tcp_output.c 95.4 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7
/*
 * INET		An implementation of the TCP/IP protocol suite for the LINUX
 *		operating system.  INET is implemented using the  BSD Socket
 *		interface as the means of communication with the user level.
 *
 *		Implementation of the Transmission Control Protocol(TCP).
 *
8
 * Authors:	Ross Biro
L
Linus Torvalds 已提交
9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36
 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
 *		Mark Evans, <evansmp@uhura.aston.ac.uk>
 *		Corey Minyard <wf-rch!minyard@relay.EU.net>
 *		Florian La Roche, <flla@stud.uni-sb.de>
 *		Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
 *		Linus Torvalds, <torvalds@cs.helsinki.fi>
 *		Alan Cox, <gw4pts@gw4pts.ampr.org>
 *		Matthew Dillon, <dillon@apollo.west.oic.com>
 *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
 *		Jorge Cwik, <jorge@laser.satlink.net>
 */

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

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

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

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

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

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

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

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

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

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

68 69 70
unsigned int sysctl_tcp_notsent_lowat __read_mostly = UINT_MAX;
EXPORT_SYMBOL(sysctl_tcp_notsent_lowat);

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

74
/* Account for new data that has been sent to the network. */
75
static void tcp_event_new_data_sent(struct sock *sk, const struct sk_buff *skb)
L
Linus Torvalds 已提交
76
{
N
Nandita Dukkipati 已提交
77
	struct inet_connection_sock *icsk = inet_csk(sk);
78
	struct tcp_sock *tp = tcp_sk(sk);
79
	unsigned int prior_packets = tp->packets_out;
80

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

84
	tp->packets_out += tcp_skb_pcount(skb);
N
Nandita Dukkipati 已提交
85
	if (!prior_packets || icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
86
	    icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
87
		tcp_rearm_rto(sk);
88
	}
89

90 91
	NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPORIGDATASENT,
		      tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
92 93 94 95 96 97 98 99
}

/* SND.NXT, if window was not shrunk.
 * If window has been shrunk, what should we make? It is not clear at all.
 * Using SND.UNA we will fail to open window, SND.NXT is out of window. :-(
 * Anything in between SND.UNA...SND.UNA+SND.WND also can be already
 * invalid. OK, let's make this for now:
 */
100
static inline __u32 tcp_acceptable_seq(const struct sock *sk)
L
Linus Torvalds 已提交
101
{
102
	const struct tcp_sock *tp = tcp_sk(sk);
103

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

/* Calculate mss to advertise in SYN segment.
 * RFC1122, RFC1063, draft-ietf-tcpimpl-pmtud-01 state that:
 *
 * 1. It is independent of path mtu.
 * 2. Ideally, it is maximal possible segment size i.e. 65535-40.
 * 3. For IPv4 it is reasonable to calculate it from maximal MTU of
 *    attached devices, because some buggy hosts are confused by
 *    large MSS.
 * 4. We do not make 3, we advertise MSS, calculated from first
 *    hop device mtu, but allow to raise it to ip_rt_min_advmss.
 *    This may be overridden via information stored in routing table.
 * 5. Value 65535 for MSS is valid in IPv6 and means "as large as possible,
 *    probably even Jumbo".
 */
static __u16 tcp_advertise_mss(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
127
	const struct dst_entry *dst = __sk_dst_get(sk);
L
Linus Torvalds 已提交
128 129
	int mss = tp->advmss;

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

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

	return (__u16)mss;
}

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

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

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

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

163
/* Congestion state accounting after a packet has been sent. */
S
Stephen Hemminger 已提交
164
static void tcp_event_data_sent(struct tcp_sock *tp,
165
				struct sock *sk)
L
Linus Torvalds 已提交
166
{
167 168
	struct inet_connection_sock *icsk = inet_csk(sk);
	const u32 now = tcp_time_stamp;
169
	const struct dst_entry *dst = __sk_dst_get(sk);
L
Linus Torvalds 已提交
170

171 172
	if (sysctl_tcp_slow_start_after_idle &&
	    (!tp->packets_out && (s32)(now - tp->lsndtime) > icsk->icsk_rto))
173
		tcp_cwnd_restart(sk, __sk_dst_get(sk));
L
Linus Torvalds 已提交
174 175 176 177 178 179

	tp->lsndtime = now;

	/* If it is a reply for ato after last received
	 * packet, enter pingpong mode.
	 */
180 181 182
	if ((u32)(now - icsk->icsk_ack.lrcvtime) < icsk->icsk_ack.ato &&
	    (!dst || !dst_metric(dst, RTAX_QUICKACK)))
			icsk->icsk_ack.pingpong = 1;
L
Linus Torvalds 已提交
183 184
}

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

192 193 194 195

u32 tcp_default_init_rwnd(u32 mss)
{
	/* Initial receive window should be twice of TCP_INIT_CWND to
W
Weiping Pan 已提交
196
	 * enable proper sending of new unsent data during fast recovery
197 198 199 200 201 202 203 204 205 206
	 * (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
207 208 209 210 211 212 213 214 215
/* 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,
216 217
			       int wscale_ok, __u8 *rcv_wscale,
			       __u32 init_rcv_wnd)
L
Linus Torvalds 已提交
218 219 220 221 222 223 224 225 226 227 228 229 230
{
	unsigned int space = (__space < 0 ? 0 : __space);

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

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

	/* NOTE: offering an initial window larger than 32767
231 232 233 234 235 236
	 * 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
237
	 */
238 239 240 241 242
	if (sysctl_tcp_workaround_signed_windows)
		(*rcv_wnd) = min(space, MAX_TCP_WINDOW);
	else
		(*rcv_wnd) = space;

L
Linus Torvalds 已提交
243 244 245
	(*rcv_wscale) = 0;
	if (wscale_ok) {
		/* Set window scaling on max possible window
246
		 * See RFC1323 for an explanation of the limit to 14
L
Linus Torvalds 已提交
247 248
		 */
		space = max_t(u32, sysctl_tcp_rmem[2], sysctl_rmem_max);
249
		space = min_t(u32, space, *window_clamp);
L
Linus Torvalds 已提交
250 251 252 253 254 255
		while (space > 65535 && (*rcv_wscale) < 14) {
			space >>= 1;
			(*rcv_wscale)++;
		}
	}

256
	if (mss > (1 << *rcv_wscale)) {
257 258 259
		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
260 261 262 263 264
	}

	/* Set the clamp no higher than max representable value */
	(*window_clamp) = min(65535U << (*rcv_wscale), *window_clamp);
}
E
Eric Dumazet 已提交
265
EXPORT_SYMBOL(tcp_select_initial_window);
L
Linus Torvalds 已提交
266 267 268 269 270 271

/* 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
272
static u16 tcp_select_window(struct sock *sk)
L
Linus Torvalds 已提交
273 274
{
	struct tcp_sock *tp = tcp_sk(sk);
275
	u32 old_win = tp->rcv_wnd;
L
Linus Torvalds 已提交
276 277 278 279
	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
280
	if (new_win < cur_win) {
L
Linus Torvalds 已提交
281 282 283 284 285 286 287
		/* 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.
		 */
288 289 290
		if (new_win == 0)
			NET_INC_STATS(sock_net(sk),
				      LINUX_MIB_TCPWANTZEROWINDOWADV);
291
		new_win = ALIGN(cur_win, 1 << tp->rx_opt.rcv_wscale);
L
Linus Torvalds 已提交
292 293 294 295 296 297 298
	}
	tp->rcv_wnd = new_win;
	tp->rcv_wup = tp->rcv_nxt;

	/* Make sure we do not exceed the maximum possible
	 * scaled window.
	 */
299
	if (!tp->rx_opt.rcv_wscale && sysctl_tcp_workaround_signed_windows)
L
Linus Torvalds 已提交
300 301 302 303 304 305 306 307
		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. */
308
	if (new_win == 0) {
L
Linus Torvalds 已提交
309
		tp->pred_flags = 0;
310 311 312 313 314 315
		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
316 317 318 319

	return new_win;
}

320
/* Packet ECN state for a SYN-ACK */
321
static inline void TCP_ECN_send_synack(const struct tcp_sock *tp, struct sk_buff *skb)
322
{
E
Eric Dumazet 已提交
323
	TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_CWR;
324
	if (!(tp->ecn_flags & TCP_ECN_OK))
E
Eric Dumazet 已提交
325
		TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_ECE;
326 327
}

328
/* Packet ECN state for a SYN.  */
329 330 331 332 333
static inline void TCP_ECN_send_syn(struct sock *sk, struct sk_buff *skb)
{
	struct tcp_sock *tp = tcp_sk(sk);

	tp->ecn_flags = 0;
334
	if (sock_net(sk)->ipv4.sysctl_tcp_ecn == 1) {
E
Eric Dumazet 已提交
335
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ECE | TCPHDR_CWR;
336 337 338 339 340
		tp->ecn_flags = TCP_ECN_OK;
	}
}

static __inline__ void
341
TCP_ECN_make_synack(const struct request_sock *req, struct tcphdr *th)
342 343 344 345 346
{
	if (inet_rsk(req)->ecn_ok)
		th->ece = 1;
}

347 348 349
/* Set up ECN state for a packet on a ESTABLISHED socket that is about to
 * be sent.
 */
350 351 352 353 354 355 356 357 358 359
static inline void TCP_ECN_send(struct sock *sk, struct sk_buff *skb,
				int tcp_header_len)
{
	struct tcp_sock *tp = tcp_sk(sk);

	if (tp->ecn_flags & TCP_ECN_OK) {
		/* Not-retransmitted data segment: set ECT and inject CWR. */
		if (skb->len != tcp_header_len &&
		    !before(TCP_SKB_CB(skb)->seq, tp->snd_nxt)) {
			INET_ECN_xmit(sk);
360
			if (tp->ecn_flags & TCP_ECN_QUEUE_CWR) {
361 362 363 364 365 366 367 368 369 370 371 372 373
				tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR;
				tcp_hdr(skb)->cwr = 1;
				skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
			}
		} else {
			/* ACK or retransmitted segment: clear ECT|CE */
			INET_ECN_dontxmit(sk);
		}
		if (tp->ecn_flags & TCP_ECN_DEMAND_CWR)
			tcp_hdr(skb)->ece = 1;
	}
}

374 375 376 377 378
/* 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)
{
379 380
	struct skb_shared_info *shinfo = skb_shinfo(skb);

381
	skb->ip_summed = CHECKSUM_PARTIAL;
382 383
	skb->csum = 0;

E
Eric Dumazet 已提交
384
	TCP_SKB_CB(skb)->tcp_flags = flags;
385 386
	TCP_SKB_CB(skb)->sacked = 0;

387 388 389
	shinfo->gso_segs = 1;
	shinfo->gso_size = 0;
	shinfo->gso_type = 0;
390 391

	TCP_SKB_CB(skb)->seq = seq;
C
Changli Gao 已提交
392
	if (flags & (TCPHDR_SYN | TCPHDR_FIN))
393 394 395 396
		seq++;
	TCP_SKB_CB(skb)->end_seq = seq;
}

E
Eric Dumazet 已提交
397
static inline bool tcp_urg_mode(const struct tcp_sock *tp)
I
Ilpo Järvinen 已提交
398 399 400 401
{
	return tp->snd_una != tp->snd_up;
}

A
Adam Langley 已提交
402 403 404
#define OPTION_SACK_ADVERTISE	(1 << 0)
#define OPTION_TS		(1 << 1)
#define OPTION_MD5		(1 << 2)
405
#define OPTION_WSCALE		(1 << 3)
Y
Yuchung Cheng 已提交
406
#define OPTION_FAST_OPEN_COOKIE	(1 << 8)
A
Adam Langley 已提交
407 408

struct tcp_out_options {
Y
Yuchung Cheng 已提交
409 410
	u16 options;		/* bit field of OPTION_* */
	u16 mss;		/* 0 to disable */
A
Adam Langley 已提交
411 412
	u8 ws;			/* window scale, 0 to disable */
	u8 num_sack_blocks;	/* number of SACK blocks to include */
413 414
	u8 hash_size;		/* bytes in hash_location */
	__u8 *hash_location;	/* temporary pointer, overloaded */
Y
Yuchung Cheng 已提交
415 416
	__u32 tsval, tsecr;	/* need to include OPTION_TS */
	struct tcp_fastopen_cookie *fastopen_cookie;	/* Fast open cookie */
A
Adam Langley 已提交
417 418
};

419 420 421
/* Write previously computed TCP options to the packet.
 *
 * Beware: Something in the Internet is very sensitive to the ordering of
422 423
 * 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
424
 * inter-operability perspective it seems that we're somewhat stuck with
425 426 427 428 429 430 431
 * 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
432
static void tcp_options_write(__be32 *ptr, struct tcp_sock *tp,
433 434
			      struct tcp_out_options *opts)
{
Y
Yuchung Cheng 已提交
435
	u16 options = opts->options;	/* mungable copy */
436 437

	if (unlikely(OPTION_MD5 & options)) {
C
Christoph Paasch 已提交
438 439
		*ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
			       (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
440 441
		/* overload cookie hash location */
		opts->hash_location = (__u8 *)ptr;
A
Adam Langley 已提交
442
		ptr += 4;
S
Stephen Hemminger 已提交
443
	}
A
Adam Langley 已提交
444

445 446 447 448 449 450
	if (unlikely(opts->mss)) {
		*ptr++ = htonl((TCPOPT_MSS << 24) |
			       (TCPOLEN_MSS << 16) |
			       opts->mss);
	}

451 452
	if (likely(OPTION_TS & options)) {
		if (unlikely(OPTION_SACK_ADVERTISE & options)) {
A
Adam Langley 已提交
453 454 455 456
			*ptr++ = htonl((TCPOPT_SACK_PERM << 24) |
				       (TCPOLEN_SACK_PERM << 16) |
				       (TCPOPT_TIMESTAMP << 8) |
				       TCPOLEN_TIMESTAMP);
457
			options &= ~OPTION_SACK_ADVERTISE;
A
Adam Langley 已提交
458 459 460 461 462 463 464 465 466 467
		} else {
			*ptr++ = htonl((TCPOPT_NOP << 24) |
				       (TCPOPT_NOP << 16) |
				       (TCPOPT_TIMESTAMP << 8) |
				       TCPOLEN_TIMESTAMP);
		}
		*ptr++ = htonl(opts->tsval);
		*ptr++ = htonl(opts->tsecr);
	}

