPTION_TS;
643
		opts->tsval = tcp_skb_timestamp(skb) + tcp_rsk(req)->ts_off;
A
Adam Langley 已提交
644
		opts->tsecr = req->ts_recent;
645
		remaining -= TCPOLEN_TSTAMP_ALIGNED;
A
Adam Langley 已提交
646 647 648
	}
	if (likely(ireq->sack_ok)) {
		opts->options |= OPTION_SACK_ADVERTISE;
E
Eric Dumazet 已提交
649
		if (unlikely(!ireq->tstamp_ok))
650
			remaining -= TCPOLEN_SACKPERM_ALIGNED;
A
Adam Langley 已提交
651
	}
652 653 654 655 656
	if (foc != NULL && foc->len >= 0) {
		u32 need = foc->len;

		need += foc->exp ? TCPOLEN_EXP_FASTOPEN_BASE :
				   TCPOLEN_FASTOPEN_BASE;
657 658 659 660 661 662 663
		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
664

665
	return MAX_TCP_OPTION_SPACE - remaining;
A
Adam Langley 已提交
666 667
}

668 669 670
/* Compute TCP options for ESTABLISHED sockets. This is not the
 * final wire format yet.
 */
671
static unsigned int tcp_established_options(struct sock *sk, struct sk_buff *skb,
A
Adam Langley 已提交
672
					struct tcp_out_options *opts,
673 674
					struct tcp_md5sig_key **md5)
{
A
Adam Langley 已提交
675
	struct tcp_sock *tp = tcp_sk(sk);
676
	unsigned int size = 0;
677
	unsigned int eff_sacks;
A
Adam Langley 已提交
678

679 680
	opts->options = 0;

A
Adam Langley 已提交
681 682 683 684 685 686 687 688 689 690 691 692
#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;
E
Eric Dumazet 已提交
693
		opts->tsval = skb ? tcp_skb_timestamp(skb) + tp->tsoffset : 0;
A
Adam Langley 已提交
694 695 696 697
		opts->tsecr = tp->rx_opt.ts_recent;
		size += TCPOLEN_TSTAMP_ALIGNED;
	}

698 699
	eff_sacks = tp->rx_opt.num_sacks + tp->rx_opt.dsack;
	if (unlikely(eff_sacks)) {
700
		const unsigned int remaining = MAX_TCP_OPTION_SPACE - size;
A
Adam Langley 已提交
701
		opts->num_sack_blocks =
702
			min_t(unsigned int, eff_sacks,
A
Adam Langley 已提交
703 704 705 706 707 708 709
			      (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
710
}
L
Linus Torvalds 已提交
711

E
Eric Dumazet 已提交
712 713 714 715 716 717 718 719 720

/* 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.
S
stephen hemminger 已提交
721
 * The invariant being skb->truesize subtracted from sk->sk_wmem_alloc
E
Eric Dumazet 已提交
722 723 724 725 726 727 728 729 730 731 732
 *
 * 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);

733 734 735 736
static void tcp_tsq_handler(struct sock *sk)
{
	if ((1 << sk->sk_state) &
	    (TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_CLOSING |
737 738 739 740 741 742 743 744
	     TCPF_CLOSE_WAIT  | TCPF_LAST_ACK)) {
		struct tcp_sock *tp = tcp_sk(sk);

		if (tp->lost_out > tp->retrans_out &&
		    tp->snd_cwnd > tcp_packets_in_flight(tp))
			tcp_xmit_retransmit_queue(sk);

		tcp_write_xmit(sk, tcp_current_mss(sk), tp->nonagle,
J
John Ogness 已提交
745
			       0, GFP_ATOMIC);
746
	}
747
}
E
Eric Dumazet 已提交
748
/*
S
stephen hemminger 已提交
749
 * One tasklet per cpu tries to send more skbs.
E
Eric Dumazet 已提交
750
 * We run in tasklet context but need to disable irqs when
S
stephen hemminger 已提交
751
 * transferring tsq->head because tcp_wfree() might
E
Eric Dumazet 已提交
752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769
 * 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);
770
		clear_bit(TSQ_QUEUED, &tp->tsq_flags);
E
Eric Dumazet 已提交
771 772

		sk = (struct sock *)tp;
773 774 775 776 777 778 779 780
		if (!sk->sk_lock.owned &&
		    test_bit(TCP_TSQ_DEFERRED, &tp->tsq_flags)) {
			bh_lock_sock(sk);
			if (!sock_owned_by_user(sk)) {
				clear_bit(TCP_TSQ_DEFERRED, &tp->tsq_flags);
				tcp_tsq_handler(sk);
			}
			bh_unlock_sock(sk);
E
Eric Dumazet 已提交
781 782 783 784 785 786
		}

		sk_free(sk);
	}
}

E
Eric Dumazet 已提交
787 788 789 790
#define TCP_DEFERRED_ALL (TCPF_TSQ_DEFERRED |		\
			  TCPF_WRITE_TIMER_DEFERRED |	\
			  TCPF_DELACK_TIMER_DEFERRED |	\
			  TCPF_MTU_REDUCED_DEFERRED)
E
Eric Dumazet 已提交
791 792 793 794 795 796 797 798 799 800
/**
 * 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);
801
	unsigned long flags, nflags;
E
Eric Dumazet 已提交
802

803 804 805 806 807 808 809 810
	/* 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);

E
Eric Dumazet 已提交
811
	if (flags & TCPF_TSQ_DEFERRED)
812 813
		tcp_tsq_handler(sk);

814 815 816 817 818 819 820 821 822 823 824
	/* Here begins the tricky part :
	 * We are called from release_sock() with :
	 * 1) BH disabled
	 * 2) sk_lock.slock spinlock held
	 * 3) socket owned by us (sk->sk_lock.owned == 1)
	 *
	 * But following code is meant to be called from BH handlers,
	 * so we should keep BH disabled, but early release socket ownership
	 */
	sock_release_ownership(sk);

E
Eric Dumazet 已提交
825
	if (flags & TCPF_WRITE_TIMER_DEFERRED) {
826
		tcp_write_timer_handler(sk);
827 828
		__sock_put(sk);
	}
E
Eric Dumazet 已提交
829
	if (flags & TCPF_DELACK_TIMER_DEFERRED) {
830
		tcp_delack_timer_handler(sk);
831 832
		__sock_put(sk);
	}
E
Eric Dumazet 已提交
833
	if (flags & TCPF_MTU_REDUCED_DEFERRED) {
834
		inet_csk(sk)->icsk_af_ops->mtu_reduced(sk);
835 836
		__sock_put(sk);
	}
E
Eric Dumazet 已提交
837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855
}
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.
S
stephen hemminger 已提交
856
 * We can't xmit new skbs from this context, as we might already
E
Eric Dumazet 已提交
857 858
 * hold qdisc lock.
 */
E
Eric Dumazet 已提交
859
void tcp_wfree(struct sk_buff *skb)
E
Eric Dumazet 已提交
860 861 862
{
	struct sock *sk = skb->sk;
	struct tcp_sock *tp = tcp_sk(sk);
863
	unsigned long flags, nval, oval;
864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879
	int wmem;

	/* Keep one reference on sk_wmem_alloc.
	 * Will be released by sk_free() from here or tcp_tasklet_func()
	 */
	wmem = atomic_sub_return(skb->truesize - 1, &sk->sk_wmem_alloc);

	/* If this softirq is serviced by ksoftirqd, we are likely under stress.
	 * Wait until our queues (qdisc + devices) are drained.
	 * This gives :
	 * - less callbacks to tcp_write_xmit(), reducing stress (batches)
	 * - chance for incoming ACK (processed by another cpu maybe)
	 *   to migrate this flow (skb->ooo_okay will be eventually set)
	 */
	if (wmem >= SKB_TRUESIZE(1) && this_cpu_ksoftirqd() == current)
		goto out;
E
Eric Dumazet 已提交
880

881
	for (oval = READ_ONCE(tp->tsq_flags);; oval = nval) {
E
Eric Dumazet 已提交
882 883
		struct tsq_tasklet *tsq;

884 885 886
		if (!(oval & TSQF_THROTTLED) || (oval & TSQF_QUEUED))
			goto out;

887
		nval = (oval & ~TSQF_THROTTLED) | TSQF_QUEUED | TCPF_TSQ_DEFERRED;
888 889 890 891
		nval = cmpxchg(&tp->tsq_flags, oval, nval);
		if (nval != oval)
			continue;

E
Eric Dumazet 已提交
892 893
		/* queue this socket to tasklet queue */
		local_irq_save(flags);
894
		tsq = this_cpu_ptr(&tsq_tasklet);
E
Eric Dumazet 已提交
895 896 897
		list_add(&tp->tsq_node, &tsq->head);
		tasklet_schedule(&tsq->tasklet);
		local_irq_restore(flags);
898
		return;
E
Eric Dumazet 已提交
899
	}
900 901
out:
	sk_free(sk);
E
Eric Dumazet 已提交
902 903
}

L
Linus Torvalds 已提交
904 905 906 907 908 909 910 911 912 913 914
/* 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.
 */
915 916
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
917
{
918 919 920 921
	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
922
	struct tcp_out_options opts;
923
	unsigned int tcp_options_size, tcp_header_size;
924
	struct tcp_md5sig_key *md5;
925 926 927 928
	struct tcphdr *th;
	int err;

	BUG_ON(!skb || !tcp_skb_pcount(skb));
929
	tp = tcp_sk(sk);
930

931
	if (clone_it) {
932
		skb_mstamp_get(&skb->skb_mstamp);
933 934
		TCP_SKB_CB(skb)->tx.in_flight = TCP_SKB_CB(skb)->end_seq
			- tp->snd_una;
935
		tcp_rate_skb_sent(sk, skb);
936

937 938 939 940 941 942 943
		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
944

945 946
	inet = inet_sk(sk);
	tcb = TCP_SKB_CB(skb);
A
Adam Langley 已提交
947
	memset(&opts, 0, sizeof(opts));
L
Linus Torvalds 已提交
948

E
Eric Dumazet 已提交
949
	if (unlikely(tcb->tcp_flags & TCPHDR_SYN))
A
Adam Langley 已提交
950 951 952 953 954
		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);
955

E
Eric Dumazet 已提交
956
	/* if no packet is in qdisc/device queue, then allow XPS to select
957 958 959 960 961
	 * another queue. We can be called from tcp_tsq_handler()
	 * which holds one reference to sk_wmem_alloc.
	 *
	 * TODO: Ideally, in-flight pure ACK packets should not matter here.
	 * One way to get this would be to set skb->truesize = 2 on them.
E
Eric Dumazet 已提交
962
	 */
963
	skb->ooo_okay = sk_wmem_alloc_get(sk) < SKB_TRUESIZE(1);
964

965 966
	skb_push(skb, tcp_header_size);
	skb_reset_transport_header(skb);
E
Eric Dumazet 已提交
967 968 969

	skb_orphan(skb);
	skb->sk = sk;
E
Eric Dumazet 已提交
970
	skb->destructor = skb_is_tcp_pure_ack(skb) ? __sock_wfree : tcp_wfree;
971
	skb_set_hash_from_sk(skb, sk);
E
Eric Dumazet 已提交
972
	atomic_add(skb->truesize, &sk->sk_wmem_alloc);
973 974

	/* Build TCP header and checksum it. */
975
	th = (struct tcphdr *)skb->data;
E
Eric Dumazet 已提交
976 977
	th->source		= inet->inet_sport;
	th->dest		= inet->inet_dport;
978 979
	th->seq			= htonl(tcb->seq);
	th->ack_seq		= htonl(tp->rcv_nxt);
980
	*(((__be16 *)th) + 6)	= htons(((tcp_header_size >> 2) << 12) |
E
Eric Dumazet 已提交
981
					tcb->tcp_flags);
982 983 984

	th->check		= 0;
	th->urg_ptr		= 0;
L
Linus Torvalds 已提交
985

I
Ilpo Järvinen 已提交
986
	/* The urg_mode check is necessary during a below snd_una win probe */
987 988 989 990 991
	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)) {
992
			th->urg_ptr = htons(0xFFFF);
993 994
			th->urg = 1;
		}
995
	}
L
Linus Torvalds 已提交
996

997
	tcp_options_write((__be32 *)(th + 1), tp, &opts);
998
	skb_shinfo(skb)->gso_type = sk->sk_gso_type;
999 1000 1001 1002 1003 1004 1005 1006 1007
	if (likely(!(tcb->tcp_flags & TCPHDR_SYN))) {
		th->window      = htons(tcp_select_window(sk));
		tcp_ecn_send(sk, skb, th, tcp_header_size);
	} else {
		/* RFC1323: The window in SYN & SYN/ACK segments
		 * is never scaled.
		 */
		th->window	= htons(min(tp->rcv_wnd, 65535U));
	}
1008 1009 1010
#ifdef CONFIG_TCP_MD5SIG
	/* Calculate the MD5 hash, as we have all we need now */
	if (md5) {
E
Eric Dumazet 已提交
1011
		sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1012
		tp->af_specific->calc_md5_hash(opts.hash_location,
1013
					       md5, sk, skb);
1014 1015 1016
	}
#endif

1017
	icsk->icsk_af_ops->send_check(sk, skb);
L
Linus Torvalds 已提交
1018

E
Eric Dumazet 已提交
1019
	if (likely(tcb->tcp_flags & TCPHDR_ACK))
1020
		tcp_event_ack_sent(sk, tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
1021

1022
	if (skb->len != tcp_header_size) {
1023
		tcp_event_data_sent(tp, sk);
1024 1025
		tp->data_segs_out += tcp_skb_pcount(skb);
	}
L
Linus Torvalds 已提交
1026

1027
	if (after(tcb->end_seq, tp->snd_nxt) || tcb->seq == tcb->end_seq)
1028 1029
		TCP_ADD_STATS(sock_net(sk), TCP_MIB_OUTSEGS,
			      tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
1030

1031
	tp->segs_out += tcp_skb_pcount(skb);
1032
	/* OK, its time to fill skb_shinfo(skb)->gso_{segs|size} */
1033
	skb_shinfo(skb)->gso_segs = tcp_skb_pcount(skb);
1034
	skb_shinfo(skb)->gso_size = tcp_skb_mss(skb);
1035

