tcp_output.c 107.0 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);
Y
Yuchung Cheng 已提交
79
	if (!prior_packets || icsk->icsk_pending == ICSK_TIME_LOSS_PROBE)
80
		tcp_rearm_rto(sk);
81

82 83
	NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPORIGDATASENT,
		      tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
84 85
}

86 87
/* SND.NXT, if window was not shrunk or the amount of shrunk was less than one
 * window scaling factor due to loss of precision.
L
Linus Torvalds 已提交
88 89 90 91 92
 * 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:
 */
93
static inline __u32 tcp_acceptable_seq(const struct sock *sk)
L
Linus Torvalds 已提交
94
{
95
	const struct tcp_sock *tp = tcp_sk(sk);
96

97 98 99
	if (!before(tcp_wnd_end(tp), tp->snd_nxt) ||
	    (tp->rx_opt.wscale_ok &&
	     ((tp->snd_nxt - tcp_wnd_end(tp)) < (1 << tp->rx_opt.rcv_wscale))))
L
Linus Torvalds 已提交
100 101
		return tp->snd_nxt;
	else
102
		return tcp_wnd_end(tp);
L
Linus Torvalds 已提交
103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121
}

/* 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);
122
	const struct dst_entry *dst = __sk_dst_get(sk);
L
Linus Torvalds 已提交
123 124
	int mss = tp->advmss;

125 126 127 128 129 130 131
	if (dst) {
		unsigned int metric = dst_metric_advmss(dst);

		if (metric < mss) {
			mss = metric;
			tp->advmss = mss;
		}
L
Linus Torvalds 已提交
132 133 134 135 136 137
	}

	return (__u16)mss;
}

/* RFC2861. Reset CWND after idle period longer RTO to "restart window".
138 139 140
 * This is the first part of cwnd validation mechanism.
 */
void tcp_cwnd_restart(struct sock *sk, s32 delta)
L
Linus Torvalds 已提交
141
{
142
	struct tcp_sock *tp = tcp_sk(sk);
143
	u32 restart_cwnd = tcp_init_cwnd(tp, __sk_dst_get(sk));
L
Linus Torvalds 已提交
144 145
	u32 cwnd = tp->snd_cwnd;

146
	tcp_ca_event(sk, CA_EVENT_CWND_RESTART);
L
Linus Torvalds 已提交
147

148
	tp->snd_ssthresh = tcp_current_ssthresh(sk);
L
Linus Torvalds 已提交
149 150
	restart_cwnd = min(restart_cwnd, cwnd);

151
	while ((delta -= inet_csk(sk)->icsk_rto) > 0 && cwnd > restart_cwnd)
L
Linus Torvalds 已提交
152 153
		cwnd >>= 1;
	tp->snd_cwnd = max(cwnd, restart_cwnd);
154
	tp->snd_cwnd_stamp = tcp_jiffies32;
L
Linus Torvalds 已提交
155 156 157
	tp->snd_cwnd_used = 0;
}

158
/* Congestion state accounting after a packet has been sent. */
S
Stephen Hemminger 已提交
159
static void tcp_event_data_sent(struct tcp_sock *tp,
160
				struct sock *sk)
L
Linus Torvalds 已提交
161
{
162
	struct inet_connection_sock *icsk = inet_csk(sk);
163
	const u32 now = tcp_jiffies32;
L
Linus Torvalds 已提交
164

165 166 167
	if (tcp_packets_in_flight(tp) == 0)
		tcp_ca_event(sk, CA_EVENT_TX_START);

L
Linus Torvalds 已提交
168 169 170 171 172
	tp->lsndtime = now;

	/* If it is a reply for ato after last received
	 * packet, enter pingpong mode.
	 */
173 174
	if ((u32)(now - icsk->icsk_ack.lrcvtime) < icsk->icsk_ack.ato)
		icsk->icsk_ack.pingpong = 1;
L
Linus Torvalds 已提交
175 176
}

177
/* Account for an ACK we sent. */
S
Stephen Hemminger 已提交
178
static inline void tcp_event_ack_sent(struct sock *sk, unsigned int pkts)
L
Linus Torvalds 已提交
179
{
180 181
	tcp_dec_quickack_mode(sk, pkts);
	inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
L
Linus Torvalds 已提交
182 183
}

184 185 186 187

u32 tcp_default_init_rwnd(u32 mss)
{
	/* Initial receive window should be twice of TCP_INIT_CWND to
W
Weiping Pan 已提交
188
	 * enable proper sending of new unsent data during fast recovery
189 190 191 192 193 194 195 196 197 198
	 * (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
199 200 201 202 203 204 205 206 207
/* 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,
208 209
			       int wscale_ok, __u8 *rcv_wscale,
			       __u32 init_rcv_wnd)
L
Linus Torvalds 已提交
210 211 212 213 214
{
	unsigned int space = (__space < 0 ? 0 : __space);

	/* If no clamp set the clamp to the max possible scaled window */
	if (*window_clamp == 0)
215
		(*window_clamp) = (U16_MAX << TCP_MAX_WSCALE);
L
Linus Torvalds 已提交
216 217 218 219
	space = min(*window_clamp, space);

	/* Quantize space offering to a multiple of mss if possible. */
	if (space > mss)
220
		space = rounddown(space, mss);
L
Linus Torvalds 已提交
221 222

	/* NOTE: offering an initial window larger than 32767
223 224 225 226 227 228
	 * 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
229
	 */
230 231 232 233 234
	if (sysctl_tcp_workaround_signed_windows)
		(*rcv_wnd) = min(space, MAX_TCP_WINDOW);
	else
		(*rcv_wnd) = space;

L
Linus Torvalds 已提交
235 236
	(*rcv_wscale) = 0;
	if (wscale_ok) {
237
		/* Set window scaling on max possible window */
238 239
		space = max_t(u32, space, sysctl_tcp_rmem[2]);
		space = max_t(u32, space, sysctl_rmem_max);
240
		space = min_t(u32, space, *window_clamp);
241
		while (space > U16_MAX && (*rcv_wscale) < TCP_MAX_WSCALE) {
L
Linus Torvalds 已提交
242 243 244 245 246
			space >>= 1;
			(*rcv_wscale)++;
		}
	}

247
	if (mss > (1 << *rcv_wscale)) {
248 249 250
		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
251 252 253
	}

	/* Set the clamp no higher than max representable value */
254
	(*window_clamp) = min_t(__u32, U16_MAX << (*rcv_wscale), *window_clamp);
L
Linus Torvalds 已提交
255
}
E
Eric Dumazet 已提交
256
EXPORT_SYMBOL(tcp_select_initial_window);
L
Linus Torvalds 已提交
257 258 259 260 261 262

/* 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
263
static u16 tcp_select_window(struct sock *sk)
L
Linus Torvalds 已提交
264 265
{
	struct tcp_sock *tp = tcp_sk(sk);
266
	u32 old_win = tp->rcv_wnd;
L
Linus Torvalds 已提交
267 268 269 270
	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
271
	if (new_win < cur_win) {
L
Linus Torvalds 已提交
272 273 274 275 276 277 278
		/* 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.
		 */
279 280 281
		if (new_win == 0)
			NET_INC_STATS(sock_net(sk),
				      LINUX_MIB_TCPWANTZEROWINDOWADV);
282
		new_win = ALIGN(cur_win, 1 << tp->rx_opt.rcv_wscale);
L
Linus Torvalds 已提交
283 284 285 286 287 288 289
	}
	tp->rcv_wnd = new_win;
	tp->rcv_wup = tp->rcv_nxt;

	/* Make sure we do not exceed the maximum possible
	 * scaled window.
	 */
290
	if (!tp->rx_opt.rcv_wscale && sysctl_tcp_workaround_signed_windows)
L
Linus Torvalds 已提交
291 292 293 294 295 296 297 298
		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. */
299
	if (new_win == 0) {
L
Linus Torvalds 已提交
300
		tp->pred_flags = 0;
301 302 303 304 305 306
		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
307 308 309 310

	return new_win;
}

311
/* Packet ECN state for a SYN-ACK */
312
static void tcp_ecn_send_synack(struct sock *sk, struct sk_buff *skb)
313
{
314 315
	const struct tcp_sock *tp = tcp_sk(sk);

E
Eric Dumazet 已提交
316
	TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_CWR;
317
	if (!(tp->ecn_flags & TCP_ECN_OK))
E
Eric Dumazet 已提交
318
		TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_ECE;
319 320
	else if (tcp_ca_needs_ecn(sk))
		INET_ECN_xmit(sk);
321 322
}

323
/* Packet ECN state for a SYN.  */
324
static void tcp_ecn_send_syn(struct sock *sk, struct sk_buff *skb)
325 326
{
	struct tcp_sock *tp = tcp_sk(sk);
327 328 329 330 331 332 333 334 335
	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;
	}
336 337

	tp->ecn_flags = 0;
338 339

	if (use_ecn) {
E
Eric Dumazet 已提交
340
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ECE | TCPHDR_CWR;
341
		tp->ecn_flags = TCP_ECN_OK;
342 343
		if (tcp_ca_needs_ecn(sk))
			INET_ECN_xmit(sk);
344 345 346
	}
}

347 348 349 350 351 352 353 354 355
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);
}

356
static void
357
tcp_ecn_make_synack(const struct request_sock *req, struct tcphdr *th)
358
{
359
	if (inet_rsk(req)->ecn_ok)
360 361 362
		th->ece = 1;
}

363 364 365
/* Set up ECN state for a packet on a ESTABLISHED socket that is about to
 * be sent.
 */
366
static void tcp_ecn_send(struct sock *sk, struct sk_buff *skb,
367
			 struct tcphdr *th, int tcp_header_len)
368 369 370 371 372 373 374 375
{
	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);
376
			if (tp->ecn_flags & TCP_ECN_QUEUE_CWR) {
377
				tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR;
378
				th->cwr = 1;
379 380
				skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
			}
381
		} else if (!tcp_ca_needs_ecn(sk)) {
382 383 384 385
			/* ACK or retransmitted segment: clear ECT|CE */
			INET_ECN_dontxmit(sk);
		}
		if (tp->ecn_flags & TCP_ECN_DEMAND_CWR)
386
			th->ece = 1;
387 388 389
	}
}

390 391 392 393 394
/* 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)
{
395
	skb->ip_summed = CHECKSUM_PARTIAL;
396 397
	skb->csum = 0;

E
Eric Dumazet 已提交
398
	TCP_SKB_CB(skb)->tcp_flags = flags;
399 400
	TCP_SKB_CB(skb)->sacked = 0;

401
	tcp_skb_pcount_set(skb, 1);
402 403

	TCP_SKB_CB(skb)->seq = seq;
C
Changli Gao 已提交
404
	if (flags & (TCPHDR_SYN | TCPHDR_FIN))
405 406 407 408
		seq++;
	TCP_SKB_CB(skb)->end_seq = seq;
}

E
Eric Dumazet 已提交
409
static inline bool tcp_urg_mode(const struct tcp_sock *tp)
I
Ilpo Järvinen 已提交
410 411 412 413
{
	return tp->snd_una != tp->snd_up;
}

A
Adam Langley 已提交
414 415 416
#define OPTION_SACK_ADVERTISE	(1 << 0)
#define OPTION_TS		(1 << 1)
#define OPTION_MD5		(1 << 2)
417
#define OPTION_WSCALE		(1 << 3)
Y
Yuchung Cheng 已提交
418
#define OPTION_FAST_OPEN_COOKIE	(1 << 8)
A
Adam Langley 已提交
419 420

struct tcp_out_options {
Y
Yuchung Cheng 已提交
421 422
	u16 options;		/* bit field of OPTION_* */
	u16 mss;		/* 0 to disable */
A
Adam Langley 已提交
423 424
	u8 ws;			/* window scale, 0 to disable */
	u8 num_sack_blocks;	/* number of SACK blocks to include */
425 426
	u8 hash_size;		/* bytes in hash_location */
	__u8 *hash_location;	/* temporary pointer, overloaded */
Y
Yuchung Cheng 已提交
427 428
	__u32 tsval, tsecr;	/* need to include OPTION_TS */
	struct tcp_fastopen_cookie *fastopen_cookie;	/* Fast open cookie */
A
Adam Langley 已提交
429 430
};

431 432 433
/* Write previously computed TCP options to the packet.
 *
 * Beware: Something in the Internet is very sensitive to the ordering of
434 435
 * 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
436
 * inter-operability perspective it seems that we're somewhat stuck with
437 438 439 440 441 442 443
 * the ordering which we have been using if we want to keep working with
 * those broken things (not that it currently hurts anybody as there isn't
 * particular reason why the ordering would need to be changed).
 *
 * At least SACK_PERM as the first option is known to lead to a disaster
 * (but it may well be that other scenarios fail similarly).
 */
A
Adam Langley 已提交
444
static void tcp_options_write(__be32 *ptr, struct tcp_sock *tp,
445 446
			      struct tcp_out_options *opts)
{
Y
Yuchung Cheng 已提交
447
	u16 options = opts->options;	/* mungable copy */
448 449

	if (unlikely(OPTION_MD5 & options)) {
C
Christoph Paasch 已提交
450 451
		*ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
			       (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
452 453
		/* overload cookie hash location */
		opts->hash_location = (__u8 *)ptr;
A
Adam Langley 已提交
454
		ptr += 4;
S
Stephen Hemminger 已提交
455
	}
A
Adam Langley 已提交
456

457 458 459 460 461 462
	if (unlikely(opts->mss)) {
		*ptr++ = htonl((TCPOPT_MSS << 24) |
			       (TCPOLEN_MSS << 16) |
			       opts->mss);
	}

463 464
	if (likely(OPTION_TS & options)) {
		if (unlikely(OPTION_SACK_ADVERTISE & options)) {
A
Adam Langley 已提交
465 466 467 468
			*ptr++ = htonl((TCPOPT_SACK_PERM << 24) |
				       (TCPOLEN_SACK_PERM << 16) |
				       (TCPOPT_TIMESTAMP << 8) |
				       TCPOLEN_TIMESTAMP);
469
			options &= ~OPTION_SACK_ADVERTISE;
A
Adam Langley 已提交
470 471 472 473 474 475 476 477 478 479
		} else {
			*ptr++ = htonl((TCPOPT_NOP << 24) |
				       (TCPOPT_NOP << 16) |
				       (TCPOPT_TIMESTAMP << 8) |
				       TCPOLEN_TIMESTAMP);
		}
		*ptr++ = htonl(opts->tsval);
		*ptr++ = htonl(opts->tsecr);
	}

480
	if (unlikely(OPTION_SACK_ADVERTISE & options)) {
A
Adam Langley 已提交
481 482 483 484 485 486
		*ptr++ = htonl((TCPOPT_NOP << 24) |
			       (TCPOPT_NOP << 16) |
			       (TCPOPT_SACK_PERM << 8) |
			       TCPOLEN_SACK_PERM);
	}

487
	if (unlikely(OPTION_WSCALE & options)) {
A
Adam Langley 已提交
488 489 490 491 492 493 494 495 496
		*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
497 498 499 500 501
		int this_sack;

