tcp_output.c 107.5 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
#include <linux/module.h>
44
#include <linux/static_key.h>
L
Linus Torvalds 已提交
45

46
#include <trace/events/tcp.h>
47

E
Eric Dumazet 已提交
48 49
static bool tcp_write_xmit(struct sock *sk, unsigned int mss_now, int nonagle,
			   int push_one, gfp_t gfp);
50

51
/* Account for new data that has been sent to the network. */
52
static void tcp_event_new_data_sent(struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
53
{
N
Nandita Dukkipati 已提交
54
	struct inet_connection_sock *icsk = inet_csk(sk);
55
	struct tcp_sock *tp = tcp_sk(sk);
56
	unsigned int prior_packets = tp->packets_out;
57

L
Linus Torvalds 已提交
58
	tp->snd_nxt = TCP_SKB_CB(skb)->end_seq;
59

60 61 62
	__skb_unlink(skb, &sk->sk_write_queue);
	tcp_rbtree_insert(&sk->tcp_rtx_queue, skb);

63
	tp->packets_out += tcp_skb_pcount(skb);
Y
Yuchung Cheng 已提交
64
	if (!prior_packets || icsk->icsk_pending == ICSK_TIME_LOSS_PROBE)
65
		tcp_rearm_rto(sk);
66

67 68
	NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPORIGDATASENT,
		      tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
69 70
}

71 72
/* SND.NXT, if window was not shrunk or the amount of shrunk was less than one
 * window scaling factor due to loss of precision.
L
Linus Torvalds 已提交
73 74 75 76 77
 * 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:
 */
78
static inline __u32 tcp_acceptable_seq(const struct sock *sk)
L
Linus Torvalds 已提交
79
{
80
	const struct tcp_sock *tp = tcp_sk(sk);
81

82 83 84
	if (!before(tcp_wnd_end(tp), tp->snd_nxt) ||
	    (tp->rx_opt.wscale_ok &&
	     ((tp->snd_nxt - tcp_wnd_end(tp)) < (1 << tp->rx_opt.rcv_wscale))))
L
Linus Torvalds 已提交
85 86
		return tp->snd_nxt;
	else
87
		return tcp_wnd_end(tp);
L
Linus Torvalds 已提交
88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106
}

/* 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);
107
	const struct dst_entry *dst = __sk_dst_get(sk);
L
Linus Torvalds 已提交
108 109
	int mss = tp->advmss;

110 111 112 113 114 115 116
	if (dst) {
		unsigned int metric = dst_metric_advmss(dst);

		if (metric < mss) {
			mss = metric;
			tp->advmss = mss;
		}
L
Linus Torvalds 已提交
117 118 119 120 121 122
	}

	return (__u16)mss;
}

/* RFC2861. Reset CWND after idle period longer RTO to "restart window".
123 124 125
 * This is the first part of cwnd validation mechanism.
 */
void tcp_cwnd_restart(struct sock *sk, s32 delta)
L
Linus Torvalds 已提交
126
{
127
	struct tcp_sock *tp = tcp_sk(sk);
128
	u32 restart_cwnd = tcp_init_cwnd(tp, __sk_dst_get(sk));
L
Linus Torvalds 已提交
129 130
	u32 cwnd = tp->snd_cwnd;

131
	tcp_ca_event(sk, CA_EVENT_CWND_RESTART);
L
Linus Torvalds 已提交
132

133
	tp->snd_ssthresh = tcp_current_ssthresh(sk);
L
Linus Torvalds 已提交
134 135
	restart_cwnd = min(restart_cwnd, cwnd);

136
	while ((delta -= inet_csk(sk)->icsk_rto) > 0 && cwnd > restart_cwnd)
L
Linus Torvalds 已提交
137 138
		cwnd >>= 1;
	tp->snd_cwnd = max(cwnd, restart_cwnd);
139
	tp->snd_cwnd_stamp = tcp_jiffies32;
L
Linus Torvalds 已提交
140 141 142
	tp->snd_cwnd_used = 0;
}

143
/* Congestion state accounting after a packet has been sent. */
S
Stephen Hemminger 已提交
144
static void tcp_event_data_sent(struct tcp_sock *tp,
145
				struct sock *sk)
L
Linus Torvalds 已提交
146
{
147
	struct inet_connection_sock *icsk = inet_csk(sk);
148
	const u32 now = tcp_jiffies32;
L
Linus Torvalds 已提交
149

150 151 152
	if (tcp_packets_in_flight(tp) == 0)
		tcp_ca_event(sk, CA_EVENT_TX_START);

L
Linus Torvalds 已提交
153 154 155 156 157
	tp->lsndtime = now;

	/* If it is a reply for ato after last received
	 * packet, enter pingpong mode.
	 */
158 159
	if ((u32)(now - icsk->icsk_ack.lrcvtime) < icsk->icsk_ack.ato)
		icsk->icsk_ack.pingpong = 1;
L
Linus Torvalds 已提交
160 161
}

162
/* Account for an ACK we sent. */
S
Stephen Hemminger 已提交
163
static inline void tcp_event_ack_sent(struct sock *sk, unsigned int pkts)
L
Linus Torvalds 已提交
164
{
E
Eric Dumazet 已提交
165 166 167
	struct tcp_sock *tp = tcp_sk(sk);

	if (unlikely(tp->compressed_ack)) {
168 169
		NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPACKCOMPRESSED,
			      tp->compressed_ack);
E
Eric Dumazet 已提交
170 171 172 173
		tp->compressed_ack = 0;
		if (hrtimer_try_to_cancel(&tp->compressed_ack_timer) == 1)
			__sock_put(sk);
	}
174 175
	tcp_dec_quickack_mode(sk, pkts);
	inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
L
Linus Torvalds 已提交
176 177
}

178 179 180 181

u32 tcp_default_init_rwnd(u32 mss)
{
	/* Initial receive window should be twice of TCP_INIT_CWND to
W
Weiping Pan 已提交
182
	 * enable proper sending of new unsent data during fast recovery
183 184 185 186 187 188 189 190 191 192
	 * (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
193 194 195 196 197 198 199
/* 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.
 */
200
void tcp_select_initial_window(const struct sock *sk, int __space, __u32 mss,
L
Linus Torvalds 已提交
201
			       __u32 *rcv_wnd, __u32 *window_clamp,
202 203
			       int wscale_ok, __u8 *rcv_wscale,
			       __u32 init_rcv_wnd)
L
Linus Torvalds 已提交
204 205 206 207 208
{
	unsigned int space = (__space < 0 ? 0 : __space);

	/* If no clamp set the clamp to the max possible scaled window */
	if (*window_clamp == 0)
209
		(*window_clamp) = (U16_MAX << TCP_MAX_WSCALE);
L
Linus Torvalds 已提交
210 211 212 213
	space = min(*window_clamp, space);

	/* Quantize space offering to a multiple of mss if possible. */
	if (space > mss)
214
		space = rounddown(space, mss);
L
Linus Torvalds 已提交
215 216

	/* NOTE: offering an initial window larger than 32767
217 218 219 220 221 222
	 * 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
223
	 */
224
	if (sock_net(sk)->ipv4.sysctl_tcp_workaround_signed_windows)
225 226 227 228
		(*rcv_wnd) = min(space, MAX_TCP_WINDOW);
	else
		(*rcv_wnd) = space;

L
Linus Torvalds 已提交
229 230
	(*rcv_wscale) = 0;
	if (wscale_ok) {
231
		/* Set window scaling on max possible window */
232
		space = max_t(u32, space, sock_net(sk)->ipv4.sysctl_tcp_rmem[2]);
233
		space = max_t(u32, space, sysctl_rmem_max);
234
		space = min_t(u32, space, *window_clamp);
235
		while (space > U16_MAX && (*rcv_wscale) < TCP_MAX_WSCALE) {
L
Linus Torvalds 已提交
236 237 238 239 240
			space >>= 1;
			(*rcv_wscale)++;
		}
	}

241 242 243
	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
244 245

	/* Set the clamp no higher than max representable value */
246
	(*window_clamp) = min_t(__u32, U16_MAX << (*rcv_wscale), *window_clamp);
L
Linus Torvalds 已提交
247
}
E
Eric Dumazet 已提交
248
EXPORT_SYMBOL(tcp_select_initial_window);
L
Linus Torvalds 已提交
249 250 251 252 253 254

/* 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
255
static u16 tcp_select_window(struct sock *sk)
L
Linus Torvalds 已提交
256 257
{
	struct tcp_sock *tp = tcp_sk(sk);
258
	u32 old_win = tp->rcv_wnd;
L
Linus Torvalds 已提交
259 260 261 262
	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
263
	if (new_win < cur_win) {
L
Linus Torvalds 已提交
264 265 266 267 268 269 270
		/* 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.
		 */
271 272 273
		if (new_win == 0)
			NET_INC_STATS(sock_net(sk),
				      LINUX_MIB_TCPWANTZEROWINDOWADV);
274
		new_win = ALIGN(cur_win, 1 << tp->rx_opt.rcv_wscale);
L
Linus Torvalds 已提交
275 276 277 278 279 280 281
	}
	tp->rcv_wnd = new_win;
	tp->rcv_wup = tp->rcv_nxt;

	/* Make sure we do not exceed the maximum possible
	 * scaled window.
	 */
282 283
	if (!tp->rx_opt.rcv_wscale &&
	    sock_net(sk)->ipv4.sysctl_tcp_workaround_signed_windows)
L
Linus Torvalds 已提交
284 285 286 287 288 289 290
		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;

291
	/* If we advertise zero window, disable fast path. */
292
	if (new_win == 0) {
293
		tp->pred_flags = 0;
294 295 296 297 298 299
		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
300 301 302 303

	return new_win;
}

304
/* Packet ECN state for a SYN-ACK */
305
static void tcp_ecn_send_synack(struct sock *sk, struct sk_buff *skb)
306
{
307 308
	const struct tcp_sock *tp = tcp_sk(sk);

E
Eric Dumazet 已提交
309
	TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_CWR;
310
	if (!(tp->ecn_flags & TCP_ECN_OK))
E
Eric Dumazet 已提交
311
		TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_ECE;
312 313
	else if (tcp_ca_needs_ecn(sk) ||
		 tcp_bpf_ca_needs_ecn(sk))
314
		INET_ECN_xmit(sk);
315 316
}

317
/* Packet ECN state for a SYN.  */
318
static void tcp_ecn_send_syn(struct sock *sk, struct sk_buff *skb)
319 320
{
	struct tcp_sock *tp = tcp_sk(sk);
321
	bool bpf_needs_ecn = tcp_bpf_ca_needs_ecn(sk);
322
	bool use_ecn = sock_net(sk)->ipv4.sysctl_tcp_ecn == 1 ||
323
		tcp_ca_needs_ecn(sk) || bpf_needs_ecn;
324 325 326 327 328 329 330

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

		if (dst && dst_feature(dst, RTAX_FEATURE_ECN))
			use_ecn = true;
	}
331 332

	tp->ecn_flags = 0;
333 334

	if (use_ecn) {
E
Eric Dumazet 已提交
335
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ECE | TCPHDR_CWR;
336
		tp->ecn_flags = TCP_ECN_OK;
337
		if (tcp_ca_needs_ecn(sk) || bpf_needs_ecn)
338
			INET_ECN_xmit(sk);
339 340 341
	}
}

342 343 344 345 346 347 348 349 350
static void tcp_ecn_clear_syn(struct sock *sk, struct sk_buff *skb)
{
	if (sock_net(sk)->ipv4.sysctl_tcp_ecn_fallback)
		/* tp->ecn_flags are cleared at a later point in time when
		 * SYN ACK is ultimatively being received.
		 */
		TCP_SKB_CB(skb)->tcp_flags &= ~(TCPHDR_ECE | TCPHDR_CWR);
}

351
static void
352
tcp_ecn_make_synack(const struct request_sock *req, struct tcphdr *th)
353
{
354
	if (inet_rsk(req)->ecn_ok)
355 356 357
		th->ece = 1;
}

358 359 360
/* Set up ECN state for a packet on a ESTABLISHED socket that is about to
 * be sent.
 */
361
static void tcp_ecn_send(struct sock *sk, struct sk_buff *skb,
362
			 struct tcphdr *th, int tcp_header_len)
363 364 365 366 367 368 369 370
{
	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);
371
			if (tp->ecn_flags & TCP_ECN_QUEUE_CWR) {
372
				tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR;
373
				th->cwr = 1;
374 375
				skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
			}
376
		} else if (!tcp_ca_needs_ecn(sk)) {
377 378 379 380
			/* ACK or retransmitted segment: clear ECT|CE */
			INET_ECN_dontxmit(sk);
		}
		if (tp->ecn_flags & TCP_ECN_DEMAND_CWR)
381
			th->ece = 1;
382 383 384
	}
}

385 386 387 388 389
/* 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)
{
390
	skb->ip_summed = CHECKSUM_PARTIAL;
391

E
Eric Dumazet 已提交
392
	TCP_SKB_CB(skb)->tcp_flags = flags;
393 394
	TCP_SKB_CB(skb)->sacked = 0;

395
	tcp_skb_pcount_set(skb, 1);
396 397

	TCP_SKB_CB(skb)->seq = seq;
C
Changli Gao 已提交
398
	if (flags & (TCPHDR_SYN | TCPHDR_FIN))
399 400 401 402
		seq++;
	TCP_SKB_CB(skb)->end_seq = seq;
}

E
Eric Dumazet 已提交
403
static inline bool tcp_urg_mode(const struct tcp_sock *tp)
I
Ilpo Järvinen 已提交
404 405 406 407
{
	return tp->snd_una != tp->snd_up;
}

A
Adam Langley 已提交
408 409 410
#define OPTION_SACK_ADVERTISE	(1 << 0)
#define OPTION_TS		(1 << 1)
#define OPTION_MD5		(1 << 2)
411
#define OPTION_WSCALE		(1 << 3)
Y
Yuchung Cheng 已提交
412
#define OPTION_FAST_OPEN_COOKIE	(1 << 8)
413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428
#define OPTION_SMC		(1 << 9)

static void smc_options_write(__be32 *ptr, u16 *options)
{
#if IS_ENABLED(CONFIG_SMC)
	if (static_branch_unlikely(&tcp_have_smc)) {
		if (unlikely(OPTION_SMC & *options)) {
			*ptr++ = htonl((TCPOPT_NOP  << 24) |
				       (TCPOPT_NOP  << 16) |
				       (TCPOPT_EXP <<  8) |
				       (TCPOLEN_EXP_SMC_BASE));
			*ptr++ = htonl(TCPOPT_SMC_MAGIC);
		}
	}
#endif
}
A
Adam Langley 已提交
429 430

struct tcp_out_options {
Y
Yuchung Cheng 已提交
431 432
	u16 options;		/* bit field of OPTION_* */
	u16 mss;		/* 0 to disable */
A
Adam Langley 已提交
433 434
	u8 ws;			/* window scale, 0 to disable */
	u8 num_sack_blocks;	/* number of SACK blocks to include */
435 436
	u8 hash_size;		/* bytes in hash_location */
	__u8 *hash_location;	/* temporary pointer, overloaded */
Y
Yuchung Cheng 已提交
437 438
	__u32 tsval, tsecr;	/* need to include OPTION_TS */
	struct tcp_fastopen_cookie *fastopen_cookie;	/* Fast open cookie */
A
Adam Langley 已提交
439 440
};

441 442 443
/* Write previously computed TCP options to the packet.
 *
 * Beware: Something in the Internet is very sensitive to the ordering of
444 445
 * TCP options, we learned this through the hard way, so be careful here.
 * Luckily we can at least blame others for their non-compliance but from
S
stephen hemminger 已提交
446
 * inter-operability perspective it seems that we're somewhat stuck with
447 448 449 450 451 452 453
 * 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
454
static void tcp_options_write(__be32 *ptr, struct tcp_sock *tp,
455 456
			      struct tcp_out_options *opts)
{
Y
Yuchung Cheng 已提交
457
	u16 options = opts->options;	/* mungable copy */
458 459

	if (unlikely(OPTION_MD5 & options)) {
C
Christoph Paasch 已提交
460 461
		*ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
			       (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
462 463
		/* overload cookie hash location */
		opts->hash_location = (__u8 *)ptr;
A
Adam Langley 已提交
464
		ptr += 4;
S
Stephen Hemminger 已提交
465
	}
A
Adam Langley 已提交
466

467 468 469 470 471 472
	if (unlikely(opts->mss)) {
		*ptr++ = htonl((TCPOPT_MSS << 24) |
			       (TCPOLEN_MSS << 16) |
			       opts->mss);
	}