E
Eric Dumazet 已提交
1036 1037
	/* Our usage of tstamp should remain private */
	skb->tstamp.tv64 = 0;
1038 1039 1040 1041 1042

	/* Cleanup our debris for IP stacks */
	memset(skb->cb, 0, max(sizeof(struct inet_skb_parm),
			       sizeof(struct inet6_skb_parm)));

1043
	err = icsk->icsk_af_ops->queue_xmit(sk, skb, &inet->cork.fl);
E
Eric Dumazet 已提交
1044

1045
	if (likely(err <= 0))
1046 1047
		return err;

1048
	tcp_enter_cwr(sk);
1049

1050
	return net_xmit_eval(err);
L
Linus Torvalds 已提交
1051 1052
}

1053
/* This routine just queues the buffer for sending.
L
Linus Torvalds 已提交
1054 1055 1056 1057 1058 1059 1060 1061 1062 1063
 *
 * 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;
1064
	__skb_header_release(skb);
1065
	tcp_add_write_queue_tail(sk, skb);
1066 1067
	sk->sk_wmem_queued += skb->truesize;
	sk_mem_charge(sk, skb->truesize);
L
Linus Torvalds 已提交
1068 1069
}

1070
/* Initialize TSO segments for a packet. */
1071
static void tcp_set_skb_tso_segs(struct sk_buff *skb, unsigned int mss_now)
1072
{
1073
	if (skb->len <= mss_now || skb->ip_summed == CHECKSUM_NONE) {
1074 1075 1076
		/* Avoid the costly divide in the normal
		 * non-TSO case.
		 */
1077
		tcp_skb_pcount_set(skb, 1);
1078
		TCP_SKB_CB(skb)->tcp_gso_size = 0;
1079
	} else {
1080
		tcp_skb_pcount_set(skb, DIV_ROUND_UP(skb->len, mss_now));
1081
		TCP_SKB_CB(skb)->tcp_gso_size = mss_now;
L
Linus Torvalds 已提交
1082 1083 1084
	}
}

I
Ilpo Järvinen 已提交
1085
/* When a modification to fackets out becomes necessary, we need to check
1086
 * skb is counted to fackets_out or not.
I
Ilpo Järvinen 已提交
1087
 */
1088
static void tcp_adjust_fackets_out(struct sock *sk, const struct sk_buff *skb,
I
Ilpo Järvinen 已提交
1089 1090
				   int decr)
{
1091 1092
	struct tcp_sock *tp = tcp_sk(sk);

1093
	if (!tp->sacked_out || tcp_is_reno(tp))
I
Ilpo Järvinen 已提交
1094 1095
		return;

1096
	if (after(tcp_highest_sack_seq(tp), TCP_SKB_CB(skb)->seq))
I
Ilpo Järvinen 已提交
1097 1098 1099
		tp->fackets_out -= decr;
}

1100 1101 1102
/* Pcount in the middle of the write queue got changed, we need to do various
 * tweaks to fix counters
 */
1103
static void tcp_adjust_pcount(struct sock *sk, const struct sk_buff *skb, int decr)
1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123
{
	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) &&
1124
	    (tcp_is_fack(tp) || (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)))
1125 1126 1127 1128 1129
		tp->lost_cnt_hint -= decr;

	tcp_verify_left_out(tp);
}

1130 1131 1132 1133 1134 1135
static bool tcp_has_tx_tstamp(const struct sk_buff *skb)
{
	return TCP_SKB_CB(skb)->txstamp_ack ||
		(skb_shinfo(skb)->tx_flags & SKBTX_ANY_TSTAMP);
}

1136 1137 1138 1139
static void tcp_fragment_tstamp(struct sk_buff *skb, struct sk_buff *skb2)
{
	struct skb_shared_info *shinfo = skb_shinfo(skb);

1140
	if (unlikely(tcp_has_tx_tstamp(skb)) &&
1141 1142 1143 1144 1145 1146 1147
	    !before(shinfo->tskey, TCP_SKB_CB(skb2)->seq)) {
		struct skb_shared_info *shinfo2 = skb_shinfo(skb2);
		u8 tsflags = shinfo->tx_flags & SKBTX_ANY_TSTAMP;

		shinfo->tx_flags &= ~tsflags;
		shinfo2->tx_flags |= tsflags;
		swap(shinfo->tskey, shinfo2->tskey);
1148 1149
		TCP_SKB_CB(skb2)->txstamp_ack = TCP_SKB_CB(skb)->txstamp_ack;
		TCP_SKB_CB(skb)->txstamp_ack = 0;
1150 1151 1152
	}
}

1153 1154 1155 1156 1157 1158
static void tcp_skb_fragment_eor(struct sk_buff *skb, struct sk_buff *skb2)
{
	TCP_SKB_CB(skb2)->eor = TCP_SKB_CB(skb)->eor;
	TCP_SKB_CB(skb)->eor = 0;
}

L
Linus Torvalds 已提交
1159 1160
/* 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
1161
 * packet to the list.  This won't be called frequently, I hope.
L
Linus Torvalds 已提交
1162 1163
 * Remember, these are still headerless SKBs at this point.
 */
1164
int tcp_fragment(struct sock *sk, struct sk_buff *skb, u32 len,
1165
		 unsigned int mss_now, gfp_t gfp)
L
Linus Torvalds 已提交
1166 1167 1168
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *buff;
1169
	int nsize, old_factor;
1170
	int nlen;
1171
	u8 flags;
L
Linus Torvalds 已提交
1172

1173 1174
	if (WARN_ON(len > skb->len))
		return -EINVAL;
1175

L
Linus Torvalds 已提交
1176 1177 1178 1179
	nsize = skb_headlen(skb) - len;
	if (nsize < 0)
		nsize = 0;

1180
	if (skb_unclone(skb, gfp))
L
Linus Torvalds 已提交
1181 1182 1183
		return -ENOMEM;

	/* Get a new skb... force flag on. */
1184
	buff = sk_stream_alloc_skb(sk, nsize, gfp, true);
1185
	if (!buff)
L
Linus Torvalds 已提交
1186
		return -ENOMEM; /* We'll just try again later. */
H
Herbert Xu 已提交
1187

1188 1189
	sk->sk_wmem_queued += buff->truesize;
	sk_mem_charge(sk, buff->truesize);
1190 1191 1192
	nlen = skb->len - len - nsize;
	buff->truesize += nlen;
	skb->truesize -= nlen;
L
Linus Torvalds 已提交
1193 1194 1195 1196 1197 1198 1199

	/* 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
1200 1201 1202
	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;
1203
	TCP_SKB_CB(buff)->sacked = TCP_SKB_CB(skb)->sacked;
1204
	tcp_skb_fragment_eor(skb, buff);
L
Linus Torvalds 已提交
1205

1206
	if (!skb_shinfo(skb)->nr_frags && skb->ip_summed != CHECKSUM_PARTIAL) {
L
Linus Torvalds 已提交
1207
		/* Copy and checksum data tail into the new buffer. */
1208 1209
		buff->csum = csum_partial_copy_nocheck(skb->data + len,
						       skb_put(buff, nsize),
L
Linus Torvalds 已提交
1210 1211 1212 1213 1214 1215
						       nsize, 0);

		skb_trim(skb, len);

		skb->csum = csum_block_sub(skb->csum, buff->csum, len);
	} else {
1216
		skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
1217 1218 1219 1220 1221
		skb_split(skb, buff, len);
	}

	buff->ip_summed = skb->ip_summed;

1222
	buff->tstamp = skb->tstamp;
1223
	tcp_fragment_tstamp(skb, buff);
L
Linus Torvalds 已提交
1224

1225 1226
	old_factor = tcp_skb_pcount(skb);

L
Linus Torvalds 已提交
1227
	/* Fix up tso_factor for both original and new SKB.  */
1228 1229
	tcp_set_skb_tso_segs(skb, mss_now);
	tcp_set_skb_tso_segs(buff, mss_now);
L
Linus Torvalds 已提交
1230

1231 1232 1233
	/* Update delivered info for the new segment */
	TCP_SKB_CB(buff)->tx = TCP_SKB_CB(skb)->tx;

1234 1235 1236
	/* If this packet has been sent out already, we must
	 * adjust the various packet counters.
	 */
1237
	if (!before(tp->snd_nxt, TCP_SKB_CB(buff)->end_seq)) {
1238 1239
		int diff = old_factor - tcp_skb_pcount(skb) -
			tcp_skb_pcount(buff);
L
Linus Torvalds 已提交
1240

1241 1242
		if (diff)
			tcp_adjust_pcount(sk, skb, diff);
L
Linus Torvalds 已提交
1243 1244 1245
	}

	/* Link BUFF into the send queue. */
1246
	__skb_header_release(buff);
1247
	tcp_insert_write_queue_after(skb, buff, sk);
L
Linus Torvalds 已提交
1248 1249 1250 1251 1252 1253 1254 1255

	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.
 */
1256
static void __pskb_trim_head(struct sk_buff *skb, int len)
L
Linus Torvalds 已提交
1257
{
1258
	struct skb_shared_info *shinfo;
L
Linus Torvalds 已提交
1259 1260
	int i, k, eat;

E
Eric Dumazet 已提交
1261 1262 1263 1264 1265 1266 1267
	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
1268 1269
	eat = len;
	k = 0;
1270 1271 1272
	shinfo = skb_shinfo(skb);
	for (i = 0; i < shinfo->nr_frags; i++) {
		int size = skb_frag_size(&shinfo->frags[i]);
E
Eric Dumazet 已提交
1273 1274

		if (size <= eat) {
1275
			skb_frag_unref(skb, i);
E
Eric Dumazet 已提交
1276
			eat -= size;
L
Linus Torvalds 已提交
1277
		} else {
1278
			shinfo->frags[k] = shinfo->frags[i];
L
Linus Torvalds 已提交
1279
			if (eat) {
1280 1281
				shinfo->frags[k].page_offset += eat;
				skb_frag_size_sub(&shinfo->frags[k], eat);
L
Linus Torvalds 已提交
1282 1283 1284 1285 1286
				eat = 0;
			}
			k++;
		}
	}
1287
	shinfo->nr_frags = k;
L
Linus Torvalds 已提交
1288

1289
	skb_reset_tail_pointer(skb);
L
Linus Torvalds 已提交
1290 1291 1292 1293
	skb->data_len -= len;
	skb->len = skb->data_len;
}

1294
/* Remove acked data from a packet in the transmit queue. */
L
Linus Torvalds 已提交
1295 1296
int tcp_trim_head(struct sock *sk, struct sk_buff *skb, u32 len)
{
1297
	if (skb_unclone(skb, GFP_ATOMIC))
L
Linus Torvalds 已提交
1298 1299
		return -ENOMEM;

E
Eric Dumazet 已提交
1300
	__pskb_trim_head(skb, len);
L
Linus Torvalds 已提交
1301 1302

	TCP_SKB_CB(skb)->seq += len;
1303
	skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
1304 1305 1306

	skb->truesize	     -= len;
	sk->sk_wmem_queued   -= len;
1307
	sk_mem_uncharge(sk, len);
L
Linus Torvalds 已提交
1308 1309
	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);

1310
	/* Any change of skb->len requires recalculation of tso factor. */
L
Linus Torvalds 已提交
1311
	if (tcp_skb_pcount(skb) > 1)
1312
		tcp_set_skb_tso_segs(skb, tcp_skb_mss(skb));
L
Linus Torvalds 已提交
1313 1314 1315 1316

	return 0;
}

1317 1318
/* 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
1319
{
1320 1321
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1322 1323 1324 1325 1326 1327 1328
	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);

1329 1330 1331 1332 1333 1334 1335 1336
	/* 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
1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349
	/* 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;
}

1350 1351 1352 1353 1354 1355 1356 1357
/* 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
1358
/* Inverse of above */
1359
int tcp_mss_to_mtu(struct sock *sk, int mss)
J
John Heffner 已提交
1360
{
1361 1362
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1363 1364 1365 1366 1367 1368 1369
	int mtu;

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

1370 1371 1372 1373 1374 1375 1376
	/* 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
1377 1378
	return mtu;
}
1379
EXPORT_SYMBOL(tcp_mss_to_mtu);
J
John Heffner 已提交
1380

1381
/* MTU probing init per socket */
J
John Heffner 已提交
1382 1383 1384 1385
void tcp_mtup_init(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct inet_connection_sock *icsk = inet_csk(sk);
F
Fan Du 已提交
1386
	struct net *net = sock_net(sk);
J
John Heffner 已提交
1387

F
Fan Du 已提交
1388
	icsk->icsk_mtup.enabled = net->ipv4.sysctl_tcp_mtu_probing > 1;
J
John Heffner 已提交
1389
	icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp + sizeof(struct tcphdr) +
1390
			       icsk->icsk_af_ops->net_header_len;
F
Fan Du 已提交
1391
	icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, net->ipv4.sysctl_tcp_base_mss);
J
John Heffner 已提交
1392
	icsk->icsk_mtup.probe_size = 0;
1393 1394
	if (icsk->icsk_mtup.enabled)
		icsk->icsk_mtup.probe_timestamp = tcp_time_stamp;
J
John Heffner 已提交
1395
}
E
Eric Dumazet 已提交
1396
EXPORT_SYMBOL(tcp_mtup_init);
J
John Heffner 已提交
1397

L
Linus Torvalds 已提交
1398 1399 1400 1401 1402 1403
/* 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
1404
   It is minimum of user_mss and mss received with SYN.
L
Linus Torvalds 已提交
1405 1406
   It also does not include TCP options.

1407
   inet_csk(sk)->icsk_pmtu_cookie is last pmtu, seen by this function.
L
Linus Torvalds 已提交
1408 1409 1410 1411 1412 1413 1414 1415 1416

   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.