468
	if (unlikely(OPTION_SACK_ADVERTISE & options)) {
A
Adam Langley 已提交
469 470 471 472 473 474
		*ptr++ = htonl((TCPOPT_NOP << 24) |
			       (TCPOPT_NOP << 16) |
			       (TCPOPT_SACK_PERM << 8) |
			       TCPOLEN_SACK_PERM);
	}

475
	if (unlikely(OPTION_WSCALE & options)) {
A
Adam Langley 已提交
476 477 478 479 480 481 482 483 484
		*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
485 486 487 488 489
		int this_sack;

		*ptr++ = htonl((TCPOPT_NOP  << 24) |
			       (TCPOPT_NOP  << 16) |
			       (TCPOPT_SACK <<  8) |
A
Adam Langley 已提交
490
			       (TCPOLEN_SACK_BASE + (opts->num_sack_blocks *
S
Stephen Hemminger 已提交
491
						     TCPOLEN_SACK_PERBLOCK)));
S
Stephen Hemminger 已提交
492

A
Adam Langley 已提交
493 494
		for (this_sack = 0; this_sack < opts->num_sack_blocks;
		     ++this_sack) {
S
Stephen Hemminger 已提交
495 496 497
			*ptr++ = htonl(sp[this_sack].start_seq);
			*ptr++ = htonl(sp[this_sack].end_seq);
		}
S
Stephen Hemminger 已提交
498

499
		tp->rx_opt.dsack = 0;
S
Stephen Hemminger 已提交
500
	}
Y
Yuchung Cheng 已提交
501 502 503 504 505 506 507 508 509 510 511 512 513 514 515

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

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

		memcpy(ptr, foc->val, foc->len);
		if ((foc->len & 3) == 2) {
			u8 *align = ((u8 *)ptr) + foc->len;
			align[0] = align[1] = TCPOPT_NOP;
		}
		ptr += (foc->len + 3) >> 2;
	}
A
Adam Langley 已提交
516 517
}

518 519 520
/* Compute TCP options for SYN packets. This is not the final
 * network wire format yet.
 */
521
static unsigned int tcp_syn_options(struct sock *sk, struct sk_buff *skb,
A
Adam Langley 已提交
522
				struct tcp_out_options *opts,
523 524
				struct tcp_md5sig_key **md5)
{
A
Adam Langley 已提交
525
	struct tcp_sock *tp = tcp_sk(sk);
526
	unsigned int remaining = MAX_TCP_OPTION_SPACE;
527
	struct tcp_fastopen_request *fastopen = tp->fastopen_req;
A
Adam Langley 已提交
528

529
#ifdef CONFIG_TCP_MD5SIG
A
Adam Langley 已提交
530 531 532
	*md5 = tp->af_specific->md5_lookup(sk, sk);
	if (*md5) {
		opts->options |= OPTION_MD5;
533
		remaining -= TCPOLEN_MD5SIG_ALIGNED;
534
	}
A
Adam Langley 已提交
535 536
#else
	*md5 = NULL;
537
#endif
A
Adam Langley 已提交
538 539 540 541 542 543 544 545 546 547 548

	/* 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);
549
	remaining -= TCPOLEN_MSS_ALIGNED;
A
Adam Langley 已提交
550

551
	if (likely(sysctl_tcp_timestamps && *md5 == NULL)) {
A
Adam Langley 已提交
552
		opts->options |= OPTION_TS;
553
		opts->tsval = TCP_SKB_CB(skb)->when + tp->tsoffset;
A
Adam Langley 已提交
554
		opts->tsecr = tp->rx_opt.ts_recent;
555
		remaining -= TCPOLEN_TSTAMP_ALIGNED;
A
Adam Langley 已提交
556
	}
557
	if (likely(sysctl_tcp_window_scaling)) {
A
Adam Langley 已提交
558
		opts->ws = tp->rx_opt.rcv_wscale;
559
		opts->options |= OPTION_WSCALE;
560
		remaining -= TCPOLEN_WSCALE_ALIGNED;
A
Adam Langley 已提交
561
	}
562
	if (likely(sysctl_tcp_sack)) {
A
Adam Langley 已提交
563
		opts->options |= OPTION_SACK_ADVERTISE;
564
		if (unlikely(!(OPTION_TS & opts->options)))
565
			remaining -= TCPOLEN_SACKPERM_ALIGNED;
A
Adam Langley 已提交
566 567
	}

568 569 570 571 572 573 574 575 576 577
	if (fastopen && fastopen->cookie.len >= 0) {
		u32 need = TCPOLEN_EXP_FASTOPEN_BASE + fastopen->cookie.len;
		need = (need + 3) & ~3U;  /* Align to 32 bits */
		if (remaining >= need) {
			opts->options |= OPTION_FAST_OPEN_COOKIE;
			opts->fastopen_cookie = &fastopen->cookie;
			remaining -= need;
			tp->syn_fastopen = 1;
		}
	}
578 579

	return MAX_TCP_OPTION_SPACE - remaining;
S
Stephen Hemminger 已提交
580 581
}

582
/* Set up TCP options for SYN-ACKs. */
583
static unsigned int tcp_synack_options(struct sock *sk,
A
Adam Langley 已提交
584
				   struct request_sock *req,
585
				   unsigned int mss, struct sk_buff *skb,
A
Adam Langley 已提交
586
				   struct tcp_out_options *opts,
587
				   struct tcp_md5sig_key **md5,
588
				   struct tcp_fastopen_cookie *foc)
589
{
A
Adam Langley 已提交
590
	struct inet_request_sock *ireq = inet_rsk(req);
591
	unsigned int remaining = MAX_TCP_OPTION_SPACE;
A
Adam Langley 已提交
592

593
#ifdef CONFIG_TCP_MD5SIG
A
Adam Langley 已提交
594 595 596
	*md5 = tcp_rsk(req)->af_specific->md5_lookup(sk, req);
	if (*md5) {
		opts->options |= OPTION_MD5;
597 598 599 600 601 602 603
		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
604
		ireq->tstamp_ok &= !ireq->sack_ok;
605
	}
A
Adam Langley 已提交
606 607
#else
	*md5 = NULL;
608
#endif
A
Adam Langley 已提交
609

610
	/* We always send an MSS option. */
A
Adam Langley 已提交
611
	opts->mss = mss;
612
	remaining -= TCPOLEN_MSS_ALIGNED;
A
Adam Langley 已提交
613 614 615

	if (likely(ireq->wscale_ok)) {
		opts->ws = ireq->rcv_wscale;
616
		opts->options |= OPTION_WSCALE;
617
		remaining -= TCPOLEN_WSCALE_ALIGNED;
A
Adam Langley 已提交
618
	}
E
Eric Dumazet 已提交
619
	if (likely(ireq->tstamp_ok)) {
A
Adam Langley 已提交
620 621 622
		opts->options |= OPTION_TS;
		opts->tsval = TCP_SKB_CB(skb)->when;
		opts->tsecr = req->ts_recent;
623
		remaining -= TCPOLEN_TSTAMP_ALIGNED;
A
Adam Langley 已提交
624 625 626
	}
	if (likely(ireq->sack_ok)) {
		opts->options |= OPTION_SACK_ADVERTISE;
E
Eric Dumazet 已提交
627
		if (unlikely(!ireq->tstamp_ok))
628
			remaining -= TCPOLEN_SACKPERM_ALIGNED;
A
Adam Langley 已提交
629
	}
630 631 632 633 634 635 636 637 638
	if (foc != NULL) {
		u32 need = TCPOLEN_EXP_FASTOPEN_BASE + foc->len;
		need = (need + 3) & ~3U;  /* Align to 32 bits */
		if (remaining >= need) {
			opts->options |= OPTION_FAST_OPEN_COOKIE;
			opts->fastopen_cookie = foc;
			remaining -= need;
		}
	}
C
Christoph Paasch 已提交
639

640
	return MAX_TCP_OPTION_SPACE - remaining;
A
Adam Langley 已提交
641 642
}

643 644 645
/* Compute TCP options for ESTABLISHED sockets. This is not the
 * final wire format yet.
 */
646
static unsigned int tcp_established_options(struct sock *sk, struct sk_buff *skb,
A
Adam Langley 已提交
647
					struct tcp_out_options *opts,
648 649
					struct tcp_md5sig_key **md5)
{
A
Adam Langley 已提交
650 651
	struct tcp_skb_cb *tcb = skb ? TCP_SKB_CB(skb) : NULL;
	struct tcp_sock *tp = tcp_sk(sk);
652
	unsigned int size = 0;
653
	unsigned int eff_sacks;
A
Adam Langley 已提交
654

655 656
	opts->options = 0;

A
Adam Langley 已提交
657 658 659 660 661 662 663 664 665 666 667 668
#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;
669
		opts->tsval = tcb ? tcb->when + tp->tsoffset : 0;
A
Adam Langley 已提交
670 671 672 673
		opts->tsecr = tp->rx_opt.ts_recent;
		size += TCPOLEN_TSTAMP_ALIGNED;
	}

674 675
	eff_sacks = tp->rx_opt.num_sacks + tp->rx_opt.dsack;
	if (unlikely(eff_sacks)) {
676
		const unsigned int remaining = MAX_TCP_OPTION_SPACE - size;
A
Adam Langley 已提交
677
		opts->num_sack_blocks =
678
			min_t(unsigned int, eff_sacks,
A
Adam Langley 已提交
679 680 681 682 683 684 685
			      (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
686
}
L
Linus Torvalds 已提交
687

E
Eric Dumazet 已提交
688 689 690 691 692 693 694 695 696

/* 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
697
 * The invariant being skb->truesize subtracted from sk->sk_wmem_alloc
E
Eric Dumazet 已提交
698 699 700 701 702 703 704 705 706 707 708
 *
 * 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);

709 710 711 712 713
static void tcp_tsq_handler(struct sock *sk)
{
	if ((1 << sk->sk_state) &
	    (TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_CLOSING |
	     TCPF_CLOSE_WAIT  | TCPF_LAST_ACK))
J
John Ogness 已提交
714 715
		tcp_write_xmit(sk, tcp_current_mss(sk), tcp_sk(sk)->nonagle,
			       0, GFP_ATOMIC);
716
}
E
Eric Dumazet 已提交
717
/*
S
stephen hemminger 已提交
718
 * One tasklet per cpu tries to send more skbs.
E
Eric Dumazet 已提交
719
 * We run in tasklet context but need to disable irqs when
S
stephen hemminger 已提交
720
 * transferring tsq->head because tcp_wfree() might
E
Eric Dumazet 已提交
721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743
 * interrupt us (non NAPI drivers)
 */
static void tcp_tasklet_func(unsigned long data)
{
	struct tsq_tasklet *tsq = (struct tsq_tasklet *)data;
	LIST_HEAD(list);
	unsigned long flags;
	struct list_head *q, *n;
	struct tcp_sock *tp;
	struct sock *sk;

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

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

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

		if (!sock_owned_by_user(sk)) {
744
			tcp_tsq_handler(sk);
E
Eric Dumazet 已提交
745 746
		} else {
			/* defer the work to tcp_release_cb() */
747
			set_bit(TCP_TSQ_DEFERRED, &tp->tsq_flags);
E
Eric Dumazet 已提交
748 749 750 751 752 753 754 755
		}
		bh_unlock_sock(sk);

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

756 757
#define TCP_DEFERRED_ALL ((1UL << TCP_TSQ_DEFERRED) |		\
			  (1UL << TCP_WRITE_TIMER_DEFERRED) |	\
758 759
			  (1UL << TCP_DELACK_TIMER_DEFERRED) |	\
			  (1UL << TCP_MTU_REDUCED_DEFERRED))
E
Eric Dumazet 已提交
760 761 762 763 764 765 766 767 768 769
/**
 * tcp_release_cb - tcp release_sock() callback
 * @sk: socket
 *
 * called from release_sock() to perform protocol dependent
 * actions before socket release.
 */
void tcp_release_cb(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
770
	unsigned long flags, nflags;
E
Eric Dumazet 已提交
771

772 773 774 775 776 777 778 779 780 781 782
	/* perform an atomic operation only if at least one flag is set */
	do {
		flags = tp->tsq_flags;
		if (!(flags & TCP_DEFERRED_ALL))
			return;
		nflags = flags & ~TCP_DEFERRED_ALL;
	} while (cmpxchg(&tp->tsq_flags, flags, nflags) != flags);

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

783 784 785 786 787 788 789 790 791 792 793
	/* 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);

794
	if (flags & (1UL << TCP_WRITE_TIMER_DEFERRED)) {
795
		tcp_write_timer_handler(sk);
796 797 798
		__sock_put(sk);
	}
	if (flags & (1UL << TCP_DELACK_TIMER_DEFERRED)) {
799
		tcp_delack_timer_handler(sk);
800 801 802
		__sock_put(sk);
	}
	if (flags & (1UL << TCP_MTU_REDUCED_DEFERRED)) {
803
		sk->sk_prot->mtu_reduced(sk);
804 805
		__sock_put(sk);
	}
E
Eric Dumazet 已提交
806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824
}
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
825
 * We can't xmit new skbs from this context, as we might already
E
Eric Dumazet 已提交
826 827
 * hold qdisc lock.
 */
E
Eric Dumazet 已提交
828
void tcp_wfree(struct sk_buff *skb)
E
Eric Dumazet 已提交
829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853
{
	struct sock *sk = skb->sk;
	struct tcp_sock *tp = tcp_sk(sk);

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

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

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

L
Linus Torvalds 已提交
854 855 856 857 858 859 860 861 862 863 864
/* 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.
 */
865 866
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
867
{
868 869 870 871
	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
872
	struct tcp_out_options opts;
873
	unsigned int tcp_options_size, tcp_header_size;
874
	struct tcp_md5sig_key *md5;
875 876 877 878 879
	struct tcphdr *th;
	int err;

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

880
	if (clone_it) {
881
		skb_mstamp_get(&skb->skb_mstamp);
882

883 884 885 886 887 888
		if (unlikely(skb_cloned(skb)))
			skb = pskb_copy(skb, gfp_mask);
		else
			skb = skb_clone(skb, gfp_mask);
		if (unlikely(!skb))
			return -ENOBUFS;
889 890
		/* Our usage of tstamp should remain private */
		skb->tstamp.tv64 = 0;
891
	}
L
Linus Torvalds 已提交
892

893 894 895
	inet = inet_sk(sk);
	tp = tcp_sk(sk);
	tcb = TCP_SKB_CB(skb);
A
Adam Langley 已提交
896
	memset(&opts, 0, sizeof(opts));
L
Linus Torvalds 已提交
897