473 474
	if (likely(OPTION_TS & options)) {
		if (unlikely(OPTION_SACK_ADVERTISE & options)) {
A
Adam Langley 已提交
475 476 477 478
			*ptr++ = htonl((TCPOPT_SACK_PERM << 24) |
				       (TCPOLEN_SACK_PERM << 16) |
				       (TCPOPT_TIMESTAMP << 8) |
				       TCPOLEN_TIMESTAMP);
479
			options &= ~OPTION_SACK_ADVERTISE;
A
Adam Langley 已提交
480 481 482 483 484 485 486 487 488 489
		} else {
			*ptr++ = htonl((TCPOPT_NOP << 24) |
				       (TCPOPT_NOP << 16) |
				       (TCPOPT_TIMESTAMP << 8) |
				       TCPOLEN_TIMESTAMP);
		}
		*ptr++ = htonl(opts->tsval);
		*ptr++ = htonl(opts->tsecr);
	}

490
	if (unlikely(OPTION_SACK_ADVERTISE & options)) {
A
Adam Langley 已提交
491 492 493 494 495 496
		*ptr++ = htonl((TCPOPT_NOP << 24) |
			       (TCPOPT_NOP << 16) |
			       (TCPOPT_SACK_PERM << 8) |
			       TCPOLEN_SACK_PERM);
	}

497
	if (unlikely(OPTION_WSCALE & options)) {
A
Adam Langley 已提交
498 499 500 501 502 503 504 505 506
		*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
507 508 509 510 511
		int this_sack;

		*ptr++ = htonl((TCPOPT_NOP  << 24) |
			       (TCPOPT_NOP  << 16) |
			       (TCPOPT_SACK <<  8) |
A
Adam Langley 已提交
512
			       (TCPOLEN_SACK_BASE + (opts->num_sack_blocks *
S
Stephen Hemminger 已提交
513
						     TCPOLEN_SACK_PERBLOCK)));
S
Stephen Hemminger 已提交
514

A
Adam Langley 已提交
515 516
		for (this_sack = 0; this_sack < opts->num_sack_blocks;
		     ++this_sack) {
S
Stephen Hemminger 已提交
517 518 519
			*ptr++ = htonl(sp[this_sack].start_seq);
			*ptr++ = htonl(sp[this_sack].end_seq);
		}
S
Stephen Hemminger 已提交
520

521
		tp->rx_opt.dsack = 0;
S
Stephen Hemminger 已提交
522
	}
Y
Yuchung Cheng 已提交
523 524 525

	if (unlikely(OPTION_FAST_OPEN_COOKIE & options)) {
		struct tcp_fastopen_cookie *foc = opts->fastopen_cookie;
526 527 528 529 530 531 532 533 534 535 536 537 538
		u8 *p = (u8 *)ptr;
		u32 len; /* Fast Open option length */

		if (foc->exp) {
			len = TCPOLEN_EXP_FASTOPEN_BASE + foc->len;
			*ptr = htonl((TCPOPT_EXP << 24) | (len << 16) |
				     TCPOPT_FASTOPEN_MAGIC);
			p += TCPOLEN_EXP_FASTOPEN_BASE;
		} else {
			len = TCPOLEN_FASTOPEN_BASE + foc->len;
			*p++ = TCPOPT_FASTOPEN;
			*p++ = len;
		}
Y
Yuchung Cheng 已提交
539

540 541 542 543
		memcpy(p, foc->val, foc->len);
		if ((len & 3) == 2) {
			p[foc->len] = TCPOPT_NOP;
			p[foc->len + 1] = TCPOPT_NOP;
Y
Yuchung Cheng 已提交
544
		}
545
		ptr += (len + 3) >> 2;
Y
Yuchung Cheng 已提交
546
	}
547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581

	smc_options_write(ptr, &options);
}

static void smc_set_option(const struct tcp_sock *tp,
			   struct tcp_out_options *opts,
			   unsigned int *remaining)
{
#if IS_ENABLED(CONFIG_SMC)
	if (static_branch_unlikely(&tcp_have_smc)) {
		if (tp->syn_smc) {
			if (*remaining >= TCPOLEN_EXP_SMC_BASE_ALIGNED) {
				opts->options |= OPTION_SMC;
				*remaining -= TCPOLEN_EXP_SMC_BASE_ALIGNED;
			}
		}
	}
#endif
}

static void smc_set_option_cond(const struct tcp_sock *tp,
				const struct inet_request_sock *ireq,
				struct tcp_out_options *opts,
				unsigned int *remaining)
{
#if IS_ENABLED(CONFIG_SMC)
	if (static_branch_unlikely(&tcp_have_smc)) {
		if (tp->syn_smc && ireq->smc_ok) {
			if (*remaining >= TCPOLEN_EXP_SMC_BASE_ALIGNED) {
				opts->options |= OPTION_SMC;
				*remaining -= TCPOLEN_EXP_SMC_BASE_ALIGNED;
			}
		}
	}
#endif
A
Adam Langley 已提交
582 583
}

584 585 586
/* Compute TCP options for SYN packets. This is not the final
 * network wire format yet.
 */
587
static unsigned int tcp_syn_options(struct sock *sk, struct sk_buff *skb,
A
Adam Langley 已提交
588
				struct tcp_out_options *opts,
589 590
				struct tcp_md5sig_key **md5)
{
A
Adam Langley 已提交
591
	struct tcp_sock *tp = tcp_sk(sk);
592
	unsigned int remaining = MAX_TCP_OPTION_SPACE;
593
	struct tcp_fastopen_request *fastopen = tp->fastopen_req;
A
Adam Langley 已提交
594

595
	*md5 = NULL;
596
#ifdef CONFIG_TCP_MD5SIG
597 598 599 600 601 602
	if (unlikely(rcu_access_pointer(tp->md5sig_info))) {
		*md5 = tp->af_specific->md5_lookup(sk, sk);
		if (*md5) {
			opts->options |= OPTION_MD5;
			remaining -= TCPOLEN_MD5SIG_ALIGNED;
		}
603 604
	}
#endif
A
Adam Langley 已提交
605 606 607 608 609 610 611 612 613 614 615

	/* 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);
616
	remaining -= TCPOLEN_MSS_ALIGNED;
A
Adam Langley 已提交
617

618
	if (likely(sock_net(sk)->ipv4.sysctl_tcp_timestamps && !*md5)) {
A
Adam Langley 已提交
619
		opts->options |= OPTION_TS;
E
Eric Dumazet 已提交
620
		opts->tsval = tcp_skb_timestamp(skb) + tp->tsoffset;
A
Adam Langley 已提交
621
		opts->tsecr = tp->rx_opt.ts_recent;
622
		remaining -= TCPOLEN_TSTAMP_ALIGNED;
A
Adam Langley 已提交
623
	}
624
	if (likely(sock_net(sk)->ipv4.sysctl_tcp_window_scaling)) {
A
Adam Langley 已提交
625
		opts->ws = tp->rx_opt.rcv_wscale;
626
		opts->options |= OPTION_WSCALE;
627
		remaining -= TCPOLEN_WSCALE_ALIGNED;
A
Adam Langley 已提交
628
	}
E
Eric Dumazet 已提交
629
	if (likely(sock_net(sk)->ipv4.sysctl_tcp_sack)) {
A
Adam Langley 已提交
630
		opts->options |= OPTION_SACK_ADVERTISE;
631
		if (unlikely(!(OPTION_TS & opts->options)))
632
			remaining -= TCPOLEN_SACKPERM_ALIGNED;
A
Adam Langley 已提交
633 634
	}

635
	if (fastopen && fastopen->cookie.len >= 0) {
636 637 638 639
		u32 need = fastopen->cookie.len;

		need += fastopen->cookie.exp ? TCPOLEN_EXP_FASTOPEN_BASE :
					       TCPOLEN_FASTOPEN_BASE;
640 641 642 643 644 645
		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;
646
			tp->syn_fastopen_exp = fastopen->cookie.exp ? 1 : 0;
647 648
		}
	}
649

650 651
	smc_set_option(tp, opts, &remaining);

652
	return MAX_TCP_OPTION_SPACE - remaining;
S
Stephen Hemminger 已提交
653 654
}

655
/* Set up TCP options for SYN-ACKs. */
656 657
static unsigned int tcp_synack_options(const struct sock *sk,
				       struct request_sock *req,
658 659 660 661
				       unsigned int mss, struct sk_buff *skb,
				       struct tcp_out_options *opts,
				       const struct tcp_md5sig_key *md5,
				       struct tcp_fastopen_cookie *foc)
662
{
A
Adam Langley 已提交
663
	struct inet_request_sock *ireq = inet_rsk(req);
664
	unsigned int remaining = MAX_TCP_OPTION_SPACE;
A
Adam Langley 已提交
665

666
#ifdef CONFIG_TCP_MD5SIG
E
Eric Dumazet 已提交
667
	if (md5) {
A
Adam Langley 已提交
668
		opts->options |= OPTION_MD5;
669 670 671 672 673 674 675
		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
676
		ireq->tstamp_ok &= !ireq->sack_ok;
677 678
	}
#endif
A
Adam Langley 已提交
679

680
	/* We always send an MSS option. */
A
Adam Langley 已提交
681
	opts->mss = mss;
682
	remaining -= TCPOLEN_MSS_ALIGNED;
A
Adam Langley 已提交
683 684 685

	if (likely(ireq->wscale_ok)) {
		opts->ws = ireq->rcv_wscale;
686
		opts->options |= OPTION_WSCALE;
687
		remaining -= TCPOLEN_WSCALE_ALIGNED;
A
Adam Langley 已提交
688
	}
E
Eric Dumazet 已提交
689
	if (likely(ireq->tstamp_ok)) {
A
Adam Langley 已提交
690
		opts->options |= OPTION_TS;
691
		opts->tsval = tcp_skb_timestamp(skb) + tcp_rsk(req)->ts_off;
A
Adam Langley 已提交
692
		opts->tsecr = req->ts_recent;
693
		remaining -= TCPOLEN_TSTAMP_ALIGNED;
A
Adam Langley 已提交
694 695 696
	}
	if (likely(ireq->sack_ok)) {
		opts->options |= OPTION_SACK_ADVERTISE;
E
Eric Dumazet 已提交
697
		if (unlikely(!ireq->tstamp_ok))
698
			remaining -= TCPOLEN_SACKPERM_ALIGNED;
A
Adam Langley 已提交
699
	}
700 701 702 703 704
	if (foc != NULL && foc->len >= 0) {
		u32 need = foc->len;

		need += foc->exp ? TCPOLEN_EXP_FASTOPEN_BASE :
				   TCPOLEN_FASTOPEN_BASE;
705 706 707 708 709 710 711
		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
712

713 714
	smc_set_option_cond(tcp_sk(sk), ireq, opts, &remaining);

715
	return MAX_TCP_OPTION_SPACE - remaining;
A
Adam Langley 已提交
716 717
}

718 719 720
/* Compute TCP options for ESTABLISHED sockets. This is not the
 * final wire format yet.
 */
721
static unsigned int tcp_established_options(struct sock *sk, struct sk_buff *skb,
A
Adam Langley 已提交
722
					struct tcp_out_options *opts,
723 724
					struct tcp_md5sig_key **md5)
{
A
Adam Langley 已提交
725
	struct tcp_sock *tp = tcp_sk(sk);
726
	unsigned int size = 0;
727
	unsigned int eff_sacks;
A
Adam Langley 已提交
728

729 730
	opts->options = 0;

731
	*md5 = NULL;
A
Adam Langley 已提交
732
#ifdef CONFIG_TCP_MD5SIG
733 734 735 736 737 738
	if (unlikely(rcu_access_pointer(tp->md5sig_info))) {
		*md5 = tp->af_specific->md5_lookup(sk, sk);
		if (*md5) {
			opts->options |= OPTION_MD5;
			size += TCPOLEN_MD5SIG_ALIGNED;
		}
A
Adam Langley 已提交
739 740 741 742 743
	}
#endif

	if (likely(tp->rx_opt.tstamp_ok)) {
		opts->options |= OPTION_TS;
E
Eric Dumazet 已提交
744
		opts->tsval = skb ? tcp_skb_timestamp(skb) + tp->tsoffset : 0;
A
Adam Langley 已提交
745 746 747 748
		opts->tsecr = tp->rx_opt.ts_recent;
		size += TCPOLEN_TSTAMP_ALIGNED;
	}

749 750
	eff_sacks = tp->rx_opt.num_sacks + tp->rx_opt.dsack;
	if (unlikely(eff_sacks)) {
751
		const unsigned int remaining = MAX_TCP_OPTION_SPACE - size;
A
Adam Langley 已提交
752
		opts->num_sack_blocks =
753
			min_t(unsigned int, eff_sacks,
A
Adam Langley 已提交
754 755 756 757 758 759 760
			      (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
761
}
L
Linus Torvalds 已提交
762

E
Eric Dumazet 已提交
763 764 765 766 767 768 769 770 771

/* 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
772
 * The invariant being skb->truesize subtracted from sk->sk_wmem_alloc
E
Eric Dumazet 已提交
773 774 775 776 777 778 779 780 781 782 783
 *
 * 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);

784
static void tcp_tsq_write(struct sock *sk)
785 786 787
{
	if ((1 << sk->sk_state) &
	    (TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_CLOSING |
788 789 790 791
	     TCPF_CLOSE_WAIT  | TCPF_LAST_ACK)) {
		struct tcp_sock *tp = tcp_sk(sk);

		if (tp->lost_out > tp->retrans_out &&
792 793
		    tp->snd_cwnd > tcp_packets_in_flight(tp)) {
			tcp_mstamp_refresh(tp);
794
			tcp_xmit_retransmit_queue(sk);
795
		}
796 797

		tcp_write_xmit(sk, tcp_current_mss(sk), tp->nonagle,
J
John Ogness 已提交
798
			       0, GFP_ATOMIC);
799
	}
800
}
801 802 803 804 805 806 807 808 809 810

static void tcp_tsq_handler(struct sock *sk)
{
	bh_lock_sock(sk);
	if (!sock_owned_by_user(sk))
		tcp_tsq_write(sk);
	else if (!test_and_set_bit(TCP_TSQ_DEFERRED, &sk->sk_tsq_flags))
		sock_hold(sk);
	bh_unlock_sock(sk);
}
E
Eric Dumazet 已提交
811
/*
S
stephen hemminger 已提交
812
 * One tasklet per cpu tries to send more skbs.
E
Eric Dumazet 已提交
813
 * We run in tasklet context but need to disable irqs when
S
stephen hemminger 已提交
814
 * transferring tsq->head because tcp_wfree() might
E
Eric Dumazet 已提交
815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834
 * 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;
835
		smp_mb__before_atomic();
836 837
		clear_bit(TSQ_QUEUED, &sk->sk_tsq_flags);

838
		tcp_tsq_handler(sk);
E
Eric Dumazet 已提交
839 840 841 842
		sk_free(sk);
	}
}

E
Eric Dumazet 已提交
843 844 845 846
#define TCP_DEFERRED_ALL (TCPF_TSQ_DEFERRED |		\
			  TCPF_WRITE_TIMER_DEFERRED |	\
			  TCPF_DELACK_TIMER_DEFERRED |	\
			  TCPF_MTU_REDUCED_DEFERRED)
E
Eric Dumazet 已提交
847 848 849 850 851 852 853 854 855
/**
 * 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)
{
856
	unsigned long flags, nflags;
E
Eric Dumazet 已提交
857

858 859
	/* perform an atomic operation only if at least one flag is set */
	do {
860
		flags = sk->sk_tsq_flags;
861 862 863
		if (!(flags & TCP_DEFERRED_ALL))
			return;
		nflags = flags & ~TCP_DEFERRED_ALL;
864
	} while (cmpxchg(&sk->sk_tsq_flags, flags, nflags) != flags);
865

866 867 868 869
	if (flags & TCPF_TSQ_DEFERRED) {
		tcp_tsq_write(sk);
		__sock_put(sk);
	}
870 871 872 873 874 875 876 877 878 879 880
	/* Here begins the tricky part :
	 * We are called from release_sock() with :
	 * 1) BH disabled
	 * 2) sk_lock.slock spinlock held
	 * 3) socket owned by us (sk->sk_lock.owned == 1)
	 *
	 * But following code is meant to be called from BH handlers,
	 * so we should keep BH disabled, but early release socket ownership
	 */
	sock_release_ownership(sk);