1417 1418
   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
1419 1420 1421 1422
 */
unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu)
{
	struct tcp_sock *tp = tcp_sk(sk);
1423
	struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1424
	int mss_now;
L
Linus Torvalds 已提交
1425

J
John Heffner 已提交
1426 1427
	if (icsk->icsk_mtup.search_high > pmtu)
		icsk->icsk_mtup.search_high = pmtu;
L
Linus Torvalds 已提交
1428

J
John Heffner 已提交
1429
	mss_now = tcp_mtu_to_mss(sk, pmtu);
1430
	mss_now = tcp_bound_to_half_wnd(tp, mss_now);
L
Linus Torvalds 已提交
1431 1432

	/* And store cached results */
1433
	icsk->icsk_pmtu_cookie = pmtu;
J
John Heffner 已提交
1434 1435
	if (icsk->icsk_mtup.enabled)
		mss_now = min(mss_now, tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low));
1436
	tp->mss_cache = mss_now;
L
Linus Torvalds 已提交
1437 1438 1439

	return mss_now;
}
E
Eric Dumazet 已提交
1440
EXPORT_SYMBOL(tcp_sync_mss);
L
Linus Torvalds 已提交
1441 1442 1443 1444

/* Compute the current effective MSS, taking SACKs and IP options,
 * and even PMTU discovery events into account.
 */
I
Ilpo Järvinen 已提交
1445
unsigned int tcp_current_mss(struct sock *sk)
L
Linus Torvalds 已提交
1446
{
1447 1448
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct dst_entry *dst = __sk_dst_get(sk);
1449
	u32 mss_now;
1450
	unsigned int header_len;
A
Adam Langley 已提交
1451 1452
	struct tcp_out_options opts;
	struct tcp_md5sig_key *md5;
1453 1454 1455

	mss_now = tp->mss_cache;

L
Linus Torvalds 已提交
1456 1457
	if (dst) {
		u32 mtu = dst_mtu(dst);
1458
		if (mtu != inet_csk(sk)->icsk_pmtu_cookie)
L
Linus Torvalds 已提交
1459 1460 1461
			mss_now = tcp_sync_mss(sk, mtu);
	}

A
Adam Langley 已提交
1462 1463 1464 1465 1466 1467 1468 1469 1470 1471
	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;
	}
1472

L
Linus Torvalds 已提交
1473 1474 1475
	return mss_now;
}

1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497
/* RFC2861, slow part. Adjust cwnd, after it was not full during one rto.
 * As additional protections, we do not touch cwnd in retransmission phases,
 * and if application hit its sndbuf limit recently.
 */
static void tcp_cwnd_application_limited(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);

	if (inet_csk(sk)->icsk_ca_state == TCP_CA_Open &&
	    sk->sk_socket && !test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
		/* Limited by application or receiver window. */
		u32 init_win = tcp_init_cwnd(tp, __sk_dst_get(sk));
		u32 win_used = max(tp->snd_cwnd_used, init_win);
		if (win_used < tp->snd_cwnd) {
			tp->snd_ssthresh = tcp_current_ssthresh(sk);
			tp->snd_cwnd = (tp->snd_cwnd + win_used) >> 1;
		}
		tp->snd_cwnd_used = 0;
	}
	tp->snd_cwnd_stamp = tcp_time_stamp;
}

1498
static void tcp_cwnd_validate(struct sock *sk, bool is_cwnd_limited)
1499
{
1500
	struct tcp_sock *tp = tcp_sk(sk);
1501

1502 1503 1504 1505 1506 1507 1508 1509 1510
	/* Track the maximum number of outstanding packets in each
	 * window, and remember whether we were cwnd-limited then.
	 */
	if (!before(tp->snd_una, tp->max_packets_seq) ||
	    tp->packets_out > tp->max_packets_out) {
		tp->max_packets_out = tp->packets_out;
		tp->max_packets_seq = tp->snd_nxt;
		tp->is_cwnd_limited = is_cwnd_limited;
	}
1511

1512
	if (tcp_is_cwnd_limited(sk)) {
1513 1514 1515 1516 1517 1518 1519 1520
		/* 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;

1521 1522
		if (sysctl_tcp_slow_start_after_idle &&
		    (s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= inet_csk(sk)->icsk_rto)
1523
			tcp_cwnd_application_limited(sk);
1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535

		/* The following conditions together indicate the starvation
		 * is caused by insufficient sender buffer:
		 * 1) just sent some data (see tcp_write_xmit)
		 * 2) not cwnd limited (this else condition)
		 * 3) no more data to send (null tcp_send_head )
		 * 4) application is hitting buffer limit (SOCK_NOSPACE)
		 */
		if (!tcp_send_head(sk) && sk->sk_socket &&
		    test_bit(SOCK_NOSPACE, &sk->sk_socket->flags) &&
		    (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))
			tcp_chrono_start(sk, TCP_CHRONO_SNDBUF_LIMITED);
1536 1537 1538
	}
}

E
Eric Dumazet 已提交
1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552
/* Minshall's variant of the Nagle send check. */
static bool tcp_minshall_check(const struct tcp_sock *tp)
{
	return after(tp->snd_sml, tp->snd_una) &&
		!after(tp->snd_sml, tp->snd_nxt);
}

/* Update snd_sml if this skb is under mss
 * Note that a TSO packet might end with a sub-mss segment
 * The test is really :
 * if ((skb->len % mss) != 0)
 *        tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
 * But we can avoid doing the divide again given we already have
 *  skb_pcount = skb->len / mss_now
1553
 */
E
Eric Dumazet 已提交
1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568
static void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss_now,
				const struct sk_buff *skb)
{
	if (skb->len < tcp_skb_pcount(skb) * mss_now)
		tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
}

/* Return false, if packet can be sent now without violation Nagle's rules:
 * 1. It is full sized. (provided by caller in %partial bool)
 * 2. Or it contains FIN. (already checked by caller)
 * 3. Or TCP_CORK is not set, and TCP_NODELAY is set.
 * 4. Or TCP_CORK is not set, and all sent packets are ACKed.
 *    With Minshall's modification: all sent small packets are ACKed.
 */
static bool tcp_nagle_check(bool partial, const struct tcp_sock *tp,
1569
			    int nonagle)
E
Eric Dumazet 已提交
1570 1571 1572 1573 1574
{
	return partial &&
		((nonagle & TCP_NAGLE_CORK) ||
		 (!nonagle && tp->packets_out && tcp_minshall_check(tp)));
}
E
Eric Dumazet 已提交
1575 1576 1577 1578

/* Return how many segs we'd like on a TSO packet,
 * to send one TSO packet per ms
 */
1579 1580
u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
		     int min_tso_segs)
E
Eric Dumazet 已提交
1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591
{
	u32 bytes, segs;

	bytes = min(sk->sk_pacing_rate >> 10,
		    sk->sk_gso_max_size - 1 - MAX_TCP_HEADER);

	/* Goal is to send at least one packet per ms,
	 * not one big TSO packet every 100 ms.
	 * This preserves ACK clocking and is consistent
	 * with tcp_tso_should_defer() heuristic.
	 */
1592
	segs = max_t(u32, bytes / mss_now, min_tso_segs);
E
Eric Dumazet 已提交
1593 1594 1595

	return min_t(u32, segs, sk->sk_gso_max_segs);
}
1596
EXPORT_SYMBOL(tcp_tso_autosize);
E
Eric Dumazet 已提交
1597

1598 1599 1600 1601 1602 1603 1604 1605
/* Return the number of segments we want in the skb we are transmitting.
 * See if congestion control module wants to decide; otherwise, autosize.
 */
static u32 tcp_tso_segs(struct sock *sk, unsigned int mss_now)
{
	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
	u32 tso_segs = ca_ops->tso_segs_goal ? ca_ops->tso_segs_goal(sk) : 0;

1606 1607
	return tso_segs ? :
		tcp_tso_autosize(sk, mss_now, sysctl_tcp_min_tso_segs);
1608 1609
}

E
Eric Dumazet 已提交
1610 1611 1612 1613 1614 1615
/* Returns the portion of skb which can be sent right away */
static unsigned int tcp_mss_split_point(const struct sock *sk,
					const struct sk_buff *skb,
					unsigned int mss_now,
					unsigned int max_segs,
					int nonagle)
1616
{
1617
	const struct tcp_sock *tp = tcp_sk(sk);
E
Eric Dumazet 已提交
1618
	u32 partial, needed, window, max_len;
1619

1620
	window = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
1621
	max_len = mss_now * max_segs;
1622

1623 1624
	if (likely(max_len <= window && skb != tcp_write_queue_tail(sk)))
		return max_len;
1625

1626 1627
	needed = min(skb->len, window);

1628 1629
	if (max_len <= needed)
		return max_len;
1630

E
Eric Dumazet 已提交
1631 1632 1633 1634 1635
	partial = needed % mss_now;
	/* If last segment is not a full MSS, check if Nagle rules allow us
	 * to include this last segment in this skb.
	 * Otherwise, we'll split the skb at last MSS boundary
	 */
1636
	if (tcp_nagle_check(partial != 0, tp, nonagle))
E
Eric Dumazet 已提交
1637 1638 1639
		return needed - partial;

	return needed;
1640 1641 1642 1643 1644
}

/* 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.
 */
1645 1646
static inline unsigned int tcp_cwnd_test(const struct tcp_sock *tp,
					 const struct sk_buff *skb)
1647
{
E
Eric Dumazet 已提交
1648
	u32 in_flight, cwnd, halfcwnd;
1649 1650

	/* Don't be strict about the congestion window for the final FIN.  */
E
Eric Dumazet 已提交
1651 1652
	if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) &&
	    tcp_skb_pcount(skb) == 1)
1653 1654 1655 1656
		return 1;

	in_flight = tcp_packets_in_flight(tp);
	cwnd = tp->snd_cwnd;
E
Eric Dumazet 已提交
1657 1658
	if (in_flight >= cwnd)
		return 0;
1659

E
Eric Dumazet 已提交
1660 1661 1662 1663 1664
	/* For better scheduling, ensure we have at least
	 * 2 GSO packets in flight.
	 */
	halfcwnd = max(cwnd >> 1, 1U);
	return min(halfcwnd, cwnd - in_flight);
1665 1666
}

1667
/* Initialize TSO state of a skb.
1668
 * This must be invoked the first time we consider transmitting
1669 1670
 * SKB onto the wire.
 */
1671
static int tcp_init_tso_segs(struct sk_buff *skb, unsigned int mss_now)
1672 1673 1674
{
	int tso_segs = tcp_skb_pcount(skb);

1675
	if (!tso_segs || (tso_segs > 1 && tcp_skb_mss(skb) != mss_now)) {
1676
		tcp_set_skb_tso_segs(skb, mss_now);
1677 1678 1679 1680 1681 1682
		tso_segs = tcp_skb_pcount(skb);
	}
	return tso_segs;
}


E
Eric Dumazet 已提交
1683
/* Return true if the Nagle test allows this packet to be
1684 1685
 * sent now.
 */
E
Eric Dumazet 已提交
1686 1687
static inline bool tcp_nagle_test(const struct tcp_sock *tp, const struct sk_buff *skb,
				  unsigned int cur_mss, int nonagle)
1688 1689 1690 1691 1692 1693 1694 1695
{
	/* 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
1696
		return true;
1697

Y
Yuchung Cheng 已提交
1698 1699
	/* 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
1700
		return true;
1701

1702
	if (!tcp_nagle_check(skb->len < cur_mss, tp, nonagle))
E
Eric Dumazet 已提交
1703
		return true;
1704

E
Eric Dumazet 已提交
1705
	return false;
1706 1707 1708
}

/* Does at least the first segment of SKB fit into the send window? */
E
Eric Dumazet 已提交
1709 1710 1711
static bool tcp_snd_wnd_test(const struct tcp_sock *tp,
			     const struct sk_buff *skb,
			     unsigned int cur_mss)
1712 1713 1714 1715 1716 1717
{
	u32 end_seq = TCP_SKB_CB(skb)->end_seq;

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

1718
	return !after(end_seq, tcp_wnd_end(tp));
1719 1720
}

1721
/* This checks if the data bearing packet SKB (usually tcp_send_head(sk))
1722 1723 1724
 * should be put on the wire right now.  If so, it returns the number of
 * packets allowed by the congestion window.
 */
1725
static unsigned int tcp_snd_test(const struct sock *sk, struct sk_buff *skb,
1726 1727
				 unsigned int cur_mss, int nonagle)
{
1728
	const struct tcp_sock *tp = tcp_sk(sk);
1729 1730
	unsigned int cwnd_quota;

1731
	tcp_init_tso_segs(skb, cur_mss);
1732 1733 1734 1735 1736

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

	cwnd_quota = tcp_cwnd_test(tp, skb);
1737
	if (cwnd_quota && !tcp_snd_wnd_test(tp, skb, cur_mss))
1738 1739 1740 1741 1742
		cwnd_quota = 0;

	return cwnd_quota;
}

1743
/* Test if sending is allowed right now. */
E
Eric Dumazet 已提交
1744
bool tcp_may_send_now(struct sock *sk)
1745
{
1746
	const struct tcp_sock *tp = tcp_sk(sk);
1747
	struct sk_buff *skb = tcp_send_head(sk);
1748

E
Eric Dumazet 已提交
1749
	return skb &&
I
Ilpo Järvinen 已提交
1750
		tcp_snd_test(sk, skb, tcp_current_mss(sk),
1751
			     (tcp_skb_is_last(sk, skb) ?
E
Eric Dumazet 已提交
1752
			      tp->nonagle : TCP_NAGLE_PUSH));
1753 1754 1755 1756 1757 1758 1759 1760 1761
}

/* 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).
 */
1762
static int tso_fragment(struct sock *sk, struct sk_buff *skb, unsigned int len,
1763
			unsigned int mss_now, gfp_t gfp)
1764 1765 1766
{
	struct sk_buff *buff;
	int nlen = skb->len - len;
1767
	u8 flags;
1768 1769

	/* All of a TSO frame must be composed of paged data.  */
1770
	if (skb->len != skb->data_len)
1771
		return tcp_fragment(sk, skb, len, mss_now, gfp);
1772

1773
	buff = sk_stream_alloc_skb(sk, 0, gfp, true);
1774
	if (unlikely(!buff))
1775 1776
		return -ENOMEM;

1777 1778
	sk->sk_wmem_queued += buff->truesize;
	sk_mem_charge(sk, buff->truesize);
1779
	buff->truesize += nlen;
1780 1781 1782 1783 1784 1785 1786 1787
	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
1788 1789 1790
	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;
1791 1792 1793 1794