		*ptr++ = htonl((TCPOPT_NOP  << 24) |
			       (TCPOPT_NOP  << 16) |
			       (TCPOPT_SACK <<  8) |
A
Adam Langley 已提交
502
			       (TCPOLEN_SACK_BASE + (opts->num_sack_blocks *
S
Stephen Hemminger 已提交
503
						     TCPOLEN_SACK_PERBLOCK)));
S
Stephen Hemminger 已提交
504

A
Adam Langley 已提交
505 506
		for (this_sack = 0; this_sack < opts->num_sack_blocks;
		     ++this_sack) {
S
Stephen Hemminger 已提交
507 508 509
			*ptr++ = htonl(sp[this_sack].start_seq);
			*ptr++ = htonl(sp[this_sack].end_seq);
		}
S
Stephen Hemminger 已提交
510

511
		tp->rx_opt.dsack = 0;
S
Stephen Hemminger 已提交
512
	}
Y
Yuchung Cheng 已提交
513 514 515

	if (unlikely(OPTION_FAST_OPEN_COOKIE & options)) {
		struct tcp_fastopen_cookie *foc = opts->fastopen_cookie;
516 517 518 519 520 521 522 523 524 525 526 527 528
		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
529

530 531 532 533
		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
534
		}
535
		ptr += (len + 3) >> 2;
Y
Yuchung Cheng 已提交
536
	}
A
Adam Langley 已提交
537 538
}

539 540 541
/* Compute TCP options for SYN packets. This is not the final
 * network wire format yet.
 */
542
static unsigned int tcp_syn_options(struct sock *sk, struct sk_buff *skb,
A
Adam Langley 已提交
543
				struct tcp_out_options *opts,
544 545
				struct tcp_md5sig_key **md5)
{
A
Adam Langley 已提交
546
	struct tcp_sock *tp = tcp_sk(sk);
547
	unsigned int remaining = MAX_TCP_OPTION_SPACE;
548
	struct tcp_fastopen_request *fastopen = tp->fastopen_req;
A
Adam Langley 已提交
549

550
#ifdef CONFIG_TCP_MD5SIG
A
Adam Langley 已提交
551 552 553
	*md5 = tp->af_specific->md5_lookup(sk, sk);
	if (*md5) {
		opts->options |= OPTION_MD5;
554
		remaining -= TCPOLEN_MD5SIG_ALIGNED;
555
	}
A
Adam Langley 已提交
556 557
#else
	*md5 = NULL;
558
#endif
A
Adam Langley 已提交
559 560 561 562 563 564 565 566 567 568 569

	/* 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);
570
	remaining -= TCPOLEN_MSS_ALIGNED;
A
Adam Langley 已提交
571

572
	if (likely(sock_net(sk)->ipv4.sysctl_tcp_timestamps && !*md5)) {
A
Adam Langley 已提交
573
		opts->options |= OPTION_TS;
E
Eric Dumazet 已提交
574
		opts->tsval = tcp_skb_timestamp(skb) + tp->tsoffset;
A
Adam Langley 已提交
575
		opts->tsecr = tp->rx_opt.ts_recent;
576
		remaining -= TCPOLEN_TSTAMP_ALIGNED;
A
Adam Langley 已提交
577
	}
578
	if (likely(sock_net(sk)->ipv4.sysctl_tcp_window_scaling)) {
A
Adam Langley 已提交
579
		opts->ws = tp->rx_opt.rcv_wscale;
580
		opts->options |= OPTION_WSCALE;
581
		remaining -= TCPOLEN_WSCALE_ALIGNED;
A
Adam Langley 已提交
582
	}
E
Eric Dumazet 已提交
583
	if (likely(sock_net(sk)->ipv4.sysctl_tcp_sack)) {
A
Adam Langley 已提交
584
		opts->options |= OPTION_SACK_ADVERTISE;
585
		if (unlikely(!(OPTION_TS & opts->options)))
586
			remaining -= TCPOLEN_SACKPERM_ALIGNED;
A
Adam Langley 已提交
587 588
	}

589
	if (fastopen && fastopen->cookie.len >= 0) {
590 591 592 593
		u32 need = fastopen->cookie.len;

		need += fastopen->cookie.exp ? TCPOLEN_EXP_FASTOPEN_BASE :
					       TCPOLEN_FASTOPEN_BASE;
594 595 596 597 598 599
		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;
600
			tp->syn_fastopen_exp = fastopen->cookie.exp ? 1 : 0;
601 602
		}
	}
603 604

	return MAX_TCP_OPTION_SPACE - remaining;
S
Stephen Hemminger 已提交
605 606
}

607
/* Set up TCP options for SYN-ACKs. */
608 609 610 611 612
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)
613
{
A
Adam Langley 已提交
614
	struct inet_request_sock *ireq = inet_rsk(req);
615
	unsigned int remaining = MAX_TCP_OPTION_SPACE;
A
Adam Langley 已提交
616

617
#ifdef CONFIG_TCP_MD5SIG
E
Eric Dumazet 已提交
618
	if (md5) {
A
Adam Langley 已提交
619
		opts->options |= OPTION_MD5;
620 621 622 623 624 625 626
		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
627
		ireq->tstamp_ok &= !ireq->sack_ok;
628 629
	}
#endif
A
Adam Langley 已提交
630

631
	/* We always send an MSS option. */
A
Adam Langley 已提交
632
	opts->mss = mss;
633
	remaining -= TCPOLEN_MSS_ALIGNED;
A
Adam Langley 已提交
634 635 636

	if (likely(ireq->wscale_ok)) {
		opts->ws = ireq->rcv_wscale;
637
		opts->options |= OPTION_WSCALE;
638
		remaining -= TCPOLEN_WSCALE_ALIGNED;
A
Adam Langley 已提交
639
	}
E
Eric Dumazet 已提交
640
	if (likely(ireq->tstamp_ok)) {
A
Adam Langley 已提交
641
		opts->options |= OPTION_TS;
642
		opts->tsval = tcp_skb_timestamp(skb) + tcp_rsk(req)->ts_off;
A
Adam Langley 已提交
643
		opts->tsecr = req->ts_recent;
644
		remaining -= TCPOLEN_TSTAMP_ALIGNED;
A
Adam Langley 已提交
645 646 647
	}
	if (likely(ireq->sack_ok)) {
		opts->options |= OPTION_SACK_ADVERTISE;
E
Eric Dumazet 已提交
648
		if (unlikely(!ireq->tstamp_ok))
649
			remaining -= TCPOLEN_SACKPERM_ALIGNED;
A
Adam Langley 已提交
650
	}
651 652 653 654 655
	if (foc != NULL && foc->len >= 0) {
		u32 need = foc->len;

		need += foc->exp ? TCPOLEN_EXP_FASTOPEN_BASE :
				   TCPOLEN_FASTOPEN_BASE;
656 657 658 659 660 661 662
		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
663

664
	return MAX_TCP_OPTION_SPACE - remaining;
A
Adam Langley 已提交
665 666
}

667 668 669
/* Compute TCP options for ESTABLISHED sockets. This is not the
 * final wire format yet.
 */
670
static unsigned int tcp_established_options(struct sock *sk, struct sk_buff *skb,
A
Adam Langley 已提交
671
					struct tcp_out_options *opts,
672 673
					struct tcp_md5sig_key **md5)
{
A
Adam Langley 已提交
674
	struct tcp_sock *tp = tcp_sk(sk);
675
	unsigned int size = 0;
676
	unsigned int eff_sacks;
A
Adam Langley 已提交
677

678 679
	opts->options = 0;

A
Adam Langley 已提交
680 681 682 683 684 685 686 687 688 689 690 691
#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
692
		opts->tsval = skb ? tcp_skb_timestamp(skb) + tp->tsoffset : 0;
A
Adam Langley 已提交
693 694 695 696
		opts->tsecr = tp->rx_opt.ts_recent;
		size += TCPOLEN_TSTAMP_ALIGNED;
	}

697 698
	eff_sacks = tp->rx_opt.num_sacks + tp->rx_opt.dsack;
	if (unlikely(eff_sacks)) {
699
		const unsigned int remaining = MAX_TCP_OPTION_SPACE - size;
A
Adam Langley 已提交
700
		opts->num_sack_blocks =
701
			min_t(unsigned int, eff_sacks,
A
Adam Langley 已提交
702 703 704 705 706 707 708
			      (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
709
}
L
Linus Torvalds 已提交
710

E
Eric Dumazet 已提交
711 712 713 714 715 716 717 718 719

/* 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
720
 * The invariant being skb->truesize subtracted from sk->sk_wmem_alloc
E
Eric Dumazet 已提交
721 722 723 724 725 726 727 728 729 730 731
 *
 * 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);

732 733 734 735
static void tcp_tsq_handler(struct sock *sk)
{
	if ((1 << sk->sk_state) &
	    (TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_CLOSING |
736 737 738 739 740 741 742 743
	     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
744
			       0, GFP_ATOMIC);
745
	}
746
}
E
Eric Dumazet 已提交
747
/*
S
stephen hemminger 已提交
748
 * One tasklet per cpu tries to send more skbs.
E
Eric Dumazet 已提交
749
 * We run in tasklet context but need to disable irqs when
S
stephen hemminger 已提交
750
 * transferring tsq->head because tcp_wfree() might
E
Eric Dumazet 已提交
751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770
 * interrupt us (non NAPI drivers)
 */
static void tcp_tasklet_func(unsigned long data)
{
	struct tsq_tasklet *tsq = (struct tsq_tasklet *)data;
	LIST_HEAD(list);
	unsigned long flags;
	struct list_head *q, *n;
	struct tcp_sock *tp;
	struct sock *sk;

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

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

		sk = (struct sock *)tp;
771
		smp_mb__before_atomic();
772 773
		clear_bit(TSQ_QUEUED, &sk->sk_tsq_flags);

774
		if (!sk->sk_lock.owned &&
775
		    test_bit(TCP_TSQ_DEFERRED, &sk->sk_tsq_flags)) {
776 777
			bh_lock_sock(sk);
			if (!sock_owned_by_user(sk)) {
778
				clear_bit(TCP_TSQ_DEFERRED, &sk->sk_tsq_flags);
779 780 781
				tcp_tsq_handler(sk);
			}
			bh_unlock_sock(sk);
E
Eric Dumazet 已提交
782 783 784 785 786 787
		}

		sk_free(sk);
	}
}

E
Eric Dumazet 已提交
788 789 790 791
#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
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)
{
801
	unsigned long flags, nflags;
E
Eric Dumazet 已提交
802

803 804
	/* perform an atomic operation only if at least one flag is set */
	do {
805
		flags = sk->sk_tsq_flags;
806 807 808
		if (!(flags & TCP_DEFERRED_ALL))
			return;
		nflags = flags & ~TCP_DEFERRED_ALL;
809
	} while (cmpxchg(&sk->sk_tsq_flags, flags, nflags) != flags);
810

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

	/* Keep one reference on sk_wmem_alloc.
	 * Will be released by sk_free() from here or tcp_tasklet_func()
	 */
868
	WARN_ON(refcount_sub_and_test(skb->truesize - 1, &sk->sk_wmem_alloc));
869 870 871 872 873 874 875 876

	/* 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)
	 */
877
	if (refcount_read(&sk->sk_wmem_alloc) >= SKB_TRUESIZE(1) && this_cpu_ksoftirqd() == current)
878
		goto out;
E
Eric Dumazet 已提交
879

880
	for (oval = READ_ONCE(sk->sk_tsq_flags);; oval = nval) {
E
Eric Dumazet 已提交
881
		struct tsq_tasklet *tsq;
882
		bool empty;
E
Eric Dumazet 已提交
883

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

887
		nval = (oval & ~TSQF_THROTTLED) | TSQF_QUEUED | TCPF_TSQ_DEFERRED;
888
		nval = cmpxchg(&sk->sk_tsq_flags, oval, nval);
889 890 891
		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);
895
		empty = list_empty(&tsq->head);
E
Eric Dumazet 已提交
896
		list_add(&tp->tsq_node, &tsq->head);
897 898
		if (empty)
			tasklet_schedule(&tsq->tasklet);
E
Eric Dumazet 已提交
899
		local_irq_restore(flags);
900
		return;
E
Eric Dumazet 已提交
901
	}
902 903
out:
	sk_free(sk);
E
Eric Dumazet 已提交
904 905
}

906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926
/* Note: Called under hard irq.
 * We can not call TCP stack right away.
 */
enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer)
{
	struct tcp_sock *tp = container_of(timer, struct tcp_sock, pacing_timer);
	struct sock *sk = (struct sock *)tp;
	unsigned long nval, oval;

	for (oval = READ_ONCE(sk->sk_tsq_flags);; oval = nval) {
		struct tsq_tasklet *tsq;
		bool empty;

		if (oval & TSQF_QUEUED)
			break;

		nval = (oval & ~TSQF_THROTTLED) | TSQF_QUEUED | TCPF_TSQ_DEFERRED;
		nval = cmpxchg(&sk->sk_tsq_flags, oval, nval);
		if (nval != oval)
			continue;

927
		if (!refcount_inc_not_zero(&sk->sk_wmem_alloc))
928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971
			break;
		/* queue this socket to tasklet queue */
		tsq = this_cpu_ptr(&tsq_tasklet);
		empty = list_empty(&tsq->head);
		list_add(&tp->tsq_node, &tsq->head);
		if (empty)
			tasklet_schedule(&tsq->tasklet);
		break;
	}
	return HRTIMER_NORESTART;
}

/* BBR congestion control needs pacing.
 * Same remark for SO_MAX_PACING_RATE.
 * sch_fq packet scheduler is efficiently handling pacing,
 * but is not always installed/used.
 * Return true if TCP stack should pace packets itself.
 */
static bool tcp_needs_internal_pacing(const struct sock *sk)
{
	return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
}

static void tcp_internal_pacing(struct sock *sk, const struct sk_buff *skb)
{
	u64 len_ns;
	u32 rate;

	if (!tcp_needs_internal_pacing(sk))
		return;
	rate = sk->sk_pacing_rate;
	if (!rate || rate == ~0U)
		return;

	/* Should account for header sizes as sch_fq does,
	 * but lets make things simple.
	 */
	len_ns = (u64)skb->len * NSEC_PER_SEC;
	do_div(len_ns, rate);
	hrtimer_start(&tcp_sk(sk)->pacing_timer,
		      ktime_add_ns(ktime_get(), len_ns),
		      HRTIMER_MODE_ABS_PINNED);
}

L
Linus Torvalds 已提交
972 973 974 975 976 977 978 979 980 981 982
/* 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.
 */
983 984
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
985
{
986 987 988 989
	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
990
	struct tcp_out_options opts;
991
	unsigned int tcp_options_size, tcp_header_size;
992
	struct tcp_md5sig_key *md5;
993 994 995 996
	struct tcphdr *th;
	int err;