E
Eric Dumazet 已提交
881
	if (flags & TCPF_WRITE_TIMER_DEFERRED) {
882
		tcp_write_timer_handler(sk);
883 884
		__sock_put(sk);
	}
E
Eric Dumazet 已提交
885
	if (flags & TCPF_DELACK_TIMER_DEFERRED) {
886
		tcp_delack_timer_handler(sk);
887 888
		__sock_put(sk);
	}
E
Eric Dumazet 已提交
889
	if (flags & TCPF_MTU_REDUCED_DEFERRED) {
890
		inet_csk(sk)->icsk_af_ops->mtu_reduced(sk);
891 892
		__sock_put(sk);
	}
E
Eric Dumazet 已提交
893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911
}
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
912
 * We can't xmit new skbs from this context, as we might already
E
Eric Dumazet 已提交
913 914
 * hold qdisc lock.
 */
E
Eric Dumazet 已提交
915
void tcp_wfree(struct sk_buff *skb)
E
Eric Dumazet 已提交
916 917 918
{
	struct sock *sk = skb->sk;
	struct tcp_sock *tp = tcp_sk(sk);
919
	unsigned long flags, nval, oval;
920 921 922 923

	/* Keep one reference on sk_wmem_alloc.
	 * Will be released by sk_free() from here or tcp_tasklet_func()
	 */
924
	WARN_ON(refcount_sub_and_test(skb->truesize - 1, &sk->sk_wmem_alloc));
925 926 927 928 929 930 931 932

	/* If this softirq is serviced by ksoftirqd, we are likely under stress.
	 * Wait until our queues (qdisc + devices) are drained.
	 * This gives :
	 * - less callbacks to tcp_write_xmit(), reducing stress (batches)
	 * - chance for incoming ACK (processed by another cpu maybe)
	 *   to migrate this flow (skb->ooo_okay will be eventually set)
	 */
933
	if (refcount_read(&sk->sk_wmem_alloc) >= SKB_TRUESIZE(1) && this_cpu_ksoftirqd() == current)
934
		goto out;
E
Eric Dumazet 已提交
935

936
	for (oval = READ_ONCE(sk->sk_tsq_flags);; oval = nval) {
E
Eric Dumazet 已提交
937
		struct tsq_tasklet *tsq;
938
		bool empty;
E
Eric Dumazet 已提交
939

940 941 942
		if (!(oval & TSQF_THROTTLED) || (oval & TSQF_QUEUED))
			goto out;

943
		nval = (oval & ~TSQF_THROTTLED) | TSQF_QUEUED;
944
		nval = cmpxchg(&sk->sk_tsq_flags, oval, nval);
945 946 947
		if (nval != oval)
			continue;

E
Eric Dumazet 已提交
948 949
		/* queue this socket to tasklet queue */
		local_irq_save(flags);
950
		tsq = this_cpu_ptr(&tsq_tasklet);
951
		empty = list_empty(&tsq->head);
E
Eric Dumazet 已提交
952
		list_add(&tp->tsq_node, &tsq->head);
953 954
		if (empty)
			tasklet_schedule(&tsq->tasklet);
E
Eric Dumazet 已提交
955
		local_irq_restore(flags);
956
		return;
E
Eric Dumazet 已提交
957
	}
958 959
out:
	sk_free(sk);
E
Eric Dumazet 已提交
960 961
}

962 963
/* Note: Called under soft irq.
 * We can call TCP stack right away, unless socket is owned by user.
964 965 966 967 968 969
 */
enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer)
{
	struct tcp_sock *tp = container_of(timer, struct tcp_sock, pacing_timer);
	struct sock *sk = (struct sock *)tp;

970 971
	tcp_tsq_handler(sk);
	sock_put(sk);
972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004

	return HRTIMER_NORESTART;
}

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

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

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

	/* Should account for header sizes as sch_fq does,
	 * but lets make things simple.
	 */
	len_ns = (u64)skb->len * NSEC_PER_SEC;
	do_div(len_ns, rate);
	hrtimer_start(&tcp_sk(sk)->pacing_timer,
		      ktime_add_ns(ktime_get(), len_ns),
1005 1006
		      HRTIMER_MODE_ABS_PINNED_SOFT);
	sock_hold(sk);
1007 1008
}

1009 1010 1011 1012 1013 1014
static void tcp_update_skb_after_send(struct tcp_sock *tp, struct sk_buff *skb)
{
	skb->skb_mstamp = tp->tcp_mstamp;
	list_move_tail(&skb->tcp_tsorted_anchor, &tp->tsorted_sent_queue);
}

L
Linus Torvalds 已提交
1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025
/* 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.
 */
1026 1027
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
1028
{
1029 1030 1031 1032
	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
1033
	struct tcp_out_options opts;
1034
	unsigned int tcp_options_size, tcp_header_size;
1035
	struct sk_buff *oskb = NULL;
1036
	struct tcp_md5sig_key *md5;
1037 1038 1039 1040
	struct tcphdr *th;
	int err;

	BUG_ON(!skb || !tcp_skb_pcount(skb));
1041
	tp = tcp_sk(sk);
1042

1043
	if (clone_it) {
1044 1045
		TCP_SKB_CB(skb)->tx.in_flight = TCP_SKB_CB(skb)->end_seq
			- tp->snd_una;
1046
		oskb = skb;
1047 1048 1049 1050 1051 1052 1053 1054

		tcp_skb_tsorted_save(oskb) {
			if (unlikely(skb_cloned(oskb)))
				skb = pskb_copy(oskb, gfp_mask);
			else
				skb = skb_clone(oskb, gfp_mask);
		} tcp_skb_tsorted_restore(oskb);

1055 1056 1057
		if (unlikely(!skb))
			return -ENOBUFS;
	}
1058
	skb->skb_mstamp = tp->tcp_mstamp;
L
Linus Torvalds 已提交
1059

1060 1061
	inet = inet_sk(sk);
	tcb = TCP_SKB_CB(skb);
A
Adam Langley 已提交
1062
	memset(&opts, 0, sizeof(opts));
L
Linus Torvalds 已提交
1063

E
Eric Dumazet 已提交
1064
	if (unlikely(tcb->tcp_flags & TCPHDR_SYN))
A
Adam Langley 已提交
1065 1066 1067 1068 1069
		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);
1070

E
Eric Dumazet 已提交
1071
	/* if no packet is in qdisc/device queue, then allow XPS to select
1072
	 * another queue. We can be called from tcp_tsq_handler()
1073
	 * which holds one reference to sk.
1074 1075 1076
	 *
	 * TODO: Ideally, in-flight pure ACK packets should not matter here.
	 * One way to get this would be to set skb->truesize = 2 on them.
E
Eric Dumazet 已提交
1077
	 */
1078
	skb->ooo_okay = sk_wmem_alloc_get(sk) < SKB_TRUESIZE(1);
1079

1080 1081 1082 1083 1084 1085 1086
	/* If we had to use memory reserve to allocate this skb,
	 * this might cause drops if packet is looped back :
	 * Other socket might not have SOCK_MEMALLOC.
	 * Packets not looped back do not care about pfmemalloc.
	 */
	skb->pfmemalloc = 0;

1087 1088
	skb_push(skb, tcp_header_size);
	skb_reset_transport_header(skb);
E
Eric Dumazet 已提交
1089 1090 1091

	skb_orphan(skb);
	skb->sk = sk;
E
Eric Dumazet 已提交
1092
	skb->destructor = skb_is_tcp_pure_ack(skb) ? __sock_wfree : tcp_wfree;
1093
	skb_set_hash_from_sk(skb, sk);
1094
	refcount_add(skb->truesize, &sk->sk_wmem_alloc);
1095

1096 1097
	skb_set_dst_pending_confirm(skb, sk->sk_dst_pending_confirm);

1098
	/* Build TCP header and checksum it. */
1099
	th = (struct tcphdr *)skb->data;
E
Eric Dumazet 已提交
1100 1101
	th->source		= inet->inet_sport;
	th->dest		= inet->inet_dport;
1102 1103
	th->seq			= htonl(tcb->seq);
	th->ack_seq		= htonl(tp->rcv_nxt);
1104
	*(((__be16 *)th) + 6)	= htons(((tcp_header_size >> 2) << 12) |
E
Eric Dumazet 已提交
1105
					tcb->tcp_flags);
1106 1107 1108

	th->check		= 0;
	th->urg_ptr		= 0;
L
Linus Torvalds 已提交
1109

I
Ilpo Järvinen 已提交
1110
	/* The urg_mode check is necessary during a below snd_una win probe */
1111 1112 1113 1114 1115
	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)) {
1116
			th->urg_ptr = htons(0xFFFF);
1117 1118
			th->urg = 1;
		}
1119
	}
L
Linus Torvalds 已提交
1120

1121
	tcp_options_write((__be32 *)(th + 1), tp, &opts);
1122
	skb_shinfo(skb)->gso_type = sk->sk_gso_type;
1123 1124 1125 1126 1127 1128 1129 1130 1131
	if (likely(!(tcb->tcp_flags & TCPHDR_SYN))) {
		th->window      = htons(tcp_select_window(sk));
		tcp_ecn_send(sk, skb, th, tcp_header_size);
	} else {
		/* RFC1323: The window in SYN & SYN/ACK segments
		 * is never scaled.
		 */
		th->window	= htons(min(tp->rcv_wnd, 65535U));
	}
1132 1133 1134
#ifdef CONFIG_TCP_MD5SIG
	/* Calculate the MD5 hash, as we have all we need now */
	if (md5) {
E
Eric Dumazet 已提交
1135
		sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1136
		tp->af_specific->calc_md5_hash(opts.hash_location,
1137
					       md5, sk, skb);
1138 1139 1140
	}
#endif

1141
	icsk->icsk_af_ops->send_check(sk, skb);
L
Linus Torvalds 已提交
1142

E
Eric Dumazet 已提交
1143
	if (likely(tcb->tcp_flags & TCPHDR_ACK))
1144
		tcp_event_ack_sent(sk, tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
1145

1146
	if (skb->len != tcp_header_size) {
1147
		tcp_event_data_sent(tp, sk);
1148
		tp->data_segs_out += tcp_skb_pcount(skb);
1149
		tcp_internal_pacing(sk, skb);
1150
	}
L
Linus Torvalds 已提交
1151

1152
	if (after(tcb->end_seq, tp->snd_nxt) || tcb->seq == tcb->end_seq)
1153 1154
		TCP_ADD_STATS(sock_net(sk), TCP_MIB_OUTSEGS,
			      tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
1155

1156
	tp->segs_out += tcp_skb_pcount(skb);
1157
	/* OK, its time to fill skb_shinfo(skb)->gso_{segs|size} */
1158
	skb_shinfo(skb)->gso_segs = tcp_skb_pcount(skb);
1159
	skb_shinfo(skb)->gso_size = tcp_skb_mss(skb);
1160

E
Eric Dumazet 已提交
1161
	/* Our usage of tstamp should remain private */
T
Thomas Gleixner 已提交
1162
	skb->tstamp = 0;
1163 1164 1165 1166 1167

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

1168
	err = icsk->icsk_af_ops->queue_xmit(sk, skb, &inet->cork.fl);
E
Eric Dumazet 已提交
1169

1170 1171 1172 1173
	if (unlikely(err > 0)) {
		tcp_enter_cwr(sk);
		err = net_xmit_eval(err);
	}
E
Eric Dumazet 已提交
1174
	if (!err && oskb) {
1175
		tcp_update_skb_after_send(tp, oskb);
E
Eric Dumazet 已提交
1176 1177
		tcp_rate_skb_sent(sk, oskb);
	}
1178
	return err;
L
Linus Torvalds 已提交
1179 1180
}

1181
/* This routine just queues the buffer for sending.
L
Linus Torvalds 已提交
1182 1183 1184 1185 1186 1187 1188 1189 1190 1191
 *
 * 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;
1192
	__skb_header_release(skb);
1193
	tcp_add_write_queue_tail(sk, skb);
1194 1195
	sk->sk_wmem_queued += skb->truesize;
	sk_mem_charge(sk, skb->truesize);
L
Linus Torvalds 已提交
1196 1197
}

1198
/* Initialize TSO segments for a packet. */
1199
static void tcp_set_skb_tso_segs(struct sk_buff *skb, unsigned int mss_now)
1200
{
1201
	if (skb->len <= mss_now) {
1202 1203 1204
		/* Avoid the costly divide in the normal
		 * non-TSO case.
		 */
1205
		tcp_skb_pcount_set(skb, 1);
1206
		TCP_SKB_CB(skb)->tcp_gso_size = 0;
1207
	} else {
1208
		tcp_skb_pcount_set(skb, DIV_ROUND_UP(skb->len, mss_now));
1209
		TCP_SKB_CB(skb)->tcp_gso_size = mss_now;
L
Linus Torvalds 已提交
1210 1211 1212
	}
}

1213 1214 1215
/* Pcount in the middle of the write queue got changed, we need to do various
 * tweaks to fix counters
 */
1216
static void tcp_adjust_pcount(struct sock *sk, const struct sk_buff *skb, int decr)
1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234
{
	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);

	if (tp->lost_skb_hint &&
	    before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(tp->lost_skb_hint)->seq) &&
Y
Yuchung Cheng 已提交
1235
	    (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED))
1236 1237 1238 1239 1240
		tp->lost_cnt_hint -= decr;

	tcp_verify_left_out(tp);
}

1241 1242 1243 1244 1245 1246
static bool tcp_has_tx_tstamp(const struct sk_buff *skb)
{
	return TCP_SKB_CB(skb)->txstamp_ack ||
		(skb_shinfo(skb)->tx_flags & SKBTX_ANY_TSTAMP);
}

1247 1248 1249 1250
static void tcp_fragment_tstamp(struct sk_buff *skb, struct sk_buff *skb2)
{
	struct skb_shared_info *shinfo = skb_shinfo(skb);

1251
	if (unlikely(tcp_has_tx_tstamp(skb)) &&
1252 1253 1254 1255 1256 1257 1258
	    !before(shinfo->tskey, TCP_SKB_CB(skb2)->seq)) {
		struct skb_shared_info *shinfo2 = skb_shinfo(skb2);
		u8 tsflags = shinfo->tx_flags & SKBTX_ANY_TSTAMP;

		shinfo->tx_flags &= ~tsflags;
		shinfo2->tx_flags |= tsflags;
		swap(shinfo->tskey, shinfo2->tskey);
1259 1260
		TCP_SKB_CB(skb2)->txstamp_ack = TCP_SKB_CB(skb)->txstamp_ack;
		TCP_SKB_CB(skb)->txstamp_ack = 0;
1261 1262 1263
	}
}

1264 1265 1266 1267 1268 1269
static void tcp_skb_fragment_eor(struct sk_buff *skb, struct sk_buff *skb2)
{
	TCP_SKB_CB(skb2)->eor = TCP_SKB_CB(skb)->eor;
	TCP_SKB_CB(skb)->eor = 0;
}

1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281
/* Insert buff after skb on the write or rtx queue of sk.  */
static void tcp_insert_write_queue_after(struct sk_buff *skb,
					 struct sk_buff *buff,
					 struct sock *sk,
					 enum tcp_queue tcp_queue)
{
	if (tcp_queue == TCP_FRAG_IN_WRITE_QUEUE)
		__skb_queue_after(&sk->sk_write_queue, skb, buff);
	else
		tcp_rbtree_insert(&sk->tcp_rtx_queue, buff);
}

L
Linus Torvalds 已提交
1282 1283
/* 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
1284
 * packet to the list.  This won't be called frequently, I hope.
L
Linus Torvalds 已提交
1285 1286
 * Remember, these are still headerless SKBs at this point.
 */
1287 1288
int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
		 struct sk_buff *skb, u32 len,
1289
		 unsigned int mss_now, gfp_t gfp)
L
Linus Torvalds 已提交
1290 1291 1292
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *buff;
1293
	int nsize, old_factor;
1294
	int nlen;
1295
	u8 flags;
L
Linus Torvalds 已提交
1296

1297 1298
	if (WARN_ON(len > skb->len))
		return -EINVAL;
1299

L
Linus Torvalds 已提交
1300 1301 1302 1303
	nsize = skb_headlen(skb) - len;
	if (nsize < 0)
		nsize = 0;

1304
	if (skb_unclone(skb, gfp))
L
Linus Torvalds 已提交
1305 1306 1307
		return -ENOMEM;

	/* Get a new skb... force flag on. */
1308
	buff = sk_stream_alloc_skb(sk, nsize, gfp, true);
1309
	if (!buff)
L
Linus Torvalds 已提交
1310
		return -ENOMEM; /* We'll just try again later. */
H
Herbert Xu 已提交
1311