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

1795 1796
	tcp_skb_fragment_eor(skb, buff);

1797
	buff->ip_summed = skb->ip_summed = CHECKSUM_PARTIAL;
1798
	skb_split(skb, buff, len);
1799
	tcp_fragment_tstamp(skb, buff);
1800 1801

	/* Fix up tso_factor for both original and new SKB.  */
1802 1803
	tcp_set_skb_tso_segs(skb, mss_now);
	tcp_set_skb_tso_segs(buff, mss_now);
1804 1805

	/* Link BUFF into the send queue. */
1806
	__skb_header_release(buff);
1807
	tcp_insert_write_queue_after(skb, buff, sk);
1808 1809 1810 1811 1812 1813 1814 1815 1816

	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.
 */
1817
static bool tcp_tso_should_defer(struct sock *sk, struct sk_buff *skb,
E
Eric Dumazet 已提交
1818
				 bool *is_cwnd_limited, u32 max_segs)
1819
{
1820
	const struct inet_connection_sock *icsk = inet_csk(sk);
1821 1822 1823 1824
	u32 age, send_win, cong_win, limit, in_flight;
	struct tcp_sock *tp = tcp_sk(sk);
	struct skb_mstamp now;
	struct sk_buff *head;
1825
	int win_divisor;
1826

E
Eric Dumazet 已提交
1827
	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
J
John Heffner 已提交
1828
		goto send_now;
1829

1830
	if (icsk->icsk_ca_state >= TCP_CA_Recovery)
J
John Heffner 已提交
1831 1832
		goto send_now;

E
Eric Dumazet 已提交
1833 1834 1835 1836
	/* Avoid bursty behavior by allowing defer
	 * only if the last write was recent.
	 */
	if ((s32)(tcp_time_stamp - tp->lsndtime) > 0)
J
John Heffner 已提交
1837
		goto send_now;
1838

1839 1840
	in_flight = tcp_packets_in_flight(tp);

1841
	BUG_ON(tcp_skb_pcount(skb) <= 1 || (tp->snd_cwnd <= in_flight));
1842

1843
	send_win = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
1844 1845 1846 1847 1848 1849

	/* 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);

1850
	/* If a full-sized TSO skb can be sent, do it. */
E
Eric Dumazet 已提交
1851
	if (limit >= max_segs * tp->mss_cache)
J
John Heffner 已提交
1852
		goto send_now;
1853

1854 1855 1856 1857
	/* 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;

1858 1859
	win_divisor = ACCESS_ONCE(sysctl_tcp_tso_win_divisor);
	if (win_divisor) {
1860 1861 1862 1863 1864
		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.
		 */
1865
		chunk /= win_divisor;
1866
		if (limit >= chunk)
J
John Heffner 已提交
1867
			goto send_now;
1868 1869 1870 1871 1872 1873
	} 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.
		 */
1874
		if (limit > tcp_max_tso_deferred_mss(tp) * tp->mss_cache)
J
John Heffner 已提交
1875
			goto send_now;
1876 1877
	}

1878 1879 1880 1881 1882 1883 1884
	head = tcp_write_queue_head(sk);
	skb_mstamp_get(&now);
	age = skb_mstamp_us_delta(&now, &head->skb_mstamp);
	/* If next ACK is likely to come too late (half srtt), do not defer */
	if (age < (tp->srtt_us >> 4))
		goto send_now;

E
Eric Dumazet 已提交
1885
	/* Ok, it looks like it is advisable to defer. */
J
John Heffner 已提交
1886

1887
	if (cong_win < send_win && cong_win <= skb->len)
1888 1889
		*is_cwnd_limited = true;

E
Eric Dumazet 已提交
1890
	return true;
J
John Heffner 已提交
1891 1892

send_now:
E
Eric Dumazet 已提交
1893
	return false;
1894 1895
}

1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920
static inline void tcp_mtu_check_reprobe(struct sock *sk)
{
	struct inet_connection_sock *icsk = inet_csk(sk);
	struct tcp_sock *tp = tcp_sk(sk);
	struct net *net = sock_net(sk);
	u32 interval;
	s32 delta;

	interval = net->ipv4.sysctl_tcp_probe_interval;
	delta = tcp_time_stamp - icsk->icsk_mtup.probe_timestamp;
	if (unlikely(delta >= interval * HZ)) {
		int mss = tcp_current_mss(sk);

		/* Update current search range */
		icsk->icsk_mtup.probe_size = 0;
		icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp +
			sizeof(struct tcphdr) +
			icsk->icsk_af_ops->net_header_len;
		icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss);

		/* Update probe time stamp */
		icsk->icsk_mtup.probe_timestamp = tcp_time_stamp;
	}
}

J
John Heffner 已提交
1921
/* Create a new MTU probe if we are ready.
1922 1923 1924 1925
 * 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
1926 1927
 * Returns 0 if we should wait to probe (no cwnd available),
 *         1 if a probe was sent,
1928 1929
 *         -1 otherwise
 */
J
John Heffner 已提交
1930 1931 1932 1933 1934
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;
1935
	struct net *net = sock_net(sk);
J
John Heffner 已提交
1936 1937
	int len;
	int probe_size;
1938
	int size_needed;
J
John Heffner 已提交
1939 1940
	int copy;
	int mss_now;
1941
	int interval;
J
John Heffner 已提交
1942 1943 1944 1945 1946 1947 1948 1949 1950

	/* 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 ||
1951
	    tp->rx_opt.num_sacks || tp->rx_opt.dsack)
J
John Heffner 已提交
1952 1953
		return -1;

1954 1955 1956 1957
	/* Use binary search for probe_size between tcp_mss_base,
	 * and current mss_clamp. if (search_high - search_low)
	 * smaller than a threshold, backoff from probing.
	 */
I
Ilpo Järvinen 已提交
1958
	mss_now = tcp_current_mss(sk);
1959 1960
	probe_size = tcp_mtu_to_mss(sk, (icsk->icsk_mtup.search_high +
				    icsk->icsk_mtup.search_low) >> 1);
1961
	size_needed = probe_size + (tp->reordering + 1) * tp->mss_cache;
1962
	interval = icsk->icsk_mtup.search_high - icsk->icsk_mtup.search_low;
1963 1964 1965 1966
	/* When misfortune happens, we are reprobing actively,
	 * and then reprobe timer has expired. We stick with current
	 * probing process by not resetting search range to its orignal.
	 */
1967
	if (probe_size > tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_high) ||
1968 1969 1970 1971 1972
		interval < net->ipv4.sysctl_tcp_probe_threshold) {
		/* Check whether enough time has elaplased for
		 * another round of probing.
		 */
		tcp_mtu_check_reprobe(sk);
J
John Heffner 已提交
1973 1974 1975 1976
		return -1;
	}

	/* Have enough data in the send queue to probe? */
1977
	if (tp->write_seq - tp->snd_nxt < size_needed)
J
John Heffner 已提交
1978 1979
		return -1;

1980 1981
	if (tp->snd_wnd < size_needed)
		return -1;
1982
	if (after(tp->snd_nxt + size_needed, tcp_wnd_end(tp)))
1983
		return 0;
J
John Heffner 已提交
1984

1985 1986 1987
	/* 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
1988 1989 1990 1991 1992 1993
			return -1;
		else
			return 0;
	}

	/* We're allowed to probe.  Build it now. */
1994
	nskb = sk_stream_alloc_skb(sk, probe_size, GFP_ATOMIC, false);
1995
	if (!nskb)
J
John Heffner 已提交
1996
		return -1;
1997 1998
	sk->sk_wmem_queued += nskb->truesize;
	sk_mem_charge(sk, nskb->truesize);
J
John Heffner 已提交
1999

2000
	skb = tcp_send_head(sk);
J
John Heffner 已提交
2001 2002 2003

	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
2004
	TCP_SKB_CB(nskb)->tcp_flags = TCPHDR_ACK;
J
John Heffner 已提交
2005 2006
	TCP_SKB_CB(nskb)->sacked = 0;
	nskb->csum = 0;
2007
	nskb->ip_summed = skb->ip_summed;
J
John Heffner 已提交
2008

2009 2010
	tcp_insert_write_queue_before(nskb, skb, sk);

J
John Heffner 已提交
2011
	len = 0;
2012
	tcp_for_write_queue_from_safe(skb, next, sk) {
J
John Heffner 已提交
2013
		copy = min_t(int, skb->len, probe_size - len);
2014
		if (nskb->ip_summed) {
J
John Heffner 已提交
2015
			skb_copy_bits(skb, 0, skb_put(nskb, copy), copy);
2016 2017 2018 2019 2020 2021
		} else {
			__wsum csum = skb_copy_and_csum_bits(skb, 0,
							     skb_put(nskb, copy),
							     copy, 0);
			nskb->csum = csum_block_add(nskb->csum, csum, len);
		}
J
John Heffner 已提交
2022 2023 2024 2025

		if (skb->len <= copy) {
			/* We've eaten all the data from this skb.
			 * Throw it away. */
E
Eric Dumazet 已提交
2026
			TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;
2027
			tcp_unlink_write_queue(skb, sk);
2028
			sk_wmem_free_skb(sk, skb);
J
John Heffner 已提交
2029
		} else {
E
Eric Dumazet 已提交
2030
			TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags &
C
Changli Gao 已提交
2031
						   ~(TCPHDR_FIN|TCPHDR_PSH);
J
John Heffner 已提交
2032 2033
			if (!skb_shinfo(skb)->nr_frags) {
				skb_pull(skb, copy);
2034
				if (skb->ip_summed != CHECKSUM_PARTIAL)
2035 2036
					skb->csum = csum_partial(skb->data,
								 skb->len, 0);
J
John Heffner 已提交
2037 2038
			} else {
				__pskb_trim_head(skb, copy);
2039
				tcp_set_skb_tso_segs(skb, mss_now);
J
John Heffner 已提交
2040 2041 2042 2043 2044
			}
			TCP_SKB_CB(skb)->seq += copy;
		}

		len += copy;
2045 2046 2047

		if (len >= probe_size)
			break;
J
John Heffner 已提交
2048
	}
2049
	tcp_init_tso_segs(nskb, nskb->len);
J
John Heffner 已提交
2050 2051

	/* We're ready to send.  If this fails, the probe will
E
Eric Dumazet 已提交
2052 2053
	 * be resegmented into mss-sized pieces by tcp_write_xmit().
	 */
J
John Heffner 已提交
2054 2055
	if (!tcp_transmit_skb(sk, nskb, 1, GFP_ATOMIC)) {
		/* Decrement cwnd here because we are sending
2056
		 * effectively two packets. */
J
John Heffner 已提交
2057
		tp->snd_cwnd--;
2058
		tcp_event_new_data_sent(sk, nskb);
J
John Heffner 已提交
2059 2060

		icsk->icsk_mtup.probe_size = tcp_mss_to_mtu(sk, nskb->len);
2061 2062
		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
2063 2064 2065 2066 2067 2068 2069

		return 1;
	}

	return -1;
}

2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102
/* TCP Small Queues :
 * Control number of packets in qdisc/devices to two packets / or ~1 ms.
 * (These limits are doubled for retransmits)
 * This allows for :
 *  - better RTT estimation and ACK scheduling
 *  - faster recovery
 *  - high rates
 * Alas, some drivers / subsystems require a fair amount
 * of queued bytes to ensure line rate.
 * One example is wifi aggregation (802.11 AMPDU)
 */
static bool tcp_small_queue_check(struct sock *sk, const struct sk_buff *skb,
				  unsigned int factor)
{
	unsigned int limit;

	limit = max(2 * skb->truesize, sk->sk_pacing_rate >> 10);
	limit = min_t(u32, limit, sysctl_tcp_limit_output_bytes);
	limit <<= factor;

	if (atomic_read(&sk->sk_wmem_alloc) > limit) {
		set_bit(TSQ_THROTTLED, &tcp_sk(sk)->tsq_flags);
		/* It is possible TX completion already happened
		 * before we set TSQ_THROTTLED, so we must
		 * test again the condition.
		 */
		smp_mb__after_atomic();
		if (atomic_read(&sk->sk_wmem_alloc) > limit)
			return true;
	}
	return false;
}

2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117
static void tcp_chrono_set(struct tcp_sock *tp, const enum tcp_chrono new)
{
	const u32 now = tcp_time_stamp;

	if (tp->chrono_type > TCP_CHRONO_UNSPEC)
		tp->chrono_stat[tp->chrono_type - 1] += now - tp->chrono_start;
	tp->chrono_start = now;
	tp->chrono_type = new;
}

void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type)
{
	struct tcp_sock *tp = tcp_sk(sk);

	/* If there are multiple conditions worthy of tracking in a
2118 2119
	 * chronograph then the highest priority enum takes precedence
	 * over the other conditions. So that if something "more interesting"
2120 2121 2122 2123 2124 2125 2126 2127 2128 2129
	 * starts happening, stop the previous chrono and start a new one.
	 */
	if (type > tp->chrono_type)
		tcp_chrono_set(tp, type);
}

void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type)
{
	struct tcp_sock *tp = tcp_sk(sk);

2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141

	/* There are multiple conditions worthy of tracking in a
	 * chronograph, so that the highest priority enum takes
	 * precedence over the other conditions (see tcp_chrono_start).
	 * If a condition stops, we only stop chrono tracking if
	 * it's the "most interesting" or current chrono we are
	 * tracking and starts busy chrono if we have pending data.
	 */
	if (tcp_write_queue_empty(sk))
		tcp_chrono_set(tp, TCP_CHRONO_UNSPEC);
	else if (type == tp->chrono_type)
		tcp_chrono_set(tp, TCP_CHRONO_BUSY);
2142 2143
}

L
Linus Torvalds 已提交
2144 2145 2146 2147
/* This routine writes packets to the network.  It advances the
 * send_head.  This happens as incoming acks open up the remote
 * window for us.
 *
2148 2149 2150 2151
 * 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
2152 2153 2154
 * 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
2155 2156
 * 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
2157
 */
E
Eric Dumazet 已提交
2158 2159
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
2160 2161
{
	struct tcp_sock *tp = tcp_sk(sk);
2162
	struct sk_buff *skb;
2163 2164
	unsigned int tso_segs, sent_pkts;
	int cwnd_quota;
J
John Heffner 已提交
2165
	int result;
2166
	bool is_cwnd_limited = false, is_rwnd_limited = false;
E
Eric Dumazet 已提交
2167
	u32 max_segs;
L
Linus Torvalds 已提交
2168