E
Eric Dumazet 已提交
898
	if (unlikely(tcb->tcp_flags & TCPHDR_SYN))
A
Adam Langley 已提交
899 900 901 902 903
		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);
904

E
Eric Dumazet 已提交
905
	if (tcp_packets_in_flight(tp) == 0)
906
		tcp_ca_event(sk, CA_EVENT_TX_START);
E
Eric Dumazet 已提交
907 908 909 910 911

	/* if no packet is in qdisc/device queue, then allow XPS to select
	 * another queue.
	 */
	skb->ooo_okay = sk_wmem_alloc_get(sk) == 0;
912

913 914
	skb_push(skb, tcp_header_size);
	skb_reset_transport_header(skb);
E
Eric Dumazet 已提交
915 916 917

	skb_orphan(skb);
	skb->sk = sk;
E
Eric Dumazet 已提交
918
	skb->destructor = tcp_wfree;
E
Eric Dumazet 已提交
919
	atomic_add(skb->truesize, &sk->sk_wmem_alloc);
920 921

	/* Build TCP header and checksum it. */
922
	th = tcp_hdr(skb);
E
Eric Dumazet 已提交
923 924
	th->source		= inet->inet_sport;
	th->dest		= inet->inet_dport;
925 926
	th->seq			= htonl(tcb->seq);
	th->ack_seq		= htonl(tp->rcv_nxt);
927
	*(((__be16 *)th) + 6)	= htons(((tcp_header_size >> 2) << 12) |
E
Eric Dumazet 已提交
928
					tcb->tcp_flags);
929

E
Eric Dumazet 已提交
930
	if (unlikely(tcb->tcp_flags & TCPHDR_SYN)) {
931 932 933
		/* RFC1323: The window in SYN & SYN/ACK segments
		 * is never scaled.
		 */
934
		th->window	= htons(min(tp->rcv_wnd, 65535U));
935 936 937 938 939
	} else {
		th->window	= htons(tcp_select_window(sk));
	}
	th->check		= 0;
	th->urg_ptr		= 0;
L
Linus Torvalds 已提交
940

I
Ilpo Järvinen 已提交
941
	/* The urg_mode check is necessary during a below snd_una win probe */
942 943 944 945 946
	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)) {
947
			th->urg_ptr = htons(0xFFFF);
948 949
			th->urg = 1;
		}
950
	}
L
Linus Torvalds 已提交
951

952
	tcp_options_write((__be32 *)(th + 1), tp, &opts);
E
Eric Dumazet 已提交
953
	if (likely((tcb->tcp_flags & TCPHDR_SYN) == 0))
954
		TCP_ECN_send(sk, skb, tcp_header_size);
L
Linus Torvalds 已提交
955

956 957 958
#ifdef CONFIG_TCP_MD5SIG
	/* Calculate the MD5 hash, as we have all we need now */
	if (md5) {
E
Eric Dumazet 已提交
959
		sk_nocaps_add(sk, NETIF_F_GSO_MASK);
960
		tp->af_specific->calc_md5_hash(opts.hash_location,
961
					       md5, sk, NULL, skb);
962 963 964
	}
#endif

965
	icsk->icsk_af_ops->send_check(sk, skb);
L
Linus Torvalds 已提交
966

E
Eric Dumazet 已提交
967
	if (likely(tcb->tcp_flags & TCPHDR_ACK))
968
		tcp_event_ack_sent(sk, tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
969

970
	if (skb->len != tcp_header_size)
971
		tcp_event_data_sent(tp, sk);
L
Linus Torvalds 已提交
972

973
	if (after(tcb->end_seq, tp->snd_nxt) || tcb->seq == tcb->end_seq)
974 975
		TCP_ADD_STATS(sock_net(sk), TCP_MIB_OUTSEGS,
			      tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
976

977
	err = icsk->icsk_af_ops->queue_xmit(sk, skb, &inet->cork.fl);
978
	if (likely(err <= 0))
979 980
		return err;

981
	tcp_enter_cwr(sk, 1);
982

983
	return net_xmit_eval(err);
L
Linus Torvalds 已提交
984 985
}

986
/* This routine just queues the buffer for sending.
L
Linus Torvalds 已提交
987 988 989 990 991 992 993 994 995 996 997
 *
 * NOTE: probe0 timer is not checked, do not forget tcp_push_pending_frames,
 * otherwise socket can stall.
 */
static void tcp_queue_skb(struct sock *sk, struct sk_buff *skb)
{
	struct tcp_sock *tp = tcp_sk(sk);

	/* Advance write_seq and place onto the write_queue. */
	tp->write_seq = TCP_SKB_CB(skb)->end_seq;
	skb_header_release(skb);
998
	tcp_add_write_queue_tail(sk, skb);
999 1000
	sk->sk_wmem_queued += skb->truesize;
	sk_mem_charge(sk, skb->truesize);
L
Linus Torvalds 已提交
1001 1002
}

1003
/* Initialize TSO segments for a packet. */
1004
static void tcp_set_skb_tso_segs(const struct sock *sk, struct sk_buff *skb,
1005
				 unsigned int mss_now)
1006
{
1007 1008
	struct skb_shared_info *shinfo = skb_shinfo(skb);

1009 1010 1011
	/* Make sure we own this skb before messing gso_size/gso_segs */
	WARN_ON_ONCE(skb_cloned(skb));

1012
	if (skb->len <= mss_now || skb->ip_summed == CHECKSUM_NONE) {
1013 1014 1015
		/* Avoid the costly divide in the normal
		 * non-TSO case.
		 */
1016 1017 1018
		shinfo->gso_segs = 1;
		shinfo->gso_size = 0;
		shinfo->gso_type = 0;
1019
	} else {
1020 1021 1022
		shinfo->gso_segs = DIV_ROUND_UP(skb->len, mss_now);
		shinfo->gso_size = mss_now;
		shinfo->gso_type = sk->sk_gso_type;
L
Linus Torvalds 已提交
1023 1024 1025
	}
}

I
Ilpo Järvinen 已提交
1026
/* When a modification to fackets out becomes necessary, we need to check
1027
 * skb is counted to fackets_out or not.
I
Ilpo Järvinen 已提交
1028
 */
1029
static void tcp_adjust_fackets_out(struct sock *sk, const struct sk_buff *skb,
I
Ilpo Järvinen 已提交
1030 1031
				   int decr)
{
1032 1033
	struct tcp_sock *tp = tcp_sk(sk);

1034
	if (!tp->sacked_out || tcp_is_reno(tp))
I
Ilpo Järvinen 已提交
1035 1036
		return;

1037
	if (after(tcp_highest_sack_seq(tp), TCP_SKB_CB(skb)->seq))
I
Ilpo Järvinen 已提交
1038 1039 1040
		tp->fackets_out -= decr;
}

1041 1042 1043
/* Pcount in the middle of the write queue got changed, we need to do various
 * tweaks to fix counters
 */
1044
static void tcp_adjust_pcount(struct sock *sk, const struct sk_buff *skb, int decr)
1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064
{
	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) &&
1065
	    (tcp_is_fack(tp) || (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)))
1066 1067 1068 1069 1070
		tp->lost_cnt_hint -= decr;

	tcp_verify_left_out(tp);
}

L
Linus Torvalds 已提交
1071 1072
/* 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
1073
 * packet to the list.  This won't be called frequently, I hope.
L
Linus Torvalds 已提交
1074 1075
 * Remember, these are still headerless SKBs at this point.
 */
1076 1077
int tcp_fragment(struct sock *sk, struct sk_buff *skb, u32 len,
		 unsigned int mss_now)
L
Linus Torvalds 已提交
1078 1079 1080
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *buff;
1081
	int nsize, old_factor;
1082
	int nlen;
1083
	u8 flags;
L
Linus Torvalds 已提交
1084

1085 1086
	if (WARN_ON(len > skb->len))
		return -EINVAL;
1087

L
Linus Torvalds 已提交
1088 1089 1090 1091
	nsize = skb_headlen(skb) - len;
	if (nsize < 0)
		nsize = 0;

1092
	if (skb_unclone(skb, GFP_ATOMIC))
L
Linus Torvalds 已提交
1093 1094 1095 1096 1097 1098
		return -ENOMEM;

	/* Get a new skb... force flag on. */
	buff = sk_stream_alloc_skb(sk, nsize, GFP_ATOMIC);
	if (buff == NULL)
		return -ENOMEM; /* We'll just try again later. */
H
Herbert Xu 已提交
1099

1100 1101
	sk->sk_wmem_queued += buff->truesize;
	sk_mem_charge(sk, buff->truesize);
1102 1103 1104
	nlen = skb->len - len - nsize;
	buff->truesize += nlen;
	skb->truesize -= nlen;
L
Linus Torvalds 已提交
1105 1106 1107 1108 1109 1110 1111

	/* 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
1112 1113 1114
	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;
1115
	TCP_SKB_CB(buff)->sacked = TCP_SKB_CB(skb)->sacked;
L
Linus Torvalds 已提交
1116

1117
	if (!skb_shinfo(skb)->nr_frags && skb->ip_summed != CHECKSUM_PARTIAL) {
L
Linus Torvalds 已提交
1118
		/* Copy and checksum data tail into the new buffer. */
1119 1120
		buff->csum = csum_partial_copy_nocheck(skb->data + len,
						       skb_put(buff, nsize),
L
Linus Torvalds 已提交
1121 1122 1123 1124 1125 1126
						       nsize, 0);

		skb_trim(skb, len);

		skb->csum = csum_block_sub(skb->csum, buff->csum, len);
	} else {
1127
		skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
1128 1129 1130 1131 1132 1133 1134 1135 1136
		skb_split(skb, buff, len);
	}

	buff->ip_summed = skb->ip_summed;

	/* Looks stupid, but our code really uses when of
	 * skbs, which it never sent before. --ANK
	 */
	TCP_SKB_CB(buff)->when = TCP_SKB_CB(skb)->when;
1137
	buff->tstamp = skb->tstamp;
L
Linus Torvalds 已提交
1138

1139 1140
	old_factor = tcp_skb_pcount(skb);

L
Linus Torvalds 已提交
1141
	/* Fix up tso_factor for both original and new SKB.  */
1142 1143
	tcp_set_skb_tso_segs(sk, skb, mss_now);
	tcp_set_skb_tso_segs(sk, buff, mss_now);
L
Linus Torvalds 已提交
1144

1145 1146 1147
	/* If this packet has been sent out already, we must
	 * adjust the various packet counters.
	 */
1148
	if (!before(tp->snd_nxt, TCP_SKB_CB(buff)->end_seq)) {
1149 1150
		int diff = old_factor - tcp_skb_pcount(skb) -
			tcp_skb_pcount(buff);
L
Linus Torvalds 已提交
1151

1152 1153
		if (diff)
			tcp_adjust_pcount(sk, skb, diff);
L
Linus Torvalds 已提交
1154 1155 1156
	}

	/* Link BUFF into the send queue. */
1157
	skb_header_release(buff);
1158
	tcp_insert_write_queue_after(skb, buff, sk);
L
Linus Torvalds 已提交
1159 1160 1161 1162 1163 1164 1165 1166

	return 0;
}

/* This is similar to __pskb_pull_head() (it will go to core/skbuff.c
 * eventually). The difference is that pulled data not copied, but
 * immediately discarded.
 */
1167
static void __pskb_trim_head(struct sk_buff *skb, int len)
L
Linus Torvalds 已提交
1168
{
1169
	struct skb_shared_info *shinfo;
L
Linus Torvalds 已提交
1170 1171
	int i, k, eat;

E
Eric Dumazet 已提交
1172 1173 1174 1175 1176 1177 1178
	eat = min_t(int, len, skb_headlen(skb));
	if (eat) {
		__skb_pull(skb, eat);
		len -= eat;
		if (!len)
			return;
	}
L
Linus Torvalds 已提交
1179 1180
	eat = len;
	k = 0;
1181 1182 1183
	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
1184 1185

		if (size <= eat) {
1186
			skb_frag_unref(skb, i);
E
Eric Dumazet 已提交
1187
			eat -= size;
L
Linus Torvalds 已提交
1188
		} else {
1189
			shinfo->frags[k] = shinfo->frags[i];
L
Linus Torvalds 已提交
1190
			if (eat) {
1191 1192
				shinfo->frags[k].page_offset += eat;
				skb_frag_size_sub(&shinfo->frags[k], eat);
L
Linus Torvalds 已提交
1193 1194 1195 1196 1197
				eat = 0;
			}
			k++;
		}
	}
1198
	shinfo->nr_frags = k;
L
Linus Torvalds 已提交
1199

1200
	skb_reset_tail_pointer(skb);
L
Linus Torvalds 已提交
1201 1202 1203 1204
	skb->data_len -= len;
	skb->len = skb->data_len;
}

1205
/* Remove acked data from a packet in the transmit queue. */
L
Linus Torvalds 已提交
1206 1207
int tcp_trim_head(struct sock *sk, struct sk_buff *skb, u32 len)
{
1208
	if (skb_unclone(skb, GFP_ATOMIC))
L
Linus Torvalds 已提交
1209 1210
		return -ENOMEM;

E
Eric Dumazet 已提交
1211
	__pskb_trim_head(skb, len);
L
Linus Torvalds 已提交
1212 1213

	TCP_SKB_CB(skb)->seq += len;
1214
	skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
1215 1216 1217

	skb->truesize	     -= len;
	sk->sk_wmem_queued   -= len;
1218
	sk_mem_uncharge(sk, len);
L
Linus Torvalds 已提交
1219 1220
	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);

1221
	/* Any change of skb->len requires recalculation of tso factor. */
L
Linus Torvalds 已提交
1222
	if (tcp_skb_pcount(skb) > 1)
1223
		tcp_set_skb_tso_segs(sk, skb, tcp_skb_mss(skb));
L
Linus Torvalds 已提交
1224 1225 1226 1227

	return 0;
}

1228 1229
/* 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
1230
{
1231 1232
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1233 1234 1235 1236 1237 1238 1239
	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);

1240 1241 1242 1243 1244 1245 1246 1247
	/* 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
1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260
	/* 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;
}

1261 1262 1263 1264 1265 1266 1267 1268
/* 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
1269
/* Inverse of above */
1270
int tcp_mss_to_mtu(struct sock *sk, int mss)
J
John Heffner 已提交
1271
{
1272 1273
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1274 1275 1276 1277 1278 1279 1280
	int mtu;

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

1281 1282 1283 1284 1285 1286 1287
	/* 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
1288 1289 1290
	return mtu;
}

1291
/* MTU probing init per socket */
J
John Heffner 已提交
1292 1293 1294 1295 1296 1297 1298
void tcp_mtup_init(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct inet_connection_sock *icsk = inet_csk(sk);

	icsk->icsk_mtup.enabled = sysctl_tcp_mtu_probing > 1;
	icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp + sizeof(struct tcphdr) +
1299
			       icsk->icsk_af_ops->net_header_len;
J
John Heffner 已提交
1300 1301 1302
	icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, sysctl_tcp_base_mss);
	icsk->icsk_mtup.probe_size = 0;
}
E
Eric Dumazet 已提交
1303
EXPORT_SYMBOL(tcp_mtup_init);
J
John Heffner 已提交
1304

L
Linus Torvalds 已提交
1305 1306 1307 1308 1309 1310
/* 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
1311
   It is minimum of user_mss and mss received with SYN.
L
Linus Torvalds 已提交
1312 1313
   It also does not include TCP options.