1312 1313
	sk->sk_wmem_queued += buff->truesize;
	sk_mem_charge(sk, buff->truesize);
1314 1315 1316
	nlen = skb->len - len - nsize;
	buff->truesize += nlen;
	skb->truesize -= nlen;
L
Linus Torvalds 已提交
1317 1318 1319 1320 1321 1322 1323

	/* 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
1324 1325 1326
	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;
1327
	TCP_SKB_CB(buff)->sacked = TCP_SKB_CB(skb)->sacked;
1328
	tcp_skb_fragment_eor(skb, buff);
L
Linus Torvalds 已提交
1329

1330
	skb_split(skb, buff, len);
L
Linus Torvalds 已提交
1331

1332
	buff->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
1333

1334
	buff->tstamp = skb->tstamp;
1335
	tcp_fragment_tstamp(skb, buff);
L
Linus Torvalds 已提交
1336

1337 1338
	old_factor = tcp_skb_pcount(skb);

L
Linus Torvalds 已提交
1339
	/* Fix up tso_factor for both original and new SKB.  */
1340 1341
	tcp_set_skb_tso_segs(skb, mss_now);
	tcp_set_skb_tso_segs(buff, mss_now);
L
Linus Torvalds 已提交
1342

1343 1344 1345
	/* Update delivered info for the new segment */
	TCP_SKB_CB(buff)->tx = TCP_SKB_CB(skb)->tx;

1346 1347 1348
	/* If this packet has been sent out already, we must
	 * adjust the various packet counters.
	 */
1349
	if (!before(tp->snd_nxt, TCP_SKB_CB(buff)->end_seq)) {
1350 1351
		int diff = old_factor - tcp_skb_pcount(skb) -
			tcp_skb_pcount(buff);
L
Linus Torvalds 已提交
1352

1353 1354
		if (diff)
			tcp_adjust_pcount(sk, skb, diff);
L
Linus Torvalds 已提交
1355 1356 1357
	}

	/* Link BUFF into the send queue. */
1358
	__skb_header_release(buff);
1359
	tcp_insert_write_queue_after(skb, buff, sk, tcp_queue);
1360 1361
	if (tcp_queue == TCP_FRAG_IN_RTX_QUEUE)
		list_add(&buff->tcp_tsorted_anchor, &skb->tcp_tsorted_anchor);
L
Linus Torvalds 已提交
1362 1363 1364 1365

	return 0;
}

1366 1367
/* This is similar to __pskb_pull_tail(). The difference is that pulled
 * data is not copied, but immediately discarded.
L
Linus Torvalds 已提交
1368
 */
1369
static int __pskb_trim_head(struct sk_buff *skb, int len)
L
Linus Torvalds 已提交
1370
{
1371
	struct skb_shared_info *shinfo;
L
Linus Torvalds 已提交
1372 1373
	int i, k, eat;

E
Eric Dumazet 已提交
1374 1375 1376 1377 1378
	eat = min_t(int, len, skb_headlen(skb));
	if (eat) {
		__skb_pull(skb, eat);
		len -= eat;
		if (!len)
1379
			return 0;
E
Eric Dumazet 已提交
1380
	}
L
Linus Torvalds 已提交
1381 1382
	eat = len;
	k = 0;
1383 1384 1385
	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
1386 1387

		if (size <= eat) {
1388
			skb_frag_unref(skb, i);
E
Eric Dumazet 已提交
1389
			eat -= size;
L
Linus Torvalds 已提交
1390
		} else {
1391
			shinfo->frags[k] = shinfo->frags[i];
L
Linus Torvalds 已提交
1392
			if (eat) {
1393 1394
				shinfo->frags[k].page_offset += eat;
				skb_frag_size_sub(&shinfo->frags[k], eat);
L
Linus Torvalds 已提交
1395 1396 1397 1398 1399
				eat = 0;
			}
			k++;
		}
	}
1400
	shinfo->nr_frags = k;
L
Linus Torvalds 已提交
1401 1402 1403

	skb->data_len -= len;
	skb->len = skb->data_len;
1404
	return len;
L
Linus Torvalds 已提交
1405 1406
}

1407
/* Remove acked data from a packet in the transmit queue. */
L
Linus Torvalds 已提交
1408 1409
int tcp_trim_head(struct sock *sk, struct sk_buff *skb, u32 len)
{
1410 1411
	u32 delta_truesize;

1412
	if (skb_unclone(skb, GFP_ATOMIC))
L
Linus Torvalds 已提交
1413 1414
		return -ENOMEM;

1415
	delta_truesize = __pskb_trim_head(skb, len);
L
Linus Torvalds 已提交
1416 1417

	TCP_SKB_CB(skb)->seq += len;
1418
	skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
1419

1420 1421 1422 1423 1424 1425
	if (delta_truesize) {
		skb->truesize	   -= delta_truesize;
		sk->sk_wmem_queued -= delta_truesize;
		sk_mem_uncharge(sk, delta_truesize);
		sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
	}
L
Linus Torvalds 已提交
1426

1427
	/* Any change of skb->len requires recalculation of tso factor. */
L
Linus Torvalds 已提交
1428
	if (tcp_skb_pcount(skb) > 1)
1429
		tcp_set_skb_tso_segs(skb, tcp_skb_mss(skb));
L
Linus Torvalds 已提交
1430 1431 1432 1433

	return 0;
}

1434 1435
/* 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
1436
{
1437 1438
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1439 1440 1441 1442 1443 1444 1445
	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);

1446 1447 1448 1449 1450 1451 1452 1453
	/* 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
1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466
	/* 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;
}

1467 1468 1469 1470 1471 1472 1473 1474
/* 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
1475
/* Inverse of above */
1476
int tcp_mss_to_mtu(struct sock *sk, int mss)
J
John Heffner 已提交
1477
{
1478 1479
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1480 1481 1482 1483 1484 1485 1486
	int mtu;

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

1487 1488 1489 1490 1491 1492 1493
	/* 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
1494 1495
	return mtu;
}
1496
EXPORT_SYMBOL(tcp_mss_to_mtu);
J
John Heffner 已提交
1497

1498
/* MTU probing init per socket */
J
John Heffner 已提交
1499 1500 1501 1502
void tcp_mtup_init(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct inet_connection_sock *icsk = inet_csk(sk);
F
Fan Du 已提交
1503
	struct net *net = sock_net(sk);
J
John Heffner 已提交
1504

F
Fan Du 已提交
1505
	icsk->icsk_mtup.enabled = net->ipv4.sysctl_tcp_mtu_probing > 1;
J
John Heffner 已提交
1506
	icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp + sizeof(struct tcphdr) +
1507
			       icsk->icsk_af_ops->net_header_len;
F
Fan Du 已提交
1508
	icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, net->ipv4.sysctl_tcp_base_mss);
J
John Heffner 已提交
1509
	icsk->icsk_mtup.probe_size = 0;
1510
	if (icsk->icsk_mtup.enabled)
1511
		icsk->icsk_mtup.probe_timestamp = tcp_jiffies32;
J
John Heffner 已提交
1512
}
E
Eric Dumazet 已提交
1513
EXPORT_SYMBOL(tcp_mtup_init);
J
John Heffner 已提交
1514

L
Linus Torvalds 已提交
1515 1516 1517 1518 1519 1520
/* 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
1521
   It is minimum of user_mss and mss received with SYN.
L
Linus Torvalds 已提交
1522 1523
   It also does not include TCP options.

1524
   inet_csk(sk)->icsk_pmtu_cookie is last pmtu, seen by this function.
L
Linus Torvalds 已提交
1525 1526 1527 1528 1529 1530 1531 1532 1533

   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.

1534 1535
   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
1536 1537 1538 1539
 */
unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu)
{
	struct tcp_sock *tp = tcp_sk(sk);
1540
	struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1541
	int mss_now;
L
Linus Torvalds 已提交
1542

J
John Heffner 已提交
1543 1544
	if (icsk->icsk_mtup.search_high > pmtu)
		icsk->icsk_mtup.search_high = pmtu;
L
Linus Torvalds 已提交
1545

J
John Heffner 已提交
1546
	mss_now = tcp_mtu_to_mss(sk, pmtu);
1547
	mss_now = tcp_bound_to_half_wnd(tp, mss_now);
L
Linus Torvalds 已提交
1548 1549

	/* And store cached results */
1550
	icsk->icsk_pmtu_cookie = pmtu;
J
John Heffner 已提交
1551 1552
	if (icsk->icsk_mtup.enabled)
		mss_now = min(mss_now, tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low));
1553
	tp->mss_cache = mss_now;
L
Linus Torvalds 已提交
1554 1555 1556

	return mss_now;
}
E
Eric Dumazet 已提交
1557
EXPORT_SYMBOL(tcp_sync_mss);
L
Linus Torvalds 已提交
1558 1559 1560 1561

/* Compute the current effective MSS, taking SACKs and IP options,
 * and even PMTU discovery events into account.
 */
I
Ilpo Järvinen 已提交
1562
unsigned int tcp_current_mss(struct sock *sk)
L
Linus Torvalds 已提交
1563
{
1564 1565
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct dst_entry *dst = __sk_dst_get(sk);
1566
	u32 mss_now;
1567
	unsigned int header_len;
A
Adam Langley 已提交
1568 1569
	struct tcp_out_options opts;
	struct tcp_md5sig_key *md5;
1570 1571 1572

	mss_now = tp->mss_cache;

L
Linus Torvalds 已提交
1573 1574
	if (dst) {
		u32 mtu = dst_mtu(dst);
1575
		if (mtu != inet_csk(sk)->icsk_pmtu_cookie)
L
Linus Torvalds 已提交
1576 1577 1578
			mss_now = tcp_sync_mss(sk, mtu);
	}

A
Adam Langley 已提交
1579 1580 1581 1582 1583 1584 1585 1586 1587 1588
	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;
	}
1589

L
Linus Torvalds 已提交
1590 1591 1592
	return mss_now;
}

1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611
/* 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;
	}
1612
	tp->snd_cwnd_stamp = tcp_jiffies32;
1613 1614
}

1615
static void tcp_cwnd_validate(struct sock *sk, bool is_cwnd_limited)
1616
{
1617
	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1618
	struct tcp_sock *tp = tcp_sk(sk);
1619

1620 1621 1622 1623 1624 1625 1626 1627 1628
	/* Track the maximum number of outstanding packets in each
	 * window, and remember whether we were cwnd-limited then.
	 */
	if (!before(tp->snd_una, tp->max_packets_seq) ||
	    tp->packets_out > tp->max_packets_out) {
		tp->max_packets_out = tp->packets_out;
		tp->max_packets_seq = tp->snd_nxt;
		tp->is_cwnd_limited = is_cwnd_limited;
	}
1629

1630
	if (tcp_is_cwnd_limited(sk)) {
1631 1632
		/* Network is feed fully. */
		tp->snd_cwnd_used = 0;
1633
		tp->snd_cwnd_stamp = tcp_jiffies32;
1634 1635 1636 1637 1638
	} else {
		/* Network starves. */
		if (tp->packets_out > tp->snd_cwnd_used)
			tp->snd_cwnd_used = tp->packets_out;

1639
		if (sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle &&
1640
		    (s32)(tcp_jiffies32 - tp->snd_cwnd_stamp) >= inet_csk(sk)->icsk_rto &&
1641
		    !ca_ops->cong_control)
1642
			tcp_cwnd_application_limited(sk);
1643 1644 1645 1646 1647

		/* The following conditions together indicate the starvation
		 * is caused by insufficient sender buffer:
		 * 1) just sent some data (see tcp_write_xmit)
		 * 2) not cwnd limited (this else condition)
1648
		 * 3) no more data to send (tcp_write_queue_empty())
1649 1650
		 * 4) application is hitting buffer limit (SOCK_NOSPACE)
		 */
1651
		if (tcp_write_queue_empty(sk) && sk->sk_socket &&
1652 1653 1654
		    test_bit(SOCK_NOSPACE, &sk->sk_socket->flags) &&
		    (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))
			tcp_chrono_start(sk, TCP_CHRONO_SNDBUF_LIMITED);
1655 1656 1657
	}
}

E
Eric Dumazet 已提交
1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671
/* 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
1672
 */
E
Eric Dumazet 已提交
1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687
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,
1688
			    int nonagle)
E
Eric Dumazet 已提交
1689 1690 1691 1692 1693
{
	return partial &&
		((nonagle & TCP_NAGLE_CORK) ||
		 (!nonagle && tp->packets_out && tcp_minshall_check(tp)));
}
E
Eric Dumazet 已提交
1694 1695 1696 1697

/* Return how many segs we'd like on a TSO packet,
 * to send one TSO packet per ms
 */
1698 1699
static u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
			    int min_tso_segs)
E
Eric Dumazet 已提交
1700 1701 1702
{
	u32 bytes, segs;

1703
	bytes = min(sk->sk_pacing_rate >> sk->sk_pacing_shift,
E
Eric Dumazet 已提交
1704 1705 1706 1707 1708 1709 1710
		    sk->sk_gso_max_size - 1 - MAX_TCP_HEADER);

	/* Goal is to send at least one packet per ms,
	 * not one big TSO packet every 100 ms.
	 * This preserves ACK clocking and is consistent
	 * with tcp_tso_should_defer() heuristic.
	 */
1711
	segs = max_t(u32, bytes / mss_now, min_tso_segs);
E
Eric Dumazet 已提交
1712

1713
	return segs;
E
Eric Dumazet 已提交
1714 1715
}

1716 1717 1718 1719 1720 1721
/* Return the number of segments we want in the skb we are transmitting.
 * See if congestion control module wants to decide; otherwise, autosize.
 */
static u32 tcp_tso_segs(struct sock *sk, unsigned int mss_now)
{
	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1722
	u32 min_tso, tso_segs;
1723

1724 1725 1726 1727 1728
	min_tso = ca_ops->min_tso_segs ?
			ca_ops->min_tso_segs(sk) :
			sock_net(sk)->ipv4.sysctl_tcp_min_tso_segs;

	tso_segs = tcp_tso_autosize(sk, mss_now, min_tso);
1729
	return min_t(u32, tso_segs, sk->sk_gso_max_segs);
1730 1731
}

E
Eric Dumazet 已提交
1732 1733 1734 1735 1736 1737
/* 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)
1738
{
1739
	const struct tcp_sock *tp = tcp_sk(sk);
E
Eric Dumazet 已提交
1740
	u32 partial, needed, window, max_len;
1741

1742
	window = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
1743
	max_len = mss_now * max_segs;
1744

1745 1746
	if (likely(max_len <= window && skb != tcp_write_queue_tail(sk)))
		return max_len;
1747

1748 1749
	needed = min(skb->len, window);

1750 1751
	if (max_len <= needed)
		return max_len;
1752

E
Eric Dumazet 已提交
1753 1754 1755 1756 1757
	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
	 */
1758
	if (tcp_nagle_check(partial != 0, tp, nonagle))
E
Eric Dumazet 已提交
1759 1760 1761
		return needed - partial;

	return needed;
1762 1763 1764 1765 1766
}

/* 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.
 */
1767 1768
static inline unsigned int tcp_cwnd_test(const struct tcp_sock *tp,
					 const struct sk_buff *skb)
1769
{
E
Eric Dumazet 已提交
1770
	u32 in_flight, cwnd, halfcwnd;
1771 1772

	/* Don't be strict about the congestion window for the final FIN.  */
E
Eric Dumazet 已提交
1773 1774
	if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) &&
	    tcp_skb_pcount(skb) == 1)
1775 1776 1777 1778
		return 1;

	in_flight = tcp_packets_in_flight(tp);
	cwnd = tp->snd_cwnd;
E
Eric Dumazet 已提交
1779 1780
	if (in_flight >= cwnd)
		return 0;
1781

E
Eric Dumazet 已提交
1782 1783 1784 1785 1786
	/* For better scheduling, ensure we have at least
	 * 2 GSO packets in flight.
	 */
	halfcwnd = max(cwnd >> 1, 1U);
	return min(halfcwnd, cwnd - in_flight);
1787 1788
}

1789
/* Initialize TSO state of a skb.
1790
 * This must be invoked the first time we consider transmitting
1791 1792
 * SKB onto the wire.
 */
1793
static int tcp_init_tso_segs(struct sk_buff *skb, unsigned int mss_now)
1794 1795 1796
{
	int tso_segs = tcp_skb_pcount(skb);