2169
	sent_pkts = 0;
J
John Heffner 已提交
2170

2171 2172 2173 2174
	if (!push_one) {
		/* Do MTU probing. */
		result = tcp_mtu_probe(sk);
		if (!result) {
E
Eric Dumazet 已提交
2175
			return false;
2176 2177 2178
		} else if (result > 0) {
			sent_pkts = 1;
		}
J
John Heffner 已提交
2179 2180
	}

2181
	max_segs = tcp_tso_segs(sk, mss_now);
2182
	while ((skb = tcp_send_head(sk))) {
2183 2184
		unsigned int limit;

2185
		tso_segs = tcp_init_tso_segs(skb, mss_now);
2186
		BUG_ON(!tso_segs);
2187

2188
		if (unlikely(tp->repair) && tp->repair_queue == TCP_SEND_QUEUE) {
E
Eric Dumazet 已提交
2189 2190
			/* "skb_mstamp" is used as a start point for the retransmit timer */
			skb_mstamp_get(&skb->skb_mstamp);
2191
			goto repair; /* Skip network transmission */
2192
		}
2193

2194
		cwnd_quota = tcp_cwnd_test(tp, skb);
N
Nandita Dukkipati 已提交
2195 2196 2197 2198 2199 2200 2201
		if (!cwnd_quota) {
			if (push_one == 2)
				/* Force out a loss probe pkt. */
				cwnd_quota = 1;
			else
				break;
		}
2202

2203 2204
		if (unlikely(!tcp_snd_wnd_test(tp, skb, mss_now))) {
			is_rwnd_limited = true;
2205
			break;
2206
		}
2207

2208
		if (tso_segs == 1) {
2209 2210 2211 2212 2213
			if (unlikely(!tcp_nagle_test(tp, skb, mss_now,
						     (tcp_skb_is_last(sk, skb) ?
						      nonagle : TCP_NAGLE_PUSH))))
				break;
		} else {
2214
			if (!push_one &&
E
Eric Dumazet 已提交
2215 2216
			    tcp_tso_should_defer(sk, skb, &is_cwnd_limited,
						 max_segs))
2217 2218
				break;
		}
2219

E
Eric Dumazet 已提交
2220
		limit = mss_now;
2221
		if (tso_segs > 1 && !tcp_urg_mode(tp))
E
Eric Dumazet 已提交
2222 2223 2224 2225 2226 2227 2228 2229 2230 2231
			limit = tcp_mss_split_point(sk, skb, mss_now,
						    min_t(unsigned int,
							  cwnd_quota,
							  max_segs),
						    nonagle);

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

2232 2233
		if (test_bit(TCP_TSQ_DEFERRED, &tp->tsq_flags))
			clear_bit(TCP_TSQ_DEFERRED, &tp->tsq_flags);
2234 2235
		if (tcp_small_queue_check(sk, skb, 0))
			break;
E
Eric Dumazet 已提交
2236

2237
		if (unlikely(tcp_transmit_skb(sk, skb, 1, gfp)))
2238
			break;
L
Linus Torvalds 已提交
2239

2240
repair:
2241 2242 2243
		/* Advance the send_head.  This one is sent out.
		 * This call will increment packets_out.
		 */
2244
		tcp_event_new_data_sent(sk, skb);
L
Linus Torvalds 已提交
2245

2246
		tcp_minshall_update(tp, mss_now, skb);
2247
		sent_pkts += tcp_skb_pcount(skb);
2248 2249 2250

		if (push_one)
			break;
2251
	}
L
Linus Torvalds 已提交
2252

2253 2254 2255 2256 2257
	if (is_rwnd_limited)
		tcp_chrono_start(sk, TCP_CHRONO_RWND_LIMITED);
	else
		tcp_chrono_stop(sk, TCP_CHRONO_RWND_LIMITED);

2258
	if (likely(sent_pkts)) {
2259 2260
		if (tcp_in_cwnd_reduction(sk))
			tp->prr_out += sent_pkts;
N
Nandita Dukkipati 已提交
2261 2262 2263 2264

		/* Send one loss probe per tail loss episode. */
		if (push_one != 2)
			tcp_schedule_loss_probe(sk);
2265
		is_cwnd_limited |= (tcp_packets_in_flight(tp) >= tp->snd_cwnd);
2266
		tcp_cwnd_validate(sk, is_cwnd_limited);
E
Eric Dumazet 已提交
2267
		return false;
L
Linus Torvalds 已提交
2268
	}
2269
	return !tp->packets_out && tcp_send_head(sk);
N
Nandita Dukkipati 已提交
2270 2271 2272 2273 2274 2275 2276
}

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;
2277
	u32 rtt = usecs_to_jiffies(tp->srtt_us >> 3);
N
Nandita Dukkipati 已提交
2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288

	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.
	 */
2289
	if (tp->fastopen_rsk)
N
Nandita Dukkipati 已提交
2290 2291 2292 2293 2294 2295 2296 2297 2298
		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.
	 */
2299
	if (sysctl_tcp_early_retrans < 3 || !tp->packets_out ||
N
Nandita Dukkipati 已提交
2300 2301 2302 2303 2304 2305 2306 2307
	    !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
2308 2309
	 * for delayed ack when there's one outstanding packet. If no RTT
	 * sample is available then probe after TCP_TIMEOUT_INIT.
N
Nandita Dukkipati 已提交
2310
	 */
2311
	timeout = rtt << 1 ? : TCP_TIMEOUT_INIT;
N
Nandita Dukkipati 已提交
2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330
	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;
}

2331 2332 2333 2334 2335 2336 2337
/* Thanks to skb fast clones, we can detect if a prior transmit of
 * a packet is still in a qdisc or driver queue.
 * In this case, there is very little point doing a retransmit !
 */
static bool skb_still_in_host_queue(const struct sock *sk,
				    const struct sk_buff *skb)
{
2338
	if (unlikely(skb_fclone_busy(sk, skb))) {
2339 2340
		NET_INC_STATS(sock_net(sk),
			      LINUX_MIB_TCPSPURIOUS_RTX_HOSTQUEUES);
2341 2342 2343 2344 2345
		return true;
	}
	return false;
}

2346
/* When probe timeout (PTO) fires, try send a new segment if possible, else
N
Nandita Dukkipati 已提交
2347 2348 2349 2350
 * retransmit the last segment.
 */
void tcp_send_loss_probe(struct sock *sk)
{
N
Nandita Dukkipati 已提交
2351
	struct tcp_sock *tp = tcp_sk(sk);
N
Nandita Dukkipati 已提交
2352 2353 2354 2355
	struct sk_buff *skb;
	int pcount;
	int mss = tcp_current_mss(sk);

2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367
	skb = tcp_send_head(sk);
	if (skb) {
		if (tcp_snd_wnd_test(tp, skb, mss)) {
			pcount = tp->packets_out;
			tcp_write_xmit(sk, mss, TCP_NAGLE_OFF, 2, GFP_ATOMIC);
			if (tp->packets_out > pcount)
				goto probe_sent;
			goto rearm_timer;
		}
		skb = tcp_write_queue_prev(sk, skb);
	} else {
		skb = tcp_write_queue_tail(sk);
N
Nandita Dukkipati 已提交
2368 2369
	}

N
Nandita Dukkipati 已提交
2370 2371 2372 2373
	/* At most one outstanding TLP retransmission. */
	if (tp->tlp_high_seq)
		goto rearm_timer;

N
Nandita Dukkipati 已提交
2374 2375 2376 2377
	/* Retransmit last segment. */
	if (WARN_ON(!skb))
		goto rearm_timer;

2378 2379 2380
	if (skb_still_in_host_queue(sk, skb))
		goto rearm_timer;

N
Nandita Dukkipati 已提交
2381 2382 2383 2384 2385
	pcount = tcp_skb_pcount(skb);
	if (WARN_ON(!pcount))
		goto rearm_timer;

	if ((pcount > 1) && (skb->len > (pcount - 1) * mss)) {
2386 2387
		if (unlikely(tcp_fragment(sk, skb, (pcount - 1) * mss, mss,
					  GFP_ATOMIC)))
N
Nandita Dukkipati 已提交
2388
			goto rearm_timer;
2389
		skb = tcp_write_queue_next(sk, skb);
N
Nandita Dukkipati 已提交
2390 2391 2392 2393 2394
	}

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

2395
	if (__tcp_retransmit_skb(sk, skb, 1))
2396
		goto rearm_timer;
N
Nandita Dukkipati 已提交
2397

N
Nandita Dukkipati 已提交
2398
	/* Record snd_nxt for loss detection. */
2399
	tp->tlp_high_seq = tp->snd_nxt;
N
Nandita Dukkipati 已提交
2400

2401
probe_sent:
2402
	NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPLOSSPROBES);
2403 2404
	/* Reset s.t. tcp_rearm_rto will restart timer from now */
	inet_csk(sk)->icsk_pending = 0;
N
Nandita Dukkipati 已提交
2405
rearm_timer:
2406
	tcp_rearm_rto(sk);
L
Linus Torvalds 已提交
2407 2408
}

2409 2410 2411 2412
/* 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.
 */
2413 2414
void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
			       int nonagle)
2415
{
2416 2417 2418 2419 2420 2421 2422
	/* 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;

2423
	if (tcp_write_xmit(sk, cur_mss, nonagle, 0,
2424
			   sk_gfp_mask(sk, GFP_ATOMIC)))
2425
		tcp_check_probe_timer(sk);
2426 2427
}

2428 2429 2430 2431 2432
/* 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)
{
2433
	struct sk_buff *skb = tcp_send_head(sk);
2434 2435 2436

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

2437
	tcp_write_xmit(sk, mss_now, TCP_NAGLE_PUSH, 1, sk->sk_allocation);
2438 2439
}

L
Linus Torvalds 已提交
2440 2441
/* This function returns the amount that we can raise the
 * usable window based on the following constraints
2442
 *
L
Linus Torvalds 已提交
2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460
 * 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.
2461
 *
L
Linus Torvalds 已提交
2462 2463
 * Of course, if the sender implements sender side SWS prevention
 * then this will not be a problem.
2464
 *
L
Linus Torvalds 已提交
2465
 * BSD seems to make the following compromise:
2466
 *
L
Linus Torvalds 已提交
2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493
 *	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)
{
2494
	struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
2495
	struct tcp_sock *tp = tcp_sk(sk);
S
Stephen Hemminger 已提交
2496
	/* MSS for the peer's data.  Previous versions used mss_clamp
L
Linus Torvalds 已提交
2497 2498 2499 2500 2501
	 * 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
	 */
2502
	int mss = icsk->icsk_ack.rcv_mss;
L
Linus Torvalds 已提交
2503
	int free_space = tcp_space(sk);
2504 2505
	int allowed_space = tcp_full_space(sk);
	int full_space = min_t(int, tp->window_clamp, allowed_space);
L
Linus Torvalds 已提交
2506 2507 2508
	int window;

	if (mss > full_space)
2509
		mss = full_space;
L
Linus Torvalds 已提交
2510

2511
	if (free_space < (full_space >> 1)) {
2512
		icsk->icsk_ack.quick = 0;
L
Linus Torvalds 已提交
2513

2514
		if (tcp_under_memory_pressure(sk))
2515 2516
			tp->rcv_ssthresh = min(tp->rcv_ssthresh,
					       4U * tp->advmss);
L
Linus Torvalds 已提交
2517

2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530
		/* free_space might become our new window, make sure we don't
		 * increase it due to wscale.
		 */
		free_space = round_down(free_space, 1 << tp->rx_opt.rcv_wscale);

		/* if free space is less than mss estimate, or is below 1/16th
		 * of the maximum allowed, try to move to zero-window, else
		 * tcp_clamp_window() will grow rcv buf up to tcp_rmem[2], and
		 * new incoming data is dropped due to memory limits.
		 * With large window, mss test triggers way too late in order
		 * to announce zero window in time before rmem limit kicks in.
		 */
		if (free_space < (allowed_space >> 4) || free_space < mss)
L
Linus Torvalds 已提交
2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560
			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)
2561
			window = (free_space / mss) * mss;
2562
		else if (mss == full_space &&
2563
			 free_space > window + (full_space >> 1))
2564
			window = free_space;
L
Linus Torvalds 已提交
2565 2566 2567 2568 2569
	}

	return window;
}

2570 2571
void tcp_skb_collapse_tstamp(struct sk_buff *skb,
			     const struct sk_buff *next_skb)
2572
{
2573 2574 2575
	if (unlikely(tcp_has_tx_tstamp(next_skb))) {
		const struct skb_shared_info *next_shinfo =
			skb_shinfo(next_skb);
2576 2577
		struct skb_shared_info *shinfo = skb_shinfo(skb);

2578
		shinfo->tx_flags |= next_shinfo->tx_flags & SKBTX_ANY_TSTAMP;
2579
		shinfo->tskey = next_shinfo->tskey;
2580 2581
		TCP_SKB_CB(skb)->txstamp_ack |=
			TCP_SKB_CB(next_skb)->txstamp_ack;
2582 2583 2584
	}
}

2585
/* Collapses two adjacent SKB's during retransmission. */
2586
static bool tcp_collapse_retrans(struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
2587 2588
{
	struct tcp_sock *tp = tcp_sk(sk);
2589
	struct sk_buff *next_skb = tcp_write_queue_next(sk, skb);
2590
	int skb_size, next_skb_size;
L
Linus Torvalds 已提交
2591

2592 2593
	skb_size = skb->len;
	next_skb_size = next_skb->len;
L
Linus Torvalds 已提交
2594

2595
	BUG_ON(tcp_skb_pcount(skb) != 1 || tcp_skb_pcount(next_skb) != 1);
2596

2597 2598 2599 2600 2601 2602 2603
	if (next_skb_size) {
		if (next_skb_size <= skb_availroom(skb))
			skb_copy_bits(next_skb, 0, skb_put(skb, next_skb_size),
				      next_skb_size);
		else if (!skb_shift(skb, next_skb, next_skb_size))
			return false;
	}
2604
	tcp_highest_sack_combine(sk, next_skb, skb);
L
Linus Torvalds 已提交
2605