	BUG_ON(!skb || !tcp_skb_pcount(skb));
997
	tp = tcp_sk(sk);
998

999
	skb->skb_mstamp = tp->tcp_mstamp;
1000
	if (clone_it) {
1001 1002
		TCP_SKB_CB(skb)->tx.in_flight = TCP_SKB_CB(skb)->end_seq
			- tp->snd_una;
1003
		tcp_rate_skb_sent(sk, skb);
1004

1005 1006 1007 1008 1009 1010 1011
		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
1012

1013 1014
	inet = inet_sk(sk);
	tcb = TCP_SKB_CB(skb);
A
Adam Langley 已提交
1015
	memset(&opts, 0, sizeof(opts));
L
Linus Torvalds 已提交
1016

E
Eric Dumazet 已提交
1017
	if (unlikely(tcb->tcp_flags & TCPHDR_SYN))
A
Adam Langley 已提交
1018 1019 1020 1021 1022
		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);
1023

E
Eric Dumazet 已提交
1024
	/* if no packet is in qdisc/device queue, then allow XPS to select
1025 1026 1027 1028 1029
	 * 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
1030
	 */
1031
	skb->ooo_okay = sk_wmem_alloc_get(sk) < SKB_TRUESIZE(1);
1032

1033 1034 1035 1036 1037 1038 1039
	/* If we had to use memory reserve to allocate this skb,
	 * this might cause drops if packet is looped back :
	 * Other socket might not have SOCK_MEMALLOC.
	 * Packets not looped back do not care about pfmemalloc.
	 */
	skb->pfmemalloc = 0;

1040 1041
	skb_push(skb, tcp_header_size);
	skb_reset_transport_header(skb);
E
Eric Dumazet 已提交
1042 1043 1044

	skb_orphan(skb);
	skb->sk = sk;
E
Eric Dumazet 已提交
1045
	skb->destructor = skb_is_tcp_pure_ack(skb) ? __sock_wfree : tcp_wfree;
1046
	skb_set_hash_from_sk(skb, sk);
1047
	refcount_add(skb->truesize, &sk->sk_wmem_alloc);
1048

1049 1050
	skb_set_dst_pending_confirm(skb, sk->sk_dst_pending_confirm);

1051
	/* Build TCP header and checksum it. */
1052
	th = (struct tcphdr *)skb->data;
E
Eric Dumazet 已提交
1053 1054
	th->source		= inet->inet_sport;
	th->dest		= inet->inet_dport;
1055 1056
	th->seq			= htonl(tcb->seq);
	th->ack_seq		= htonl(tp->rcv_nxt);
1057
	*(((__be16 *)th) + 6)	= htons(((tcp_header_size >> 2) << 12) |
E
Eric Dumazet 已提交
1058
					tcb->tcp_flags);
1059 1060 1061

	th->check		= 0;
	th->urg_ptr		= 0;
L
Linus Torvalds 已提交
1062

I
Ilpo Järvinen 已提交
1063
	/* The urg_mode check is necessary during a below snd_una win probe */
1064 1065 1066 1067 1068
	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)) {
1069
			th->urg_ptr = htons(0xFFFF);
1070 1071
			th->urg = 1;
		}
1072
	}
L
Linus Torvalds 已提交
1073

1074
	tcp_options_write((__be32 *)(th + 1), tp, &opts);
1075
	skb_shinfo(skb)->gso_type = sk->sk_gso_type;
1076 1077 1078 1079 1080 1081 1082 1083 1084
	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));
	}
1085 1086 1087
#ifdef CONFIG_TCP_MD5SIG
	/* Calculate the MD5 hash, as we have all we need now */
	if (md5) {
E
Eric Dumazet 已提交
1088
		sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1089
		tp->af_specific->calc_md5_hash(opts.hash_location,
1090
					       md5, sk, skb);
1091 1092 1093
	}
#endif

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

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

1099
	if (skb->len != tcp_header_size) {
1100
		tcp_event_data_sent(tp, sk);
1101
		tp->data_segs_out += tcp_skb_pcount(skb);
1102
		tcp_internal_pacing(sk, skb);
1103
	}
L
Linus Torvalds 已提交
1104

1105
	if (after(tcb->end_seq, tp->snd_nxt) || tcb->seq == tcb->end_seq)
1106 1107
		TCP_ADD_STATS(sock_net(sk), TCP_MIB_OUTSEGS,
			      tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
1108

1109
	tp->segs_out += tcp_skb_pcount(skb);
1110
	/* OK, its time to fill skb_shinfo(skb)->gso_{segs|size} */
1111
	skb_shinfo(skb)->gso_segs = tcp_skb_pcount(skb);
1112
	skb_shinfo(skb)->gso_size = tcp_skb_mss(skb);
1113

E
Eric Dumazet 已提交
1114
	/* Our usage of tstamp should remain private */
T
Thomas Gleixner 已提交
1115
	skb->tstamp = 0;
1116 1117 1118 1119 1120

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

1121
	err = icsk->icsk_af_ops->queue_xmit(sk, skb, &inet->cork.fl);
E
Eric Dumazet 已提交
1122

1123
	if (likely(err <= 0))
1124 1125
		return err;

1126
	tcp_enter_cwr(sk);
1127

1128
	return net_xmit_eval(err);
L
Linus Torvalds 已提交
1129 1130
}

1131
/* This routine just queues the buffer for sending.
L
Linus Torvalds 已提交
1132 1133 1134 1135 1136 1137 1138 1139 1140 1141
 *
 * 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;
1142
	__skb_header_release(skb);
1143
	tcp_add_write_queue_tail(sk, skb);
1144 1145
	sk->sk_wmem_queued += skb->truesize;
	sk_mem_charge(sk, skb->truesize);
L
Linus Torvalds 已提交
1146 1147
}

1148
/* Initialize TSO segments for a packet. */
1149
static void tcp_set_skb_tso_segs(struct sk_buff *skb, unsigned int mss_now)
1150
{
1151
	if (skb->len <= mss_now || skb->ip_summed == CHECKSUM_NONE) {
1152 1153 1154
		/* Avoid the costly divide in the normal
		 * non-TSO case.
		 */
1155
		tcp_skb_pcount_set(skb, 1);
1156
		TCP_SKB_CB(skb)->tcp_gso_size = 0;
1157
	} else {
1158
		tcp_skb_pcount_set(skb, DIV_ROUND_UP(skb->len, mss_now));
1159
		TCP_SKB_CB(skb)->tcp_gso_size = mss_now;
L
Linus Torvalds 已提交
1160 1161 1162
	}
}

I
Ilpo Järvinen 已提交
1163
/* When a modification to fackets out becomes necessary, we need to check
1164
 * skb is counted to fackets_out or not.
I
Ilpo Järvinen 已提交
1165
 */
1166
static void tcp_adjust_fackets_out(struct sock *sk, const struct sk_buff *skb,
I
Ilpo Järvinen 已提交
1167 1168
				   int decr)
{
1169 1170
	struct tcp_sock *tp = tcp_sk(sk);

1171
	if (!tp->sacked_out || tcp_is_reno(tp))
I
Ilpo Järvinen 已提交
1172 1173
		return;

1174
	if (after(tcp_highest_sack_seq(tp), TCP_SKB_CB(skb)->seq))
I
Ilpo Järvinen 已提交
1175 1176 1177
		tp->fackets_out -= decr;
}

1178 1179 1180
/* Pcount in the middle of the write queue got changed, we need to do various
 * tweaks to fix counters
 */
1181
static void tcp_adjust_pcount(struct sock *sk, const struct sk_buff *skb, int decr)
1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201
{
	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) &&
1202
	    (tcp_is_fack(tp) || (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)))
1203 1204 1205 1206 1207
		tp->lost_cnt_hint -= decr;

	tcp_verify_left_out(tp);
}

1208 1209 1210 1211 1212 1213
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);
}

1214 1215 1216 1217
static void tcp_fragment_tstamp(struct sk_buff *skb, struct sk_buff *skb2)
{
	struct skb_shared_info *shinfo = skb_shinfo(skb);

1218
	if (unlikely(tcp_has_tx_tstamp(skb)) &&
1219 1220 1221 1222 1223 1224 1225
	    !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);
1226 1227
		TCP_SKB_CB(skb2)->txstamp_ack = TCP_SKB_CB(skb)->txstamp_ack;
		TCP_SKB_CB(skb)->txstamp_ack = 0;
1228 1229 1230
	}
}

1231 1232 1233 1234 1235 1236
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
1237 1238
/* 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
1239
 * packet to the list.  This won't be called frequently, I hope.
L
Linus Torvalds 已提交
1240 1241
 * Remember, these are still headerless SKBs at this point.
 */
1242
int tcp_fragment(struct sock *sk, struct sk_buff *skb, u32 len,
1243
		 unsigned int mss_now, gfp_t gfp)
L
Linus Torvalds 已提交
1244 1245 1246
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *buff;
1247
	int nsize, old_factor;
1248
	int nlen;
1249
	u8 flags;
L
Linus Torvalds 已提交
1250

1251 1252
	if (WARN_ON(len > skb->len))
		return -EINVAL;
1253

L
Linus Torvalds 已提交
1254 1255 1256 1257
	nsize = skb_headlen(skb) - len;
	if (nsize < 0)
		nsize = 0;

1258
	if (skb_unclone(skb, gfp))
L
Linus Torvalds 已提交
1259 1260 1261
		return -ENOMEM;

	/* Get a new skb... force flag on. */
1262
	buff = sk_stream_alloc_skb(sk, nsize, gfp, true);
1263
	if (!buff)
L
Linus Torvalds 已提交
1264
		return -ENOMEM; /* We'll just try again later. */
H
Herbert Xu 已提交
1265

1266 1267
	sk->sk_wmem_queued += buff->truesize;
	sk_mem_charge(sk, buff->truesize);
1268 1269 1270
	nlen = skb->len - len - nsize;
	buff->truesize += nlen;
	skb->truesize -= nlen;
L
Linus Torvalds 已提交
1271 1272 1273 1274 1275 1276 1277

	/* 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
1278 1279 1280
	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;
1281
	TCP_SKB_CB(buff)->sacked = TCP_SKB_CB(skb)->sacked;
1282
	tcp_skb_fragment_eor(skb, buff);
L
Linus Torvalds 已提交
1283

1284
	if (!skb_shinfo(skb)->nr_frags && skb->ip_summed != CHECKSUM_PARTIAL) {
L
Linus Torvalds 已提交
1285
		/* Copy and checksum data tail into the new buffer. */
1286 1287
		buff->csum = csum_partial_copy_nocheck(skb->data + len,
						       skb_put(buff, nsize),
L
Linus Torvalds 已提交
1288 1289 1290 1291 1292 1293
						       nsize, 0);

		skb_trim(skb, len);

		skb->csum = csum_block_sub(skb->csum, buff->csum, len);
	} else {
1294
		skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
1295 1296 1297 1298 1299
		skb_split(skb, buff, len);
	}

	buff->ip_summed = skb->ip_summed;

1300
	buff->tstamp = skb->tstamp;
1301
	tcp_fragment_tstamp(skb, buff);
L
Linus Torvalds 已提交
1302

1303 1304
	old_factor = tcp_skb_pcount(skb);

L
Linus Torvalds 已提交
1305
	/* Fix up tso_factor for both original and new SKB.  */
1306 1307
	tcp_set_skb_tso_segs(skb, mss_now);
	tcp_set_skb_tso_segs(buff, mss_now);
L
Linus Torvalds 已提交
1308

1309 1310 1311
	/* Update delivered info for the new segment */
	TCP_SKB_CB(buff)->tx = TCP_SKB_CB(skb)->tx;

1312 1313 1314
	/* If this packet has been sent out already, we must
	 * adjust the various packet counters.
	 */
1315
	if (!before(tp->snd_nxt, TCP_SKB_CB(buff)->end_seq)) {
1316 1317
		int diff = old_factor - tcp_skb_pcount(skb) -
			tcp_skb_pcount(buff);
L
Linus Torvalds 已提交
1318

1319 1320
		if (diff)
			tcp_adjust_pcount(sk, skb, diff);
L
Linus Torvalds 已提交
1321 1322 1323
	}

	/* Link BUFF into the send queue. */
1324
	__skb_header_release(buff);
1325
	tcp_insert_write_queue_after(skb, buff, sk);
L
Linus Torvalds 已提交
1326 1327 1328 1329

	return 0;
}

1330 1331
/* This is similar to __pskb_pull_tail(). The difference is that pulled
 * data is not copied, but immediately discarded.
L
Linus Torvalds 已提交
1332
 */
1333
static int __pskb_trim_head(struct sk_buff *skb, int len)
L
Linus Torvalds 已提交
1334
{
1335
	struct skb_shared_info *shinfo;
L
Linus Torvalds 已提交
1336 1337
	int i, k, eat;

E
Eric Dumazet 已提交
1338 1339 1340 1341 1342
	eat = min_t(int, len, skb_headlen(skb));
	if (eat) {
		__skb_pull(skb, eat);
		len -= eat;
		if (!len)
1343
			return 0;
E
Eric Dumazet 已提交
1344
	}
L
Linus Torvalds 已提交
1345 1346
	eat = len;
	k = 0;
1347 1348 1349
	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
1350 1351

		if (size <= eat) {
1352
			skb_frag_unref(skb, i);
E
Eric Dumazet 已提交
1353
			eat -= size;
L
Linus Torvalds 已提交
1354
		} else {
1355
			shinfo->frags[k] = shinfo->frags[i];
L
Linus Torvalds 已提交
1356
			if (eat) {
1357 1358
				shinfo->frags[k].page_offset += eat;
				skb_frag_size_sub(&shinfo->frags[k], eat);
L
Linus Torvalds 已提交
1359 1360 1361 1362 1363
				eat = 0;
			}
			k++;
		}
	}
1364
	shinfo->nr_frags = k;
L
Linus Torvalds 已提交
1365 1366 1367

	skb->data_len -= len;
	skb->len = skb->data_len;
1368
	return len;
L
Linus Torvalds 已提交
1369 1370
}

1371
/* Remove acked data from a packet in the transmit queue. */
L
Linus Torvalds 已提交
1372 1373
int tcp_trim_head(struct sock *sk, struct sk_buff *skb, u32 len)
{
1374 1375
	u32 delta_truesize;

1376
	if (skb_unclone(skb, GFP_ATOMIC))
L
Linus Torvalds 已提交
1377 1378
		return -ENOMEM;

1379
	delta_truesize = __pskb_trim_head(skb, len);
L
Linus Torvalds 已提交
1380 1381

	TCP_SKB_CB(skb)->seq += len;
1382
	skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
1383

1384 1385 1386 1387 1388 1389
	if (delta_truesize) {
		skb->truesize	   -= delta_truesize;
		sk->sk_wmem_queued -= delta_truesize;
		sk_mem_uncharge(sk, delta_truesize);
		sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
	}
L
Linus Torvalds 已提交
1390

1391
	/* Any change of skb->len requires recalculation of tso factor. */
L
Linus Torvalds 已提交
1392
	if (tcp_skb_pcount(skb) > 1)
1393
		tcp_set_skb_tso_segs(skb, tcp_skb_mss(skb));
L
Linus Torvalds 已提交
1394 1395 1396 1397

	return 0;
}

1398 1399
/* 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
1400
{
1401 1402
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1403 1404 1405 1406 1407 1408 1409
	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);

1410 1411 1412 1413 1414 1415 1416 1417
	/* 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
1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430
	/* 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;
}

1431 1432 1433 1434 1435 1436 1437 1438
/* 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
1439
/* Inverse of above */
1440
int tcp_mss_to_mtu(struct sock *sk, int mss)
J
John Heffner 已提交
1441
{
1442 1443
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1444 1445 1446 1447 1448 1449 1450
	int mtu;

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

1451 1452 1453 1454 1455 1456 1457
	/* 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
1458 1459
	return mtu;
}
1460
EXPORT_SYMBOL(tcp_mss_to_mtu);
J
John Heffner 已提交
1461

1462
/* MTU probing init per socket */
J
John Heffner 已提交
1463 1464 1465 1466
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
1467
	struct net *net = sock_net(sk);
J
John Heffner 已提交
1468

F
Fan Du 已提交
1469
	icsk->icsk_mtup.enabled = net->ipv4.sysctl_tcp_mtu_probing > 1;
J
John Heffner 已提交
1470
	icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp + sizeof(struct tcphdr) +
1471
			       icsk->icsk_af_ops->net_header_len;
F
Fan Du 已提交
1472
	icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, net->ipv4.sysctl_tcp_base_mss);
J
John Heffner 已提交
1473
	icsk->icsk_mtup.probe_size = 0;
1474
	if (icsk->icsk_mtup.enabled)
1475
		icsk->icsk_mtup.probe_timestamp = tcp_jiffies32;
J
John Heffner 已提交
1476
}
E
Eric Dumazet 已提交
1477
EXPORT_SYMBOL(tcp_mtup_init);
J
John Heffner 已提交
1478

L
Linus Torvalds 已提交
1479 1480 1481 1482 1483 1484
/* 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
1485
   It is minimum of user_mss and mss received with SYN.
L
Linus Torvalds 已提交
1486 1487
   It also does not include TCP options.