1314
   inet_csk(sk)->icsk_pmtu_cookie is last pmtu, seen by this function.
L
Linus Torvalds 已提交
1315 1316 1317 1318 1319 1320 1321 1322 1323

   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.

1324 1325
   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
1326 1327 1328 1329
 */
unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu)
{
	struct tcp_sock *tp = tcp_sk(sk);
1330
	struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1331
	int mss_now;
L
Linus Torvalds 已提交
1332

J
John Heffner 已提交
1333 1334
	if (icsk->icsk_mtup.search_high > pmtu)
		icsk->icsk_mtup.search_high = pmtu;
L
Linus Torvalds 已提交
1335

J
John Heffner 已提交
1336
	mss_now = tcp_mtu_to_mss(sk, pmtu);
1337
	mss_now = tcp_bound_to_half_wnd(tp, mss_now);
L
Linus Torvalds 已提交
1338 1339

	/* And store cached results */
1340
	icsk->icsk_pmtu_cookie = pmtu;
J
John Heffner 已提交
1341 1342
	if (icsk->icsk_mtup.enabled)
		mss_now = min(mss_now, tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low));
1343
	tp->mss_cache = mss_now;
L
Linus Torvalds 已提交
1344 1345 1346

	return mss_now;
}
E
Eric Dumazet 已提交
1347
EXPORT_SYMBOL(tcp_sync_mss);
L
Linus Torvalds 已提交
1348 1349 1350 1351

/* Compute the current effective MSS, taking SACKs and IP options,
 * and even PMTU discovery events into account.
 */
I
Ilpo Järvinen 已提交
1352
unsigned int tcp_current_mss(struct sock *sk)
L
Linus Torvalds 已提交
1353
{
1354 1355
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct dst_entry *dst = __sk_dst_get(sk);
1356
	u32 mss_now;
1357
	unsigned int header_len;
A
Adam Langley 已提交
1358 1359
	struct tcp_out_options opts;
	struct tcp_md5sig_key *md5;
1360 1361 1362

	mss_now = tp->mss_cache;

L
Linus Torvalds 已提交
1363 1364
	if (dst) {
		u32 mtu = dst_mtu(dst);
1365
		if (mtu != inet_csk(sk)->icsk_pmtu_cookie)
L
Linus Torvalds 已提交
1366 1367 1368
			mss_now = tcp_sync_mss(sk, mtu);
	}

A
Adam Langley 已提交
1369 1370 1371 1372 1373 1374 1375 1376 1377 1378
	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;
	}
1379

L
Linus Torvalds 已提交
1380 1381 1382
	return mss_now;
}

1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404
/* RFC2861, slow part. Adjust cwnd, after it was not full during one rto.
 * As additional protections, we do not touch cwnd in retransmission phases,
 * and if application hit its sndbuf limit recently.
 */
static void tcp_cwnd_application_limited(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);

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

1405
static void tcp_cwnd_validate(struct sock *sk, u32 unsent_segs)
1406
{
1407
	struct tcp_sock *tp = tcp_sk(sk);
1408

1409 1410 1411
	tp->lsnd_pending = tp->packets_out + unsent_segs;

	if (tcp_is_cwnd_limited(sk, 0)) {
1412 1413 1414 1415 1416 1417 1418 1419
		/* Network is feed fully. */
		tp->snd_cwnd_used = 0;
		tp->snd_cwnd_stamp = tcp_time_stamp;
	} else {
		/* Network starves. */
		if (tp->packets_out > tp->snd_cwnd_used)
			tp->snd_cwnd_used = tp->packets_out;

1420 1421
		if (sysctl_tcp_slow_start_after_idle &&
		    (s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= inet_csk(sk)->icsk_rto)
1422 1423 1424 1425
			tcp_cwnd_application_limited(sk);
	}
}

E
Eric Dumazet 已提交
1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439
/* 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
1440
 */
E
Eric Dumazet 已提交
1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455
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,
1456
			    int nonagle)
E
Eric Dumazet 已提交
1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467
{
	return partial &&
		((nonagle & TCP_NAGLE_CORK) ||
		 (!nonagle && tp->packets_out && tcp_minshall_check(tp)));
}
/* 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)
1468
{
1469
	const struct tcp_sock *tp = tcp_sk(sk);
E
Eric Dumazet 已提交
1470
	u32 partial, needed, window, max_len;
1471

1472
	window = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
1473
	max_len = mss_now * max_segs;
1474

1475 1476
	if (likely(max_len <= window && skb != tcp_write_queue_tail(sk)))
		return max_len;
1477

1478 1479
	needed = min(skb->len, window);

1480 1481
	if (max_len <= needed)
		return max_len;
1482

E
Eric Dumazet 已提交
1483 1484 1485 1486 1487
	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
	 */
1488
	if (tcp_nagle_check(partial != 0, tp, nonagle))
E
Eric Dumazet 已提交
1489 1490 1491
		return needed - partial;

	return needed;
1492 1493 1494 1495 1496
}

/* 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.
 */
1497 1498
static inline unsigned int tcp_cwnd_test(const struct tcp_sock *tp,
					 const struct sk_buff *skb)
1499 1500 1501 1502
{
	u32 in_flight, cwnd;

	/* Don't be strict about the congestion window for the final FIN.  */
E
Eric Dumazet 已提交
1503 1504
	if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) &&
	    tcp_skb_pcount(skb) == 1)
1505 1506 1507 1508 1509 1510 1511 1512 1513 1514
		return 1;

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

	return 0;
}

1515
/* Initialize TSO state of a skb.
1516
 * This must be invoked the first time we consider transmitting
1517 1518
 * SKB onto the wire.
 */
1519
static int tcp_init_tso_segs(const struct sock *sk, struct sk_buff *skb,
1520
			     unsigned int mss_now)
1521 1522 1523
{
	int tso_segs = tcp_skb_pcount(skb);

1524
	if (!tso_segs || (tso_segs > 1 && tcp_skb_mss(skb) != mss_now)) {
1525
		tcp_set_skb_tso_segs(sk, skb, mss_now);
1526 1527 1528 1529 1530 1531
		tso_segs = tcp_skb_pcount(skb);
	}
	return tso_segs;
}


E
Eric Dumazet 已提交
1532
/* Return true if the Nagle test allows this packet to be
1533 1534
 * sent now.
 */
E
Eric Dumazet 已提交
1535 1536
static inline bool tcp_nagle_test(const struct tcp_sock *tp, const struct sk_buff *skb,
				  unsigned int cur_mss, int nonagle)
1537 1538 1539 1540 1541 1542 1543 1544
{
	/* 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
1545
		return true;
1546

Y
Yuchung Cheng 已提交
1547 1548
	/* 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
1549
		return true;
1550

1551
	if (!tcp_nagle_check(skb->len < cur_mss, tp, nonagle))
E
Eric Dumazet 已提交
1552
		return true;
1553

E
Eric Dumazet 已提交
1554
	return false;
1555 1556 1557
}

/* Does at least the first segment of SKB fit into the send window? */
E
Eric Dumazet 已提交
1558 1559 1560
static bool tcp_snd_wnd_test(const struct tcp_sock *tp,
			     const struct sk_buff *skb,
			     unsigned int cur_mss)
1561 1562 1563 1564 1565 1566
{
	u32 end_seq = TCP_SKB_CB(skb)->end_seq;

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

1567
	return !after(end_seq, tcp_wnd_end(tp));
1568 1569
}

1570
/* This checks if the data bearing packet SKB (usually tcp_send_head(sk))
1571 1572 1573
 * should be put on the wire right now.  If so, it returns the number of
 * packets allowed by the congestion window.
 */
1574
static unsigned int tcp_snd_test(const struct sock *sk, struct sk_buff *skb,
1575 1576
				 unsigned int cur_mss, int nonagle)
{
1577
	const struct tcp_sock *tp = tcp_sk(sk);
1578 1579
	unsigned int cwnd_quota;

1580
	tcp_init_tso_segs(sk, skb, cur_mss);
1581 1582 1583 1584 1585

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

	cwnd_quota = tcp_cwnd_test(tp, skb);
1586
	if (cwnd_quota && !tcp_snd_wnd_test(tp, skb, cur_mss))
1587 1588 1589 1590 1591
		cwnd_quota = 0;

	return cwnd_quota;
}

1592
/* Test if sending is allowed right now. */
E
Eric Dumazet 已提交
1593
bool tcp_may_send_now(struct sock *sk)
1594
{
1595
	const struct tcp_sock *tp = tcp_sk(sk);
1596
	struct sk_buff *skb = tcp_send_head(sk);
1597

E
Eric Dumazet 已提交
1598
	return skb &&
I
Ilpo Järvinen 已提交
1599
		tcp_snd_test(sk, skb, tcp_current_mss(sk),
1600
			     (tcp_skb_is_last(sk, skb) ?
E
Eric Dumazet 已提交
1601
			      tp->nonagle : TCP_NAGLE_PUSH));
1602 1603 1604 1605 1606 1607 1608 1609 1610
}

/* 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).
 */
1611
static int tso_fragment(struct sock *sk, struct sk_buff *skb, unsigned int len,
1612
			unsigned int mss_now, gfp_t gfp)
1613 1614 1615
{
	struct sk_buff *buff;
	int nlen = skb->len - len;
1616
	u8 flags;
1617 1618

	/* All of a TSO frame must be composed of paged data.  */
1619 1620
	if (skb->len != skb->data_len)
		return tcp_fragment(sk, skb, len, mss_now);
1621

1622
	buff = sk_stream_alloc_skb(sk, 0, gfp);
1623 1624 1625
	if (unlikely(buff == NULL))
		return -ENOMEM;

1626 1627
	sk->sk_wmem_queued += buff->truesize;
	sk_mem_charge(sk, buff->truesize);
1628
	buff->truesize += nlen;
1629 1630 1631 1632 1633 1634 1635 1636
	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
1637 1638 1639
	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;
1640 1641 1642 1643

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

1644
	buff->ip_summed = skb->ip_summed = CHECKSUM_PARTIAL;
1645 1646 1647
	skb_split(skb, buff, len);

	/* Fix up tso_factor for both original and new SKB.  */
1648 1649
	tcp_set_skb_tso_segs(sk, skb, mss_now);
	tcp_set_skb_tso_segs(sk, buff, mss_now);
1650 1651 1652

	/* Link BUFF into the send queue. */
	skb_header_release(buff);
1653
	tcp_insert_write_queue_after(skb, buff, sk);
1654 1655 1656 1657 1658 1659 1660 1661 1662

	return 0;
}

/* Try to defer sending, if possible, in order to minimize the amount
 * of TSO splitting we do.  View it as a kind of TSO Nagle test.
 *
 * This algorithm is from John Heffner.
 */
E
Eric Dumazet 已提交
1663
static bool tcp_tso_should_defer(struct sock *sk, struct sk_buff *skb)
1664
{
1665
	struct tcp_sock *tp = tcp_sk(sk);
1666
	const struct inet_connection_sock *icsk = inet_csk(sk);
1667
	u32 send_win, cong_win, limit, in_flight;
1668
	int win_divisor;
1669

E
Eric Dumazet 已提交
1670
	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
J
John Heffner 已提交
1671
		goto send_now;
1672

1673
	if (icsk->icsk_ca_state != TCP_CA_Open)
J
John Heffner 已提交
1674 1675 1676
		goto send_now;

	/* Defer for less than two clock ticks. */
I
Ilpo Järvinen 已提交
1677
	if (tp->tso_deferred &&
1678
	    (((u32)jiffies << 1) >> 1) - (tp->tso_deferred >> 1) > 1)
J
John Heffner 已提交
1679
		goto send_now;
1680

1681 1682
	in_flight = tcp_packets_in_flight(tp);

1683
	BUG_ON(tcp_skb_pcount(skb) <= 1 || (tp->snd_cwnd <= in_flight));
1684

1685
	send_win = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
1686 1687 1688 1689 1690 1691

	/* 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);

1692
	/* If a full-sized TSO skb can be sent, do it. */
1693
	if (limit >= min_t(unsigned int, sk->sk_gso_max_size,
E
Eric Dumazet 已提交
1694
			   tp->xmit_size_goal_segs * tp->mss_cache))
J
John Heffner 已提交
1695
		goto send_now;
1696

1697 1698 1699 1700
	/* 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;

1701 1702
	win_divisor = ACCESS_ONCE(sysctl_tcp_tso_win_divisor);
	if (win_divisor) {
1703 1704 1705 1706 1707
		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.
		 */
1708
		chunk /= win_divisor;
1709
		if (limit >= chunk)
J
John Heffner 已提交
1710
			goto send_now;
1711 1712 1713 1714 1715 1716
	} 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.
		 */
1717
		if (limit > tcp_max_tso_deferred_mss(tp) * tp->mss_cache)
J
John Heffner 已提交
1718
			goto send_now;
1719 1720
	}

E
Eric Dumazet 已提交
1721 1722 1723 1724 1725
	/* Ok, it looks like it is advisable to defer.
	 * Do not rearm the timer if already set to not break TCP ACK clocking.
	 */
	if (!tp->tso_deferred)
		tp->tso_deferred = 1 | (jiffies << 1);
J
John Heffner 已提交
1726

E
Eric Dumazet 已提交
1727
	return true;
J
John Heffner 已提交
1728 1729 1730

send_now:
	tp->tso_deferred = 0;
E
Eric Dumazet 已提交
1731
	return false;
1732 1733
}