2606
	tcp_unlink_write_queue(next_skb, sk);
L
Linus Torvalds 已提交
2607

2608 2609
	if (next_skb->ip_summed == CHECKSUM_PARTIAL)
		skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
2610

2611 2612
	if (skb->ip_summed != CHECKSUM_PARTIAL)
		skb->csum = csum_block_add(skb->csum, next_skb->csum, skb_size);
L
Linus Torvalds 已提交
2613

2614 2615
	/* 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
2616

2617
	/* Merge over control information. This moves PSH/FIN etc. over */
E
Eric Dumazet 已提交
2618
	TCP_SKB_CB(skb)->tcp_flags |= TCP_SKB_CB(next_skb)->tcp_flags;
2619 2620 2621 2622 2623

	/* 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;
2624
	TCP_SKB_CB(skb)->eor = TCP_SKB_CB(next_skb)->eor;
2625 2626

	/* changed transmit queue under us so clear hints */
2627 2628 2629
	tcp_clear_retrans_hints_partial(tp);
	if (next_skb == tp->retransmit_skb_hint)
		tp->retransmit_skb_hint = skb;
2630

2631 2632
	tcp_adjust_pcount(sk, next_skb, tcp_skb_pcount(next_skb));

2633 2634
	tcp_skb_collapse_tstamp(skb, next_skb);

2635
	sk_wmem_free_skb(sk, next_skb);
2636
	return true;
L
Linus Torvalds 已提交
2637 2638
}

2639
/* Check if coalescing SKBs is legal. */
E
Eric Dumazet 已提交
2640
static bool tcp_can_collapse(const struct sock *sk, const struct sk_buff *skb)
2641 2642
{
	if (tcp_skb_pcount(skb) > 1)
E
Eric Dumazet 已提交
2643
		return false;
2644
	if (skb_cloned(skb))
E
Eric Dumazet 已提交
2645
		return false;
2646
	if (skb == tcp_send_head(sk))
E
Eric Dumazet 已提交
2647
		return false;
E
Eric Dumazet 已提交
2648
	/* Some heuristics for collapsing over SACK'd could be invented */
2649
	if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)
E
Eric Dumazet 已提交
2650
		return false;
2651

E
Eric Dumazet 已提交
2652
	return true;
2653 2654
}

2655 2656 2657
/* Collapse packets in the retransmit queue to make to create
 * less packets on the wire. This is only done on retransmission.
 */
2658 2659 2660 2661 2662
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
2663
	bool first = true;
2664 2665 2666

	if (!sysctl_tcp_retrans_collapse)
		return;
E
Eric Dumazet 已提交
2667
	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)
2668 2669 2670 2671 2672 2673
		return;

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

2674 2675 2676
		if (!tcp_skb_can_collapse_to(to))
			break;

2677 2678 2679
		space -= skb->len;

		if (first) {
E
Eric Dumazet 已提交
2680
			first = false;
2681 2682 2683 2684 2685 2686 2687 2688 2689
			continue;
		}

		if (space < 0)
			break;

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

2690 2691
		if (!tcp_collapse_retrans(sk, to))
			break;
2692 2693 2694
	}
}

L
Linus Torvalds 已提交
2695 2696 2697 2698
/* 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.
 */
2699
int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs)
L
Linus Torvalds 已提交
2700
{
J
John Heffner 已提交
2701
	struct inet_connection_sock *icsk = inet_csk(sk);
2702
	struct tcp_sock *tp = tcp_sk(sk);
2703
	unsigned int cur_mss;
2704 2705
	int diff, len, err;

L
Linus Torvalds 已提交
2706

2707 2708
	/* Inconclusive MTU probe */
	if (icsk->icsk_mtup.probe_size)
J
John Heffner 已提交
2709 2710
		icsk->icsk_mtup.probe_size = 0;

L
Linus Torvalds 已提交
2711
	/* Do not sent more than we queued. 1/4 is reserved for possible
S
Stephen Hemminger 已提交
2712
	 * copying overhead: fragmentation, tunneling, mangling etc.
L
Linus Torvalds 已提交
2713 2714
	 */
	if (atomic_read(&sk->sk_wmem_alloc) >
2715 2716
	    min_t(u32, sk->sk_wmem_queued + (sk->sk_wmem_queued >> 2),
		  sk->sk_sndbuf))
L
Linus Torvalds 已提交
2717 2718
		return -EAGAIN;

2719 2720 2721
	if (skb_still_in_host_queue(sk, skb))
		return -EBUSY;

L
Linus Torvalds 已提交
2722 2723 2724 2725 2726 2727 2728
	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;
	}

2729 2730 2731
	if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk))
		return -EHOSTUNREACH; /* Routing failure or similar. */

I
Ilpo Järvinen 已提交
2732
	cur_mss = tcp_current_mss(sk);
2733

L
Linus Torvalds 已提交
2734 2735 2736 2737 2738
	/* 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.
	 */
2739 2740
	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
2741 2742
		return -EAGAIN;

2743 2744 2745
	len = cur_mss * segs;
	if (skb->len > len) {
		if (tcp_fragment(sk, skb, len, cur_mss, GFP_ATOMIC))
L
Linus Torvalds 已提交
2746
			return -ENOMEM; /* We'll try again later. */
2747
	} else {
2748 2749
		if (skb_unclone(skb, GFP_ATOMIC))
			return -ENOMEM;
2750

2751 2752 2753 2754 2755 2756 2757
		diff = tcp_skb_pcount(skb);
		tcp_set_skb_tso_segs(skb, cur_mss);
		diff -= tcp_skb_pcount(skb);
		if (diff)
			tcp_adjust_pcount(sk, skb, diff);
		if (skb->len < cur_mss)
			tcp_retrans_try_collapse(sk, skb, cur_mss);
L
Linus Torvalds 已提交
2758 2759
	}

2760 2761 2762 2763
	/* RFC3168, section 6.1.1.1. ECN fallback */
	if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN_ECN) == TCPHDR_SYN_ECN)
		tcp_ecn_clear_syn(sk, skb);

2764 2765 2766 2767 2768 2769
	/* 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)) {
2770 2771 2772 2773
		struct sk_buff *nskb;

		skb_mstamp_get(&skb->skb_mstamp);
		nskb = __pskb_copy(skb, MAX_TCP_HEADER, GFP_ATOMIC);
2774 2775
		err = nskb ? tcp_transmit_skb(sk, nskb, 0, GFP_ATOMIC) :
			     -ENOBUFS;
E
Eric Dumazet 已提交
2776
	} else {
2777
		err = tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
E
Eric Dumazet 已提交
2778
	}
2779

2780
	if (likely(!err)) {
2781 2782
		segs = tcp_skb_pcount(skb);

2783
		TCP_SKB_CB(skb)->sacked |= TCPCB_EVER_RETRANS;
2784
		/* Update global TCP statistics. */
2785
		TCP_ADD_STATS(sock_net(sk), TCP_MIB_RETRANSSEGS, segs);
2786
		if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)
2787
			__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPSYNRETRANS);
2788
		tp->total_retrans += segs;
2789
	}
2790
	return err;
2791 2792
}

2793
int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs)
2794 2795
{
	struct tcp_sock *tp = tcp_sk(sk);
2796
	int err = __tcp_retransmit_skb(sk, skb, segs);
L
Linus Torvalds 已提交
2797 2798 2799

	if (err == 0) {
#if FASTRETRANS_DEBUG > 0
2800
		if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) {
2801
			net_dbg_ratelimited("retrans_out leaked\n");
L
Linus Torvalds 已提交
2802 2803 2804 2805 2806 2807 2808
		}
#endif
		TCP_SKB_CB(skb)->sacked |= TCPCB_RETRANS;
		tp->retrans_out += tcp_skb_pcount(skb);

		/* Save stamp of the first retransmit. */
		if (!tp->retrans_stamp)
E
Eric Dumazet 已提交
2809
			tp->retrans_stamp = tcp_skb_timestamp(skb);
L
Linus Torvalds 已提交
2810

2811
	} else if (err != -EBUSY) {
2812
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPRETRANSFAIL);
L
Linus Torvalds 已提交
2813
	}
Y
Yuchung Cheng 已提交
2814 2815 2816 2817

	if (tp->undo_retrans < 0)
		tp->undo_retrans = 0;
	tp->undo_retrans += tcp_skb_pcount(skb);
L
Linus Torvalds 已提交
2818 2819 2820
	return err;
}

2821 2822 2823
/* Check if we forward retransmits are possible in the current
 * window/congestion state.
 */
E
Eric Dumazet 已提交
2824
static bool tcp_can_forward_retransmit(struct sock *sk)
2825 2826
{
	const struct inet_connection_sock *icsk = inet_csk(sk);
2827
	const struct tcp_sock *tp = tcp_sk(sk);
2828 2829 2830

	/* Forward retransmissions are possible only during Recovery. */
	if (icsk->icsk_ca_state != TCP_CA_Recovery)
E
Eric Dumazet 已提交
2831
		return false;
2832 2833 2834

	/* No forward retransmissions in Reno are possible. */
	if (tcp_is_reno(tp))
E
Eric Dumazet 已提交
2835
		return false;
2836 2837 2838 2839 2840 2841 2842 2843 2844 2845

	/* 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
2846
		return false;
2847

E
Eric Dumazet 已提交
2848
	return true;
2849 2850
}

L
Linus Torvalds 已提交
2851 2852 2853 2854 2855 2856 2857 2858 2859 2860
/* 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)
{
2861
	const struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
2862 2863
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb;
2864
	struct sk_buff *hole = NULL;
2865
	u32 max_segs, last_lost;
2866
	int mib_idx;
2867
	int fwd_rexmitting = 0;
2868

2869 2870 2871
	if (!tp->packets_out)
		return;

2872 2873 2874
	if (!tp->lost_out)
		tp->retransmit_high = tp->snd_una;

2875
	if (tp->retransmit_skb_hint) {
2876
		skb = tp->retransmit_skb_hint;
2877 2878 2879 2880
		last_lost = TCP_SKB_CB(skb)->end_seq;
		if (after(last_lost, tp->retransmit_high))
			last_lost = tp->retransmit_high;
	} else {
2881
		skb = tcp_write_queue_head(sk);
2882 2883
		last_lost = tp->snd_una;
	}
L
Linus Torvalds 已提交
2884

2885
	max_segs = tcp_tso_segs(sk, tcp_current_mss(sk));
2886
	tcp_for_write_queue_from(skb, sk) {
E
Eric Dumazet 已提交
2887
		__u8 sacked;
2888
		int segs;
L
Linus Torvalds 已提交
2889

2890 2891 2892
		if (skb == tcp_send_head(sk))
			break;
		/* we could do better than to assign each time */
2893
		if (!hole)
2894
			tp->retransmit_skb_hint = skb;
2895

2896 2897
		segs = tp->snd_cwnd - tcp_packets_in_flight(tp);
		if (segs <= 0)
2898
			return;
E
Eric Dumazet 已提交
2899
		sacked = TCP_SKB_CB(skb)->sacked;
2900 2901 2902 2903
		/* In case tcp_shift_skb_data() have aggregated large skbs,
		 * we need to make sure not sending too bigs TSO packets
		 */
		segs = min_t(int, segs, max_segs);
L
Linus Torvalds 已提交
2904

2905 2906 2907 2908 2909
		if (fwd_rexmitting) {
begin_fwd:
			if (!before(TCP_SKB_CB(skb)->seq, tcp_highest_sack_seq(tp)))
				break;
			mib_idx = LINUX_MIB_TCPFORWARDRETRANS;
2910

2911
		} else if (!before(TCP_SKB_CB(skb)->seq, tp->retransmit_high)) {
2912
			tp->retransmit_high = last_lost;
2913 2914 2915
			if (!tcp_can_forward_retransmit(sk))
				break;
			/* Backtrack if necessary to non-L'ed skb */
2916
			if (hole) {
2917 2918 2919 2920 2921
				skb = hole;
				hole = NULL;
			}
			fwd_rexmitting = 1;
			goto begin_fwd;
L
Linus Torvalds 已提交
2922

2923
		} else if (!(sacked & TCPCB_LOST)) {
2924
			if (!hole && !(sacked & (TCPCB_SACKED_RETRANS|TCPCB_SACKED_ACKED)))
2925 2926
				hole = skb;
			continue;
L
Linus Torvalds 已提交
2927

2928
		} else {
2929
			last_lost = TCP_SKB_CB(skb)->end_seq;
2930 2931 2932 2933 2934
			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
2935

2936
		if (sacked & (TCPCB_SACKED_ACKED|TCPCB_SACKED_RETRANS))
L
Linus Torvalds 已提交
2937 2938
			continue;

2939 2940 2941
		if (tcp_small_queue_check(sk, skb, 1))
			return;

2942
		if (tcp_retransmit_skb(sk, skb, segs))
2943
			return;
2944

2945
		NET_ADD_STATS(sock_net(sk), mib_idx, tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
2946

2947
		if (tcp_in_cwnd_reduction(sk))
2948 2949
			tp->prr_out += tcp_skb_pcount(skb);

2950
		if (skb == tcp_write_queue_head(sk))
2951 2952 2953
			inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
						  inet_csk(sk)->icsk_rto,
						  TCP_RTO_MAX);
L
Linus Torvalds 已提交
2954 2955 2956
	}
}

2957 2958
/* We allow to exceed memory limits for FIN packets to expedite
 * connection tear down and (memory) recovery.
2959 2960
 * Otherwise tcp_send_fin() could be tempted to either delay FIN
 * or even be forced to close flow without any FIN.
2961 2962
 * In general, we want to allow one skb per socket to avoid hangs
 * with edge trigger epoll()
2963
 */
2964
void sk_forced_mem_schedule(struct sock *sk, int size)
2965
{
2966
	int amt;
2967 2968 2969 2970 2971

	if (size <= sk->sk_forward_alloc)
		return;
	amt = sk_mem_pages(size);
	sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2972 2973
	sk_memory_allocated_add(sk, amt);