1488
   inet_csk(sk)->icsk_pmtu_cookie is last pmtu, seen by this function.
L
Linus Torvalds 已提交
1489 1490 1491 1492 1493 1494 1495 1496 1497

   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.

1498 1499
   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
1500 1501 1502 1503
 */
unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu)
{
	struct tcp_sock *tp = tcp_sk(sk);
1504
	struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1505
	int mss_now;
L
Linus Torvalds 已提交
1506

J
John Heffner 已提交
1507 1508
	if (icsk->icsk_mtup.search_high > pmtu)
		icsk->icsk_mtup.search_high = pmtu;
L
Linus Torvalds 已提交
1509

J
John Heffner 已提交
1510
	mss_now = tcp_mtu_to_mss(sk, pmtu);
1511
	mss_now = tcp_bound_to_half_wnd(tp, mss_now);
L
Linus Torvalds 已提交
1512 1513

	/* And store cached results */
1514
	icsk->icsk_pmtu_cookie = pmtu;
J
John Heffner 已提交
1515 1516
	if (icsk->icsk_mtup.enabled)
		mss_now = min(mss_now, tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low));
1517
	tp->mss_cache = mss_now;
L
Linus Torvalds 已提交
1518 1519 1520

	return mss_now;
}
E
Eric Dumazet 已提交
1521
EXPORT_SYMBOL(tcp_sync_mss);
L
Linus Torvalds 已提交
1522 1523 1524 1525

/* Compute the current effective MSS, taking SACKs and IP options,
 * and even PMTU discovery events into account.
 */
I
Ilpo Järvinen 已提交
1526
unsigned int tcp_current_mss(struct sock *sk)
L
Linus Torvalds 已提交
1527
{
1528 1529
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct dst_entry *dst = __sk_dst_get(sk);
1530
	u32 mss_now;
1531
	unsigned int header_len;
A
Adam Langley 已提交
1532 1533
	struct tcp_out_options opts;
	struct tcp_md5sig_key *md5;
1534 1535 1536

	mss_now = tp->mss_cache;

L
Linus Torvalds 已提交
1537 1538
	if (dst) {
		u32 mtu = dst_mtu(dst);
1539
		if (mtu != inet_csk(sk)->icsk_pmtu_cookie)
L
Linus Torvalds 已提交
1540 1541 1542
			mss_now = tcp_sync_mss(sk, mtu);
	}

A
Adam Langley 已提交
1543 1544 1545 1546 1547 1548 1549 1550 1551 1552
	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;
	}
1553

L
Linus Torvalds 已提交
1554 1555 1556
	return mss_now;
}

1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575
/* 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;
	}
1576
	tp->snd_cwnd_stamp = tcp_jiffies32;
1577 1578
}

1579
static void tcp_cwnd_validate(struct sock *sk, bool is_cwnd_limited)
1580
{
1581
	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1582
	struct tcp_sock *tp = tcp_sk(sk);
1583

1584 1585 1586 1587 1588 1589 1590 1591 1592
	/* 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;
	}
1593

1594
	if (tcp_is_cwnd_limited(sk)) {
1595 1596
		/* Network is feed fully. */
		tp->snd_cwnd_used = 0;
1597
		tp->snd_cwnd_stamp = tcp_jiffies32;
1598 1599 1600 1601 1602
	} else {
		/* Network starves. */
		if (tp->packets_out > tp->snd_cwnd_used)
			tp->snd_cwnd_used = tp->packets_out;

1603
		if (sysctl_tcp_slow_start_after_idle &&
1604
		    (s32)(tcp_jiffies32 - tp->snd_cwnd_stamp) >= inet_csk(sk)->icsk_rto &&
1605
		    !ca_ops->cong_control)
1606
			tcp_cwnd_application_limited(sk);
1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618

		/* 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);
1619 1620 1621
	}
}

E
Eric Dumazet 已提交
1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635
/* 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
1636
 */
E
Eric Dumazet 已提交
1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651
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,
1652
			    int nonagle)
E
Eric Dumazet 已提交
1653 1654 1655 1656 1657
{
	return partial &&
		((nonagle & TCP_NAGLE_CORK) ||
		 (!nonagle && tp->packets_out && tcp_minshall_check(tp)));
}
E
Eric Dumazet 已提交
1658 1659 1660 1661

/* Return how many segs we'd like on a TSO packet,
 * to send one TSO packet per ms
 */
1662 1663
u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
		     int min_tso_segs)
E
Eric Dumazet 已提交
1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674
{
	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.
	 */
1675
	segs = max_t(u32, bytes / mss_now, min_tso_segs);
E
Eric Dumazet 已提交
1676 1677 1678

	return min_t(u32, segs, sk->sk_gso_max_segs);
}
1679
EXPORT_SYMBOL(tcp_tso_autosize);
E
Eric Dumazet 已提交
1680

1681 1682 1683 1684 1685 1686 1687 1688
/* 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;

1689 1690
	return tso_segs ? :
		tcp_tso_autosize(sk, mss_now, sysctl_tcp_min_tso_segs);
1691 1692
}

E
Eric Dumazet 已提交
1693 1694 1695 1696 1697 1698
/* 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)
1699
{
1700
	const struct tcp_sock *tp = tcp_sk(sk);
E
Eric Dumazet 已提交
1701
	u32 partial, needed, window, max_len;
1702

1703
	window = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
1704
	max_len = mss_now * max_segs;
1705

1706 1707
	if (likely(max_len <= window && skb != tcp_write_queue_tail(sk)))
		return max_len;
1708

1709 1710
	needed = min(skb->len, window);

1711 1712
	if (max_len <= needed)
		return max_len;
1713

E
Eric Dumazet 已提交
1714 1715 1716 1717 1718
	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
	 */
1719
	if (tcp_nagle_check(partial != 0, tp, nonagle))
E
Eric Dumazet 已提交
1720 1721 1722
		return needed - partial;

	return needed;
1723 1724 1725 1726 1727
}

/* 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.
 */
1728 1729
static inline unsigned int tcp_cwnd_test(const struct tcp_sock *tp,
					 const struct sk_buff *skb)
1730
{
E
Eric Dumazet 已提交
1731
	u32 in_flight, cwnd, halfcwnd;
1732 1733

	/* Don't be strict about the congestion window for the final FIN.  */
E
Eric Dumazet 已提交
1734 1735
	if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) &&
	    tcp_skb_pcount(skb) == 1)
1736 1737 1738 1739
		return 1;

	in_flight = tcp_packets_in_flight(tp);
	cwnd = tp->snd_cwnd;
E
Eric Dumazet 已提交
1740 1741
	if (in_flight >= cwnd)
		return 0;
1742

E
Eric Dumazet 已提交
1743 1744 1745 1746 1747
	/* For better scheduling, ensure we have at least
	 * 2 GSO packets in flight.
	 */
	halfcwnd = max(cwnd >> 1, 1U);
	return min(halfcwnd, cwnd - in_flight);
1748 1749
}

1750
/* Initialize TSO state of a skb.
1751
 * This must be invoked the first time we consider transmitting
1752 1753
 * SKB onto the wire.
 */
1754
static int tcp_init_tso_segs(struct sk_buff *skb, unsigned int mss_now)
1755 1756 1757
{
	int tso_segs = tcp_skb_pcount(skb);

1758
	if (!tso_segs || (tso_segs > 1 && tcp_skb_mss(skb) != mss_now)) {
1759
		tcp_set_skb_tso_segs(skb, mss_now);
1760 1761 1762 1763 1764 1765
		tso_segs = tcp_skb_pcount(skb);
	}
	return tso_segs;
}


E
Eric Dumazet 已提交
1766
/* Return true if the Nagle test allows this packet to be
1767 1768
 * sent now.
 */
E
Eric Dumazet 已提交
1769 1770
static inline bool tcp_nagle_test(const struct tcp_sock *tp, const struct sk_buff *skb,
				  unsigned int cur_mss, int nonagle)
1771 1772 1773 1774 1775 1776 1777 1778
{
	/* 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
1779
		return true;
1780

Y
Yuchung Cheng 已提交
1781 1782
	/* 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
1783
		return true;
1784

1785
	if (!tcp_nagle_check(skb->len < cur_mss, tp, nonagle))
E
Eric Dumazet 已提交
1786
		return true;
1787

E
Eric Dumazet 已提交
1788
	return false;
1789 1790 1791
}

/* Does at least the first segment of SKB fit into the send window? */
E
Eric Dumazet 已提交
1792 1793 1794
static bool tcp_snd_wnd_test(const struct tcp_sock *tp,
			     const struct sk_buff *skb,
			     unsigned int cur_mss)
1795 1796 1797 1798 1799 1800
{
	u32 end_seq = TCP_SKB_CB(skb)->end_seq;

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

1801
	return !after(end_seq, tcp_wnd_end(tp));
1802 1803
}

1804
/* This checks if the data bearing packet SKB (usually tcp_send_head(sk))
1805 1806 1807
 * should be put on the wire right now.  If so, it returns the number of
 * packets allowed by the congestion window.
 */
1808
static unsigned int tcp_snd_test(const struct sock *sk, struct sk_buff *skb,
1809 1810
				 unsigned int cur_mss, int nonagle)
{
1811
	const struct tcp_sock *tp = tcp_sk(sk);
1812 1813
	unsigned int cwnd_quota;

1814
	tcp_init_tso_segs(skb, cur_mss);
1815 1816 1817 1818 1819

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

	cwnd_quota = tcp_cwnd_test(tp, skb);
1820
	if (cwnd_quota && !tcp_snd_wnd_test(tp, skb, cur_mss))
1821 1822 1823 1824 1825
		cwnd_quota = 0;

	return cwnd_quota;
}

1826
/* Test if sending is allowed right now. */
E
Eric Dumazet 已提交
1827
bool tcp_may_send_now(struct sock *sk)
1828
{
1829
	const struct tcp_sock *tp = tcp_sk(sk);
1830
	struct sk_buff *skb = tcp_send_head(sk);
1831

E
Eric Dumazet 已提交
1832
	return skb &&
I
Ilpo Järvinen 已提交
1833
		tcp_snd_test(sk, skb, tcp_current_mss(sk),
1834
			     (tcp_skb_is_last(sk, skb) ?
E
Eric Dumazet 已提交
1835
			      tp->nonagle : TCP_NAGLE_PUSH));
1836 1837 1838 1839 1840 1841 1842 1843 1844
}

/* 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).
 */
1845
static int tso_fragment(struct sock *sk, struct sk_buff *skb, unsigned int len,
1846
			unsigned int mss_now, gfp_t gfp)
1847 1848 1849
{
	struct sk_buff *buff;
	int nlen = skb->len - len;
1850
	u8 flags;
1851 1852

	/* All of a TSO frame must be composed of paged data.  */
1853
	if (skb->len != skb->data_len)
1854
		return tcp_fragment(sk, skb, len, mss_now, gfp);
1855

1856
	buff = sk_stream_alloc_skb(sk, 0, gfp, true);
1857
	if (unlikely(!buff))
1858 1859
		return -ENOMEM;

1860 1861
	sk->sk_wmem_queued += buff->truesize;
	sk_mem_charge(sk, buff->truesize);
1862
	buff->truesize += nlen;
1863 1864 1865 1866 1867 1868 1869 1870
	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
1871 1872 1873
	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;
1874 1875 1876 1877

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

1878 1879
	tcp_skb_fragment_eor(skb, buff);

1880
	buff->ip_summed = skb->ip_summed = CHECKSUM_PARTIAL;
1881
	skb_split(skb, buff, len);
1882
	tcp_fragment_tstamp(skb, buff);
1883 1884

	/* Fix up tso_factor for both original and new SKB.  */
1885 1886
	tcp_set_skb_tso_segs(skb, mss_now);
	tcp_set_skb_tso_segs(buff, mss_now);
1887 1888

	/* Link BUFF into the send queue. */
1889
	__skb_header_release(buff);
1890
	tcp_insert_write_queue_after(skb, buff, sk);
1891 1892 1893 1894 1895 1896 1897 1898 1899

	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.
 */
1900
static bool tcp_tso_should_defer(struct sock *sk, struct sk_buff *skb,
E
Eric Dumazet 已提交
1901
				 bool *is_cwnd_limited, u32 max_segs)
1902
{
1903
	const struct inet_connection_sock *icsk = inet_csk(sk);
1904 1905 1906
	u32 age, send_win, cong_win, limit, in_flight;
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *head;
1907
	int win_divisor;
1908

E
Eric Dumazet 已提交
1909
	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
J
John Heffner 已提交
1910
		goto send_now;
1911

1912
	if (icsk->icsk_ca_state >= TCP_CA_Recovery)
J
John Heffner 已提交
1913 1914
		goto send_now;

E
Eric Dumazet 已提交
1915 1916 1917
	/* Avoid bursty behavior by allowing defer
	 * only if the last write was recent.
	 */
1918
	if ((s32)(tcp_jiffies32 - tp->lsndtime) > 0)
J
John Heffner 已提交
1919
		goto send_now;
1920

1921 1922
	in_flight = tcp_packets_in_flight(tp);