J
John Heffner 已提交
1734
/* Create a new MTU probe if we are ready.
1735 1736 1737 1738
 * 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
1739 1740
 * Returns 0 if we should wait to probe (no cwnd available),
 *         1 if a probe was sent,
1741 1742
 *         -1 otherwise
 */
J
John Heffner 已提交
1743 1744 1745 1746 1747 1748 1749
static int tcp_mtu_probe(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct inet_connection_sock *icsk = inet_csk(sk);
	struct sk_buff *skb, *nskb, *next;
	int len;
	int probe_size;
1750
	int size_needed;
J
John Heffner 已提交
1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761
	int copy;
	int mss_now;

	/* Not currently probing/verifying,
	 * not in recovery,
	 * have enough cwnd, and
	 * not SACKing (the variable headers throw things off) */
	if (!icsk->icsk_mtup.enabled ||
	    icsk->icsk_mtup.probe_size ||
	    inet_csk(sk)->icsk_ca_state != TCP_CA_Open ||
	    tp->snd_cwnd < 11 ||
1762
	    tp->rx_opt.num_sacks || tp->rx_opt.dsack)
J
John Heffner 已提交
1763 1764 1765
		return -1;

	/* Very simple search strategy: just double the MSS. */
I
Ilpo Järvinen 已提交
1766
	mss_now = tcp_current_mss(sk);
1767
	probe_size = 2 * tp->mss_cache;
1768
	size_needed = probe_size + (tp->reordering + 1) * tp->mss_cache;
J
John Heffner 已提交
1769 1770 1771 1772 1773 1774
	if (probe_size > tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_high)) {
		/* TODO: set timer for probe_converge_event */
		return -1;
	}

	/* Have enough data in the send queue to probe? */
1775
	if (tp->write_seq - tp->snd_nxt < size_needed)
J
John Heffner 已提交
1776 1777
		return -1;

1778 1779
	if (tp->snd_wnd < size_needed)
		return -1;
1780
	if (after(tp->snd_nxt + size_needed, tcp_wnd_end(tp)))
1781
		return 0;
J
John Heffner 已提交
1782

1783 1784 1785
	/* 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
1786 1787 1788 1789 1790 1791 1792 1793
			return -1;
		else
			return 0;
	}

	/* We're allowed to probe.  Build it now. */
	if ((nskb = sk_stream_alloc_skb(sk, probe_size, GFP_ATOMIC)) == NULL)
		return -1;
1794 1795
	sk->sk_wmem_queued += nskb->truesize;
	sk_mem_charge(sk, nskb->truesize);
J
John Heffner 已提交
1796

1797
	skb = tcp_send_head(sk);
J
John Heffner 已提交
1798 1799 1800

	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
1801
	TCP_SKB_CB(nskb)->tcp_flags = TCPHDR_ACK;
J
John Heffner 已提交
1802 1803
	TCP_SKB_CB(nskb)->sacked = 0;
	nskb->csum = 0;
1804
	nskb->ip_summed = skb->ip_summed;
J
John Heffner 已提交
1805

1806 1807
	tcp_insert_write_queue_before(nskb, skb, sk);

J
John Heffner 已提交
1808
	len = 0;
1809
	tcp_for_write_queue_from_safe(skb, next, sk) {
J
John Heffner 已提交
1810 1811 1812 1813 1814
		copy = min_t(int, skb->len, probe_size - len);
		if (nskb->ip_summed)
			skb_copy_bits(skb, 0, skb_put(nskb, copy), copy);
		else
			nskb->csum = skb_copy_and_csum_bits(skb, 0,
1815 1816
							    skb_put(nskb, copy),
							    copy, nskb->csum);
J
John Heffner 已提交
1817 1818 1819 1820

		if (skb->len <= copy) {
			/* We've eaten all the data from this skb.
			 * Throw it away. */
E
Eric Dumazet 已提交
1821
			TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;
1822
			tcp_unlink_write_queue(skb, sk);
1823
			sk_wmem_free_skb(sk, skb);
J
John Heffner 已提交
1824
		} else {
E
Eric Dumazet 已提交
1825
			TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags &
C
Changli Gao 已提交
1826
						   ~(TCPHDR_FIN|TCPHDR_PSH);
J
John Heffner 已提交
1827 1828
			if (!skb_shinfo(skb)->nr_frags) {
				skb_pull(skb, copy);
1829
				if (skb->ip_summed != CHECKSUM_PARTIAL)
1830 1831
					skb->csum = csum_partial(skb->data,
								 skb->len, 0);
J
John Heffner 已提交
1832 1833 1834 1835 1836 1837 1838 1839
			} else {
				__pskb_trim_head(skb, copy);
				tcp_set_skb_tso_segs(sk, skb, mss_now);
			}
			TCP_SKB_CB(skb)->seq += copy;
		}

		len += copy;
1840 1841 1842

		if (len >= probe_size)
			break;
J
John Heffner 已提交
1843 1844 1845 1846 1847 1848 1849 1850
	}
	tcp_init_tso_segs(sk, nskb, nskb->len);

	/* We're ready to send.  If this fails, the probe will
	 * be resegmented into mss-sized pieces by tcp_write_xmit(). */
	TCP_SKB_CB(nskb)->when = tcp_time_stamp;
	if (!tcp_transmit_skb(sk, nskb, 1, GFP_ATOMIC)) {
		/* Decrement cwnd here because we are sending
1851
		 * effectively two packets. */
J
John Heffner 已提交
1852
		tp->snd_cwnd--;
1853
		tcp_event_new_data_sent(sk, nskb);
J
John Heffner 已提交
1854 1855

		icsk->icsk_mtup.probe_size = tcp_mss_to_mtu(sk, nskb->len);
1856 1857
		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
1858 1859 1860 1861 1862 1863 1864

		return 1;
	}

	return -1;
}

L
Linus Torvalds 已提交
1865 1866 1867 1868
/* This routine writes packets to the network.  It advances the
 * send_head.  This happens as incoming acks open up the remote
 * window for us.
 *
1869 1870 1871 1872
 * 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
1873 1874 1875
 * 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
1876 1877
 * 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
1878
 */
E
Eric Dumazet 已提交
1879 1880
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
1881 1882
{
	struct tcp_sock *tp = tcp_sk(sk);
1883
	struct sk_buff *skb;
1884
	unsigned int tso_segs, sent_pkts, unsent_segs = 0;
1885
	int cwnd_quota;
J
John Heffner 已提交
1886
	int result;
L
Linus Torvalds 已提交
1887

1888
	sent_pkts = 0;
J
John Heffner 已提交
1889

1890 1891 1892 1893
	if (!push_one) {
		/* Do MTU probing. */
		result = tcp_mtu_probe(sk);
		if (!result) {
E
Eric Dumazet 已提交
1894
			return false;
1895 1896 1897
		} else if (result > 0) {
			sent_pkts = 1;
		}
J
John Heffner 已提交
1898 1899
	}

1900
	while ((skb = tcp_send_head(sk))) {
1901 1902
		unsigned int limit;

1903
		tso_segs = tcp_init_tso_segs(sk, skb, mss_now);
1904
		BUG_ON(!tso_segs);
1905

1906 1907 1908
		if (unlikely(tp->repair) && tp->repair_queue == TCP_SEND_QUEUE)
			goto repair; /* Skip network transmission */

1909
		cwnd_quota = tcp_cwnd_test(tp, skb);
N
Nandita Dukkipati 已提交
1910 1911 1912 1913 1914 1915 1916
		if (!cwnd_quota) {
			if (push_one == 2)
				/* Force out a loss probe pkt. */
				cwnd_quota = 1;
			else
				break;
		}
1917 1918 1919 1920

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

1921 1922 1923 1924 1925 1926
		if (tso_segs == 1) {
			if (unlikely(!tcp_nagle_test(tp, skb, mss_now,
						     (tcp_skb_is_last(sk, skb) ?
						      nonagle : TCP_NAGLE_PUSH))))
				break;
		} else {
1927
			if (!push_one && tcp_tso_should_defer(sk, skb))
1928
				goto compute_unsent_segs;
1929
		}
1930

E
Eric Dumazet 已提交
1931 1932 1933 1934 1935 1936
		/* TCP Small Queues :
		 * Control number of packets in qdisc/devices to two packets / or ~1 ms.
		 * This allows for :
		 *  - better RTT estimation and ACK scheduling
		 *  - faster recovery
		 *  - high rates
1937 1938 1939
		 * Alas, some drivers / subsystems require a fair amount
		 * of queued bytes to ensure line rate.
		 * One example is wifi aggregation (802.11 AMPDU)
E
Eric Dumazet 已提交
1940
		 */
1941 1942
		limit = max_t(unsigned int, sysctl_tcp_limit_output_bytes,
			      sk->sk_pacing_rate >> 10);
E
Eric Dumazet 已提交
1943 1944

		if (atomic_read(&sk->sk_wmem_alloc) > limit) {
E
Eric Dumazet 已提交
1945
			set_bit(TSQ_THROTTLED, &tp->tsq_flags);
J
John Ogness 已提交
1946 1947 1948 1949 1950 1951 1952
			/* It is possible TX completion already happened
			 * before we set TSQ_THROTTLED, so we must
			 * test again the condition.
			 * We abuse smp_mb__after_clear_bit() because
			 * there is no smp_mb__after_set_bit() yet
			 */
			smp_mb__after_clear_bit();
1953 1954 1955 1956 1957 1958
			if (atomic_read(&sk->sk_wmem_alloc) > limit) {
				u32 unsent_bytes;

compute_unsent_segs:
				unsent_bytes = tp->write_seq - tp->snd_nxt;
				unsent_segs = DIV_ROUND_UP(unsent_bytes, mss_now);
J
John Ogness 已提交
1959
				break;
1960
			}
E
Eric Dumazet 已提交
1961
		}
E
Eric Dumazet 已提交
1962

1963
		limit = mss_now;
1964
		if (tso_segs > 1 && !tcp_urg_mode(tp))
1965
			limit = tcp_mss_split_point(sk, skb, mss_now,
1966 1967
						    min_t(unsigned int,
							  cwnd_quota,
E
Eric Dumazet 已提交
1968 1969
							  sk->sk_gso_max_segs),
						    nonagle);
L
Linus Torvalds 已提交
1970

1971
		if (skb->len > limit &&
1972
		    unlikely(tso_fragment(sk, skb, limit, mss_now, gfp)))
1973 1974
			break;

1975
		TCP_SKB_CB(skb)->when = tcp_time_stamp;
1976

1977
		if (unlikely(tcp_transmit_skb(sk, skb, 1, gfp)))
1978
			break;
L
Linus Torvalds 已提交
1979

1980
repair:
1981 1982 1983
		/* Advance the send_head.  This one is sent out.
		 * This call will increment packets_out.
		 */
1984
		tcp_event_new_data_sent(sk, skb);
L
Linus Torvalds 已提交
1985

1986
		tcp_minshall_update(tp, mss_now, skb);
1987
		sent_pkts += tcp_skb_pcount(skb);
1988 1989 1990

		if (push_one)
			break;
1991
	}
L
Linus Torvalds 已提交
1992

1993
	if (likely(sent_pkts)) {
1994 1995
		if (tcp_in_cwnd_reduction(sk))
			tp->prr_out += sent_pkts;
N
Nandita Dukkipati 已提交
1996 1997 1998 1999

		/* Send one loss probe per tail loss episode. */
		if (push_one != 2)
			tcp_schedule_loss_probe(sk);
2000
		tcp_cwnd_validate(sk, unsent_segs);
E
Eric Dumazet 已提交
2001
		return false;
L
Linus Torvalds 已提交
2002
	}
N
Nandita Dukkipati 已提交
2003 2004 2005 2006 2007 2008 2009 2010
	return (push_one == 2) || (!tp->packets_out && tcp_send_head(sk));
}

bool tcp_schedule_loss_probe(struct sock *sk)
{
	struct inet_connection_sock *icsk = inet_csk(sk);
	struct tcp_sock *tp = tcp_sk(sk);
	u32 timeout, tlp_time_stamp, rto_time_stamp;
2011
	u32 rtt = usecs_to_jiffies(tp->srtt_us >> 3);
N
Nandita Dukkipati 已提交
2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032

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

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

	/* Schedule a loss probe in 2*RTT for SACK capable connections
	 * in Open state, that are either limited by cwnd or application.
	 */
2033
	if (sysctl_tcp_early_retrans < 3 || !tp->srtt_us || !tp->packets_out ||
N
Nandita Dukkipati 已提交
2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063
	    !tcp_is_sack(tp) || inet_csk(sk)->icsk_ca_state != TCP_CA_Open)
		return false;

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

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

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

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

2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082
/* 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 !
 * Note: This is called from BH context only.
 */
static bool skb_still_in_host_queue(const struct sock *sk,
				    const struct sk_buff *skb)
{
	const struct sk_buff *fclone = skb + 1;

	if (unlikely(skb->fclone == SKB_FCLONE_ORIG &&
		     fclone->fclone == SKB_FCLONE_CLONE)) {
		NET_INC_STATS_BH(sock_net(sk),
				 LINUX_MIB_TCPSPURIOUS_RTX_HOSTQUEUES);
		return true;
	}
	return false;
}

N
Nandita Dukkipati 已提交
2083 2084 2085 2086 2087
/* When probe timeout (PTO) fires, send a new segment if one exists, else
 * retransmit the last segment.
 */
void tcp_send_loss_probe(struct sock *sk)
{
N
Nandita Dukkipati 已提交
2088
	struct tcp_sock *tp = tcp_sk(sk);
N
Nandita Dukkipati 已提交
2089 2090 2091 2092 2093 2094 2095 2096 2097 2098
	struct sk_buff *skb;
	int pcount;
	int mss = tcp_current_mss(sk);
	int err = -1;

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

N
Nandita Dukkipati 已提交
2099 2100 2101 2102
	/* At most one outstanding TLP retransmission. */
	if (tp->tlp_high_seq)
		goto rearm_timer;

N
Nandita Dukkipati 已提交
2103 2104 2105 2106 2107
	/* Retransmit last segment. */
	skb = tcp_write_queue_tail(sk);
	if (WARN_ON(!skb))
		goto rearm_timer;

2108 2109 2110
	if (skb_still_in_host_queue(sk, skb))
		goto rearm_timer;

N
Nandita Dukkipati 已提交
2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127
	pcount = tcp_skb_pcount(skb);
	if (WARN_ON(!pcount))
		goto rearm_timer;