1797
	if (!tso_segs || (tso_segs > 1 && tcp_skb_mss(skb) != mss_now)) {
1798
		tcp_set_skb_tso_segs(skb, mss_now);
1799 1800 1801 1802 1803 1804
		tso_segs = tcp_skb_pcount(skb);
	}
	return tso_segs;
}


E
Eric Dumazet 已提交
1805
/* Return true if the Nagle test allows this packet to be
1806 1807
 * sent now.
 */
E
Eric Dumazet 已提交
1808 1809
static inline bool tcp_nagle_test(const struct tcp_sock *tp, const struct sk_buff *skb,
				  unsigned int cur_mss, int nonagle)
1810 1811 1812 1813 1814 1815 1816 1817
{
	/* 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
1818
		return true;
1819

Y
Yuchung Cheng 已提交
1820 1821
	/* 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
1822
		return true;
1823

1824
	if (!tcp_nagle_check(skb->len < cur_mss, tp, nonagle))
E
Eric Dumazet 已提交
1825
		return true;
1826

E
Eric Dumazet 已提交
1827
	return false;
1828 1829 1830
}

/* Does at least the first segment of SKB fit into the send window? */
E
Eric Dumazet 已提交
1831 1832 1833
static bool tcp_snd_wnd_test(const struct tcp_sock *tp,
			     const struct sk_buff *skb,
			     unsigned int cur_mss)
1834 1835 1836 1837 1838 1839
{
	u32 end_seq = TCP_SKB_CB(skb)->end_seq;

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

1840
	return !after(end_seq, tcp_wnd_end(tp));
1841 1842 1843 1844 1845 1846 1847 1848 1849
}

/* 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).
 */
1850 1851
static int tso_fragment(struct sock *sk, enum tcp_queue tcp_queue,
			struct sk_buff *skb, unsigned int len,
1852
			unsigned int mss_now, gfp_t gfp)
1853 1854 1855
{
	struct sk_buff *buff;
	int nlen = skb->len - len;
1856
	u8 flags;
1857 1858

	/* All of a TSO frame must be composed of paged data.  */
1859
	if (skb->len != skb->data_len)
1860
		return tcp_fragment(sk, tcp_queue, skb, len, mss_now, gfp);
1861

1862
	buff = sk_stream_alloc_skb(sk, 0, gfp, true);
1863
	if (unlikely(!buff))
1864 1865
		return -ENOMEM;

1866 1867
	sk->sk_wmem_queued += buff->truesize;
	sk_mem_charge(sk, buff->truesize);
1868
	buff->truesize += nlen;
1869 1870 1871 1872 1873 1874 1875 1876
	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
1877 1878 1879
	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;
1880 1881 1882 1883

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

1884 1885
	tcp_skb_fragment_eor(skb, buff);

1886
	buff->ip_summed = CHECKSUM_PARTIAL;
1887
	skb_split(skb, buff, len);
1888
	tcp_fragment_tstamp(skb, buff);
1889 1890

	/* Fix up tso_factor for both original and new SKB.  */
1891 1892
	tcp_set_skb_tso_segs(skb, mss_now);
	tcp_set_skb_tso_segs(buff, mss_now);
1893 1894

	/* Link BUFF into the send queue. */
1895
	__skb_header_release(buff);
1896
	tcp_insert_write_queue_after(skb, buff, sk, tcp_queue);
1897 1898 1899 1900 1901 1902 1903 1904 1905

	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.
 */
1906
static bool tcp_tso_should_defer(struct sock *sk, struct sk_buff *skb,
E
Eric Dumazet 已提交
1907
				 bool *is_cwnd_limited, u32 max_segs)
1908
{
1909
	const struct inet_connection_sock *icsk = inet_csk(sk);
1910 1911 1912
	u32 age, send_win, cong_win, limit, in_flight;
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *head;
1913
	int win_divisor;
1914

E
Eric Dumazet 已提交
1915
	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
J
John Heffner 已提交
1916
		goto send_now;
1917

1918
	if (icsk->icsk_ca_state >= TCP_CA_Recovery)
J
John Heffner 已提交
1919 1920
		goto send_now;

E
Eric Dumazet 已提交
1921 1922 1923
	/* Avoid bursty behavior by allowing defer
	 * only if the last write was recent.
	 */
1924
	if ((s32)(tcp_jiffies32 - tp->lsndtime) > 0)
J
John Heffner 已提交
1925
		goto send_now;
1926

1927 1928
	in_flight = tcp_packets_in_flight(tp);

1929 1930
	BUG_ON(tcp_skb_pcount(skb) <= 1);
	BUG_ON(tp->snd_cwnd <= in_flight);
1931

1932
	send_win = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
1933 1934 1935 1936 1937 1938

	/* 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);

1939
	/* If a full-sized TSO skb can be sent, do it. */
E
Eric Dumazet 已提交
1940
	if (limit >= max_segs * tp->mss_cache)
J
John Heffner 已提交
1941
		goto send_now;
1942

1943 1944 1945 1946
	/* 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;

1947
	win_divisor = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_tso_win_divisor);
1948
	if (win_divisor) {
1949 1950 1951 1952 1953
		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.
		 */
1954
		chunk /= win_divisor;
1955
		if (limit >= chunk)
J
John Heffner 已提交
1956
			goto send_now;
1957 1958 1959 1960 1961 1962
	} 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.
		 */
1963
		if (limit > tcp_max_tso_deferred_mss(tp) * tp->mss_cache)
J
John Heffner 已提交
1964
			goto send_now;
1965 1966
	}

1967 1968 1969 1970
	/* TODO : use tsorted_sent_queue ? */
	head = tcp_rtx_queue_head(sk);
	if (!head)
		goto send_now;
1971
	age = tcp_stamp_us_delta(tp->tcp_mstamp, head->skb_mstamp);
1972 1973 1974 1975
	/* If next ACK is likely to come too late (half srtt), do not defer */
	if (age < (tp->srtt_us >> 4))
		goto send_now;

E
Eric Dumazet 已提交
1976
	/* Ok, it looks like it is advisable to defer. */
J
John Heffner 已提交
1977

1978
	if (cong_win < send_win && cong_win <= skb->len)
1979 1980
		*is_cwnd_limited = true;

E
Eric Dumazet 已提交
1981
	return true;
J
John Heffner 已提交
1982 1983

send_now:
E
Eric Dumazet 已提交
1984
	return false;
1985 1986
}

1987 1988 1989 1990 1991 1992 1993 1994 1995
static inline void tcp_mtu_check_reprobe(struct sock *sk)
{
	struct inet_connection_sock *icsk = inet_csk(sk);
	struct tcp_sock *tp = tcp_sk(sk);
	struct net *net = sock_net(sk);
	u32 interval;
	s32 delta;

	interval = net->ipv4.sysctl_tcp_probe_interval;
1996
	delta = tcp_jiffies32 - icsk->icsk_mtup.probe_timestamp;
1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007
	if (unlikely(delta >= interval * HZ)) {
		int mss = tcp_current_mss(sk);

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

		/* Update probe time stamp */
2008
		icsk->icsk_mtup.probe_timestamp = tcp_jiffies32;
2009 2010 2011
	}
}

2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029
static bool tcp_can_coalesce_send_queue_head(struct sock *sk, int len)
{
	struct sk_buff *skb, *next;

	skb = tcp_send_head(sk);
	tcp_for_write_queue_from_safe(skb, next, sk) {
		if (len <= skb->len)
			break;

		if (unlikely(TCP_SKB_CB(skb)->eor))
			return false;

		len -= skb->len;
	}

	return true;
}

J
John Heffner 已提交
2030
/* Create a new MTU probe if we are ready.
2031 2032 2033 2034
 * 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
2035 2036
 * Returns 0 if we should wait to probe (no cwnd available),
 *         1 if a probe was sent,
2037 2038
 *         -1 otherwise
 */
J
John Heffner 已提交
2039 2040 2041
static int tcp_mtu_probe(struct sock *sk)
{
	struct inet_connection_sock *icsk = inet_csk(sk);
2042
	struct tcp_sock *tp = tcp_sk(sk);
J
John Heffner 已提交
2043
	struct sk_buff *skb, *nskb, *next;
2044
	struct net *net = sock_net(sk);
J
John Heffner 已提交
2045
	int probe_size;
2046
	int size_needed;
2047
	int copy, len;
J
John Heffner 已提交
2048
	int mss_now;
2049
	int interval;
J
John Heffner 已提交
2050 2051 2052 2053

	/* Not currently probing/verifying,
	 * not in recovery,
	 * have enough cwnd, and
2054 2055 2056 2057 2058 2059 2060
	 * not SACKing (the variable headers throw things off)
	 */
	if (likely(!icsk->icsk_mtup.enabled ||
		   icsk->icsk_mtup.probe_size ||
		   inet_csk(sk)->icsk_ca_state != TCP_CA_Open ||
		   tp->snd_cwnd < 11 ||
		   tp->rx_opt.num_sacks || tp->rx_opt.dsack))
J
John Heffner 已提交
2061 2062
		return -1;

2063 2064 2065 2066
	/* Use binary search for probe_size between tcp_mss_base,
	 * and current mss_clamp. if (search_high - search_low)
	 * smaller than a threshold, backoff from probing.
	 */
I
Ilpo Järvinen 已提交
2067
	mss_now = tcp_current_mss(sk);
2068 2069
	probe_size = tcp_mtu_to_mss(sk, (icsk->icsk_mtup.search_high +
				    icsk->icsk_mtup.search_low) >> 1);
2070
	size_needed = probe_size + (tp->reordering + 1) * tp->mss_cache;
2071
	interval = icsk->icsk_mtup.search_high - icsk->icsk_mtup.search_low;
2072 2073 2074 2075
	/* When misfortune happens, we are reprobing actively,
	 * and then reprobe timer has expired. We stick with current
	 * probing process by not resetting search range to its orignal.
	 */
2076
	if (probe_size > tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_high) ||
2077 2078 2079 2080 2081
		interval < net->ipv4.sysctl_tcp_probe_threshold) {
		/* Check whether enough time has elaplased for
		 * another round of probing.
		 */
		tcp_mtu_check_reprobe(sk);
J
John Heffner 已提交
2082 2083 2084 2085
		return -1;
	}

	/* Have enough data in the send queue to probe? */
2086
	if (tp->write_seq - tp->snd_nxt < size_needed)
J
John Heffner 已提交
2087 2088
		return -1;

2089 2090
	if (tp->snd_wnd < size_needed)
		return -1;
2091
	if (after(tp->snd_nxt + size_needed, tcp_wnd_end(tp)))
2092
		return 0;
J
John Heffner 已提交
2093

2094 2095 2096
	/* 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
2097 2098 2099 2100 2101
			return -1;
		else
			return 0;
	}

2102 2103 2104
	if (!tcp_can_coalesce_send_queue_head(sk, probe_size))
		return -1;

J
John Heffner 已提交
2105
	/* We're allowed to probe.  Build it now. */
2106
	nskb = sk_stream_alloc_skb(sk, probe_size, GFP_ATOMIC, false);
2107
	if (!nskb)
J
John Heffner 已提交
2108
		return -1;
2109 2110
	sk->sk_wmem_queued += nskb->truesize;
	sk_mem_charge(sk, nskb->truesize);
J
John Heffner 已提交
2111

2112
	skb = tcp_send_head(sk);
J
John Heffner 已提交
2113 2114 2115

	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
2116
	TCP_SKB_CB(nskb)->tcp_flags = TCPHDR_ACK;
J
John Heffner 已提交
2117 2118
	TCP_SKB_CB(nskb)->sacked = 0;
	nskb->csum = 0;
2119
	nskb->ip_summed = CHECKSUM_PARTIAL;
J
John Heffner 已提交
2120

2121
	tcp_insert_write_queue_before(nskb, skb, sk);
2122
	tcp_highest_sack_replace(sk, skb, nskb);
2123

J
John Heffner 已提交
2124
	len = 0;
2125
	tcp_for_write_queue_from_safe(skb, next, sk) {
J
John Heffner 已提交
2126
		copy = min_t(int, skb->len, probe_size - len);
2127
		skb_copy_bits(skb, 0, skb_put(nskb, copy), copy);
J
John Heffner 已提交
2128 2129 2130 2131

		if (skb->len <= copy) {
			/* We've eaten all the data from this skb.
			 * Throw it away. */
E
Eric Dumazet 已提交
2132
			TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;
2133 2134 2135 2136
			/* If this is the last SKB we copy and eor is set
			 * we need to propagate it to the new skb.
			 */
			TCP_SKB_CB(nskb)->eor = TCP_SKB_CB(skb)->eor;
2137
			tcp_unlink_write_queue(skb, sk);
2138
			sk_wmem_free_skb(sk, skb);
J
John Heffner 已提交
2139
		} else {
E
Eric Dumazet 已提交
2140
			TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags &
C
Changli Gao 已提交
2141
						   ~(TCPHDR_FIN|TCPHDR_PSH);
J
John Heffner 已提交
2142 2143 2144 2145
			if (!skb_shinfo(skb)->nr_frags) {
				skb_pull(skb, copy);
			} else {
				__pskb_trim_head(skb, copy);
2146
				tcp_set_skb_tso_segs(skb, mss_now);
J
John Heffner 已提交
2147 2148 2149 2150 2151
			}
			TCP_SKB_CB(skb)->seq += copy;
		}

		len += copy;
2152 2153 2154

		if (len >= probe_size)
			break;
J
John Heffner 已提交
2155
	}
2156
	tcp_init_tso_segs(nskb, nskb->len);
J
John Heffner 已提交
2157 2158

	/* We're ready to send.  If this fails, the probe will
E
Eric Dumazet 已提交
2159 2160
	 * be resegmented into mss-sized pieces by tcp_write_xmit().
	 */
J
John Heffner 已提交
2161 2162
	if (!tcp_transmit_skb(sk, nskb, 1, GFP_ATOMIC)) {
		/* Decrement cwnd here because we are sending
2163
		 * effectively two packets. */
J
John Heffner 已提交
2164
		tp->snd_cwnd--;
2165
		tcp_event_new_data_sent(sk, nskb);
J
John Heffner 已提交
2166 2167

		icsk->icsk_mtup.probe_size = tcp_mss_to_mtu(sk, nskb->len);
2168 2169
		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
2170 2171 2172 2173 2174 2175 2176

		return 1;
	}

	return -1;
}

2177 2178 2179
static bool tcp_pacing_check(const struct sock *sk)
{
	return tcp_needs_internal_pacing(sk) &&
2180
	       hrtimer_is_queued(&tcp_sk(sk)->pacing_timer);
2181 2182
}

2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198
/* TCP Small Queues :
 * Control number of packets in qdisc/devices to two packets / or ~1 ms.
 * (These limits are doubled for retransmits)
 * This allows for :
 *  - better RTT estimation and ACK scheduling
 *  - faster recovery
 *  - high rates
 * Alas, some drivers / subsystems require a fair amount
 * of queued bytes to ensure line rate.
 * One example is wifi aggregation (802.11 AMPDU)
 */
static bool tcp_small_queue_check(struct sock *sk, const struct sk_buff *skb,
				  unsigned int factor)
{
	unsigned int limit;

2199
	limit = max(2 * skb->truesize, sk->sk_pacing_rate >> sk->sk_pacing_shift);
2200 2201
	limit = min_t(u32, limit,
		      sock_net(sk)->ipv4.sysctl_tcp_limit_output_bytes);
2202 2203
	limit <<= factor;

2204
	if (refcount_read(&sk->sk_wmem_alloc) > limit) {
2205
		/* Always send skb if rtx queue is empty.
2206 2207 2208 2209
		 * No need to wait for TX completion to call us back,
		 * after softirq/tasklet schedule.
		 * This helps when TX completions are delayed too much.
		 */
2210
		if (tcp_rtx_queue_empty(sk))
2211 2212
			return false;

2213
		set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags);
2214 2215 2216 2217 2218
		/* It is possible TX completion already happened
		 * before we set TSQ_THROTTLED, so we must
		 * test again the condition.
		 */
		smp_mb__after_atomic();
2219
		if (refcount_read(&sk->sk_wmem_alloc) > limit)
2220 2221 2222 2223 2224
			return true;
	}
	return false;
}

2225 2226
static void tcp_chrono_set(struct tcp_sock *tp, const enum tcp_chrono new)
{
2227
	const u32 now = tcp_jiffies32;
2228
	enum tcp_chrono old = tp->chrono_type;
2229

2230 2231
	if (old > TCP_CHRONO_UNSPEC)
		tp->chrono_stat[old - 1] += now - tp->chrono_start;
2232 2233 2234 2235 2236 2237 2238 2239 2240
	tp->chrono_start = now;
	tp->chrono_type = new;
}