1923
	BUG_ON(tcp_skb_pcount(skb) <= 1 || (tp->snd_cwnd <= in_flight));
1924

1925
	send_win = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
1926 1927 1928 1929 1930 1931

	/* 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);

1932
	/* If a full-sized TSO skb can be sent, do it. */
E
Eric Dumazet 已提交
1933
	if (limit >= max_segs * tp->mss_cache)
J
John Heffner 已提交
1934
		goto send_now;
1935

1936 1937 1938 1939
	/* 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;

1940 1941
	win_divisor = ACCESS_ONCE(sysctl_tcp_tso_win_divisor);
	if (win_divisor) {
1942 1943 1944 1945 1946
		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.
		 */
1947
		chunk /= win_divisor;
1948
		if (limit >= chunk)
J
John Heffner 已提交
1949
			goto send_now;
1950 1951 1952 1953 1954 1955
	} 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.
		 */
1956
		if (limit > tcp_max_tso_deferred_mss(tp) * tp->mss_cache)
J
John Heffner 已提交
1957
			goto send_now;
1958 1959
	}

1960
	head = tcp_write_queue_head(sk);
1961

1962
	age = tcp_stamp_us_delta(tp->tcp_mstamp, head->skb_mstamp);
1963 1964 1965 1966
	/* 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
1967
	/* Ok, it looks like it is advisable to defer. */
J
John Heffner 已提交
1968

1969
	if (cong_win < send_win && cong_win <= skb->len)
1970 1971
		*is_cwnd_limited = true;

E
Eric Dumazet 已提交
1972
	return true;
J
John Heffner 已提交
1973 1974

send_now:
E
Eric Dumazet 已提交
1975
	return false;
1976 1977
}

1978 1979 1980 1981 1982 1983 1984 1985 1986
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;
1987
	delta = tcp_jiffies32 - icsk->icsk_mtup.probe_timestamp;
1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998
	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 */
1999
		icsk->icsk_mtup.probe_timestamp = tcp_jiffies32;
2000 2001 2002
	}
}

J
John Heffner 已提交
2003
/* Create a new MTU probe if we are ready.
2004 2005 2006 2007
 * 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
2008 2009
 * Returns 0 if we should wait to probe (no cwnd available),
 *         1 if a probe was sent,
2010 2011
 *         -1 otherwise
 */
J
John Heffner 已提交
2012 2013 2014
static int tcp_mtu_probe(struct sock *sk)
{
	struct inet_connection_sock *icsk = inet_csk(sk);
2015
	struct tcp_sock *tp = tcp_sk(sk);
J
John Heffner 已提交
2016
	struct sk_buff *skb, *nskb, *next;
2017
	struct net *net = sock_net(sk);
J
John Heffner 已提交
2018
	int probe_size;
2019
	int size_needed;
2020
	int copy, len;
J
John Heffner 已提交
2021
	int mss_now;
2022
	int interval;
J
John Heffner 已提交
2023 2024 2025 2026

	/* Not currently probing/verifying,
	 * not in recovery,
	 * have enough cwnd, and
2027 2028 2029 2030 2031 2032 2033
	 * not SACKing (the variable headers throw things off)
	 */
	if (likely(!icsk->icsk_mtup.enabled ||
		   icsk->icsk_mtup.probe_size ||
		   inet_csk(sk)->icsk_ca_state != TCP_CA_Open ||
		   tp->snd_cwnd < 11 ||
		   tp->rx_opt.num_sacks || tp->rx_opt.dsack))
J
John Heffner 已提交
2034 2035
		return -1;

2036 2037 2038 2039
	/* 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
2040
	mss_now = tcp_current_mss(sk);
2041 2042
	probe_size = tcp_mtu_to_mss(sk, (icsk->icsk_mtup.search_high +
				    icsk->icsk_mtup.search_low) >> 1);
2043
	size_needed = probe_size + (tp->reordering + 1) * tp->mss_cache;
2044
	interval = icsk->icsk_mtup.search_high - icsk->icsk_mtup.search_low;
2045 2046 2047 2048
	/* 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.
	 */
2049
	if (probe_size > tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_high) ||
2050 2051 2052 2053 2054
		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
2055 2056 2057 2058
		return -1;
	}

	/* Have enough data in the send queue to probe? */
2059
	if (tp->write_seq - tp->snd_nxt < size_needed)
J
John Heffner 已提交
2060 2061
		return -1;

2062 2063
	if (tp->snd_wnd < size_needed)
		return -1;
2064
	if (after(tp->snd_nxt + size_needed, tcp_wnd_end(tp)))
2065
		return 0;
J
John Heffner 已提交
2066

2067 2068 2069
	/* 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
2070 2071 2072 2073 2074 2075
			return -1;
		else
			return 0;
	}

	/* We're allowed to probe.  Build it now. */
2076
	nskb = sk_stream_alloc_skb(sk, probe_size, GFP_ATOMIC, false);
2077
	if (!nskb)
J
John Heffner 已提交
2078
		return -1;
2079 2080
	sk->sk_wmem_queued += nskb->truesize;
	sk_mem_charge(sk, nskb->truesize);
J
John Heffner 已提交
2081

2082
	skb = tcp_send_head(sk);
J
John Heffner 已提交
2083 2084 2085

	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
2086
	TCP_SKB_CB(nskb)->tcp_flags = TCPHDR_ACK;
J
John Heffner 已提交
2087 2088
	TCP_SKB_CB(nskb)->sacked = 0;
	nskb->csum = 0;
2089
	nskb->ip_summed = skb->ip_summed;
J
John Heffner 已提交
2090

2091 2092
	tcp_insert_write_queue_before(nskb, skb, sk);

J
John Heffner 已提交
2093
	len = 0;
2094
	tcp_for_write_queue_from_safe(skb, next, sk) {
J
John Heffner 已提交
2095
		copy = min_t(int, skb->len, probe_size - len);
2096
		if (nskb->ip_summed) {
J
John Heffner 已提交
2097
			skb_copy_bits(skb, 0, skb_put(nskb, copy), copy);
2098 2099 2100 2101 2102 2103
		} 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
2104 2105 2106 2107

		if (skb->len <= copy) {
			/* We've eaten all the data from this skb.
			 * Throw it away. */
E
Eric Dumazet 已提交
2108
			TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;
2109
			tcp_unlink_write_queue(skb, sk);
2110
			sk_wmem_free_skb(sk, skb);
J
John Heffner 已提交
2111
		} else {
E
Eric Dumazet 已提交
2112
			TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags &
C
Changli Gao 已提交
2113
						   ~(TCPHDR_FIN|TCPHDR_PSH);
J
John Heffner 已提交
2114 2115
			if (!skb_shinfo(skb)->nr_frags) {
				skb_pull(skb, copy);
2116
				if (skb->ip_summed != CHECKSUM_PARTIAL)
2117 2118
					skb->csum = csum_partial(skb->data,
								 skb->len, 0);
J
John Heffner 已提交
2119 2120
			} else {
				__pskb_trim_head(skb, copy);
2121
				tcp_set_skb_tso_segs(skb, mss_now);
J
John Heffner 已提交
2122 2123 2124 2125 2126
			}
			TCP_SKB_CB(skb)->seq += copy;
		}

		len += copy;
2127 2128 2129

		if (len >= probe_size)
			break;
J
John Heffner 已提交
2130
	}
2131
	tcp_init_tso_segs(nskb, nskb->len);
J
John Heffner 已提交
2132 2133

	/* We're ready to send.  If this fails, the probe will
E
Eric Dumazet 已提交
2134 2135
	 * be resegmented into mss-sized pieces by tcp_write_xmit().
	 */
J
John Heffner 已提交
2136 2137
	if (!tcp_transmit_skb(sk, nskb, 1, GFP_ATOMIC)) {
		/* Decrement cwnd here because we are sending
2138
		 * effectively two packets. */
J
John Heffner 已提交
2139
		tp->snd_cwnd--;
2140
		tcp_event_new_data_sent(sk, nskb);
J
John Heffner 已提交
2141 2142

		icsk->icsk_mtup.probe_size = tcp_mss_to_mtu(sk, nskb->len);
2143 2144
		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
2145 2146 2147 2148 2149 2150 2151

		return 1;
	}

	return -1;
}

2152 2153 2154 2155 2156 2157
static bool tcp_pacing_check(const struct sock *sk)
{
	return tcp_needs_internal_pacing(sk) &&
	       hrtimer_active(&tcp_sk(sk)->pacing_timer);
}

2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177
/* 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;

2178
	if (refcount_read(&sk->sk_wmem_alloc) > limit) {
2179 2180 2181 2182 2183 2184 2185 2186 2187
		/* Always send the 1st or 2nd skb in write queue.
		 * No need to wait for TX completion to call us back,
		 * after softirq/tasklet schedule.
		 * This helps when TX completions are delayed too much.
		 */
		if (skb == sk->sk_write_queue.next ||
		    skb->prev == sk->sk_write_queue.next)
			return false;

2188
		set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags);
2189 2190 2191 2192 2193
		/* It is possible TX completion already happened
		 * before we set TSQ_THROTTLED, so we must
		 * test again the condition.
		 */
		smp_mb__after_atomic();
2194
		if (refcount_read(&sk->sk_wmem_alloc) > limit)
2195 2196 2197 2198 2199
			return true;
	}
	return false;
}

2200 2201
static void tcp_chrono_set(struct tcp_sock *tp, const enum tcp_chrono new)
{
2202
	const u32 now = tcp_jiffies32;
2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214

	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
2215 2216
	 * chronograph then the highest priority enum takes precedence
	 * over the other conditions. So that if something "more interesting"
2217 2218 2219 2220 2221 2222 2223 2224 2225 2226
	 * 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);

2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238

	/* 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);
2239 2240
}

L
Linus Torvalds 已提交
2241 2242 2243 2244
/* This routine writes packets to the network.  It advances the
 * send_head.  This happens as incoming acks open up the remote
 * window for us.
 *
2245 2246 2247 2248
 * 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
2249 2250 2251
 * 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
2252 2253
 * 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
2254
 */
E
Eric Dumazet 已提交
2255 2256
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
2257 2258
{
	struct tcp_sock *tp = tcp_sk(sk);
2259
	struct sk_buff *skb;
2260 2261
	unsigned int tso_segs, sent_pkts;
	int cwnd_quota;
J
John Heffner 已提交
2262
	int result;
2263
	bool is_cwnd_limited = false, is_rwnd_limited = false;
E
Eric Dumazet 已提交
2264
	u32 max_segs;
L
Linus Torvalds 已提交
2265

2266
	sent_pkts = 0;
J
John Heffner 已提交
2267

2268 2269 2270 2271
	if (!push_one) {
		/* Do MTU probing. */
		result = tcp_mtu_probe(sk);
		if (!result) {
E
Eric Dumazet 已提交
2272
			return false;
2273 2274 2275
		} else if (result > 0) {
			sent_pkts = 1;
		}
J
John Heffner 已提交
2276 2277
	}

2278
	max_segs = tcp_tso_segs(sk, mss_now);
2279
	tcp_mstamp_refresh(tp);
2280
	while ((skb = tcp_send_head(sk))) {
2281 2282
		unsigned int limit;

2283 2284 2285
		if (tcp_pacing_check(sk))
			break;

2286
		tso_segs = tcp_init_tso_segs(skb, mss_now);
2287
		BUG_ON(!tso_segs);
2288

2289
		if (unlikely(tp->repair) && tp->repair_queue == TCP_SEND_QUEUE) {
E
Eric Dumazet 已提交
2290
			/* "skb_mstamp" is used as a start point for the retransmit timer */
2291
			skb->skb_mstamp = tp->tcp_mstamp;
2292
			goto repair; /* Skip network transmission */
2293
		}
2294

2295
		cwnd_quota = tcp_cwnd_test(tp, skb);
N
Nandita Dukkipati 已提交
2296 2297 2298 2299 2300 2301 2302
		if (!cwnd_quota) {
			if (push_one == 2)
				/* Force out a loss probe pkt. */
				cwnd_quota = 1;
			else
				break;
		}
2303

2304 2305
		if (unlikely(!tcp_snd_wnd_test(tp, skb, mss_now))) {
			is_rwnd_limited = true;
2306
			break;
2307
		}
2308

2309
		if (tso_segs == 1) {
2310 2311 2312 2313 2314
			if (unlikely(!tcp_nagle_test(tp, skb, mss_now,
						     (tcp_skb_is_last(sk, skb) ?
						      nonagle : TCP_NAGLE_PUSH))))
				break;
		} else {
2315
			if (!push_one &&
E
Eric Dumazet 已提交
2316 2317
			    tcp_tso_should_defer(sk, skb, &is_cwnd_limited,
						 max_segs))
2318 2319
				break;
		}
2320

E
Eric Dumazet 已提交
2321
		limit = mss_now;
2322
		if (tso_segs > 1 && !tcp_urg_mode(tp))
E
Eric Dumazet 已提交
2323 2324 2325 2326 2327 2328 2329 2330 2331 2332
			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;

2333 2334
		if (test_bit(TCP_TSQ_DEFERRED, &sk->sk_tsq_flags))
			clear_bit(TCP_TSQ_DEFERRED, &sk->sk_tsq_flags);
2335 2336
		if (tcp_small_queue_check(sk, skb, 0))
			break;
E
Eric Dumazet 已提交
2337

2338
		if (unlikely(tcp_transmit_skb(sk, skb, 1, gfp)))
2339
			break;
L
Linus Torvalds 已提交
2340

2341
repair:
2342 2343 2344
		/* Advance the send_head.  This one is sent out.
		 * This call will increment packets_out.
		 */
2345
		tcp_event_new_data_sent(sk, skb);
L
Linus Torvalds 已提交
2346

2347
		tcp_minshall_update(tp, mss_now, skb);
2348
		sent_pkts += tcp_skb_pcount(skb);
2349 2350 2351

		if (push_one)
			break;
2352
	}
L
Linus Torvalds 已提交
2353

2354 2355 2356 2357 2358
	if (is_rwnd_limited)
		tcp_chrono_start(sk, TCP_CHRONO_RWND_LIMITED);
	else
		tcp_chrono_stop(sk, TCP_CHRONO_RWND_LIMITED);

2359
	if (likely(sent_pkts)) {
2360 2361
		if (tcp_in_cwnd_reduction(sk))
			tp->prr_out += sent_pkts;
N
Nandita Dukkipati 已提交
2362 2363 2364 2365

		/* Send one loss probe per tail loss episode. */
		if (push_one != 2)
			tcp_schedule_loss_probe(sk);
2366
		is_cwnd_limited |= (tcp_packets_in_flight(tp) >= tp->snd_cwnd);
2367
		tcp_cwnd_validate(sk, is_cwnd_limited);
E
Eric Dumazet 已提交
2368
		return false;
L
Linus Torvalds 已提交
2369
	}
2370
	return !tp->packets_out && tcp_send_head(sk);
N
Nandita Dukkipati 已提交
2371 2372 2373 2374 2375 2376 2377
}

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;
2378
	u32 rtt = usecs_to_jiffies(tp->srtt_us >> 3);
N
Nandita Dukkipati 已提交
2379 2380 2381 2382 2383 2384 2385 2386 2387