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

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

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

N
Nandita Dukkipati 已提交
2128 2129 2130 2131
	/* Record snd_nxt for loss detection. */
	if (likely(!err))
		tp->tlp_high_seq = tp->snd_nxt;

N
Nandita Dukkipati 已提交
2132 2133 2134 2135 2136 2137 2138 2139
rearm_timer:
	inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
				  inet_csk(sk)->icsk_rto,
				  TCP_RTO_MAX);

	if (likely(!err))
		NET_INC_STATS_BH(sock_net(sk),
				 LINUX_MIB_TCPLOSSPROBES);
L
Linus Torvalds 已提交
2140 2141
}

2142 2143 2144 2145
/* 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.
 */
2146 2147
void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
			       int nonagle)
2148
{
2149 2150 2151 2152 2153 2154 2155
	/* 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;

2156 2157
	if (tcp_write_xmit(sk, cur_mss, nonagle, 0,
			   sk_gfp_atomic(sk, GFP_ATOMIC)))
2158
		tcp_check_probe_timer(sk);
2159 2160
}

2161 2162 2163 2164 2165
/* 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)
{
2166
	struct sk_buff *skb = tcp_send_head(sk);
2167 2168 2169

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

2170
	tcp_write_xmit(sk, mss_now, TCP_NAGLE_PUSH, 1, sk->sk_allocation);
2171 2172
}

L
Linus Torvalds 已提交
2173 2174
/* This function returns the amount that we can raise the
 * usable window based on the following constraints
2175
 *
L
Linus Torvalds 已提交
2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193
 * 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.
2194
 *
L
Linus Torvalds 已提交
2195 2196
 * Of course, if the sender implements sender side SWS prevention
 * then this will not be a problem.
2197
 *
L
Linus Torvalds 已提交
2198
 * BSD seems to make the following compromise:
2199
 *
L
Linus Torvalds 已提交
2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226
 *	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)
{
2227
	struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
2228
	struct tcp_sock *tp = tcp_sk(sk);
S
Stephen Hemminger 已提交
2229
	/* MSS for the peer's data.  Previous versions used mss_clamp
L
Linus Torvalds 已提交
2230 2231 2232 2233 2234
	 * 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
	 */
2235
	int mss = icsk->icsk_ack.rcv_mss;
L
Linus Torvalds 已提交
2236
	int free_space = tcp_space(sk);
2237 2238
	int allowed_space = tcp_full_space(sk);
	int full_space = min_t(int, tp->window_clamp, allowed_space);
L
Linus Torvalds 已提交
2239 2240 2241
	int window;

	if (mss > full_space)
2242
		mss = full_space;
L
Linus Torvalds 已提交
2243

2244
	if (free_space < (full_space >> 1)) {
2245
		icsk->icsk_ack.quick = 0;
L
Linus Torvalds 已提交
2246

2247
		if (sk_under_memory_pressure(sk))
2248 2249
			tp->rcv_ssthresh = min(tp->rcv_ssthresh,
					       4U * tp->advmss);
L
Linus Torvalds 已提交
2250

2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263
		/* 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
2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293
			return 0;
	}

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

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

		/* Advertise enough space so that it won't get scaled away.
		 * Import case: prevent zero window announcement if
		 * 1<<rcv_wscale > mss.
		 */
		if (((window >> tp->rx_opt.rcv_wscale) << tp->rx_opt.rcv_wscale) != window)
			window = (((window >> tp->rx_opt.rcv_wscale) + 1)
				  << tp->rx_opt.rcv_wscale);
	} else {
		/* Get the largest window that is a nice multiple of mss.
		 * Window clamp already applied above.
		 * If our current window offering is within 1 mss of the
		 * free space we just keep it. This prevents the divide
		 * and multiply from happening most of the time.
		 * We also don't do any window rounding when the free space
		 * is too small.
		 */
		if (window <= free_space - mss || window > free_space)
2294
			window = (free_space / mss) * mss;
2295
		else if (mss == full_space &&
2296
			 free_space > window + (full_space >> 1))
2297
			window = free_space;
L
Linus Torvalds 已提交
2298 2299 2300 2301 2302
	}

	return window;
}

2303 2304
/* Collapses two adjacent SKB's during retransmission. */
static void tcp_collapse_retrans(struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
2305 2306
{
	struct tcp_sock *tp = tcp_sk(sk);
2307
	struct sk_buff *next_skb = tcp_write_queue_next(sk, skb);
2308
	int skb_size, next_skb_size;
L
Linus Torvalds 已提交
2309

2310 2311
	skb_size = skb->len;
	next_skb_size = next_skb->len;
L
Linus Torvalds 已提交
2312

2313
	BUG_ON(tcp_skb_pcount(skb) != 1 || tcp_skb_pcount(next_skb) != 1);
2314

2315
	tcp_highest_sack_combine(sk, next_skb, skb);
L
Linus Torvalds 已提交
2316

2317
	tcp_unlink_write_queue(next_skb, sk);
L
Linus Torvalds 已提交
2318

2319 2320
	skb_copy_from_linear_data(next_skb, skb_put(skb, next_skb_size),
				  next_skb_size);
L
Linus Torvalds 已提交
2321

2322 2323
	if (next_skb->ip_summed == CHECKSUM_PARTIAL)
		skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
2324

2325 2326
	if (skb->ip_summed != CHECKSUM_PARTIAL)
		skb->csum = csum_block_add(skb->csum, next_skb->csum, skb_size);
L
Linus Torvalds 已提交
2327

2328 2329
	/* 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
2330

2331
	/* Merge over control information. This moves PSH/FIN etc. over */
E
Eric Dumazet 已提交
2332
	TCP_SKB_CB(skb)->tcp_flags |= TCP_SKB_CB(next_skb)->tcp_flags;
2333 2334 2335 2336 2337 2338 2339

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

	/* changed transmit queue under us so clear hints */
2340 2341 2342
	tcp_clear_retrans_hints_partial(tp);
	if (next_skb == tp->retransmit_skb_hint)
		tp->retransmit_skb_hint = skb;
2343

2344 2345
	tcp_adjust_pcount(sk, next_skb, tcp_skb_pcount(next_skb));

2346
	sk_wmem_free_skb(sk, next_skb);
L
Linus Torvalds 已提交
2347 2348
}

2349
/* Check if coalescing SKBs is legal. */
E
Eric Dumazet 已提交
2350
static bool tcp_can_collapse(const struct sock *sk, const struct sk_buff *skb)
2351 2352
{
	if (tcp_skb_pcount(skb) > 1)
E
Eric Dumazet 已提交
2353
		return false;
2354 2355
	/* TODO: SACK collapsing could be used to remove this condition */
	if (skb_shinfo(skb)->nr_frags != 0)
E
Eric Dumazet 已提交
2356
		return false;
2357
	if (skb_cloned(skb))
E
Eric Dumazet 已提交
2358
		return false;
2359
	if (skb == tcp_send_head(sk))
E
Eric Dumazet 已提交
2360
		return false;
2361 2362
	/* Some heurestics for collapsing over SACK'd could be invented */
	if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)
E
Eric Dumazet 已提交
2363
		return false;
2364

E
Eric Dumazet 已提交
2365
	return true;
2366 2367
}

2368 2369 2370
/* Collapse packets in the retransmit queue to make to create
 * less packets on the wire. This is only done on retransmission.
 */
2371 2372 2373 2374 2375
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
2376
	bool first = true;
2377 2378 2379

	if (!sysctl_tcp_retrans_collapse)
		return;
E
Eric Dumazet 已提交
2380
	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)
2381 2382 2383 2384 2385 2386 2387 2388 2389
		return;

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

		space -= skb->len;

		if (first) {
E
Eric Dumazet 已提交
2390
			first = false;
2391 2392 2393 2394 2395 2396 2397 2398
			continue;
		}

		if (space < 0)
			break;
		/* Punt if not enough space exists in the first SKB for
		 * the data in the second
		 */
2399
		if (skb->len > skb_availroom(to))
2400 2401 2402 2403 2404 2405 2406 2407 2408
			break;

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

		tcp_collapse_retrans(sk, to);
	}
}

L
Linus Torvalds 已提交
2409 2410 2411 2412
/* 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.
 */
2413
int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
2414 2415
{
	struct tcp_sock *tp = tcp_sk(sk);
J
John Heffner 已提交
2416
	struct inet_connection_sock *icsk = inet_csk(sk);
2417
	unsigned int cur_mss;
2418
	int err;
L
Linus Torvalds 已提交
2419

J
John Heffner 已提交
2420 2421 2422 2423 2424
	/* Inconslusive MTU probe */
	if (icsk->icsk_mtup.probe_size) {
		icsk->icsk_mtup.probe_size = 0;
	}

L
Linus Torvalds 已提交
2425
	/* Do not sent more than we queued. 1/4 is reserved for possible
S
Stephen Hemminger 已提交
2426
	 * copying overhead: fragmentation, tunneling, mangling etc.
L
Linus Torvalds 已提交
2427 2428 2429 2430 2431
	 */
	if (atomic_read(&sk->sk_wmem_alloc) >
	    min(sk->sk_wmem_queued + (sk->sk_wmem_queued >> 2), sk->sk_sndbuf))
		return -EAGAIN;

2432 2433 2434
	if (skb_still_in_host_queue(sk, skb))
		return -EBUSY;

L
Linus Torvalds 已提交
2435 2436 2437 2438 2439 2440 2441
	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;
	}

2442 2443 2444
	if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk))
		return -EHOSTUNREACH; /* Routing failure or similar. */

I
Ilpo Järvinen 已提交
2445
	cur_mss = tcp_current_mss(sk);
2446

L
Linus Torvalds 已提交
2447 2448 2449 2450 2451
	/* 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.
	 */
2452 2453
	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
2454 2455 2456
		return -EAGAIN;

	if (skb->len > cur_mss) {
2457
		if (tcp_fragment(sk, skb, cur_mss, cur_mss))
L
Linus Torvalds 已提交
2458
			return -ENOMEM; /* We'll try again later. */
2459
	} else {
2460 2461 2462
		int oldpcount = tcp_skb_pcount(skb);

		if (unlikely(oldpcount > 1)) {
2463 2464
			if (skb_unclone(skb, GFP_ATOMIC))
				return -ENOMEM;
2465 2466 2467
			tcp_init_tso_segs(sk, skb, cur_mss);
			tcp_adjust_pcount(sk, skb, oldpcount - tcp_skb_pcount(skb));
		}
L
Linus Torvalds 已提交
2468 2469
	}

2470
	tcp_retrans_try_collapse(sk, skb, cur_mss);
L
Linus Torvalds 已提交
2471 2472 2473 2474 2475 2476

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

2477 2478 2479 2480 2481 2482
	/* make sure skb->data is aligned on arches that require it
	 * and check if ack-trimming & collapsing extended the headroom
	 * beyond what csum_start can cover.
	 */
	if (unlikely((NET_IP_ALIGN && ((unsigned long)skb->data & 3)) ||
		     skb_headroom(skb) >= 0xFFFF)) {
E
Eric Dumazet 已提交
2483 2484
		struct sk_buff *nskb = __pskb_copy(skb, MAX_TCP_HEADER,
						   GFP_ATOMIC);
2485 2486
		err = nskb ? tcp_transmit_skb(sk, nskb, 0, GFP_ATOMIC) :
			     -ENOBUFS;
E
Eric Dumazet 已提交
2487
	} else {
2488
		err = tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
E
Eric Dumazet 已提交
2489
	}
2490 2491 2492 2493

	if (likely(!err))
		TCP_SKB_CB(skb)->sacked |= TCPCB_EVER_RETRANS;
	return err;
2494 2495 2496 2497 2498 2499
}

int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
{
	struct tcp_sock *tp = tcp_sk(sk);
	int err = __tcp_retransmit_skb(sk, skb);
L
Linus Torvalds 已提交
2500 2501 2502

	if (err == 0) {
		/* Update global TCP statistics. */
P
Pavel Emelyanov 已提交
2503
		TCP_INC_STATS(sock_net(sk), TCP_MIB_RETRANSSEGS);
2504 2505
		if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)
			NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSYNRETRANS);
L
Linus Torvalds 已提交
2506 2507 2508
		tp->total_retrans++;

#if FASTRETRANS_DEBUG > 0
2509
		if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) {
2510
			net_dbg_ratelimited("retrans_out leaked\n");
L
Linus Torvalds 已提交
2511 2512
		}
#endif
2513 2514
		if (!tp->retrans_out)
			tp->lost_retrans_low = tp->snd_nxt;
L
Linus Torvalds 已提交
2515 2516 2517 2518 2519 2520 2521
		TCP_SKB_CB(skb)->sacked |= TCPCB_RETRANS;
		tp->retrans_out += tcp_skb_pcount(skb);

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

Y
Yuchung Cheng 已提交
2522
		tp->undo_retrans += tcp_skb_pcount(skb);
L
Linus Torvalds 已提交
2523 2524 2525 2526 2527

		/* snd_nxt is stored to detect loss of retransmitted segment,
		 * see tcp_input.c tcp_sacktag_write_queue().
		 */
		TCP_SKB_CB(skb)->ack_seq = tp->snd_nxt;
2528
	} else if (err != -EBUSY) {
2529
		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPRETRANSFAIL);
L
Linus Torvalds 已提交
2530 2531 2532 2533
	}
	return err;
}

2534 2535 2536
/* Check if we forward retransmits are possible in the current
 * window/congestion state.
 */
E
Eric Dumazet 已提交
2537
static bool tcp_can_forward_retransmit(struct sock *sk)
2538 2539
{
	const struct inet_connection_sock *icsk = inet_csk(sk);
2540
	const struct tcp_sock *tp = tcp_sk(sk);
2541 2542 2543

	/* Forward retransmissions are possible only during Recovery. */
	if (icsk->icsk_ca_state != TCP_CA_Recovery)
E
Eric Dumazet 已提交
2544
		return false;
2545 2546 2547

	/* No forward retransmissions in Reno are possible. */
	if (tcp_is_reno(tp))
E
Eric Dumazet 已提交
2548
		return false;
2549 2550 2551 2552 2553 2554 2555 2556 2557 2558

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

	if (tcp_may_send_now(sk))
E
Eric Dumazet 已提交
2559
		return false;
2560

E
Eric Dumazet 已提交
2561
	return true;
2562 2563
}

L
Linus Torvalds 已提交
2564 2565 2566 2567 2568 2569 2570 2571 2572 2573
/* 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)
{
2574
	const struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
2575 2576
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb;
2577
	struct sk_buff *hole = NULL;
2578
	u32 last_lost;
2579
	int mib_idx;
2580
	int fwd_rexmitting = 0;
2581