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

	/* If there are multiple conditions worthy of tracking in a
2241 2242
	 * chronograph then the highest priority enum takes precedence
	 * over the other conditions. So that if something "more interesting"
2243 2244 2245 2246 2247 2248 2249 2250 2251 2252
	 * starts happening, stop the previous chrono and start a new one.
	 */
	if (type > tp->chrono_type)
		tcp_chrono_set(tp, type);
}

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

2253 2254 2255 2256 2257 2258 2259 2260

	/* There are multiple conditions worthy of tracking in a
	 * chronograph, so that the highest priority enum takes
	 * precedence over the other conditions (see tcp_chrono_start).
	 * If a condition stops, we only stop chrono tracking if
	 * it's the "most interesting" or current chrono we are
	 * tracking and starts busy chrono if we have pending data.
	 */
2261
	if (tcp_rtx_and_write_queues_empty(sk))
2262 2263 2264
		tcp_chrono_set(tp, TCP_CHRONO_UNSPEC);
	else if (type == tp->chrono_type)
		tcp_chrono_set(tp, TCP_CHRONO_BUSY);
2265 2266
}

L
Linus Torvalds 已提交
2267 2268 2269 2270
/* This routine writes packets to the network.  It advances the
 * send_head.  This happens as incoming acks open up the remote
 * window for us.
 *
2271 2272 2273 2274
 * 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
2275 2276 2277
 * 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
2278 2279
 * 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
2280
 */
E
Eric Dumazet 已提交
2281 2282
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
2283 2284
{
	struct tcp_sock *tp = tcp_sk(sk);
2285
	struct sk_buff *skb;
2286 2287
	unsigned int tso_segs, sent_pkts;
	int cwnd_quota;
J
John Heffner 已提交
2288
	int result;
2289
	bool is_cwnd_limited = false, is_rwnd_limited = false;
E
Eric Dumazet 已提交
2290
	u32 max_segs;
L
Linus Torvalds 已提交
2291

2292
	sent_pkts = 0;
J
John Heffner 已提交
2293

2294
	tcp_mstamp_refresh(tp);
2295 2296 2297 2298
	if (!push_one) {
		/* Do MTU probing. */
		result = tcp_mtu_probe(sk);
		if (!result) {
E
Eric Dumazet 已提交
2299
			return false;
2300 2301 2302
		} else if (result > 0) {
			sent_pkts = 1;
		}
J
John Heffner 已提交
2303 2304
	}

2305
	max_segs = tcp_tso_segs(sk, mss_now);
2306
	while ((skb = tcp_send_head(sk))) {
2307 2308
		unsigned int limit;

2309 2310 2311
		if (tcp_pacing_check(sk))
			break;

2312
		tso_segs = tcp_init_tso_segs(skb, mss_now);
2313
		BUG_ON(!tso_segs);
2314

2315
		if (unlikely(tp->repair) && tp->repair_queue == TCP_SEND_QUEUE) {
E
Eric Dumazet 已提交
2316
			/* "skb_mstamp" is used as a start point for the retransmit timer */
2317
			tcp_update_skb_after_send(tp, skb);
2318
			goto repair; /* Skip network transmission */
2319
		}
2320

2321
		cwnd_quota = tcp_cwnd_test(tp, skb);
N
Nandita Dukkipati 已提交
2322 2323 2324 2325 2326 2327 2328
		if (!cwnd_quota) {
			if (push_one == 2)
				/* Force out a loss probe pkt. */
				cwnd_quota = 1;
			else
				break;
		}
2329

2330 2331
		if (unlikely(!tcp_snd_wnd_test(tp, skb, mss_now))) {
			is_rwnd_limited = true;
2332
			break;
2333
		}
2334

2335
		if (tso_segs == 1) {
2336 2337 2338 2339 2340
			if (unlikely(!tcp_nagle_test(tp, skb, mss_now,
						     (tcp_skb_is_last(sk, skb) ?
						      nonagle : TCP_NAGLE_PUSH))))
				break;
		} else {
2341
			if (!push_one &&
E
Eric Dumazet 已提交
2342 2343
			    tcp_tso_should_defer(sk, skb, &is_cwnd_limited,
						 max_segs))
2344 2345
				break;
		}
2346

E
Eric Dumazet 已提交
2347
		limit = mss_now;
2348
		if (tso_segs > 1 && !tcp_urg_mode(tp))
E
Eric Dumazet 已提交
2349 2350 2351 2352 2353 2354 2355
			limit = tcp_mss_split_point(sk, skb, mss_now,
						    min_t(unsigned int,
							  cwnd_quota,
							  max_segs),
						    nonagle);

		if (skb->len > limit &&
2356 2357
		    unlikely(tso_fragment(sk, TCP_FRAG_IN_WRITE_QUEUE,
					  skb, limit, mss_now, gfp)))
E
Eric Dumazet 已提交
2358 2359
			break;

2360 2361
		if (tcp_small_queue_check(sk, skb, 0))
			break;
E
Eric Dumazet 已提交
2362

2363
		if (unlikely(tcp_transmit_skb(sk, skb, 1, gfp)))
2364
			break;
L
Linus Torvalds 已提交
2365

2366
repair:
2367 2368 2369
		/* Advance the send_head.  This one is sent out.
		 * This call will increment packets_out.
		 */
2370
		tcp_event_new_data_sent(sk, skb);
L
Linus Torvalds 已提交
2371

2372
		tcp_minshall_update(tp, mss_now, skb);
2373
		sent_pkts += tcp_skb_pcount(skb);
2374 2375 2376

		if (push_one)
			break;
2377
	}
L
Linus Torvalds 已提交
2378

2379 2380 2381 2382 2383
	if (is_rwnd_limited)
		tcp_chrono_start(sk, TCP_CHRONO_RWND_LIMITED);
	else
		tcp_chrono_stop(sk, TCP_CHRONO_RWND_LIMITED);

2384
	if (likely(sent_pkts)) {
2385 2386
		if (tcp_in_cwnd_reduction(sk))
			tp->prr_out += sent_pkts;
N
Nandita Dukkipati 已提交
2387 2388 2389

		/* Send one loss probe per tail loss episode. */
		if (push_one != 2)
2390
			tcp_schedule_loss_probe(sk, false);
2391
		is_cwnd_limited |= (tcp_packets_in_flight(tp) >= tp->snd_cwnd);
2392
		tcp_cwnd_validate(sk, is_cwnd_limited);
E
Eric Dumazet 已提交
2393
		return false;
L
Linus Torvalds 已提交
2394
	}
2395
	return !tp->packets_out && !tcp_write_queue_empty(sk);
N
Nandita Dukkipati 已提交
2396 2397
}

2398
bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto)
N
Nandita Dukkipati 已提交
2399 2400 2401
{
	struct inet_connection_sock *icsk = inet_csk(sk);
	struct tcp_sock *tp = tcp_sk(sk);
2402
	u32 timeout, rto_delta_us;
2403
	int early_retrans;
N
Nandita Dukkipati 已提交
2404 2405 2406 2407

	/* Don't do any loss probe on a Fast Open connection before 3WHS
	 * finishes.
	 */
2408
	if (tp->fastopen_rsk)
N
Nandita Dukkipati 已提交
2409 2410
		return false;

2411
	early_retrans = sock_net(sk)->ipv4.sysctl_tcp_early_retrans;
N
Nandita Dukkipati 已提交
2412
	/* Schedule a loss probe in 2*RTT for SACK capable connections
N
Neal Cardwell 已提交
2413
	 * not in loss recovery, that are either limited by cwnd or application.
N
Nandita Dukkipati 已提交
2414
	 */
2415
	if ((early_retrans != 3 && early_retrans != 4) ||
Y
Yuchung Cheng 已提交
2416
	    !tp->packets_out || !tcp_is_sack(tp) ||
N
Neal Cardwell 已提交
2417 2418
	    (icsk->icsk_ca_state != TCP_CA_Open &&
	     icsk->icsk_ca_state != TCP_CA_CWR))
N
Nandita Dukkipati 已提交
2419 2420
		return false;

2421
	/* Probe timeout is 2*rtt. Add minimum RTO to account
2422 2423
	 * for delayed ack when there's one outstanding packet. If no RTT
	 * sample is available then probe after TCP_TIMEOUT_INIT.
N
Nandita Dukkipati 已提交
2424
	 */
2425 2426 2427 2428 2429 2430 2431 2432 2433
	if (tp->srtt_us) {
		timeout = usecs_to_jiffies(tp->srtt_us >> 2);
		if (tp->packets_out == 1)
			timeout += TCP_RTO_MIN;
		else
			timeout += TCP_TIMEOUT_MIN;
	} else {
		timeout = TCP_TIMEOUT_INIT;
	}
N
Nandita Dukkipati 已提交
2434

2435
	/* If the RTO formula yields an earlier time, then use that time. */
2436 2437 2438
	rto_delta_us = advancing_rto ?
			jiffies_to_usecs(inet_csk(sk)->icsk_rto) :
			tcp_rto_delta_us(sk);  /* How far in future is RTO? */
2439 2440
	if (rto_delta_us > 0)
		timeout = min_t(u32, timeout, usecs_to_jiffies(rto_delta_us));
N
Nandita Dukkipati 已提交
2441 2442 2443 2444 2445 2446

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

2447 2448 2449 2450 2451 2452 2453
/* Thanks to skb fast clones, we can detect if a prior transmit of
 * a packet is still in a qdisc or driver queue.
 * In this case, there is very little point doing a retransmit !
 */
static bool skb_still_in_host_queue(const struct sock *sk,
				    const struct sk_buff *skb)
{
2454
	if (unlikely(skb_fclone_busy(sk, skb))) {
2455 2456
		NET_INC_STATS(sock_net(sk),
			      LINUX_MIB_TCPSPURIOUS_RTX_HOSTQUEUES);
2457 2458 2459 2460 2461
		return true;
	}
	return false;
}

2462
/* When probe timeout (PTO) fires, try send a new segment if possible, else
N
Nandita Dukkipati 已提交
2463 2464 2465 2466
 * retransmit the last segment.
 */
void tcp_send_loss_probe(struct sock *sk)
{
N
Nandita Dukkipati 已提交
2467
	struct tcp_sock *tp = tcp_sk(sk);
N
Nandita Dukkipati 已提交
2468 2469 2470 2471
	struct sk_buff *skb;
	int pcount;
	int mss = tcp_current_mss(sk);

2472
	skb = tcp_send_head(sk);
2473 2474 2475 2476 2477 2478
	if (skb && tcp_snd_wnd_test(tp, skb, mss)) {
		pcount = tp->packets_out;
		tcp_write_xmit(sk, mss, TCP_NAGLE_OFF, 2, GFP_ATOMIC);
		if (tp->packets_out > pcount)
			goto probe_sent;
		goto rearm_timer;
N
Nandita Dukkipati 已提交
2479
	}
2480
	skb = skb_rb_last(&sk->tcp_rtx_queue);
N
Nandita Dukkipati 已提交
2481

N
Nandita Dukkipati 已提交
2482 2483 2484 2485
	/* At most one outstanding TLP retransmission. */
	if (tp->tlp_high_seq)
		goto rearm_timer;

N
Nandita Dukkipati 已提交
2486 2487 2488 2489
	/* Retransmit last segment. */
	if (WARN_ON(!skb))
		goto rearm_timer;

2490 2491 2492
	if (skb_still_in_host_queue(sk, skb))
		goto rearm_timer;

N
Nandita Dukkipati 已提交
2493 2494 2495 2496 2497
	pcount = tcp_skb_pcount(skb);
	if (WARN_ON(!pcount))
		goto rearm_timer;

	if ((pcount > 1) && (skb->len > (pcount - 1) * mss)) {
2498 2499
		if (unlikely(tcp_fragment(sk, TCP_FRAG_IN_RTX_QUEUE, skb,
					  (pcount - 1) * mss, mss,
2500
					  GFP_ATOMIC)))
N
Nandita Dukkipati 已提交
2501
			goto rearm_timer;
2502
		skb = skb_rb_next(skb);
N
Nandita Dukkipati 已提交
2503 2504 2505 2506 2507
	}

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

2508
	if (__tcp_retransmit_skb(sk, skb, 1))
2509
		goto rearm_timer;
N
Nandita Dukkipati 已提交
2510

N
Nandita Dukkipati 已提交
2511
	/* Record snd_nxt for loss detection. */
2512
	tp->tlp_high_seq = tp->snd_nxt;
N
Nandita Dukkipati 已提交
2513

2514
probe_sent:
2515
	NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPLOSSPROBES);
2516 2517
	/* Reset s.t. tcp_rearm_rto will restart timer from now */
	inet_csk(sk)->icsk_pending = 0;
N
Nandita Dukkipati 已提交
2518
rearm_timer:
2519
	tcp_rearm_rto(sk);
L
Linus Torvalds 已提交
2520 2521
}

2522 2523 2524 2525
/* 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.
 */
2526 2527
void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
			       int nonagle)
2528
{
2529 2530 2531 2532 2533 2534 2535
	/* 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;

2536
	if (tcp_write_xmit(sk, cur_mss, nonagle, 0,
2537
			   sk_gfp_mask(sk, GFP_ATOMIC)))
2538
		tcp_check_probe_timer(sk);
2539 2540
}

2541 2542 2543 2544 2545
/* 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)
{
2546
	struct sk_buff *skb = tcp_send_head(sk);
2547 2548 2549

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

2550
	tcp_write_xmit(sk, mss_now, TCP_NAGLE_PUSH, 1, sk->sk_allocation);
2551 2552
}

L
Linus Torvalds 已提交
2553 2554
/* This function returns the amount that we can raise the
 * usable window based on the following constraints
2555
 *
L
Linus Torvalds 已提交
2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573
 * 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.
2574
 *
L
Linus Torvalds 已提交
2575 2576
 * Of course, if the sender implements sender side SWS prevention
 * then this will not be a problem.
2577
 *
L
Linus Torvalds 已提交
2578
 * BSD seems to make the following compromise:
2579
 *
L
Linus Torvalds 已提交
2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606
 *	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)
{
2607
	struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
2608
	struct tcp_sock *tp = tcp_sk(sk);
S
Stephen Hemminger 已提交
2609
	/* MSS for the peer's data.  Previous versions used mss_clamp
L
Linus Torvalds 已提交
2610 2611 2612 2613 2614
	 * 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
	 */
2615
	int mss = icsk->icsk_ack.rcv_mss;
L
Linus Torvalds 已提交
2616
	int free_space = tcp_space(sk);
2617 2618
	int allowed_space = tcp_full_space(sk);
	int full_space = min_t(int, tp->window_clamp, allowed_space);
L
Linus Torvalds 已提交
2619 2620
	int window;

2621
	if (unlikely(mss > full_space)) {
2622
		mss = full_space;
2623 2624 2625
		if (mss <= 0)
			return 0;
	}
2626
	if (free_space < (full_space >> 1)) {
2627
		icsk->icsk_ack.quick = 0;
L
Linus Torvalds 已提交
2628

2629
		if (tcp_under_memory_pressure(sk))
2630 2631
			tp->rcv_ssthresh = min(tp->rcv_ssthresh,
					       4U * tp->advmss);
L
Linus Torvalds 已提交
2632

2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645
		/* 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
2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661
			return 0;
	}

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

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

		/* Advertise enough space so that it won't get scaled away.
		 * Import case: prevent zero window announcement if
		 * 1<<rcv_wscale > mss.
		 */
2662
		window = ALIGN(window, (1 << tp->rx_opt.rcv_wscale));
L
Linus Torvalds 已提交
2663
	} else {
2664
		window = tp->rcv_wnd;
L
Linus Torvalds 已提交
2665 2666 2667 2668 2669 2670 2671 2672 2673
		/* 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)
2674
			window = rounddown(free_space, mss);
2675
		else if (mss == full_space &&
2676
			 free_space > window + (full_space >> 1))
2677
			window = free_space;
L
Linus Torvalds 已提交
2678 2679 2680 2681 2682
	}

	return window;
}

2683 2684
void tcp_skb_collapse_tstamp(struct sk_buff *skb,
			     const struct sk_buff *next_skb)
2685
{
2686 2687 2688
	if (unlikely(tcp_has_tx_tstamp(next_skb))) {
		const struct skb_shared_info *next_shinfo =
			skb_shinfo(next_skb);
2689 2690
		struct skb_shared_info *shinfo = skb_shinfo(skb);