	/* 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.
	 */
2388
	if (tp->fastopen_rsk)
N
Nandita Dukkipati 已提交
2389 2390 2391 2392 2393 2394 2395 2396 2397
		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.
	 */
Y
Yuchung Cheng 已提交
2398 2399 2400
	if ((sysctl_tcp_early_retrans != 3 && sysctl_tcp_early_retrans != 4) ||
	    !tp->packets_out || !tcp_is_sack(tp) ||
	    icsk->icsk_ca_state != TCP_CA_Open)
N
Nandita Dukkipati 已提交
2401 2402 2403 2404 2405 2406 2407
		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
2408 2409
	 * 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
2410
	 */
2411
	timeout = rtt << 1 ? : TCP_TIMEOUT_INIT;
N
Nandita Dukkipati 已提交
2412 2413 2414 2415 2416 2417
	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. */
2418
	tlp_time_stamp = tcp_jiffies32 + timeout;
N
Nandita Dukkipati 已提交
2419 2420
	rto_time_stamp = (u32)inet_csk(sk)->icsk_timeout;
	if ((s32)(tlp_time_stamp - rto_time_stamp) > 0) {
2421
		s32 delta = rto_time_stamp - tcp_jiffies32;
N
Nandita Dukkipati 已提交
2422 2423 2424 2425 2426 2427 2428 2429 2430
		if (delta > 0)
			timeout = delta;
	}

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

2431 2432 2433 2434 2435 2436 2437
/* 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)
{
2438
	if (unlikely(skb_fclone_busy(sk, skb))) {
2439 2440
		NET_INC_STATS(sock_net(sk),
			      LINUX_MIB_TCPSPURIOUS_RTX_HOSTQUEUES);
2441 2442 2443 2444 2445
		return true;
	}
	return false;
}

2446
/* When probe timeout (PTO) fires, try send a new segment if possible, else
N
Nandita Dukkipati 已提交
2447 2448 2449 2450
 * retransmit the last segment.
 */
void tcp_send_loss_probe(struct sock *sk)
{
N
Nandita Dukkipati 已提交
2451
	struct tcp_sock *tp = tcp_sk(sk);
N
Nandita Dukkipati 已提交
2452 2453 2454 2455
	struct sk_buff *skb;
	int pcount;
	int mss = tcp_current_mss(sk);

2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467
	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
2468 2469
	}

N
Nandita Dukkipati 已提交
2470 2471 2472 2473
	/* At most one outstanding TLP retransmission. */
	if (tp->tlp_high_seq)
		goto rearm_timer;

N
Nandita Dukkipati 已提交
2474 2475 2476 2477
	/* Retransmit last segment. */
	if (WARN_ON(!skb))
		goto rearm_timer;

2478 2479 2480
	if (skb_still_in_host_queue(sk, skb))
		goto rearm_timer;

N
Nandita Dukkipati 已提交
2481 2482 2483 2484 2485
	pcount = tcp_skb_pcount(skb);
	if (WARN_ON(!pcount))
		goto rearm_timer;

	if ((pcount > 1) && (skb->len > (pcount - 1) * mss)) {
2486 2487
		if (unlikely(tcp_fragment(sk, skb, (pcount - 1) * mss, mss,
					  GFP_ATOMIC)))
N
Nandita Dukkipati 已提交
2488
			goto rearm_timer;
2489
		skb = tcp_write_queue_next(sk, skb);
N
Nandita Dukkipati 已提交
2490 2491 2492 2493 2494
	}

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

2495
	if (__tcp_retransmit_skb(sk, skb, 1))
2496
		goto rearm_timer;
N
Nandita Dukkipati 已提交
2497

N
Nandita Dukkipati 已提交
2498
	/* Record snd_nxt for loss detection. */
2499
	tp->tlp_high_seq = tp->snd_nxt;
N
Nandita Dukkipati 已提交
2500

2501
probe_sent:
2502
	NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPLOSSPROBES);
2503 2504
	/* Reset s.t. tcp_rearm_rto will restart timer from now */
	inet_csk(sk)->icsk_pending = 0;
N
Nandita Dukkipati 已提交
2505
rearm_timer:
2506
	tcp_rearm_rto(sk);
L
Linus Torvalds 已提交
2507 2508
}

2509 2510 2511 2512
/* 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.
 */
2513 2514
void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
			       int nonagle)
2515
{
2516 2517 2518 2519 2520 2521 2522
	/* 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;

2523
	if (tcp_write_xmit(sk, cur_mss, nonagle, 0,
2524
			   sk_gfp_mask(sk, GFP_ATOMIC)))
2525
		tcp_check_probe_timer(sk);
2526 2527
}

2528 2529 2530 2531 2532
/* 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)
{
2533
	struct sk_buff *skb = tcp_send_head(sk);
2534 2535 2536

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

2537
	tcp_write_xmit(sk, mss_now, TCP_NAGLE_PUSH, 1, sk->sk_allocation);
2538 2539
}

L
Linus Torvalds 已提交
2540 2541
/* This function returns the amount that we can raise the
 * usable window based on the following constraints
2542
 *
L
Linus Torvalds 已提交
2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560
 * 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.
2561
 *
L
Linus Torvalds 已提交
2562 2563
 * Of course, if the sender implements sender side SWS prevention
 * then this will not be a problem.
2564
 *
L
Linus Torvalds 已提交
2565
 * BSD seems to make the following compromise:
2566
 *
L
Linus Torvalds 已提交
2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593
 *	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)
{
2594
	struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
2595
	struct tcp_sock *tp = tcp_sk(sk);
S
Stephen Hemminger 已提交
2596
	/* MSS for the peer's data.  Previous versions used mss_clamp
L
Linus Torvalds 已提交
2597 2598 2599 2600 2601
	 * 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
	 */
2602
	int mss = icsk->icsk_ack.rcv_mss;
L
Linus Torvalds 已提交
2603
	int free_space = tcp_space(sk);
2604 2605
	int allowed_space = tcp_full_space(sk);
	int full_space = min_t(int, tp->window_clamp, allowed_space);
L
Linus Torvalds 已提交
2606 2607
	int window;

2608
	if (unlikely(mss > full_space)) {
2609
		mss = full_space;
2610 2611 2612
		if (mss <= 0)
			return 0;
	}
2613
	if (free_space < (full_space >> 1)) {
2614
		icsk->icsk_ack.quick = 0;
L
Linus Torvalds 已提交
2615

2616
		if (tcp_under_memory_pressure(sk))
2617 2618
			tp->rcv_ssthresh = min(tp->rcv_ssthresh,
					       4U * tp->advmss);
L
Linus Torvalds 已提交
2619

2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632
		/* 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
2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648
			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.
	 */
	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.
		 */
2649
		window = ALIGN(window, (1 << tp->rx_opt.rcv_wscale));
L
Linus Torvalds 已提交
2650
	} else {
2651
		window = tp->rcv_wnd;
L
Linus Torvalds 已提交
2652 2653 2654 2655 2656 2657 2658 2659 2660
		/* 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)
2661
			window = rounddown(free_space, mss);
2662
		else if (mss == full_space &&
2663
			 free_space > window + (full_space >> 1))
2664
			window = free_space;
L
Linus Torvalds 已提交
2665 2666 2667 2668 2669
	}

	return window;
}

2670 2671
void tcp_skb_collapse_tstamp(struct sk_buff *skb,
			     const struct sk_buff *next_skb)
2672
{
2673 2674 2675
	if (unlikely(tcp_has_tx_tstamp(next_skb))) {
		const struct skb_shared_info *next_shinfo =
			skb_shinfo(next_skb);
2676 2677
		struct skb_shared_info *shinfo = skb_shinfo(skb);

2678
		shinfo->tx_flags |= next_shinfo->tx_flags & SKBTX_ANY_TSTAMP;
2679
		shinfo->tskey = next_shinfo->tskey;
2680 2681
		TCP_SKB_CB(skb)->txstamp_ack |=
			TCP_SKB_CB(next_skb)->txstamp_ack;
2682 2683 2684
	}
}

2685
/* Collapses two adjacent SKB's during retransmission. */
2686
static bool tcp_collapse_retrans(struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
2687 2688
{
	struct tcp_sock *tp = tcp_sk(sk);
2689
	struct sk_buff *next_skb = tcp_write_queue_next(sk, skb);
2690
	int skb_size, next_skb_size;
L
Linus Torvalds 已提交
2691

2692 2693
	skb_size = skb->len;
	next_skb_size = next_skb->len;
L
Linus Torvalds 已提交
2694

2695
	BUG_ON(tcp_skb_pcount(skb) != 1 || tcp_skb_pcount(next_skb) != 1);
2696

2697 2698 2699 2700 2701 2702 2703
	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;
	}
2704
	tcp_highest_sack_combine(sk, next_skb, skb);
L
Linus Torvalds 已提交
2705

2706
	tcp_unlink_write_queue(next_skb, sk);
L
Linus Torvalds 已提交
2707

2708 2709
	if (next_skb->ip_summed == CHECKSUM_PARTIAL)
		skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
2710

2711 2712
	if (skb->ip_summed != CHECKSUM_PARTIAL)
		skb->csum = csum_block_add(skb->csum, next_skb->csum, skb_size);
L
Linus Torvalds 已提交
2713

2714 2715
	/* 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
2716

2717
	/* Merge over control information. This moves PSH/FIN etc. over */
E
Eric Dumazet 已提交
2718
	TCP_SKB_CB(skb)->tcp_flags |= TCP_SKB_CB(next_skb)->tcp_flags;
2719 2720 2721 2722 2723

	/* 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;
2724
	TCP_SKB_CB(skb)->eor = TCP_SKB_CB(next_skb)->eor;
2725 2726

	/* changed transmit queue under us so clear hints */
2727 2728 2729
	tcp_clear_retrans_hints_partial(tp);
	if (next_skb == tp->retransmit_skb_hint)
		tp->retransmit_skb_hint = skb;
2730

2731 2732
	tcp_adjust_pcount(sk, next_skb, tcp_skb_pcount(next_skb));

2733 2734
	tcp_skb_collapse_tstamp(skb, next_skb);

2735
	sk_wmem_free_skb(sk, next_skb);
2736
	return true;
L
Linus Torvalds 已提交
2737 2738
}

2739
/* Check if coalescing SKBs is legal. */
E
Eric Dumazet 已提交
2740
static bool tcp_can_collapse(const struct sock *sk, const struct sk_buff *skb)
2741 2742
{
	if (tcp_skb_pcount(skb) > 1)
E
Eric Dumazet 已提交
2743
		return false;
2744
	if (skb_cloned(skb))
E
Eric Dumazet 已提交
2745
		return false;
2746
	if (skb == tcp_send_head(sk))
E
Eric Dumazet 已提交
2747
		return false;
E
Eric Dumazet 已提交
2748
	/* Some heuristics for collapsing over SACK'd could be invented */
2749
	if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)
E
Eric Dumazet 已提交
2750
		return false;
2751

E
Eric Dumazet 已提交
2752
	return true;
2753 2754
}

2755 2756 2757
/* Collapse packets in the retransmit queue to make to create
 * less packets on the wire. This is only done on retransmission.
 */
2758 2759 2760 2761 2762
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
2763
	bool first = true;
2764 2765 2766

	if (!sysctl_tcp_retrans_collapse)
		return;
E
Eric Dumazet 已提交
2767
	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)
2768 2769 2770 2771 2772 2773
		return;

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

2774 2775 2776
		if (!tcp_skb_can_collapse_to(to))
			break;

2777 2778 2779
		space -= skb->len;

		if (first) {
E
Eric Dumazet 已提交
2780
			first = false;
2781 2782 2783 2784 2785 2786 2787 2788 2789
			continue;
		}

		if (space < 0)
			break;

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

2790 2791
		if (!tcp_collapse_retrans(sk, to))
			break;
2792 2793 2794
	}
}

L
Linus Torvalds 已提交
2795 2796 2797 2798
/* 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.
 */
2799
int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs)
L
Linus Torvalds 已提交
2800
{
J
John Heffner 已提交
2801
	struct inet_connection_sock *icsk = inet_csk(sk);
2802
	struct tcp_sock *tp = tcp_sk(sk);
2803
	unsigned int cur_mss;
2804 2805
	int diff, len, err;

L
Linus Torvalds 已提交
2806

2807 2808
	/* Inconclusive MTU probe */
	if (icsk->icsk_mtup.probe_size)
J
John Heffner 已提交
2809 2810
		icsk->icsk_mtup.probe_size = 0;

L
Linus Torvalds 已提交
2811
	/* Do not sent more than we queued. 1/4 is reserved for possible
S
Stephen Hemminger 已提交
2812
	 * copying overhead: fragmentation, tunneling, mangling etc.
L
Linus Torvalds 已提交
2813
	 */
2814
	if (refcount_read(&sk->sk_wmem_alloc) >
2815 2816
	    min_t(u32, sk->sk_wmem_queued + (sk->sk_wmem_queued >> 2),
		  sk->sk_sndbuf))
L
Linus Torvalds 已提交
2817 2818
		return -EAGAIN;

2819 2820 2821
	if (skb_still_in_host_queue(sk, skb))
		return -EBUSY;

L
Linus Torvalds 已提交
2822 2823 2824 2825 2826 2827 2828
	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;
	}

2829 2830 2831
	if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk))
		return -EHOSTUNREACH; /* Routing failure or similar. */

I
Ilpo Järvinen 已提交
2832
	cur_mss = tcp_current_mss(sk);
2833

L
Linus Torvalds 已提交
2834 2835 2836 2837 2838
	/* 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.
	 */
2839 2840
	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
2841 2842
		return -EAGAIN;

2843 2844 2845
	len = cur_mss * segs;
	if (skb->len > len) {
		if (tcp_fragment(sk, skb, len, cur_mss, GFP_ATOMIC))
L
Linus Torvalds 已提交
2846
			return -ENOMEM; /* We'll try again later. */
2847
	} else {
2848 2849
		if (skb_unclone(skb, GFP_ATOMIC))
			return -ENOMEM;
2850

2851 2852 2853 2854 2855 2856 2857
		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
2858 2859
	}

2860 2861 2862 2863
	/* 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);

2864 2865 2866 2867 2868 2869 2870
	/* Update global and local TCP statistics. */
	segs = tcp_skb_pcount(skb);
	TCP_ADD_STATS(sock_net(sk), TCP_MIB_RETRANSSEGS, segs);
	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)
		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPSYNRETRANS);
	tp->total_retrans += segs;

2871 2872 2873 2874 2875 2876
	/* 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)) {
2877 2878
		struct sk_buff *nskb;

2879
		skb->skb_mstamp = tp->tcp_mstamp;
2880
		nskb = __pskb_copy(skb, MAX_TCP_HEADER, GFP_ATOMIC);
2881 2882
		err = nskb ? tcp_transmit_skb(sk, nskb, 0, GFP_ATOMIC) :
			     -ENOBUFS;
E
Eric Dumazet 已提交
2883
	} else {
2884
		err = tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
E
Eric Dumazet 已提交
2885
	}
2886