2974 2975
	if (mem_cgroup_sockets_enabled && sk->sk_memcg)
		mem_cgroup_charge_skmem(sk->sk_memcg, amt);
2976 2977
}

2978 2979
/* Send a FIN. The caller locks the socket for us.
 * We should try to send a FIN packet really hard, but eventually give up.
L
Linus Torvalds 已提交
2980 2981 2982
 */
void tcp_send_fin(struct sock *sk)
{
2983
	struct sk_buff *skb, *tskb = tcp_write_queue_tail(sk);
2984 2985
	struct tcp_sock *tp = tcp_sk(sk);

2986 2987 2988 2989
	/* Optimization, tack on the FIN if we have one skb in write queue and
	 * this skb was not yet sent, or we are under memory pressure.
	 * Note: in the latter case, FIN packet will be sent after a timeout,
	 * as TCP stack thinks it has already been transmitted.
L
Linus Torvalds 已提交
2990
	 */
2991
	if (tskb && (tcp_send_head(sk) || tcp_under_memory_pressure(sk))) {
2992 2993 2994
coalesce:
		TCP_SKB_CB(tskb)->tcp_flags |= TCPHDR_FIN;
		TCP_SKB_CB(tskb)->end_seq++;
L
Linus Torvalds 已提交
2995
		tp->write_seq++;
2996 2997 2998 2999 3000 3001 3002 3003 3004 3005
		if (!tcp_send_head(sk)) {
			/* This means tskb was already sent.
			 * Pretend we included the FIN on previous transmit.
			 * We need to set tp->snd_nxt to the value it would have
			 * if FIN had been sent. This is because retransmit path
			 * does not change tp->snd_nxt.
			 */
			tp->snd_nxt++;
			return;
		}
L
Linus Torvalds 已提交
3006
	} else {
3007 3008 3009 3010 3011
		skb = alloc_skb_fclone(MAX_TCP_HEADER, sk->sk_allocation);
		if (unlikely(!skb)) {
			if (tskb)
				goto coalesce;
			return;
L
Linus Torvalds 已提交
3012
		}
3013
		skb_reserve(skb, MAX_TCP_HEADER);
3014
		sk_forced_mem_schedule(sk, skb->truesize);
L
Linus Torvalds 已提交
3015
		/* FIN eats a sequence byte, write_seq advanced by tcp_queue_skb(). */
3016
		tcp_init_nondata_skb(skb, tp->write_seq,
C
Changli Gao 已提交
3017
				     TCPHDR_ACK | TCPHDR_FIN);
L
Linus Torvalds 已提交
3018 3019
		tcp_queue_skb(sk, skb);
	}
3020
	__tcp_push_pending_frames(sk, tcp_current_mss(sk), TCP_NAGLE_OFF);
L
Linus Torvalds 已提交
3021 3022 3023 3024 3025
}

/* 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
3026
 * by RFC 2525, section 2.17.  -DaveM
L
Linus Torvalds 已提交
3027
 */
A
Al Viro 已提交
3028
void tcp_send_active_reset(struct sock *sk, gfp_t priority)
L
Linus Torvalds 已提交
3029 3030 3031 3032 3033 3034
{
	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
3035
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTFAILED);
L
Linus Torvalds 已提交
3036 3037 3038 3039 3040
		return;
	}

	/* Reserve space for headers and prepare control bits. */
	skb_reserve(skb, MAX_TCP_HEADER);
3041
	tcp_init_nondata_skb(skb, tcp_acceptable_seq(sk),
C
Changli Gao 已提交
3042
			     TCPHDR_ACK | TCPHDR_RST);
3043
	skb_mstamp_get(&skb->skb_mstamp);
L
Linus Torvalds 已提交
3044
	/* Send it off. */
3045
	if (tcp_transmit_skb(sk, skb, 0, priority))
P
Pavel Emelyanov 已提交
3046
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTFAILED);
3047

P
Pavel Emelyanov 已提交
3048
	TCP_INC_STATS(sock_net(sk), TCP_MIB_OUTRSTS);
L
Linus Torvalds 已提交
3049 3050
}

3051 3052
/* 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
3053 3054 3055 3056 3057 3058
 * 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)
{
3059
	struct sk_buff *skb;
L
Linus Torvalds 已提交
3060

3061
	skb = tcp_write_queue_head(sk);
3062
	if (!skb || !(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
3063
		pr_debug("%s: wrong queue state\n", __func__);
L
Linus Torvalds 已提交
3064 3065
		return -EFAULT;
	}
E
Eric Dumazet 已提交
3066
	if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)) {
L
Linus Torvalds 已提交
3067 3068
		if (skb_cloned(skb)) {
			struct sk_buff *nskb = skb_copy(skb, GFP_ATOMIC);
3069
			if (!nskb)
L
Linus Torvalds 已提交
3070
				return -ENOMEM;
3071
			tcp_unlink_write_queue(skb, sk);
3072
			__skb_header_release(nskb);
3073
			__tcp_add_write_queue_head(sk, nskb);
3074 3075 3076
			sk_wmem_free_skb(sk, skb);
			sk->sk_wmem_queued += nskb->truesize;
			sk_mem_charge(sk, nskb->truesize);
L
Linus Torvalds 已提交
3077 3078 3079
			skb = nskb;
		}

E
Eric Dumazet 已提交
3080
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ACK;
3081
		tcp_ecn_send_synack(sk, skb);
L
Linus Torvalds 已提交
3082
	}
3083
	return tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
L
Linus Torvalds 已提交
3084 3085
}

3086 3087 3088 3089 3090 3091 3092 3093 3094
/**
 * 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.
 */
3095
struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
3096
				struct request_sock *req,
3097
				struct tcp_fastopen_cookie *foc,
3098
				enum tcp_synack_type synack_type)
L
Linus Torvalds 已提交
3099
{
3100
	struct inet_request_sock *ireq = inet_rsk(req);
3101
	const struct tcp_sock *tp = tcp_sk(sk);
E
Eric Dumazet 已提交
3102
	struct tcp_md5sig_key *md5 = NULL;
3103 3104
	struct tcp_out_options opts;
	struct sk_buff *skb;
3105
	int tcp_header_size;
3106 3107
	struct tcphdr *th;
	u16 user_mss;
3108
	int mss;
L
Linus Torvalds 已提交
3109

3110
	skb = alloc_skb(MAX_TCP_HEADER, GFP_ATOMIC);
3111 3112
	if (unlikely(!skb)) {
		dst_release(dst);
L
Linus Torvalds 已提交
3113
		return NULL;
3114
	}
L
Linus Torvalds 已提交
3115 3116 3117
	/* Reserve space for headers. */
	skb_reserve(skb, MAX_TCP_HEADER);

3118 3119
	switch (synack_type) {
	case TCP_SYNACK_NORMAL:
3120
		skb_set_owner_w(skb, req_to_sk(req));
3121 3122 3123 3124 3125 3126 3127
		break;
	case TCP_SYNACK_COOKIE:
		/* Under synflood, we do not attach skb to a socket,
		 * to avoid false sharing.
		 */
		break;
	case TCP_SYNACK_FASTOPEN:
3128 3129 3130 3131 3132
		/* sk is a const pointer, because we want to express multiple
		 * cpu might call us concurrently.
		 * sk->sk_wmem_alloc in an atomic, we can promote to rw.
		 */
		skb_set_owner_w(skb, (struct sock *)sk);
3133
		break;
3134
	}
3135
	skb_dst_set(skb, dst);
L
Linus Torvalds 已提交
3136

3137
	mss = dst_metric_advmss(dst);
3138 3139 3140
	user_mss = READ_ONCE(tp->rx_opt.user_mss);
	if (user_mss && user_mss < mss)
		mss = user_mss;
3141

A
Adam Langley 已提交
3142
	memset(&opts, 0, sizeof(opts));
3143 3144
#ifdef CONFIG_SYN_COOKIES
	if (unlikely(req->cookie_ts))
E
Eric Dumazet 已提交
3145
		skb->skb_mstamp.stamp_jiffies = cookie_init_timestamp(req);
3146 3147
	else
#endif
E
Eric Dumazet 已提交
3148
	skb_mstamp_get(&skb->skb_mstamp);
E
Eric Dumazet 已提交
3149 3150 3151

#ifdef CONFIG_TCP_MD5SIG
	rcu_read_lock();
3152
	md5 = tcp_rsk(req)->af_specific->req_md5_lookup(sk, req_to_sk(req));
E
Eric Dumazet 已提交
3153
#endif
3154
	skb_set_hash(skb, tcp_rsk(req)->txhash, PKT_HASH_TYPE_L4);
3155 3156
	tcp_header_size = tcp_synack_options(req, mss, skb, &opts, md5, foc) +
			  sizeof(*th);
3157

3158 3159
	skb_push(skb, tcp_header_size);
	skb_reset_transport_header(skb);
L
Linus Torvalds 已提交
3160

3161
	th = (struct tcphdr *)skb->data;
L
Linus Torvalds 已提交
3162 3163 3164
	memset(th, 0, sizeof(struct tcphdr));
	th->syn = 1;
	th->ack = 1;
3165
	tcp_ecn_make_synack(req, th);
E
Eric Dumazet 已提交
3166
	th->source = htons(ireq->ir_num);
3167
	th->dest = ireq->ir_rmt_port;
3168 3169 3170 3171
	/* 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
3172
			     TCPHDR_SYN | TCPHDR_ACK);
3173

L
Linus Torvalds 已提交
3174
	th->seq = htonl(TCP_SKB_CB(skb)->seq);
3175 3176
	/* 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
3177 3178

	/* RFC1323: The window in SYN & SYN/ACK segments is never scaled. */
3179
	th->window = htons(min(req->rsk_rcv_wnd, 65535U));
3180
	tcp_options_write((__be32 *)(th + 1), NULL, &opts);
L
Linus Torvalds 已提交
3181
	th->doff = (tcp_header_size >> 2);
E
Eric Dumazet 已提交
3182
	__TCP_INC_STATS(sock_net(sk), TCP_MIB_OUTSEGS);
3183 3184 3185

#ifdef CONFIG_TCP_MD5SIG
	/* Okay, we have all we need - do the md5 hash if needed */
E
Eric Dumazet 已提交
3186
	if (md5)
3187
		tcp_rsk(req)->af_specific->calc_md5_hash(opts.hash_location,
3188
					       md5, req_to_sk(req), skb);
E
Eric Dumazet 已提交
3189
	rcu_read_unlock();
3190 3191
#endif

3192 3193
	/* Do not fool tcpdump (if any), clean our debris */
	skb->tstamp.tv64 = 0;
L
Linus Torvalds 已提交
3194 3195
	return skb;
}
E
Eric Dumazet 已提交
3196
EXPORT_SYMBOL(tcp_make_synack);
L
Linus Torvalds 已提交
3197

3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216
static void tcp_ca_dst_init(struct sock *sk, const struct dst_entry *dst)
{
	struct inet_connection_sock *icsk = inet_csk(sk);
	const struct tcp_congestion_ops *ca;
	u32 ca_key = dst_metric(dst, RTAX_CC_ALGO);

	if (ca_key == TCP_CA_UNSPEC)
		return;

	rcu_read_lock();
	ca = tcp_ca_find_key(ca_key);
	if (likely(ca && try_module_get(ca->owner))) {
		module_put(icsk->icsk_ca_ops->owner);
		icsk->icsk_ca_dst_locked = tcp_ca_dst_locked(dst);
		icsk->icsk_ca_ops = ca;
	}
	rcu_read_unlock();
}

3217
/* Do all connect socket setups that can be done AF independent. */
3218
static void tcp_connect_init(struct sock *sk)
L
Linus Torvalds 已提交
3219
{
3220
	const struct dst_entry *dst = __sk_dst_get(sk);
L
Linus Torvalds 已提交
3221 3222 3223 3224 3225 3226 3227
	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) +
3228
		(sysctl_tcp_timestamps ? TCPOLEN_TSTAMP_ALIGNED : 0);
L
Linus Torvalds 已提交
3229

3230
#ifdef CONFIG_TCP_MD5SIG
3231
	if (tp->af_specific->md5_lookup(sk, sk))
3232 3233 3234
		tp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
#endif

L
Linus Torvalds 已提交
3235 3236 3237 3238
	/* 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
3239
	tcp_mtup_init(sk);
L
Linus Torvalds 已提交
3240 3241
	tcp_sync_mss(sk, dst_mtu(dst));

3242 3243
	tcp_ca_dst_init(sk, dst);

L
Linus Torvalds 已提交
3244 3245
	if (!tp->window_clamp)
		tp->window_clamp = dst_metric(dst, RTAX_WINDOW);
3246
	tp->advmss = dst_metric_advmss(dst);
3247 3248 3249
	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
3250 3251
	tcp_initialize_rcv_mss(sk);

3252 3253 3254 3255 3256
	/* 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
3257 3258 3259 3260
	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,
3261
				  sysctl_tcp_window_scaling,
3262 3263
				  &rcv_wscale,
				  dst_metric(dst, RTAX_INITRWND));
L
Linus Torvalds 已提交
3264 3265 3266 3267 3268 3269 3270

	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;
3271
	tcp_init_wl(tp, 0);
L
Linus Torvalds 已提交
3272 3273
	tp->snd_una = tp->write_seq;
	tp->snd_sml = tp->write_seq;
I
Ilpo Järvinen 已提交
3274
	tp->snd_up = tp->write_seq;
P
Pavel Emelyanov 已提交
3275
	tp->snd_nxt = tp->write_seq;
P
Pavel Emelyanov 已提交
3276 3277 3278

	if (likely(!tp->repair))
		tp->rcv_nxt = 0;
3279 3280
	else
		tp->rcv_tstamp = tcp_time_stamp;
P
Pavel Emelyanov 已提交
3281 3282
	tp->rcv_wup = tp->rcv_nxt;
	tp->copied_seq = tp->rcv_nxt;
L
Linus Torvalds 已提交
3283

3284 3285
	inet_csk(sk)->icsk_rto = TCP_TIMEOUT_INIT;
	inet_csk(sk)->icsk_retransmits = 0;
L
Linus Torvalds 已提交
3286 3287 3288
	tcp_clear_retrans(tp);
}