2691
		shinfo->tx_flags |= next_shinfo->tx_flags & SKBTX_ANY_TSTAMP;
2692
		shinfo->tskey = next_shinfo->tskey;
2693 2694
		TCP_SKB_CB(skb)->txstamp_ack |=
			TCP_SKB_CB(next_skb)->txstamp_ack;
2695 2696 2697
	}
}

2698
/* Collapses two adjacent SKB's during retransmission. */
2699
static bool tcp_collapse_retrans(struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
2700 2701
{
	struct tcp_sock *tp = tcp_sk(sk);
2702
	struct sk_buff *next_skb = skb_rb_next(skb);
2703
	int skb_size, next_skb_size;
L
Linus Torvalds 已提交
2704

2705 2706
	skb_size = skb->len;
	next_skb_size = next_skb->len;
L
Linus Torvalds 已提交
2707

2708
	BUG_ON(tcp_skb_pcount(skb) != 1 || tcp_skb_pcount(next_skb) != 1);
2709

2710 2711 2712 2713 2714 2715 2716
	if (next_skb_size) {
		if (next_skb_size <= skb_availroom(skb))
			skb_copy_bits(next_skb, 0, skb_put(skb, next_skb_size),
				      next_skb_size);
		else if (!skb_shift(skb, next_skb, next_skb_size))
			return false;
	}
2717
	tcp_highest_sack_replace(sk, next_skb, skb);
L
Linus Torvalds 已提交
2718

2719 2720
	/* 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
2721

2722
	/* Merge over control information. This moves PSH/FIN etc. over */
E
Eric Dumazet 已提交
2723
	TCP_SKB_CB(skb)->tcp_flags |= TCP_SKB_CB(next_skb)->tcp_flags;
2724 2725 2726 2727 2728

	/* 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;
2729
	TCP_SKB_CB(skb)->eor = TCP_SKB_CB(next_skb)->eor;
2730 2731

	/* changed transmit queue under us so clear hints */
2732 2733 2734
	tcp_clear_retrans_hints_partial(tp);
	if (next_skb == tp->retransmit_skb_hint)
		tp->retransmit_skb_hint = skb;
2735

2736 2737
	tcp_adjust_pcount(sk, next_skb, tcp_skb_pcount(next_skb));

2738 2739
	tcp_skb_collapse_tstamp(skb, next_skb);

2740
	tcp_rtx_queue_unlink_and_free(next_skb, sk);
2741
	return true;
L
Linus Torvalds 已提交
2742 2743
}

2744
/* Check if coalescing SKBs is legal. */
E
Eric Dumazet 已提交
2745
static bool tcp_can_collapse(const struct sock *sk, const struct sk_buff *skb)
2746 2747
{
	if (tcp_skb_pcount(skb) > 1)
E
Eric Dumazet 已提交
2748
		return false;
2749
	if (skb_cloned(skb))
E
Eric Dumazet 已提交
2750
		return false;
E
Eric Dumazet 已提交
2751
	/* Some heuristics for collapsing over SACK'd could be invented */
2752
	if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)
E
Eric Dumazet 已提交
2753
		return false;
2754

E
Eric Dumazet 已提交
2755
	return true;
2756 2757
}

2758 2759 2760
/* Collapse packets in the retransmit queue to make to create
 * less packets on the wire. This is only done on retransmission.
 */
2761 2762 2763 2764 2765
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
2766
	bool first = true;
2767

2768
	if (!sock_net(sk)->ipv4.sysctl_tcp_retrans_collapse)
2769
		return;
E
Eric Dumazet 已提交
2770
	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)
2771 2772
		return;

2773
	skb_rbtree_walk_from_safe(skb, tmp) {
2774 2775 2776
		if (!tcp_can_collapse(sk, skb))
			break;

2777 2778 2779
		if (!tcp_skb_can_collapse_to(to))
			break;

2780 2781 2782
		space -= skb->len;

		if (first) {
E
Eric Dumazet 已提交
2783
			first = false;
2784 2785 2786 2787 2788 2789 2790 2791 2792
			continue;
		}

		if (space < 0)
			break;

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

2793 2794
		if (!tcp_collapse_retrans(sk, to))
			break;
2795 2796 2797
	}
}

L
Linus Torvalds 已提交
2798 2799 2800 2801
/* 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.
 */
2802
int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs)
L
Linus Torvalds 已提交
2803
{
J
John Heffner 已提交
2804
	struct inet_connection_sock *icsk = inet_csk(sk);
2805
	struct tcp_sock *tp = tcp_sk(sk);
2806
	unsigned int cur_mss;
2807 2808
	int diff, len, err;

L
Linus Torvalds 已提交
2809

2810 2811
	/* Inconclusive MTU probe */
	if (icsk->icsk_mtup.probe_size)
J
John Heffner 已提交
2812 2813
		icsk->icsk_mtup.probe_size = 0;

L
Linus Torvalds 已提交
2814
	/* Do not sent more than we queued. 1/4 is reserved for possible
S
Stephen Hemminger 已提交
2815
	 * copying overhead: fragmentation, tunneling, mangling etc.
L
Linus Torvalds 已提交
2816
	 */
2817
	if (refcount_read(&sk->sk_wmem_alloc) >
2818 2819
	    min_t(u32, sk->sk_wmem_queued + (sk->sk_wmem_queued >> 2),
		  sk->sk_sndbuf))
L
Linus Torvalds 已提交
2820 2821
		return -EAGAIN;

2822 2823 2824
	if (skb_still_in_host_queue(sk, skb))
		return -EBUSY;

L
Linus Torvalds 已提交
2825
	if (before(TCP_SKB_CB(skb)->seq, tp->snd_una)) {
2826 2827 2828 2829
		if (unlikely(before(TCP_SKB_CB(skb)->end_seq, tp->snd_una))) {
			WARN_ON_ONCE(1);
			return -EINVAL;
		}
L
Linus Torvalds 已提交
2830 2831 2832 2833
		if (tcp_trim_head(sk, skb, tp->snd_una - TCP_SKB_CB(skb)->seq))
			return -ENOMEM;
	}

2834 2835 2836
	if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk))
		return -EHOSTUNREACH; /* Routing failure or similar. */

I
Ilpo Järvinen 已提交
2837
	cur_mss = tcp_current_mss(sk);
2838

L
Linus Torvalds 已提交
2839 2840 2841 2842 2843
	/* 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.
	 */
2844 2845
	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
2846 2847
		return -EAGAIN;

2848 2849
	len = cur_mss * segs;
	if (skb->len > len) {
2850 2851
		if (tcp_fragment(sk, TCP_FRAG_IN_RTX_QUEUE, skb, len,
				 cur_mss, GFP_ATOMIC))
L
Linus Torvalds 已提交
2852
			return -ENOMEM; /* We'll try again later. */
2853
	} else {
2854 2855
		if (skb_unclone(skb, GFP_ATOMIC))
			return -ENOMEM;
2856

2857 2858 2859 2860 2861 2862 2863
		diff = tcp_skb_pcount(skb);
		tcp_set_skb_tso_segs(skb, cur_mss);
		diff -= tcp_skb_pcount(skb);
		if (diff)
			tcp_adjust_pcount(sk, skb, diff);
		if (skb->len < cur_mss)
			tcp_retrans_try_collapse(sk, skb, cur_mss);
L
Linus Torvalds 已提交
2864 2865
	}

2866 2867 2868 2869
	/* RFC3168, section 6.1.1.1. ECN fallback */
	if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN_ECN) == TCPHDR_SYN_ECN)
		tcp_ecn_clear_syn(sk, skb);

2870 2871 2872 2873 2874 2875 2876
	/* Update global and local TCP statistics. */
	segs = tcp_skb_pcount(skb);
	TCP_ADD_STATS(sock_net(sk), TCP_MIB_RETRANSSEGS, segs);
	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)
		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPSYNRETRANS);
	tp->total_retrans += segs;

2877 2878 2879 2880 2881 2882
	/* 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)) {
2883 2884
		struct sk_buff *nskb;

2885 2886 2887 2888 2889 2890
		tcp_skb_tsorted_save(skb) {
			nskb = __pskb_copy(skb, MAX_TCP_HEADER, GFP_ATOMIC);
			err = nskb ? tcp_transmit_skb(sk, nskb, 0, GFP_ATOMIC) :
				     -ENOBUFS;
		} tcp_skb_tsorted_restore(skb);

2891
		if (!err) {
2892
			tcp_update_skb_after_send(tp, skb);
2893 2894
			tcp_rate_skb_sent(sk, skb);
		}
E
Eric Dumazet 已提交
2895
	} else {
2896
		err = tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
E
Eric Dumazet 已提交
2897
	}
2898

2899 2900 2901 2902
	if (BPF_SOCK_OPS_TEST_FLAG(tp, BPF_SOCK_OPS_RETRANS_CB_FLAG))
		tcp_call_bpf_3arg(sk, BPF_SOCK_OPS_RETRANS_CB,
				  TCP_SKB_CB(skb)->seq, segs, err);

2903
	if (likely(!err)) {
2904
		TCP_SKB_CB(skb)->sacked |= TCPCB_EVER_RETRANS;
2905
		trace_tcp_retransmit_skb(sk, skb);
2906 2907
	} else if (err != -EBUSY) {
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPRETRANSFAIL);
2908
	}
2909
	return err;
2910 2911
}

2912
int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs)
2913 2914
{
	struct tcp_sock *tp = tcp_sk(sk);
2915
	int err = __tcp_retransmit_skb(sk, skb, segs);
L
Linus Torvalds 已提交
2916 2917 2918

	if (err == 0) {
#if FASTRETRANS_DEBUG > 0
2919
		if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) {
2920
			net_dbg_ratelimited("retrans_out leaked\n");
L
Linus Torvalds 已提交
2921 2922 2923 2924 2925 2926 2927
		}
#endif
		TCP_SKB_CB(skb)->sacked |= TCPCB_RETRANS;
		tp->retrans_out += tcp_skb_pcount(skb);

		/* Save stamp of the first retransmit. */
		if (!tp->retrans_stamp)
E
Eric Dumazet 已提交
2928
			tp->retrans_stamp = tcp_skb_timestamp(skb);
L
Linus Torvalds 已提交
2929 2930

	}
Y
Yuchung Cheng 已提交
2931 2932 2933 2934

	if (tp->undo_retrans < 0)
		tp->undo_retrans = 0;
	tp->undo_retrans += tcp_skb_pcount(skb);
L
Linus Torvalds 已提交
2935 2936 2937 2938 2939 2940 2941 2942 2943 2944
	return err;
}

/* This gets called after a retransmit timeout, and the initially
 * retransmitted data is acknowledged.  It tries to continue
 * resending the rest of the retransmit queue, until either
 * we've sent it all or the congestion window limit is reached.
 */
void tcp_xmit_retransmit_queue(struct sock *sk)
{
2945
	const struct inet_connection_sock *icsk = inet_csk(sk);
2946
	struct sk_buff *skb, *rtx_head, *hole = NULL;
L
Linus Torvalds 已提交
2947
	struct tcp_sock *tp = tcp_sk(sk);
2948
	u32 max_segs;
2949
	int mib_idx;
2950

2951 2952 2953
	if (!tp->packets_out)
		return;

2954 2955
	rtx_head = tcp_rtx_queue_head(sk);
	skb = tp->retransmit_skb_hint ?: rtx_head;
2956
	max_segs = tcp_tso_segs(sk, tcp_current_mss(sk));
2957
	skb_rbtree_walk_from(skb) {
E
Eric Dumazet 已提交
2958
		__u8 sacked;
2959
		int segs;
L
Linus Torvalds 已提交
2960

2961 2962 2963
		if (tcp_pacing_check(sk))
			break;

2964
		/* we could do better than to assign each time */
2965
		if (!hole)
2966
			tp->retransmit_skb_hint = skb;
2967

2968 2969
		segs = tp->snd_cwnd - tcp_packets_in_flight(tp);
		if (segs <= 0)
2970
			return;
E
Eric Dumazet 已提交
2971
		sacked = TCP_SKB_CB(skb)->sacked;
2972 2973 2974 2975
		/* In case tcp_shift_skb_data() have aggregated large skbs,
		 * we need to make sure not sending too bigs TSO packets
		 */
		segs = min_t(int, segs, max_segs);
L
Linus Torvalds 已提交
2976

2977 2978
		if (tp->retrans_out >= tp->lost_out) {
			break;
2979
		} else if (!(sacked & TCPCB_LOST)) {
2980
			if (!hole && !(sacked & (TCPCB_SACKED_RETRANS|TCPCB_SACKED_ACKED)))
2981 2982
				hole = skb;
			continue;
L
Linus Torvalds 已提交
2983

2984 2985 2986 2987 2988 2989
		} else {
			if (icsk->icsk_ca_state != TCP_CA_Loss)
				mib_idx = LINUX_MIB_TCPFASTRETRANS;
			else
				mib_idx = LINUX_MIB_TCPSLOWSTARTRETRANS;
		}
L
Linus Torvalds 已提交
2990

2991
		if (sacked & (TCPCB_SACKED_ACKED|TCPCB_SACKED_RETRANS))
L
Linus Torvalds 已提交
2992 2993
			continue;

2994 2995 2996
		if (tcp_small_queue_check(sk, skb, 1))
			return;

2997
		if (tcp_retransmit_skb(sk, skb, segs))
2998
			return;
2999

3000
		NET_ADD_STATS(sock_net(sk), mib_idx, tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
3001

3002
		if (tcp_in_cwnd_reduction(sk))
3003 3004
			tp->prr_out += tcp_skb_pcount(skb);

3005
		if (skb == rtx_head &&
3006
		    icsk->icsk_pending != ICSK_TIME_REO_TIMEOUT)
3007 3008 3009
			inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
						  inet_csk(sk)->icsk_rto,
						  TCP_RTO_MAX);
L
Linus Torvalds 已提交
3010 3011 3012
	}
}

3013 3014
/* We allow to exceed memory limits for FIN packets to expedite
 * connection tear down and (memory) recovery.
3015 3016
 * Otherwise tcp_send_fin() could be tempted to either delay FIN
 * or even be forced to close flow without any FIN.
3017 3018
 * In general, we want to allow one skb per socket to avoid hangs
 * with edge trigger epoll()
3019
 */
3020
void sk_forced_mem_schedule(struct sock *sk, int size)
3021
{
3022
	int amt;
3023 3024 3025 3026 3027

	if (size <= sk->sk_forward_alloc)
		return;
	amt = sk_mem_pages(size);
	sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
3028 3029
	sk_memory_allocated_add(sk, amt);

3030 3031
	if (mem_cgroup_sockets_enabled && sk->sk_memcg)
		mem_cgroup_charge_skmem(sk->sk_memcg, amt);
3032 3033
}

3034 3035
/* Send a FIN. The caller locks the socket for us.
 * We should try to send a FIN packet really hard, but eventually give up.
L
Linus Torvalds 已提交
3036 3037 3038
 */
void tcp_send_fin(struct sock *sk)
{
3039
	struct sk_buff *skb, *tskb = tcp_write_queue_tail(sk);
3040 3041
	struct tcp_sock *tp = tcp_sk(sk);

3042 3043 3044 3045
	/* Optimization, tack on the FIN if we have one skb in write queue and
	 * this skb was not yet sent, or we are under memory pressure.
	 * Note: in the latter case, FIN packet will be sent after a timeout,
	 * as TCP stack thinks it has already been transmitted.
L
Linus Torvalds 已提交
3046
	 */
3047 3048 3049 3050
	if (!tskb && tcp_under_memory_pressure(sk))
		tskb = skb_rb_last(&sk->tcp_rtx_queue);

	if (tskb) {
3051 3052 3053
coalesce:
		TCP_SKB_CB(tskb)->tcp_flags |= TCPHDR_FIN;
		TCP_SKB_CB(tskb)->end_seq++;
L
Linus Torvalds 已提交
3054
		tp->write_seq++;
3055
		if (tcp_write_queue_empty(sk)) {
3056 3057 3058 3059 3060 3061 3062 3063 3064
			/* This means tskb was already sent.
			 * Pretend we included the FIN on previous transmit.
			 * We need to set tp->snd_nxt to the value it would have
			 * if FIN had been sent. This is because retransmit path
			 * does not change tp->snd_nxt.
			 */
			tp->snd_nxt++;
			return;
		}
L
Linus Torvalds 已提交
3065
	} else {
3066 3067 3068 3069 3070
		skb = alloc_skb_fclone(MAX_TCP_HEADER, sk->sk_allocation);
		if (unlikely(!skb)) {
			if (tskb)
				goto coalesce;
			return;
L
Linus Torvalds 已提交
3071
		}
3072
		INIT_LIST_HEAD(&skb->tcp_tsorted_anchor);
3073
		skb_reserve(skb, MAX_TCP_HEADER);
3074
		sk_forced_mem_schedule(sk, skb->truesize);
L
Linus Torvalds 已提交
3075
		/* FIN eats a sequence byte, write_seq advanced by tcp_queue_skb(). */
3076
		tcp_init_nondata_skb(skb, tp->write_seq,
C
Changli Gao 已提交
3077
				     TCPHDR_ACK | TCPHDR_FIN);
L
Linus Torvalds 已提交
3078 3079
		tcp_queue_skb(sk, skb);
	}
3080
	__tcp_push_pending_frames(sk, tcp_current_mss(sk), TCP_NAGLE_OFF);
L
Linus Torvalds 已提交
3081 3082 3083 3084 3085
}