2887
	if (likely(!err)) {
2888
		TCP_SKB_CB(skb)->sacked |= TCPCB_EVER_RETRANS;
2889 2890
	} else if (err != -EBUSY) {
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPRETRANSFAIL);
2891
	}
2892
	return err;
2893 2894
}

2895
int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs)
2896 2897
{
	struct tcp_sock *tp = tcp_sk(sk);
2898
	int err = __tcp_retransmit_skb(sk, skb, segs);
L
Linus Torvalds 已提交
2899 2900 2901

	if (err == 0) {
#if FASTRETRANS_DEBUG > 0
2902
		if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) {
2903
			net_dbg_ratelimited("retrans_out leaked\n");
L
Linus Torvalds 已提交
2904 2905 2906 2907 2908 2909 2910
		}
#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
2911
			tp->retrans_stamp = tcp_skb_timestamp(skb);
L
Linus Torvalds 已提交
2912 2913

	}
Y
Yuchung Cheng 已提交
2914 2915 2916 2917

	if (tp->undo_retrans < 0)
		tp->undo_retrans = 0;
	tp->undo_retrans += tcp_skb_pcount(skb);
L
Linus Torvalds 已提交
2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930
	return err;
}

/* 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)
{
2931
	const struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
2932 2933
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb;
2934
	struct sk_buff *hole = NULL;
2935
	u32 max_segs;
2936
	int mib_idx;
2937

2938 2939 2940
	if (!tp->packets_out)
		return;

2941
	if (tp->retransmit_skb_hint) {
2942
		skb = tp->retransmit_skb_hint;
2943
	} else {
2944
		skb = tcp_write_queue_head(sk);
2945
	}
L
Linus Torvalds 已提交
2946

2947
	max_segs = tcp_tso_segs(sk, tcp_current_mss(sk));
2948
	tcp_for_write_queue_from(skb, sk) {
E
Eric Dumazet 已提交
2949
		__u8 sacked;
2950
		int segs;
L
Linus Torvalds 已提交
2951

2952 2953
		if (skb == tcp_send_head(sk))
			break;
2954 2955 2956 2957

		if (tcp_pacing_check(sk))
			break;

2958
		/* we could do better than to assign each time */
2959
		if (!hole)
2960
			tp->retransmit_skb_hint = skb;
2961

2962 2963
		segs = tp->snd_cwnd - tcp_packets_in_flight(tp);
		if (segs <= 0)
2964
			return;
E
Eric Dumazet 已提交
2965
		sacked = TCP_SKB_CB(skb)->sacked;
2966 2967 2968 2969
		/* 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
2970

2971 2972
		if (tp->retrans_out >= tp->lost_out) {
			break;
2973
		} else if (!(sacked & TCPCB_LOST)) {
2974
			if (!hole && !(sacked & (TCPCB_SACKED_RETRANS|TCPCB_SACKED_ACKED)))
2975 2976
				hole = skb;
			continue;
L
Linus Torvalds 已提交
2977

2978 2979 2980 2981 2982 2983
		} else {
			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
2984

2985
		if (sacked & (TCPCB_SACKED_ACKED|TCPCB_SACKED_RETRANS))
L
Linus Torvalds 已提交
2986 2987
			continue;

2988 2989 2990
		if (tcp_small_queue_check(sk, skb, 1))
			return;

2991
		if (tcp_retransmit_skb(sk, skb, segs))
2992
			return;
2993

2994
		NET_ADD_STATS(sock_net(sk), mib_idx, tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
2995

2996
		if (tcp_in_cwnd_reduction(sk))
2997 2998
			tp->prr_out += tcp_skb_pcount(skb);

2999 3000
		if (skb == tcp_write_queue_head(sk) &&
		    icsk->icsk_pending != ICSK_TIME_REO_TIMEOUT)
3001 3002 3003
			inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
						  inet_csk(sk)->icsk_rto,
						  TCP_RTO_MAX);
L
Linus Torvalds 已提交
3004 3005 3006
	}
}

3007 3008
/* We allow to exceed memory limits for FIN packets to expedite
 * connection tear down and (memory) recovery.
3009 3010
 * Otherwise tcp_send_fin() could be tempted to either delay FIN
 * or even be forced to close flow without any FIN.
3011 3012
 * In general, we want to allow one skb per socket to avoid hangs
 * with edge trigger epoll()
3013
 */
3014
void sk_forced_mem_schedule(struct sock *sk, int size)
3015
{
3016
	int amt;
3017 3018 3019 3020 3021

	if (size <= sk->sk_forward_alloc)
		return;
	amt = sk_mem_pages(size);
	sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
3022 3023
	sk_memory_allocated_add(sk, amt);

3024 3025
	if (mem_cgroup_sockets_enabled && sk->sk_memcg)
		mem_cgroup_charge_skmem(sk->sk_memcg, amt);
3026 3027
}

3028 3029
/* 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
3030 3031 3032
 */
void tcp_send_fin(struct sock *sk)
{
3033
	struct sk_buff *skb, *tskb = tcp_write_queue_tail(sk);
3034 3035
	struct tcp_sock *tp = tcp_sk(sk);

3036 3037 3038 3039
	/* 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
3040
	 */
3041
	if (tskb && (tcp_send_head(sk) || tcp_under_memory_pressure(sk))) {
3042 3043 3044
coalesce:
		TCP_SKB_CB(tskb)->tcp_flags |= TCPHDR_FIN;
		TCP_SKB_CB(tskb)->end_seq++;
L
Linus Torvalds 已提交
3045
		tp->write_seq++;
3046 3047 3048 3049 3050 3051 3052 3053 3054 3055
		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
3056
	} else {
3057 3058 3059 3060 3061
		skb = alloc_skb_fclone(MAX_TCP_HEADER, sk->sk_allocation);
		if (unlikely(!skb)) {
			if (tskb)
				goto coalesce;
			return;
L
Linus Torvalds 已提交
3062
		}
3063
		skb_reserve(skb, MAX_TCP_HEADER);
3064
		sk_forced_mem_schedule(sk, skb->truesize);
L
Linus Torvalds 已提交
3065
		/* FIN eats a sequence byte, write_seq advanced by tcp_queue_skb(). */
3066
		tcp_init_nondata_skb(skb, tp->write_seq,
C
Changli Gao 已提交
3067
				     TCPHDR_ACK | TCPHDR_FIN);
L
Linus Torvalds 已提交
3068 3069
		tcp_queue_skb(sk, skb);
	}
3070
	__tcp_push_pending_frames(sk, tcp_current_mss(sk), TCP_NAGLE_OFF);
L
Linus Torvalds 已提交
3071 3072 3073 3074 3075
}

/* 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
3076
 * by RFC 2525, section 2.17.  -DaveM
L
Linus Torvalds 已提交
3077
 */
A
Al Viro 已提交
3078
void tcp_send_active_reset(struct sock *sk, gfp_t priority)
L
Linus Torvalds 已提交
3079 3080 3081
{
	struct sk_buff *skb;

3082 3083
	TCP_INC_STATS(sock_net(sk), TCP_MIB_OUTRSTS);

L
Linus Torvalds 已提交
3084 3085 3086
	/* 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
3087
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTFAILED);
L
Linus Torvalds 已提交
3088 3089 3090 3091 3092
		return;
	}

	/* Reserve space for headers and prepare control bits. */
	skb_reserve(skb, MAX_TCP_HEADER);
3093
	tcp_init_nondata_skb(skb, tcp_acceptable_seq(sk),
C
Changli Gao 已提交
3094
			     TCPHDR_ACK | TCPHDR_RST);
3095
	tcp_mstamp_refresh(tcp_sk(sk));
L
Linus Torvalds 已提交
3096
	/* Send it off. */
3097
	if (tcp_transmit_skb(sk, skb, 0, priority))
P
Pavel Emelyanov 已提交
3098
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTFAILED);
L
Linus Torvalds 已提交
3099 3100
}

3101 3102
/* 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
3103 3104 3105 3106 3107 3108
 * 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)
{
3109
	struct sk_buff *skb;
L
Linus Torvalds 已提交
3110

3111
	skb = tcp_write_queue_head(sk);
3112
	if (!skb || !(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
3113
		pr_debug("%s: wrong queue state\n", __func__);
L
Linus Torvalds 已提交
3114 3115
		return -EFAULT;
	}
E
Eric Dumazet 已提交
3116
	if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)) {
L
Linus Torvalds 已提交
3117 3118
		if (skb_cloned(skb)) {
			struct sk_buff *nskb = skb_copy(skb, GFP_ATOMIC);
3119
			if (!nskb)
L
Linus Torvalds 已提交
3120
				return -ENOMEM;
3121
			tcp_unlink_write_queue(skb, sk);
3122
			__skb_header_release(nskb);
3123
			__tcp_add_write_queue_head(sk, nskb);
3124 3125 3126
			sk_wmem_free_skb(sk, skb);
			sk->sk_wmem_queued += nskb->truesize;
			sk_mem_charge(sk, nskb->truesize);
L
Linus Torvalds 已提交
3127 3128 3129
			skb = nskb;
		}

E
Eric Dumazet 已提交
3130
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ACK;
3131
		tcp_ecn_send_synack(sk, skb);
L
Linus Torvalds 已提交
3132
	}
3133
	return tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
L
Linus Torvalds 已提交
3134 3135
}

3136 3137 3138 3139 3140 3141 3142 3143 3144
/**
 * 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.
 */
3145
struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
3146
				struct request_sock *req,
3147
				struct tcp_fastopen_cookie *foc,
3148
				enum tcp_synack_type synack_type)
L
Linus Torvalds 已提交
3149
{
3150
	struct inet_request_sock *ireq = inet_rsk(req);
3151
	const struct tcp_sock *tp = tcp_sk(sk);
E
Eric Dumazet 已提交
3152
	struct tcp_md5sig_key *md5 = NULL;
3153 3154
	struct tcp_out_options opts;
	struct sk_buff *skb;
3155
	int tcp_header_size;
3156
	struct tcphdr *th;
3157
	int mss;
L
Linus Torvalds 已提交
3158

3159
	skb = alloc_skb(MAX_TCP_HEADER, GFP_ATOMIC);
3160 3161
	if (unlikely(!skb)) {
		dst_release(dst);
L
Linus Torvalds 已提交
3162
		return NULL;
3163
	}
L
Linus Torvalds 已提交
3164 3165 3166
	/* Reserve space for headers. */
	skb_reserve(skb, MAX_TCP_HEADER);

3167 3168
	switch (synack_type) {
	case TCP_SYNACK_NORMAL:
3169
		skb_set_owner_w(skb, req_to_sk(req));
3170 3171 3172 3173 3174 3175 3176
		break;
	case TCP_SYNACK_COOKIE:
		/* Under synflood, we do not attach skb to a socket,
		 * to avoid false sharing.
		 */
		break;
	case TCP_SYNACK_FASTOPEN:
3177 3178 3179 3180 3181
		/* 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);
3182
		break;
3183
	}
3184
	skb_dst_set(skb, dst);
L
Linus Torvalds 已提交
3185

E
Eric Dumazet 已提交
3186
	mss = tcp_mss_clamp(tp, dst_metric_advmss(dst));
3187

A
Adam Langley 已提交
3188
	memset(&opts, 0, sizeof(opts));
3189 3190
#ifdef CONFIG_SYN_COOKIES
	if (unlikely(req->cookie_ts))
3191
		skb->skb_mstamp = cookie_init_timestamp(req);
3192 3193
	else
#endif
3194
		skb->skb_mstamp = tcp_clock_us();
E
Eric Dumazet 已提交
3195 3196 3197

#ifdef CONFIG_TCP_MD5SIG
	rcu_read_lock();
3198
	md5 = tcp_rsk(req)->af_specific->req_md5_lookup(sk, req_to_sk(req));
E
Eric Dumazet 已提交
3199
#endif
3200
	skb_set_hash(skb, tcp_rsk(req)->txhash, PKT_HASH_TYPE_L4);
3201 3202
	tcp_header_size = tcp_synack_options(req, mss, skb, &opts, md5, foc) +
			  sizeof(*th);
3203

3204 3205
	skb_push(skb, tcp_header_size);
	skb_reset_transport_header(skb);
L
Linus Torvalds 已提交
3206

3207
	th = (struct tcphdr *)skb->data;
L
Linus Torvalds 已提交
3208 3209 3210
	memset(th, 0, sizeof(struct tcphdr));
	th->syn = 1;
	th->ack = 1;
3211
	tcp_ecn_make_synack(req, th);
E
Eric Dumazet 已提交
3212
	th->source = htons(ireq->ir_num);
3213
	th->dest = ireq->ir_rmt_port;
3214 3215 3216 3217
	/* 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
3218
			     TCPHDR_SYN | TCPHDR_ACK);
3219

L
Linus Torvalds 已提交
3220
	th->seq = htonl(TCP_SKB_CB(skb)->seq);
3221 3222
	/* 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
3223 3224

	/* RFC1323: The window in SYN & SYN/ACK segments is never scaled. */
3225
	th->window = htons(min(req->rsk_rcv_wnd, 65535U));
3226
	tcp_options_write((__be32 *)(th + 1), NULL, &opts);
L
Linus Torvalds 已提交
3227
	th->doff = (tcp_header_size >> 2);
E
Eric Dumazet 已提交
3228
	__TCP_INC_STATS(sock_net(sk), TCP_MIB_OUTSEGS);
3229 3230 3231

#ifdef CONFIG_TCP_MD5SIG
	/* Okay, we have all we need - do the md5 hash if needed */
E
Eric Dumazet 已提交
3232
	if (md5)
3233
		tcp_rsk(req)->af_specific->calc_md5_hash(opts.hash_location,
3234
					       md5, req_to_sk(req), skb);
E
Eric Dumazet 已提交
3235
	rcu_read_unlock();
3236 3237
#endif

3238
	/* Do not fool tcpdump (if any), clean our debris */
T
Thomas Gleixner 已提交
3239
	skb->tstamp = 0;
L
Linus Torvalds 已提交
3240 3241
	return skb;
}
E
Eric Dumazet 已提交
3242
EXPORT_SYMBOL(tcp_make_synack);
L
Linus Torvalds 已提交
3243

3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262
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();
}

3263
/* Do all connect socket setups that can be done AF independent. */
3264
static void tcp_connect_init(struct sock *sk)
L
Linus Torvalds 已提交
3265
{
3266
	const struct dst_entry *dst = __sk_dst_get(sk);
L
Linus Torvalds 已提交
3267 3268
	struct tcp_sock *tp = tcp_sk(sk);
	__u8 rcv_wscale;
3269
	u32 rcv_wnd;
L
Linus Torvalds 已提交
3270 3271 3272 3273

	/* 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.
	 */
3274 3275 3276
	tp->tcp_header_len = sizeof(struct tcphdr);
	if (sock_net(sk)->ipv4.sysctl_tcp_timestamps)
		tp->tcp_header_len += TCPOLEN_TSTAMP_ALIGNED;
L
Linus Torvalds 已提交
3277