2582 2583 2584
	if (!tp->packets_out)
		return;

2585 2586 2587
	if (!tp->lost_out)
		tp->retransmit_high = tp->snd_una;

2588
	if (tp->retransmit_skb_hint) {
2589
		skb = tp->retransmit_skb_hint;
2590 2591 2592 2593
		last_lost = TCP_SKB_CB(skb)->end_seq;
		if (after(last_lost, tp->retransmit_high))
			last_lost = tp->retransmit_high;
	} else {
2594
		skb = tcp_write_queue_head(sk);
2595 2596
		last_lost = tp->snd_una;
	}
L
Linus Torvalds 已提交
2597

2598 2599
	tcp_for_write_queue_from(skb, sk) {
		__u8 sacked = TCP_SKB_CB(skb)->sacked;
L
Linus Torvalds 已提交
2600

2601 2602 2603
		if (skb == tcp_send_head(sk))
			break;
		/* we could do better than to assign each time */
2604 2605
		if (hole == NULL)
			tp->retransmit_skb_hint = skb;
2606 2607 2608 2609 2610 2611 2612 2613 2614 2615

		/* Assume this retransmit will generate
		 * only one packet for congestion window
		 * calculation purposes.  This works because
		 * tcp_retransmit_skb() will chop up the
		 * packet to be MSS sized and all the
		 * packet counting works out.
		 */
		if (tcp_packets_in_flight(tp) >= tp->snd_cwnd)
			return;
L
Linus Torvalds 已提交
2616

2617 2618 2619 2620 2621
		if (fwd_rexmitting) {
begin_fwd:
			if (!before(TCP_SKB_CB(skb)->seq, tcp_highest_sack_seq(tp)))
				break;
			mib_idx = LINUX_MIB_TCPFORWARDRETRANS;
2622

2623
		} else if (!before(TCP_SKB_CB(skb)->seq, tp->retransmit_high)) {
2624
			tp->retransmit_high = last_lost;
2625 2626 2627 2628 2629 2630 2631 2632 2633
			if (!tcp_can_forward_retransmit(sk))
				break;
			/* Backtrack if necessary to non-L'ed skb */
			if (hole != NULL) {
				skb = hole;
				hole = NULL;
			}
			fwd_rexmitting = 1;
			goto begin_fwd;
L
Linus Torvalds 已提交
2634

2635
		} else if (!(sacked & TCPCB_LOST)) {
2636
			if (hole == NULL && !(sacked & (TCPCB_SACKED_RETRANS|TCPCB_SACKED_ACKED)))
2637 2638
				hole = skb;
			continue;
L
Linus Torvalds 已提交
2639

2640
		} else {
2641
			last_lost = TCP_SKB_CB(skb)->end_seq;
2642 2643 2644 2645 2646
			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
2647

2648
		if (sacked & (TCPCB_SACKED_ACKED|TCPCB_SACKED_RETRANS))
L
Linus Torvalds 已提交
2649 2650
			continue;

2651
		if (tcp_retransmit_skb(sk, skb))
2652
			return;
2653

2654
		NET_INC_STATS_BH(sock_net(sk), mib_idx);
L
Linus Torvalds 已提交
2655

2656
		if (tcp_in_cwnd_reduction(sk))
2657 2658
			tp->prr_out += tcp_skb_pcount(skb);

2659
		if (skb == tcp_write_queue_head(sk))
2660 2661 2662
			inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
						  inet_csk(sk)->icsk_rto,
						  TCP_RTO_MAX);
L
Linus Torvalds 已提交
2663 2664 2665 2666 2667 2668 2669 2670
	}
}

/* Send a fin.  The caller locks the socket for us.  This cannot be
 * allowed to fail queueing a FIN frame under any circumstances.
 */
void tcp_send_fin(struct sock *sk)
{
2671
	struct tcp_sock *tp = tcp_sk(sk);
2672
	struct sk_buff *skb = tcp_write_queue_tail(sk);
L
Linus Torvalds 已提交
2673
	int mss_now;
2674

L
Linus Torvalds 已提交
2675 2676 2677 2678
	/* Optimization, tack on the FIN if we have a queue of
	 * unsent frames.  But be careful about outgoing SACKS
	 * and IP options.
	 */
I
Ilpo Järvinen 已提交
2679
	mss_now = tcp_current_mss(sk);
L
Linus Torvalds 已提交
2680

2681
	if (tcp_send_head(sk) != NULL) {
E
Eric Dumazet 已提交
2682
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_FIN;
L
Linus Torvalds 已提交
2683 2684 2685 2686 2687
		TCP_SKB_CB(skb)->end_seq++;
		tp->write_seq++;
	} else {
		/* Socket is locked, keep trying until memory is available. */
		for (;;) {
2688 2689
			skb = alloc_skb_fclone(MAX_TCP_HEADER,
					       sk->sk_allocation);
L
Linus Torvalds 已提交
2690 2691 2692 2693 2694 2695 2696 2697
			if (skb)
				break;
			yield();
		}

		/* Reserve space for headers and prepare control bits. */
		skb_reserve(skb, MAX_TCP_HEADER);
		/* FIN eats a sequence byte, write_seq advanced by tcp_queue_skb(). */
2698
		tcp_init_nondata_skb(skb, tp->write_seq,
C
Changli Gao 已提交
2699
				     TCPHDR_ACK | TCPHDR_FIN);
L
Linus Torvalds 已提交
2700 2701
		tcp_queue_skb(sk, skb);
	}
2702
	__tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_OFF);
L
Linus Torvalds 已提交
2703 2704 2705 2706 2707
}

/* 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
2708
 * by RFC 2525, section 2.17.  -DaveM
L
Linus Torvalds 已提交
2709
 */
A
Al Viro 已提交
2710
void tcp_send_active_reset(struct sock *sk, gfp_t priority)
L
Linus Torvalds 已提交
2711 2712 2713 2714 2715 2716
{
	struct sk_buff *skb;

	/* NOTE: No TCP options attached and we never retransmit this. */
	skb = alloc_skb(MAX_TCP_HEADER, priority);
	if (!skb) {
P
Pavel Emelyanov 已提交
2717
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTFAILED);
L
Linus Torvalds 已提交
2718 2719 2720 2721 2722
		return;
	}

	/* Reserve space for headers and prepare control bits. */
	skb_reserve(skb, MAX_TCP_HEADER);
2723
	tcp_init_nondata_skb(skb, tcp_acceptable_seq(sk),
C
Changli Gao 已提交
2724
			     TCPHDR_ACK | TCPHDR_RST);
L
Linus Torvalds 已提交
2725 2726
	/* Send it off. */
	TCP_SKB_CB(skb)->when = tcp_time_stamp;
2727
	if (tcp_transmit_skb(sk, skb, 0, priority))
P
Pavel Emelyanov 已提交
2728
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTFAILED);
2729

P
Pavel Emelyanov 已提交
2730
	TCP_INC_STATS(sock_net(sk), TCP_MIB_OUTRSTS);
L
Linus Torvalds 已提交
2731 2732
}

2733 2734
/* 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
2735 2736 2737 2738 2739 2740
 * 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)
{
2741
	struct sk_buff *skb;
L
Linus Torvalds 已提交
2742

2743
	skb = tcp_write_queue_head(sk);
E
Eric Dumazet 已提交
2744
	if (skb == NULL || !(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
2745
		pr_debug("%s: wrong queue state\n", __func__);
L
Linus Torvalds 已提交
2746 2747
		return -EFAULT;
	}
E
Eric Dumazet 已提交
2748
	if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)) {
L
Linus Torvalds 已提交
2749 2750 2751 2752
		if (skb_cloned(skb)) {
			struct sk_buff *nskb = skb_copy(skb, GFP_ATOMIC);
			if (nskb == NULL)
				return -ENOMEM;
2753
			tcp_unlink_write_queue(skb, sk);
L
Linus Torvalds 已提交
2754
			skb_header_release(nskb);
2755
			__tcp_add_write_queue_head(sk, nskb);
2756 2757 2758
			sk_wmem_free_skb(sk, skb);
			sk->sk_wmem_queued += nskb->truesize;
			sk_mem_charge(sk, nskb->truesize);
L
Linus Torvalds 已提交
2759 2760 2761
			skb = nskb;
		}

E
Eric Dumazet 已提交
2762
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ACK;
L
Linus Torvalds 已提交
2763 2764 2765
		TCP_ECN_send_synack(tcp_sk(sk), skb);
	}
	TCP_SKB_CB(skb)->when = tcp_time_stamp;
2766
	return tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
L
Linus Torvalds 已提交
2767 2768
}

2769 2770 2771 2772 2773 2774 2775 2776 2777
/**
 * 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.
 */
2778
struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
2779
				struct request_sock *req,
2780
				struct tcp_fastopen_cookie *foc)
L
Linus Torvalds 已提交
2781
{
2782
	struct tcp_out_options opts;
2783
	struct inet_request_sock *ireq = inet_rsk(req);
L
Linus Torvalds 已提交
2784 2785 2786
	struct tcp_sock *tp = tcp_sk(sk);
	struct tcphdr *th;
	struct sk_buff *skb;
2787
	struct tcp_md5sig_key *md5;
2788
	int tcp_header_size;
2789
	int mss;
L
Linus Torvalds 已提交
2790

2791
	skb = sock_wmalloc(sk, MAX_TCP_HEADER, 1, GFP_ATOMIC);
2792 2793
	if (unlikely(!skb)) {
		dst_release(dst);
L
Linus Torvalds 已提交
2794
		return NULL;
2795
	}
L
Linus Torvalds 已提交
2796 2797 2798
	/* Reserve space for headers. */
	skb_reserve(skb, MAX_TCP_HEADER);

2799
	skb_dst_set(skb, dst);
2800
	security_skb_owned_by(skb, sk);
L
Linus Torvalds 已提交
2801

2802
	mss = dst_metric_advmss(dst);
2803 2804 2805
	if (tp->rx_opt.user_mss && tp->rx_opt.user_mss < mss)
		mss = tp->rx_opt.user_mss;

A
Adam Langley 已提交
2806 2807 2808 2809
	if (req->rcv_wnd == 0) { /* ignored for retransmitted syns */
		__u8 rcv_wscale;
		/* Set this up on the first call only */
		req->window_clamp = tp->window_clamp ? : dst_metric(dst, RTAX_WINDOW);
2810 2811 2812 2813 2814 2815

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

A
Adam Langley 已提交
2816 2817
		/* tcp_full_space because it is guaranteed to be the first packet */
		tcp_select_initial_window(tcp_full_space(sk),
2818
			mss - (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0),
A
Adam Langley 已提交
2819 2820 2821
			&req->rcv_wnd,
			&req->window_clamp,
			ireq->wscale_ok,
2822 2823
			&rcv_wscale,
			dst_metric(dst, RTAX_INITRWND));
A
Adam Langley 已提交
2824 2825 2826 2827
		ireq->rcv_wscale = rcv_wscale;
	}

	memset(&opts, 0, sizeof(opts));
2828 2829 2830 2831 2832
#ifdef CONFIG_SYN_COOKIES
	if (unlikely(req->cookie_ts))
		TCP_SKB_CB(skb)->when = cookie_init_timestamp(req);
	else
#endif
A
Adam Langley 已提交
2833
	TCP_SKB_CB(skb)->when = tcp_time_stamp;
C
Christoph Paasch 已提交
2834 2835
	tcp_header_size = tcp_synack_options(sk, req, mss, skb, &opts, &md5,
					     foc) + sizeof(*th);
2836

2837 2838
	skb_push(skb, tcp_header_size);
	skb_reset_transport_header(skb);
L
Linus Torvalds 已提交
2839

2840
	th = tcp_hdr(skb);
L
Linus Torvalds 已提交
2841 2842 2843 2844
	memset(th, 0, sizeof(struct tcphdr));
	th->syn = 1;
	th->ack = 1;
	TCP_ECN_make_synack(req, th);
E
Eric Dumazet 已提交
2845
	th->source = htons(ireq->ir_num);
2846
	th->dest = ireq->ir_rmt_port;
2847 2848 2849 2850
	/* 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
2851
			     TCPHDR_SYN | TCPHDR_ACK);
2852

L
Linus Torvalds 已提交
2853
	th->seq = htonl(TCP_SKB_CB(skb)->seq);
2854 2855
	/* 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
2856 2857

	/* RFC1323: The window in SYN & SYN/ACK segments is never scaled. */
2858
	th->window = htons(min(req->rcv_wnd, 65535U));
2859
	tcp_options_write((__be32 *)(th + 1), tp, &opts);
L
Linus Torvalds 已提交
2860
	th->doff = (tcp_header_size >> 2);
2861
	TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_OUTSEGS);
2862 2863 2864 2865

#ifdef CONFIG_TCP_MD5SIG
	/* Okay, we have all we need - do the md5 hash if needed */
	if (md5) {
2866
		tcp_rsk(req)->af_specific->calc_md5_hash(opts.hash_location,
2867
					       md5, NULL, req, skb);
2868 2869 2870
	}
#endif

L
Linus Torvalds 已提交
2871 2872
	return skb;
}
E
Eric Dumazet 已提交
2873
EXPORT_SYMBOL(tcp_make_synack);
L
Linus Torvalds 已提交
2874

2875
/* Do all connect socket setups that can be done AF independent. */
2876
static void tcp_connect_init(struct sock *sk)
L
Linus Torvalds 已提交
2877
{
2878
	const struct dst_entry *dst = __sk_dst_get(sk);
L
Linus Torvalds 已提交
2879 2880 2881 2882 2883 2884 2885
	struct tcp_sock *tp = tcp_sk(sk);
	__u8 rcv_wscale;

	/* We'll fix this up when we get a response from the other end.
	 * See tcp_input.c:tcp_rcv_state_process case TCP_SYN_SENT.
	 */
	tp->tcp_header_len = sizeof(struct tcphdr) +
2886
		(sysctl_tcp_timestamps ? TCPOLEN_TSTAMP_ALIGNED : 0);
L
Linus Torvalds 已提交
2887

2888 2889 2890 2891 2892
#ifdef CONFIG_TCP_MD5SIG
	if (tp->af_specific->md5_lookup(sk, sk) != NULL)
		tp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
#endif