3289 3290 3291 3292 3293 3294
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;
3295
	__skb_header_release(skb);
3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313
	__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;
3314
	int syn_loss = 0, space, err = 0;
3315
	unsigned long last_syn_loss = 0;
3316
	struct sk_buff *syn_data;
3317

3318
	tp->rx_opt.mss_clamp = tp->advmss;  /* If MSS is not cached */
3319 3320 3321 3322 3323 3324 3325 3326
	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;
	}
3327

3328 3329 3330
	if (sysctl_tcp_fastopen & TFO_CLIENT_NO_COOKIE)
		fo->cookie.len = -1;
	else if (fo->cookie.len <= 0)
3331 3332 3333 3334 3335 3336 3337 3338
		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;
3339
	space = __tcp_mtu_to_mss(sk, inet_csk(sk)->icsk_pmtu_cookie) -
3340 3341
		MAX_TCP_OPTION_SPACE;

3342 3343 3344 3345 3346
	space = min_t(size_t, space, fo->size);

	/* limit to order-0 allocations */
	space = min_t(size_t, space, SKB_MAX_HEAD(MAX_TCP_HEADER));

3347
	syn_data = sk_stream_alloc_skb(sk, space, sk->sk_allocation, false);
3348
	if (!syn_data)
3349
		goto fallback;
3350 3351
	syn_data->ip_summed = CHECKSUM_PARTIAL;
	memcpy(syn_data->cb, syn->cb, sizeof(syn->cb));
3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362
	if (space) {
		int copied = copy_from_iter(skb_put(syn_data, space), space,
					    &fo->data->msg_iter);
		if (unlikely(!copied)) {
			kfree_skb(syn_data);
			goto fallback;
		}
		if (copied != space) {
			skb_trim(syn_data, copied);
			space = copied;
		}
3363
	}
3364 3365 3366 3367
	/* No more data pending in inet_wait_for_connect() */
	if (space == fo->size)
		fo->data = NULL;
	fo->copied = space;
3368

3369
	tcp_connect_queue_skb(sk, syn_data);
3370 3371
	if (syn_data->len)
		tcp_chrono_start(sk, TCP_CHRONO_BUSY);
3372

3373
	err = tcp_transmit_skb(sk, syn_data, 1, sk->sk_allocation);
3374

3375
	syn->skb_mstamp = syn_data->skb_mstamp;
E
Eric Dumazet 已提交
3376

3377 3378 3379 3380 3381 3382 3383 3384
	/* Now full SYN+DATA was cloned and sent (or not),
	 * remove the SYN from the original skb (syn_data)
	 * we keep in write queue in case of a retransmit, as we
	 * also have the SYN packet (with no data) in the same queue.
	 */
	TCP_SKB_CB(syn_data)->seq++;
	TCP_SKB_CB(syn_data)->tcp_flags = TCPHDR_ACK | TCPHDR_PSH;
	if (!err) {
3385
		tp->syn_data = (fo->copied > 0);
3386
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPORIGDATASENT);
3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401
		goto done;
	}

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;
done:
	fo->cookie.len = -1;  /* Exclude Fast Open option for SYN retries */
	return err;
}

3402
/* Build a SYN and send it off. */
L
Linus Torvalds 已提交
3403 3404 3405 3406
int tcp_connect(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *buff;
3407
	int err;
L
Linus Torvalds 已提交
3408 3409 3410

	tcp_connect_init(sk);

A
Andrey Vagin 已提交
3411 3412 3413 3414 3415
	if (unlikely(tp->repair)) {
		tcp_finish_connect(sk, NULL);
		return 0;
	}

3416
	buff = sk_stream_alloc_skb(sk, 0, sk->sk_allocation, true);
3417
	if (unlikely(!buff))
L
Linus Torvalds 已提交
3418 3419
		return -ENOBUFS;

C
Changli Gao 已提交
3420
	tcp_init_nondata_skb(buff, tp->write_seq++, TCPHDR_SYN);
E
Eric Dumazet 已提交
3421
	tp->retrans_stamp = tcp_time_stamp;
3422
	tcp_connect_queue_skb(sk, buff);
3423
	tcp_ecn_send_syn(sk, buff);
L
Linus Torvalds 已提交
3424

3425 3426 3427
	/* 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);
3428 3429
	if (err == -ECONNREFUSED)
		return err;
3430 3431 3432 3433 3434 3435

	/* 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
3436
	TCP_INC_STATS(sock_net(sk), TCP_MIB_ACTIVEOPENS);
L
Linus Torvalds 已提交
3437 3438

	/* Timer for repeating the SYN until an answer. */
3439 3440
	inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
				  inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
L
Linus Torvalds 已提交
3441 3442
	return 0;
}
E
Eric Dumazet 已提交
3443
EXPORT_SYMBOL(tcp_connect);
L
Linus Torvalds 已提交
3444 3445 3446 3447 3448 3449 3450

/* 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)
{
3451 3452
	struct inet_connection_sock *icsk = inet_csk(sk);
	int ato = icsk->icsk_ack.ato;
L
Linus Torvalds 已提交
3453 3454
	unsigned long timeout;

3455 3456
	tcp_ca_event(sk, CA_EVENT_DELAYED_ACK);

L
Linus Torvalds 已提交
3457
	if (ato > TCP_DELACK_MIN) {
3458
		const struct tcp_sock *tp = tcp_sk(sk);
3459
		int max_ato = HZ / 2;
L
Linus Torvalds 已提交
3460

3461 3462
		if (icsk->icsk_ack.pingpong ||
		    (icsk->icsk_ack.pending & ICSK_ACK_PUSHED))
L
Linus Torvalds 已提交
3463 3464 3465 3466 3467
			max_ato = TCP_DELACK_MAX;

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

		/* If some rtt estimate is known, use it to bound delayed ack.
3468
		 * Do not use inet_csk(sk)->icsk_rto here, use results of rtt measurements
L
Linus Torvalds 已提交
3469 3470
		 * directly.
		 */
3471 3472 3473
		if (tp->srtt_us) {
			int rtt = max_t(int, usecs_to_jiffies(tp->srtt_us >> 3),
					TCP_DELACK_MIN);
L
Linus Torvalds 已提交
3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485

			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. */
3486
	if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
L
Linus Torvalds 已提交
3487 3488 3489
		/* If delack timer was blocked or is about to expire,
		 * send ACK now.
		 */
3490 3491
		if (icsk->icsk_ack.blocked ||
		    time_before_eq(icsk->icsk_ack.timeout, jiffies + (ato >> 2))) {
L
Linus Torvalds 已提交
3492 3493 3494 3495
			tcp_send_ack(sk);
			return;
		}

3496 3497
		if (!time_before(timeout, icsk->icsk_ack.timeout))
			timeout = icsk->icsk_ack.timeout;
L
Linus Torvalds 已提交
3498
	}
3499 3500 3501
	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
3502 3503 3504 3505 3506
}

/* This routine sends an ack and also updates the window. */
void tcp_send_ack(struct sock *sk)
{
3507
	struct sk_buff *buff;
L
Linus Torvalds 已提交
3508

3509 3510 3511
	/* If we have been reset, we may not send again. */
	if (sk->sk_state == TCP_CLOSE)
		return;
L
Linus Torvalds 已提交
3512

3513 3514
	tcp_ca_event(sk, CA_EVENT_NON_DELAYED_ACK);

3515 3516 3517 3518
	/* We are not putting this on the write queue, so
	 * tcp_transmit_skb() will set the ownership to this
	 * sock.
	 */
3519 3520 3521
	buff = alloc_skb(MAX_TCP_HEADER,
			 sk_gfp_mask(sk, GFP_ATOMIC | __GFP_NOWARN));
	if (unlikely(!buff)) {
3522 3523 3524 3525 3526
		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
3527
	}
3528 3529 3530

	/* Reserve space for headers and prepare control bits. */
	skb_reserve(buff, MAX_TCP_HEADER);
C
Changli Gao 已提交
3531
	tcp_init_nondata_skb(buff, tcp_acceptable_seq(sk), TCPHDR_ACK);
3532

E
Eric Dumazet 已提交
3533 3534 3535 3536 3537 3538 3539 3540
	/* We do not want pure acks influencing TCP Small Queues or fq/pacing
	 * too much.
	 * SKB_TRUESIZE(max(1 .. 66, MAX_TCP_HEADER)) is unfortunately ~784
	 * We also avoid tcp_wfree() overhead (cache line miss accessing
	 * tp->tsq_flags) by using regular sock_wfree()
	 */
	skb_set_tcp_pure_ack(buff);

3541
	/* Send it off, this clears delayed acks for us. */
E
Eric Dumazet 已提交
3542
	skb_mstamp_get(&buff->skb_mstamp);
3543
	tcp_transmit_skb(sk, buff, 0, (__force gfp_t)0);
L
Linus Torvalds 已提交
3544
}
3545
EXPORT_SYMBOL_GPL(tcp_send_ack);
L
Linus Torvalds 已提交
3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557

/* 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.
 */
3558
static int tcp_xmit_probe_skb(struct sock *sk, int urgent, int mib)
L
Linus Torvalds 已提交
3559 3560 3561 3562 3563
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb;

	/* We don't queue it, tcp_transmit_skb() sets ownership. */
3564 3565
	skb = alloc_skb(MAX_TCP_HEADER,
			sk_gfp_mask(sk, GFP_ATOMIC | __GFP_NOWARN));
3566
	if (!skb)
L
Linus Torvalds 已提交
3567 3568 3569 3570 3571 3572 3573 3574
		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
3575
	tcp_init_nondata_skb(skb, tp->snd_una - !urgent, TCPHDR_ACK);
E
Eric Dumazet 已提交
3576
	skb_mstamp_get(&skb->skb_mstamp);
3577
	NET_INC_STATS(sock_net(sk), mib);
3578
	return tcp_transmit_skb(sk, skb, 0, (__force gfp_t)0);
L
Linus Torvalds 已提交
3579 3580
}

P
Pavel Emelyanov 已提交
3581 3582 3583 3584
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;
3585
		tcp_xmit_probe_skb(sk, 0, LINUX_MIB_TCPWINPROBE);
P
Pavel Emelyanov 已提交
3586 3587 3588
	}
}

3589
/* Initiate keepalive or window probe from timer. */
3590
int tcp_write_wakeup(struct sock *sk, int mib)
L
Linus Torvalds 已提交
3591
{
3592 3593
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb;
L
Linus Torvalds 已提交
3594

3595 3596 3597
	if (sk->sk_state == TCP_CLOSE)
		return -1;

3598 3599
	skb = tcp_send_head(sk);
	if (skb && before(TCP_SKB_CB(skb)->seq, tcp_wnd_end(tp))) {
3600
		int err;
I
Ilpo Järvinen 已提交
3601
		unsigned int mss = tcp_current_mss(sk);
3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613
		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
3614
			TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
3615
			if (tcp_fragment(sk, skb, seg_size, mss, GFP_ATOMIC))
3616 3617
				return -1;
		} else if (!tcp_skb_pcount(skb))
3618
			tcp_set_skb_tso_segs(skb, mss);
3619

E
Eric Dumazet 已提交
3620
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
3621 3622 3623 3624 3625
		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
3626
		if (between(tp->snd_up, tp->snd_una + 1, tp->snd_una + 0xFFFF))
3627 3628
			tcp_xmit_probe_skb(sk, 1, mib);
		return tcp_xmit_probe_skb(sk, 0, mib);
L
Linus Torvalds 已提交
3629 3630 3631 3632 3633 3634 3635 3636
	}
}

/* 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)
{
3637
	struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
3638
	struct tcp_sock *tp = tcp_sk(sk);
3639
	struct net *net = sock_net(sk);
3640
	unsigned long probe_max;
L
Linus Torvalds 已提交
3641 3642
	int err;

3643
	err = tcp_write_wakeup(sk, LINUX_MIB_TCPWINPROBE);
L
Linus Torvalds 已提交
3644

3645
	if (tp->packets_out || !tcp_send_head(sk)) {
L
Linus Torvalds 已提交
3646
		/* Cancel probe timer, if it is not required. */
3647
		icsk->icsk_probes_out = 0;
3648
		icsk->icsk_backoff = 0;
L
Linus Torvalds 已提交
3649 3650 3651 3652
		return;
	}

	if (err <= 0) {
3653
		if (icsk->icsk_backoff < net->ipv4.sysctl_tcp_retries2)
3654
			icsk->icsk_backoff++;
3655
		icsk->icsk_probes_out++;
3656
		probe_max = TCP_RTO_MAX;
L
Linus Torvalds 已提交
3657 3658
	} else {
		/* If packet was not sent due to local congestion,
3659
		 * do not backoff and do not remember icsk_probes_out.
L
Linus Torvalds 已提交
3660 3661 3662 3663
		 * Let local senders to fight for local resources.
		 *
		 * Use accumulated backoff yet.
		 */
3664 3665
		if (!icsk->icsk_probes_out)
			icsk->icsk_probes_out = 1;
3666
		probe_max = TCP_RESOURCE_PROBE_INTERVAL;
L
Linus Torvalds 已提交
3667
	}
3668
	inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
3669
				  tcp_probe0_when(sk, probe_max),
3670
				  TCP_RTO_MAX);
L
Linus Torvalds 已提交
3671
}
3672

3673
int tcp_rtx_synack(const struct sock *sk, struct request_sock *req)
3674 3675 3676 3677 3678
{
	const struct tcp_request_sock_ops *af_ops = tcp_rsk(req)->af_specific;
	struct flowi fl;
	int res;

3679
	tcp_rsk(req)->txhash = net_tx_rndhash();
3680
	res = af_ops->send_synack(sk, NULL, &fl, req, NULL, TCP_SYNACK_NORMAL);
3681
	if (!res) {
E
Eric Dumazet 已提交
3682
		__TCP_INC_STATS(sock_net(sk), TCP_MIB_RETRANSSEGS);
3683
		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPSYNRETRANS);
3684 3685
		if (unlikely(tcp_passive_fastopen(sk)))
			tcp_sk(sk)->total_retrans++;
3686 3687 3688 3689
	}
	return res;
}
EXPORT_SYMBOL(tcp_rtx_synack);