3278
#ifdef CONFIG_TCP_MD5SIG
3279
	if (tp->af_specific->md5_lookup(sk, sk))
3280 3281 3282
		tp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
#endif

L
Linus Torvalds 已提交
3283 3284 3285 3286
	/* 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
3287
	tcp_mtup_init(sk);
L
Linus Torvalds 已提交
3288 3289
	tcp_sync_mss(sk, dst_mtu(dst));

3290 3291
	tcp_ca_dst_init(sk, dst);

L
Linus Torvalds 已提交
3292 3293
	if (!tp->window_clamp)
		tp->window_clamp = dst_metric(dst, RTAX_WINDOW);
E
Eric Dumazet 已提交
3294
	tp->advmss = tcp_mss_clamp(tp, dst_metric_advmss(dst));
3295

L
Linus Torvalds 已提交
3296 3297
	tcp_initialize_rcv_mss(sk);

3298 3299 3300 3301 3302
	/* 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);

3303 3304 3305 3306
	rcv_wnd = tcp_rwnd_init_bpf(sk);
	if (rcv_wnd == 0)
		rcv_wnd = dst_metric(dst, RTAX_INITRWND);

L
Linus Torvalds 已提交
3307 3308 3309 3310
	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,
3311
				  sock_net(sk)->ipv4.sysctl_tcp_window_scaling,
3312
				  &rcv_wscale,
3313
				  rcv_wnd);
L
Linus Torvalds 已提交
3314 3315 3316 3317 3318 3319 3320

	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;
3321
	tcp_init_wl(tp, 0);
L
Linus Torvalds 已提交
3322 3323
	tp->snd_una = tp->write_seq;
	tp->snd_sml = tp->write_seq;
I
Ilpo Järvinen 已提交
3324
	tp->snd_up = tp->write_seq;
P
Pavel Emelyanov 已提交
3325
	tp->snd_nxt = tp->write_seq;
P
Pavel Emelyanov 已提交
3326 3327 3328

	if (likely(!tp->repair))
		tp->rcv_nxt = 0;
3329
	else
3330
		tp->rcv_tstamp = tcp_jiffies32;
P
Pavel Emelyanov 已提交
3331 3332
	tp->rcv_wup = tp->rcv_nxt;
	tp->copied_seq = tp->rcv_nxt;
L
Linus Torvalds 已提交
3333

3334
	inet_csk(sk)->icsk_rto = tcp_timeout_init(sk);
3335
	inet_csk(sk)->icsk_retransmits = 0;
L
Linus Torvalds 已提交
3336 3337 3338
	tcp_clear_retrans(tp);
}

3339 3340 3341 3342 3343 3344
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;
3345
	__skb_header_release(skb);
3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363
	__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;
3364
	int space, err = 0;
3365
	struct sk_buff *syn_data;
3366

3367
	tp->rx_opt.mss_clamp = tp->advmss;  /* If MSS is not cached */
3368
	if (!tcp_fastopen_cookie_check(sk, &tp->rx_opt.mss_clamp, &fo->cookie))
3369 3370 3371 3372 3373 3374
		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 :(
	 */
E
Eric Dumazet 已提交
3375 3376
	tp->rx_opt.mss_clamp = tcp_mss_clamp(tp, tp->rx_opt.mss_clamp);

3377
	space = __tcp_mtu_to_mss(sk, inet_csk(sk)->icsk_pmtu_cookie) -
3378 3379
		MAX_TCP_OPTION_SPACE;

3380 3381 3382 3383 3384
	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));

3385
	syn_data = sk_stream_alloc_skb(sk, space, sk->sk_allocation, false);
3386
	if (!syn_data)
3387
		goto fallback;
3388 3389
	syn_data->ip_summed = CHECKSUM_PARTIAL;
	memcpy(syn_data->cb, syn->cb, sizeof(syn->cb));
3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400
	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;
		}
3401
	}
3402 3403 3404 3405
	/* No more data pending in inet_wait_for_connect() */
	if (space == fo->size)
		fo->data = NULL;
	fo->copied = space;
3406

3407
	tcp_connect_queue_skb(sk, syn_data);
3408 3409
	if (syn_data->len)
		tcp_chrono_start(sk, TCP_CHRONO_BUSY);
3410

3411
	err = tcp_transmit_skb(sk, syn_data, 1, sk->sk_allocation);
3412

3413
	syn->skb_mstamp = syn_data->skb_mstamp;
E
Eric Dumazet 已提交
3414

3415 3416 3417 3418 3419 3420 3421 3422
	/* 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) {
3423
		tp->syn_data = (fo->copied > 0);
3424
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPORIGDATASENT);
3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439
		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;
}

3440
/* Build a SYN and send it off. */
L
Linus Torvalds 已提交
3441 3442 3443 3444
int tcp_connect(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *buff;
3445
	int err;
L
Linus Torvalds 已提交
3446 3447 3448

	tcp_connect_init(sk);

A
Andrey Vagin 已提交
3449 3450 3451 3452 3453
	if (unlikely(tp->repair)) {
		tcp_finish_connect(sk, NULL);
		return 0;
	}

3454
	buff = sk_stream_alloc_skb(sk, 0, sk->sk_allocation, true);
3455
	if (unlikely(!buff))
L
Linus Torvalds 已提交
3456 3457
		return -ENOBUFS;

C
Changli Gao 已提交
3458
	tcp_init_nondata_skb(buff, tp->write_seq++, TCPHDR_SYN);
3459 3460
	tcp_mstamp_refresh(tp);
	tp->retrans_stamp = tcp_time_stamp(tp);
3461
	tcp_connect_queue_skb(sk, buff);
3462
	tcp_ecn_send_syn(sk, buff);
L
Linus Torvalds 已提交
3463

3464 3465 3466
	/* 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);
3467 3468
	if (err == -ECONNREFUSED)
		return err;
3469 3470 3471 3472 3473 3474

	/* 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
3475
	TCP_INC_STATS(sock_net(sk), TCP_MIB_ACTIVEOPENS);
L
Linus Torvalds 已提交
3476 3477

	/* Timer for repeating the SYN until an answer. */
3478 3479
	inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
				  inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
L
Linus Torvalds 已提交
3480 3481
	return 0;
}
E
Eric Dumazet 已提交
3482
EXPORT_SYMBOL(tcp_connect);
L
Linus Torvalds 已提交
3483 3484 3485 3486 3487 3488 3489

/* 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)
{
3490 3491
	struct inet_connection_sock *icsk = inet_csk(sk);
	int ato = icsk->icsk_ack.ato;
L
Linus Torvalds 已提交
3492 3493
	unsigned long timeout;

3494 3495
	tcp_ca_event(sk, CA_EVENT_DELAYED_ACK);

L
Linus Torvalds 已提交
3496
	if (ato > TCP_DELACK_MIN) {
3497
		const struct tcp_sock *tp = tcp_sk(sk);
3498
		int max_ato = HZ / 2;
L
Linus Torvalds 已提交
3499

3500 3501
		if (icsk->icsk_ack.pingpong ||
		    (icsk->icsk_ack.pending & ICSK_ACK_PUSHED))
L
Linus Torvalds 已提交
3502 3503 3504 3505 3506
			max_ato = TCP_DELACK_MAX;

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

		/* If some rtt estimate is known, use it to bound delayed ack.
3507
		 * Do not use inet_csk(sk)->icsk_rto here, use results of rtt measurements
L
Linus Torvalds 已提交
3508 3509
		 * directly.
		 */
3510 3511 3512
		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
3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524

			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. */
3525
	if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
L
Linus Torvalds 已提交
3526 3527 3528
		/* If delack timer was blocked or is about to expire,
		 * send ACK now.
		 */
3529 3530
		if (icsk->icsk_ack.blocked ||
		    time_before_eq(icsk->icsk_ack.timeout, jiffies + (ato >> 2))) {
L
Linus Torvalds 已提交
3531 3532 3533 3534
			tcp_send_ack(sk);
			return;
		}

3535 3536
		if (!time_before(timeout, icsk->icsk_ack.timeout))
			timeout = icsk->icsk_ack.timeout;
L
Linus Torvalds 已提交
3537
	}
3538 3539 3540
	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
3541 3542 3543 3544 3545
}

/* This routine sends an ack and also updates the window. */
void tcp_send_ack(struct sock *sk)
{
3546
	struct sk_buff *buff;
L
Linus Torvalds 已提交
3547

3548 3549 3550
	/* If we have been reset, we may not send again. */
	if (sk->sk_state == TCP_CLOSE)
		return;
L
Linus Torvalds 已提交
3551

3552 3553
	tcp_ca_event(sk, CA_EVENT_NON_DELAYED_ACK);

3554 3555 3556 3557
	/* We are not putting this on the write queue, so
	 * tcp_transmit_skb() will set the ownership to this
	 * sock.
	 */
3558 3559 3560
	buff = alloc_skb(MAX_TCP_HEADER,
			 sk_gfp_mask(sk, GFP_ATOMIC | __GFP_NOWARN));
	if (unlikely(!buff)) {
3561 3562 3563 3564 3565
		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
3566
	}
3567 3568 3569

	/* Reserve space for headers and prepare control bits. */
	skb_reserve(buff, MAX_TCP_HEADER);
C
Changli Gao 已提交
3570
	tcp_init_nondata_skb(buff, tcp_acceptable_seq(sk), TCPHDR_ACK);
3571

E
Eric Dumazet 已提交
3572 3573 3574 3575 3576 3577
	/* 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
	 */
	skb_set_tcp_pure_ack(buff);

3578
	/* Send it off, this clears delayed acks for us. */
3579
	tcp_transmit_skb(sk, buff, 0, (__force gfp_t)0);
L
Linus Torvalds 已提交
3580
}
3581
EXPORT_SYMBOL_GPL(tcp_send_ack);
L
Linus Torvalds 已提交
3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593

/* 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.
 */
3594
static int tcp_xmit_probe_skb(struct sock *sk, int urgent, int mib)
L
Linus Torvalds 已提交
3595 3596 3597 3598 3599
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb;

	/* We don't queue it, tcp_transmit_skb() sets ownership. */
3600 3601
	skb = alloc_skb(MAX_TCP_HEADER,
			sk_gfp_mask(sk, GFP_ATOMIC | __GFP_NOWARN));
3602
	if (!skb)
L
Linus Torvalds 已提交
3603 3604 3605 3606 3607 3608 3609 3610
		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
3611
	tcp_init_nondata_skb(skb, tp->snd_una - !urgent, TCPHDR_ACK);
3612
	NET_INC_STATS(sock_net(sk), mib);
3613
	return tcp_transmit_skb(sk, skb, 0, (__force gfp_t)0);
L
Linus Torvalds 已提交
3614 3615
}

3616
/* Called from setsockopt( ... TCP_REPAIR ) */
P
Pavel Emelyanov 已提交
3617 3618 3619 3620
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;
3621
		tcp_mstamp_refresh(tcp_sk(sk));
3622
		tcp_xmit_probe_skb(sk, 0, LINUX_MIB_TCPWINPROBE);
P
Pavel Emelyanov 已提交
3623 3624 3625
	}
}

3626
/* Initiate keepalive or window probe from timer. */
3627
int tcp_write_wakeup(struct sock *sk, int mib)
L
Linus Torvalds 已提交
3628
{
3629 3630
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb;
L
Linus Torvalds 已提交
3631

3632 3633 3634
	if (sk->sk_state == TCP_CLOSE)
		return -1;

3635 3636
	skb = tcp_send_head(sk);
	if (skb && before(TCP_SKB_CB(skb)->seq, tcp_wnd_end(tp))) {
3637
		int err;
I
Ilpo Järvinen 已提交
3638
		unsigned int mss = tcp_current_mss(sk);
3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650
		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
3651
			TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
3652
			if (tcp_fragment(sk, skb, seg_size, mss, GFP_ATOMIC))
3653 3654
				return -1;
		} else if (!tcp_skb_pcount(skb))
3655
			tcp_set_skb_tso_segs(skb, mss);
3656

E
Eric Dumazet 已提交
3657
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
3658 3659 3660 3661 3662
		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
3663
		if (between(tp->snd_up, tp->snd_una + 1, tp->snd_una + 0xFFFF))
3664 3665
			tcp_xmit_probe_skb(sk, 1, mib);
		return tcp_xmit_probe_skb(sk, 0, mib);
L
Linus Torvalds 已提交
3666 3667 3668 3669 3670 3671 3672 3673
	}
}

/* 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)
{
3674
	struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
3675
	struct tcp_sock *tp = tcp_sk(sk);
3676
	struct net *net = sock_net(sk);
3677
	unsigned long probe_max;
L
Linus Torvalds 已提交
3678 3679
	int err;

3680
	err = tcp_write_wakeup(sk, LINUX_MIB_TCPWINPROBE);
L
Linus Torvalds 已提交
3681

3682
	if (tp->packets_out || !tcp_send_head(sk)) {
L
Linus Torvalds 已提交
3683
		/* Cancel probe timer, if it is not required. */
3684
		icsk->icsk_probes_out = 0;
3685
		icsk->icsk_backoff = 0;
L
Linus Torvalds 已提交
3686 3687 3688 3689
		return;
	}

	if (err <= 0) {
3690
		if (icsk->icsk_backoff < net->ipv4.sysctl_tcp_retries2)
3691
			icsk->icsk_backoff++;
3692
		icsk->icsk_probes_out++;
3693
		probe_max = TCP_RTO_MAX;
L
Linus Torvalds 已提交
3694 3695
	} else {
		/* If packet was not sent due to local congestion,
3696
		 * do not backoff and do not remember icsk_probes_out.
L
Linus Torvalds 已提交
3697 3698 3699 3700
		 * Let local senders to fight for local resources.
		 *
		 * Use accumulated backoff yet.
		 */
3701 3702
		if (!icsk->icsk_probes_out)
			icsk->icsk_probes_out = 1;
3703
		probe_max = TCP_RESOURCE_PROBE_INTERVAL;
L
Linus Torvalds 已提交
3704
	}
3705
	inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
3706
				  tcp_probe0_when(sk, probe_max),
3707
				  TCP_RTO_MAX);
L
Linus Torvalds 已提交
3708
}
3709

3710
int tcp_rtx_synack(const struct sock *sk, struct request_sock *req)
3711 3712 3713 3714 3715
{
	const struct tcp_request_sock_ops *af_ops = tcp_rsk(req)->af_specific;
	struct flowi fl;
	int res;

3716
	tcp_rsk(req)->txhash = net_tx_rndhash();
3717
	res = af_ops->send_synack(sk, NULL, &fl, req, NULL, TCP_SYNACK_NORMAL);
3718
	if (!res) {
E
Eric Dumazet 已提交
3719
		__TCP_INC_STATS(sock_net(sk), TCP_MIB_RETRANSSEGS);
3720
		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPSYNRETRANS);
3721 3722
		if (unlikely(tcp_passive_fastopen(sk)))
			tcp_sk(sk)->total_retrans++;
3723 3724 3725 3726
	}
	return res;
}
EXPORT_SYMBOL(tcp_rtx_synack);