L
Linus Torvalds 已提交
2893 2894 2895 2896
	/* 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
2897
	tcp_mtup_init(sk);
L
Linus Torvalds 已提交
2898 2899 2900 2901
	tcp_sync_mss(sk, dst_mtu(dst));

	if (!tp->window_clamp)
		tp->window_clamp = dst_metric(dst, RTAX_WINDOW);
2902
	tp->advmss = dst_metric_advmss(dst);
2903 2904 2905
	if (tp->rx_opt.user_mss && tp->rx_opt.user_mss < tp->advmss)
		tp->advmss = tp->rx_opt.user_mss;

L
Linus Torvalds 已提交
2906 2907
	tcp_initialize_rcv_mss(sk);

2908 2909 2910 2911 2912
	/* limit the window selection if the user enforce a smaller rx buffer */
	if (sk->sk_userlocks & SOCK_RCVBUF_LOCK &&
	    (tp->window_clamp > tcp_full_space(sk) || tp->window_clamp == 0))
		tp->window_clamp = tcp_full_space(sk);

L
Linus Torvalds 已提交
2913 2914 2915 2916
	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,
2917
				  sysctl_tcp_window_scaling,
2918 2919
				  &rcv_wscale,
				  dst_metric(dst, RTAX_INITRWND));
L
Linus Torvalds 已提交
2920 2921 2922 2923 2924 2925 2926

	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;
2927
	tcp_init_wl(tp, 0);
L
Linus Torvalds 已提交
2928 2929
	tp->snd_una = tp->write_seq;
	tp->snd_sml = tp->write_seq;
I
Ilpo Järvinen 已提交
2930
	tp->snd_up = tp->write_seq;
P
Pavel Emelyanov 已提交
2931
	tp->snd_nxt = tp->write_seq;
P
Pavel Emelyanov 已提交
2932 2933 2934

	if (likely(!tp->repair))
		tp->rcv_nxt = 0;
2935 2936
	else
		tp->rcv_tstamp = tcp_time_stamp;
P
Pavel Emelyanov 已提交
2937 2938
	tp->rcv_wup = tp->rcv_nxt;
	tp->copied_seq = tp->rcv_nxt;
L
Linus Torvalds 已提交
2939

2940 2941
	inet_csk(sk)->icsk_rto = TCP_TIMEOUT_INIT;
	inet_csk(sk)->icsk_retransmits = 0;
L
Linus Torvalds 已提交
2942 2943 2944
	tcp_clear_retrans(tp);
}

2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969
static void tcp_connect_queue_skb(struct sock *sk, struct sk_buff *skb)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);

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

/* Build and send a SYN with data and (cached) Fast Open cookie. However,
 * queue a data-only packet after the regular SYN, such that regular SYNs
 * are retransmitted on timeouts. Also if the remote SYN-ACK acknowledges
 * only the SYN sequence, the data are retransmitted in the first ACK.
 * If cookie is not cached or other error occurs, falls back to send a
 * regular SYN with Fast Open cookie request option.
 */
static int tcp_send_syn_data(struct sock *sk, struct sk_buff *syn)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct tcp_fastopen_request *fo = tp->fastopen_req;
2970
	int syn_loss = 0, space, i, err = 0, iovlen = fo->data->msg_iovlen;
2971
	struct sk_buff *syn_data = NULL, *data;
2972 2973
	unsigned long last_syn_loss = 0;

2974
	tp->rx_opt.mss_clamp = tp->advmss;  /* If MSS is not cached */
2975 2976 2977 2978 2979 2980 2981 2982
	tcp_fastopen_cache_get(sk, &tp->rx_opt.mss_clamp, &fo->cookie,
			       &syn_loss, &last_syn_loss);
	/* Recurring FO SYN losses: revert to regular handshake temporarily */
	if (syn_loss > 1 &&
	    time_before(jiffies, last_syn_loss + (60*HZ << syn_loss))) {
		fo->cookie.len = -1;
		goto fallback;
	}
2983

2984 2985 2986
	if (sysctl_tcp_fastopen & TFO_CLIENT_NO_COOKIE)
		fo->cookie.len = -1;
	else if (fo->cookie.len <= 0)
2987 2988 2989 2990 2991 2992 2993 2994
		goto fallback;

	/* MSS for SYN-data is based on cached MSS and bounded by PMTU and
	 * user-MSS. Reserve maximum option space for middleboxes that add
	 * private TCP options. The cost is reduced data space in SYN :(
	 */
	if (tp->rx_opt.user_mss && tp->rx_opt.user_mss < tp->rx_opt.mss_clamp)
		tp->rx_opt.mss_clamp = tp->rx_opt.user_mss;
2995
	space = __tcp_mtu_to_mss(sk, inet_csk(sk)->icsk_pmtu_cookie) -
2996 2997
		MAX_TCP_OPTION_SPACE;

2998 2999 3000 3001 3002 3003
	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));

	syn_data = skb_copy_expand(syn, MAX_TCP_HEADER, space,
3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032
				   sk->sk_allocation);
	if (syn_data == NULL)
		goto fallback;

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

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

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

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

E
Eric Dumazet 已提交
3033 3034 3035 3036 3037 3038
	/* syn_data is about to be sent, we need to take current time stamps
	 * for the packets that are in write queue : SYN packet and DATA
	 */
	skb_mstamp_get(&syn->skb_mstamp);
	data->skb_mstamp = syn->skb_mstamp;

3039
	if (tcp_transmit_skb(sk, syn_data, 0, sk->sk_allocation) == 0) {
3040
		tp->syn_data = (fo->copied > 0);
3041
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPORIGDATASENT);
3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058
		goto done;
	}
	syn_data = NULL;

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

3059
/* Build a SYN and send it off. */
L
Linus Torvalds 已提交
3060 3061 3062 3063
int tcp_connect(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *buff;
3064
	int err;
L
Linus Torvalds 已提交
3065 3066 3067

	tcp_connect_init(sk);

A
Andrey Vagin 已提交
3068 3069 3070 3071 3072
	if (unlikely(tp->repair)) {
		tcp_finish_connect(sk, NULL);
		return 0;
	}

3073
	buff = alloc_skb_fclone(MAX_TCP_HEADER + 15, sk->sk_allocation);
L
Linus Torvalds 已提交
3074 3075 3076 3077 3078 3079
	if (unlikely(buff == NULL))
		return -ENOBUFS;

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

C
Changli Gao 已提交
3080
	tcp_init_nondata_skb(buff, tp->write_seq++, TCPHDR_SYN);
3081 3082
	tp->retrans_stamp = TCP_SKB_CB(buff)->when = tcp_time_stamp;
	tcp_connect_queue_skb(sk, buff);
3083
	TCP_ECN_send_syn(sk, buff);
L
Linus Torvalds 已提交
3084

3085 3086 3087
	/* 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);
3088 3089
	if (err == -ECONNREFUSED)
		return err;
3090 3091 3092 3093 3094 3095

	/* 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
3096
	TCP_INC_STATS(sock_net(sk), TCP_MIB_ACTIVEOPENS);
L
Linus Torvalds 已提交
3097 3098

	/* Timer for repeating the SYN until an answer. */
3099 3100
	inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
				  inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
L
Linus Torvalds 已提交
3101 3102
	return 0;
}
E
Eric Dumazet 已提交
3103
EXPORT_SYMBOL(tcp_connect);
L
Linus Torvalds 已提交
3104 3105 3106 3107 3108 3109 3110

/* 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)
{
3111 3112
	struct inet_connection_sock *icsk = inet_csk(sk);
	int ato = icsk->icsk_ack.ato;
L
Linus Torvalds 已提交
3113 3114 3115
	unsigned long timeout;

	if (ato > TCP_DELACK_MIN) {
3116
		const struct tcp_sock *tp = tcp_sk(sk);
3117
		int max_ato = HZ / 2;
L
Linus Torvalds 已提交
3118

3119 3120
		if (icsk->icsk_ack.pingpong ||
		    (icsk->icsk_ack.pending & ICSK_ACK_PUSHED))
L
Linus Torvalds 已提交
3121 3122 3123 3124 3125
			max_ato = TCP_DELACK_MAX;

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

		/* If some rtt estimate is known, use it to bound delayed ack.
3126
		 * Do not use inet_csk(sk)->icsk_rto here, use results of rtt measurements
L
Linus Torvalds 已提交
3127 3128
		 * directly.
		 */
3129 3130 3131
		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
3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143

			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. */
3144
	if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
L
Linus Torvalds 已提交
3145 3146 3147
		/* If delack timer was blocked or is about to expire,
		 * send ACK now.
		 */
3148 3149
		if (icsk->icsk_ack.blocked ||
		    time_before_eq(icsk->icsk_ack.timeout, jiffies + (ato >> 2))) {
L
Linus Torvalds 已提交
3150 3151 3152 3153
			tcp_send_ack(sk);
			return;
		}

3154 3155
		if (!time_before(timeout, icsk->icsk_ack.timeout))
			timeout = icsk->icsk_ack.timeout;
L
Linus Torvalds 已提交
3156
	}
3157 3158 3159
	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
3160 3161 3162 3163 3164
}

/* This routine sends an ack and also updates the window. */
void tcp_send_ack(struct sock *sk)
{
3165
	struct sk_buff *buff;
L
Linus Torvalds 已提交
3166

3167 3168 3169
	/* If we have been reset, we may not send again. */
	if (sk->sk_state == TCP_CLOSE)
		return;
L
Linus Torvalds 已提交
3170

3171 3172 3173 3174
	/* We are not putting this on the write queue, so
	 * tcp_transmit_skb() will set the ownership to this
	 * sock.
	 */
3175
	buff = alloc_skb(MAX_TCP_HEADER, sk_gfp_atomic(sk, GFP_ATOMIC));
3176 3177 3178 3179 3180 3181
	if (buff == NULL) {
		inet_csk_schedule_ack(sk);
		inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
		inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
					  TCP_DELACK_MAX, TCP_RTO_MAX);
		return;
L
Linus Torvalds 已提交
3182
	}
3183 3184 3185

	/* Reserve space for headers and prepare control bits. */
	skb_reserve(buff, MAX_TCP_HEADER);
C
Changli Gao 已提交
3186
	tcp_init_nondata_skb(buff, tcp_acceptable_seq(sk), TCPHDR_ACK);
3187 3188 3189

	/* Send it off, this clears delayed acks for us. */
	TCP_SKB_CB(buff)->when = tcp_time_stamp;
3190
	tcp_transmit_skb(sk, buff, 0, sk_gfp_atomic(sk, GFP_ATOMIC));
L
Linus Torvalds 已提交
3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209
}

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

	/* We don't queue it, tcp_transmit_skb() sets ownership. */
3210
	skb = alloc_skb(MAX_TCP_HEADER, sk_gfp_atomic(sk, GFP_ATOMIC));
3211
	if (skb == NULL)
L
Linus Torvalds 已提交
3212 3213 3214 3215 3216 3217 3218 3219
		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
3220
	tcp_init_nondata_skb(skb, tp->snd_una - !urgent, TCPHDR_ACK);
L
Linus Torvalds 已提交
3221
	TCP_SKB_CB(skb)->when = tcp_time_stamp;
3222
	return tcp_transmit_skb(sk, skb, 0, GFP_ATOMIC);
L
Linus Torvalds 已提交
3223 3224
}

P
Pavel Emelyanov 已提交
3225 3226 3227 3228 3229 3230 3231 3232
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;
		tcp_xmit_probe_skb(sk, 0);
	}
}

3233
/* Initiate keepalive or window probe from timer. */
L
Linus Torvalds 已提交
3234 3235
int tcp_write_wakeup(struct sock *sk)
{
3236 3237
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb;
L
Linus Torvalds 已提交
3238

3239 3240 3241 3242 3243 3244
	if (sk->sk_state == TCP_CLOSE)
		return -1;

	if ((skb = tcp_send_head(sk)) != NULL &&
	    before(TCP_SKB_CB(skb)->seq, tcp_wnd_end(tp))) {
		int err;
I
Ilpo Järvinen 已提交
3245
		unsigned int mss = tcp_current_mss(sk);
3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257
		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
3258
			TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
3259 3260 3261 3262 3263
			if (tcp_fragment(sk, skb, seg_size, mss))
				return -1;
		} else if (!tcp_skb_pcount(skb))
			tcp_set_skb_tso_segs(sk, skb, mss);

E
Eric Dumazet 已提交
3264
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
3265 3266 3267 3268 3269 3270
		TCP_SKB_CB(skb)->when = tcp_time_stamp;
		err = tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
		if (!err)
			tcp_event_new_data_sent(sk, skb);
		return err;
	} else {
I
Ilpo Järvinen 已提交
3271
		if (between(tp->snd_up, tp->snd_una + 1, tp->snd_una + 0xFFFF))
3272 3273
			tcp_xmit_probe_skb(sk, 1);
		return tcp_xmit_probe_skb(sk, 0);
L
Linus Torvalds 已提交
3274 3275 3276 3277 3278 3279 3280 3281
	}
}

/* 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)
{
3282
	struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
3283 3284 3285 3286 3287
	struct tcp_sock *tp = tcp_sk(sk);
	int err;

	err = tcp_write_wakeup(sk);

3288
	if (tp->packets_out || !tcp_send_head(sk)) {
L
Linus Torvalds 已提交
3289
		/* Cancel probe timer, if it is not required. */
3290
		icsk->icsk_probes_out = 0;
3291
		icsk->icsk_backoff = 0;
L
Linus Torvalds 已提交
3292 3293 3294 3295
		return;
	}

	if (err <= 0) {
3296 3297
		if (icsk->icsk_backoff < sysctl_tcp_retries2)
			icsk->icsk_backoff++;
3298
		icsk->icsk_probes_out++;
3299
		inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
3300 3301
					  min(icsk->icsk_rto << icsk->icsk_backoff, TCP_RTO_MAX),
					  TCP_RTO_MAX);
L
Linus Torvalds 已提交
3302 3303
	} else {
		/* If packet was not sent due to local congestion,
3304
		 * do not backoff and do not remember icsk_probes_out.
L
Linus Torvalds 已提交
3305 3306 3307 3308
		 * Let local senders to fight for local resources.
		 *
		 * Use accumulated backoff yet.
		 */
3309 3310
		if (!icsk->icsk_probes_out)
			icsk->icsk_probes_out = 1;
3311
		inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
3312
					  min(icsk->icsk_rto << icsk->icsk_backoff,
3313 3314
					      TCP_RESOURCE_PROBE_INTERVAL),
					  TCP_RTO_MAX);
L
Linus Torvalds 已提交
3315 3316
	}
}