/* 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
3086
 * by RFC 2525, section 2.17.  -DaveM
L
Linus Torvalds 已提交
3087
 */
A
Al Viro 已提交
3088
void tcp_send_active_reset(struct sock *sk, gfp_t priority)
L
Linus Torvalds 已提交
3089 3090 3091
{
	struct sk_buff *skb;

3092 3093
	TCP_INC_STATS(sock_net(sk), TCP_MIB_OUTRSTS);

L
Linus Torvalds 已提交
3094 3095 3096
	/* 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
3097
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTFAILED);
L
Linus Torvalds 已提交
3098 3099 3100 3101 3102
		return;
	}

	/* Reserve space for headers and prepare control bits. */
	skb_reserve(skb, MAX_TCP_HEADER);
3103
	tcp_init_nondata_skb(skb, tcp_acceptable_seq(sk),
C
Changli Gao 已提交
3104
			     TCPHDR_ACK | TCPHDR_RST);
3105
	tcp_mstamp_refresh(tcp_sk(sk));
L
Linus Torvalds 已提交
3106
	/* Send it off. */
3107
	if (tcp_transmit_skb(sk, skb, 0, priority))
P
Pavel Emelyanov 已提交
3108
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTFAILED);
3109 3110 3111 3112 3113

	/* skb of trace_tcp_send_reset() keeps the skb that caused RST,
	 * skb here is different to the troublesome skb, so use NULL
	 */
	trace_tcp_send_reset(sk, NULL);
L
Linus Torvalds 已提交
3114 3115
}

3116 3117
/* 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
3118 3119 3120 3121 3122 3123
 * 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)
{
3124
	struct sk_buff *skb;
L
Linus Torvalds 已提交
3125

3126
	skb = tcp_rtx_queue_head(sk);
3127
	if (!skb || !(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
3128
		pr_err("%s: wrong queue state\n", __func__);
L
Linus Torvalds 已提交
3129 3130
		return -EFAULT;
	}
E
Eric Dumazet 已提交
3131
	if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)) {
L
Linus Torvalds 已提交
3132
		if (skb_cloned(skb)) {
3133 3134 3135 3136 3137
			struct sk_buff *nskb;

			tcp_skb_tsorted_save(skb) {
				nskb = skb_copy(skb, GFP_ATOMIC);
			} tcp_skb_tsorted_restore(skb);
3138
			if (!nskb)
L
Linus Torvalds 已提交
3139
				return -ENOMEM;
3140
			INIT_LIST_HEAD(&nskb->tcp_tsorted_anchor);
3141
			tcp_rtx_queue_unlink_and_free(skb, sk);
3142
			__skb_header_release(nskb);
3143
			tcp_rbtree_insert(&sk->tcp_rtx_queue, nskb);
3144 3145
			sk->sk_wmem_queued += nskb->truesize;
			sk_mem_charge(sk, nskb->truesize);
L
Linus Torvalds 已提交
3146 3147 3148
			skb = nskb;
		}

E
Eric Dumazet 已提交
3149
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ACK;
3150
		tcp_ecn_send_synack(sk, skb);
L
Linus Torvalds 已提交
3151
	}
3152
	return tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
L
Linus Torvalds 已提交
3153 3154
}

3155 3156 3157 3158 3159 3160 3161 3162 3163
/**
 * 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.
 */
3164
struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
3165
				struct request_sock *req,
3166
				struct tcp_fastopen_cookie *foc,
3167
				enum tcp_synack_type synack_type)
L
Linus Torvalds 已提交
3168
{
3169
	struct inet_request_sock *ireq = inet_rsk(req);
3170
	const struct tcp_sock *tp = tcp_sk(sk);
E
Eric Dumazet 已提交
3171
	struct tcp_md5sig_key *md5 = NULL;
3172 3173
	struct tcp_out_options opts;
	struct sk_buff *skb;
3174
	int tcp_header_size;
3175
	struct tcphdr *th;
3176
	int mss;
L
Linus Torvalds 已提交
3177

3178
	skb = alloc_skb(MAX_TCP_HEADER, GFP_ATOMIC);
3179 3180
	if (unlikely(!skb)) {
		dst_release(dst);
L
Linus Torvalds 已提交
3181
		return NULL;
3182
	}
L
Linus Torvalds 已提交
3183 3184 3185
	/* Reserve space for headers. */
	skb_reserve(skb, MAX_TCP_HEADER);

3186 3187
	switch (synack_type) {
	case TCP_SYNACK_NORMAL:
3188
		skb_set_owner_w(skb, req_to_sk(req));
3189 3190 3191 3192 3193 3194 3195
		break;
	case TCP_SYNACK_COOKIE:
		/* Under synflood, we do not attach skb to a socket,
		 * to avoid false sharing.
		 */
		break;
	case TCP_SYNACK_FASTOPEN:
3196 3197 3198 3199 3200
		/* sk is a const pointer, because we want to express multiple
		 * cpu might call us concurrently.
		 * sk->sk_wmem_alloc in an atomic, we can promote to rw.
		 */
		skb_set_owner_w(skb, (struct sock *)sk);
3201
		break;
3202
	}
3203
	skb_dst_set(skb, dst);
L
Linus Torvalds 已提交
3204

E
Eric Dumazet 已提交
3205
	mss = tcp_mss_clamp(tp, dst_metric_advmss(dst));
3206

A
Adam Langley 已提交
3207
	memset(&opts, 0, sizeof(opts));
3208 3209
#ifdef CONFIG_SYN_COOKIES
	if (unlikely(req->cookie_ts))
3210
		skb->skb_mstamp = cookie_init_timestamp(req);
3211 3212
	else
#endif
3213
		skb->skb_mstamp = tcp_clock_us();
E
Eric Dumazet 已提交
3214 3215 3216

#ifdef CONFIG_TCP_MD5SIG
	rcu_read_lock();
3217
	md5 = tcp_rsk(req)->af_specific->req_md5_lookup(sk, req_to_sk(req));
E
Eric Dumazet 已提交
3218
#endif
3219
	skb_set_hash(skb, tcp_rsk(req)->txhash, PKT_HASH_TYPE_L4);
3220 3221
	tcp_header_size = tcp_synack_options(sk, req, mss, skb, &opts, md5,
					     foc) + sizeof(*th);
3222

3223 3224
	skb_push(skb, tcp_header_size);
	skb_reset_transport_header(skb);
L
Linus Torvalds 已提交
3225

3226
	th = (struct tcphdr *)skb->data;
L
Linus Torvalds 已提交
3227 3228 3229
	memset(th, 0, sizeof(struct tcphdr));
	th->syn = 1;
	th->ack = 1;
3230
	tcp_ecn_make_synack(req, th);
E
Eric Dumazet 已提交
3231
	th->source = htons(ireq->ir_num);
3232
	th->dest = ireq->ir_rmt_port;
3233
	skb->mark = ireq->ir_mark;
3234 3235
	skb->ip_summed = CHECKSUM_PARTIAL;
	th->seq = htonl(tcp_rsk(req)->snt_isn);
3236 3237
	/* 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
3238 3239

	/* RFC1323: The window in SYN & SYN/ACK segments is never scaled. */
3240
	th->window = htons(min(req->rsk_rcv_wnd, 65535U));
3241
	tcp_options_write((__be32 *)(th + 1), NULL, &opts);
L
Linus Torvalds 已提交
3242
	th->doff = (tcp_header_size >> 2);
E
Eric Dumazet 已提交
3243
	__TCP_INC_STATS(sock_net(sk), TCP_MIB_OUTSEGS);
3244 3245 3246

#ifdef CONFIG_TCP_MD5SIG
	/* Okay, we have all we need - do the md5 hash if needed */
E
Eric Dumazet 已提交
3247
	if (md5)
3248
		tcp_rsk(req)->af_specific->calc_md5_hash(opts.hash_location,
3249
					       md5, req_to_sk(req), skb);
E
Eric Dumazet 已提交
3250
	rcu_read_unlock();
3251 3252
#endif

3253
	/* Do not fool tcpdump (if any), clean our debris */
T
Thomas Gleixner 已提交
3254
	skb->tstamp = 0;
L
Linus Torvalds 已提交
3255 3256
	return skb;
}
E
Eric Dumazet 已提交
3257
EXPORT_SYMBOL(tcp_make_synack);
L
Linus Torvalds 已提交
3258

3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277
static void tcp_ca_dst_init(struct sock *sk, const struct dst_entry *dst)
{
	struct inet_connection_sock *icsk = inet_csk(sk);
	const struct tcp_congestion_ops *ca;
	u32 ca_key = dst_metric(dst, RTAX_CC_ALGO);

	if (ca_key == TCP_CA_UNSPEC)
		return;

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

3278
/* Do all connect socket setups that can be done AF independent. */
3279
static void tcp_connect_init(struct sock *sk)
L
Linus Torvalds 已提交
3280
{
3281
	const struct dst_entry *dst = __sk_dst_get(sk);
L
Linus Torvalds 已提交
3282 3283
	struct tcp_sock *tp = tcp_sk(sk);
	__u8 rcv_wscale;
3284
	u32 rcv_wnd;
L
Linus Torvalds 已提交
3285 3286 3287 3288

	/* 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.
	 */
3289 3290 3291
	tp->tcp_header_len = sizeof(struct tcphdr);
	if (sock_net(sk)->ipv4.sysctl_tcp_timestamps)
		tp->tcp_header_len += TCPOLEN_TSTAMP_ALIGNED;
L
Linus Torvalds 已提交
3292

3293
#ifdef CONFIG_TCP_MD5SIG
3294
	if (tp->af_specific->md5_lookup(sk, sk))
3295 3296 3297
		tp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
#endif

L
Linus Torvalds 已提交
3298 3299 3300 3301
	/* 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
3302
	tcp_mtup_init(sk);
L
Linus Torvalds 已提交
3303 3304
	tcp_sync_mss(sk, dst_mtu(dst));

3305 3306
	tcp_ca_dst_init(sk, dst);

L
Linus Torvalds 已提交
3307 3308
	if (!tp->window_clamp)
		tp->window_clamp = dst_metric(dst, RTAX_WINDOW);
E
Eric Dumazet 已提交
3309
	tp->advmss = tcp_mss_clamp(tp, dst_metric_advmss(dst));
3310

L
Linus Torvalds 已提交
3311 3312
	tcp_initialize_rcv_mss(sk);

3313 3314 3315 3316 3317
	/* 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);

3318 3319 3320 3321
	rcv_wnd = tcp_rwnd_init_bpf(sk);
	if (rcv_wnd == 0)
		rcv_wnd = dst_metric(dst, RTAX_INITRWND);

3322
	tcp_select_initial_window(sk, tcp_full_space(sk),
L
Linus Torvalds 已提交
3323 3324 3325
				  tp->advmss - (tp->rx_opt.ts_recent_stamp ? tp->tcp_header_len - sizeof(struct tcphdr) : 0),
				  &tp->rcv_wnd,
				  &tp->window_clamp,
3326
				  sock_net(sk)->ipv4.sysctl_tcp_window_scaling,
3327
				  &rcv_wscale,
3328
				  rcv_wnd);
L
Linus Torvalds 已提交
3329 3330 3331 3332 3333 3334 3335

	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;
3336
	tcp_init_wl(tp, 0);
3337
	tcp_write_queue_purge(sk);
L
Linus Torvalds 已提交
3338 3339
	tp->snd_una = tp->write_seq;
	tp->snd_sml = tp->write_seq;
I
Ilpo Järvinen 已提交
3340
	tp->snd_up = tp->write_seq;
P
Pavel Emelyanov 已提交
3341
	tp->snd_nxt = tp->write_seq;
P
Pavel Emelyanov 已提交
3342 3343 3344

	if (likely(!tp->repair))
		tp->rcv_nxt = 0;
3345
	else
3346
		tp->rcv_tstamp = tcp_jiffies32;
P
Pavel Emelyanov 已提交
3347 3348
	tp->rcv_wup = tp->rcv_nxt;
	tp->copied_seq = tp->rcv_nxt;
L
Linus Torvalds 已提交
3349

3350
	inet_csk(sk)->icsk_rto = tcp_timeout_init(sk);
3351
	inet_csk(sk)->icsk_retransmits = 0;
L
Linus Torvalds 已提交
3352 3353 3354
	tcp_clear_retrans(tp);
}

3355 3356 3357 3358 3359 3360
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;
3361
	__skb_header_release(skb);
3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378
	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;
3379
	int space, err = 0;
3380
	struct sk_buff *syn_data;
3381

3382
	tp->rx_opt.mss_clamp = tp->advmss;  /* If MSS is not cached */
3383
	if (!tcp_fastopen_cookie_check(sk, &tp->rx_opt.mss_clamp, &fo->cookie))
3384 3385 3386 3387 3388 3389
		goto fallback;

	/* MSS for SYN-data is based on cached MSS and bounded by PMTU and
	 * user-MSS. Reserve maximum option space for middleboxes that add
	 * private TCP options. The cost is reduced data space in SYN :(
	 */
E
Eric Dumazet 已提交
3390 3391
	tp->rx_opt.mss_clamp = tcp_mss_clamp(tp, tp->rx_opt.mss_clamp);

3392
	space = __tcp_mtu_to_mss(sk, inet_csk(sk)->icsk_pmtu_cookie) -
3393 3394
		MAX_TCP_OPTION_SPACE;

3395 3396 3397 3398 3399
	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));

3400
	syn_data = sk_stream_alloc_skb(sk, space, sk->sk_allocation, false);
3401
	if (!syn_data)
3402
		goto fallback;
3403 3404
	syn_data->ip_summed = CHECKSUM_PARTIAL;
	memcpy(syn_data->cb, syn->cb, sizeof(syn->cb));
3405 3406 3407 3408
	if (space) {
		int copied = copy_from_iter(skb_put(syn_data, space), space,
					    &fo->data->msg_iter);
		if (unlikely(!copied)) {
E
Eric Dumazet 已提交
3409
			tcp_skb_tsorted_anchor_cleanup(syn_data);
3410 3411 3412 3413 3414 3415 3416
			kfree_skb(syn_data);
			goto fallback;
		}
		if (copied != space) {
			skb_trim(syn_data, copied);
			space = copied;
		}
3417
	}
3418 3419 3420 3421
	/* No more data pending in inet_wait_for_connect() */
	if (space == fo->size)
		fo->data = NULL;
	fo->copied = space;
3422

3423
	tcp_connect_queue_skb(sk, syn_data);
3424 3425
	if (syn_data->len)
		tcp_chrono_start(sk, TCP_CHRONO_BUSY);
3426

3427
	err = tcp_transmit_skb(sk, syn_data, 1, sk->sk_allocation);
3428

3429
	syn->skb_mstamp = syn_data->skb_mstamp;
E
Eric Dumazet 已提交
3430

3431 3432 3433 3434 3435 3436 3437 3438
	/* Now full SYN+DATA was cloned and sent (or not),
	 * remove the SYN from the original skb (syn_data)
	 * we keep in write queue in case of a retransmit, as we
	 * also have the SYN packet (with no data) in the same queue.
	 */
	TCP_SKB_CB(syn_data)->seq++;
	TCP_SKB_CB(syn_data)->tcp_flags = TCPHDR_ACK | TCPHDR_PSH;
	if (!err) {
3439
		tp->syn_data = (fo->copied > 0);
3440
		tcp_rbtree_insert(&sk->tcp_rtx_queue, syn_data);
3441
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPORIGDATASENT);
3442 3443 3444
		goto done;
	}

3445 3446
	/* data was not sent, put it in write_queue */
	__skb_queue_tail(&sk->sk_write_queue, syn_data);
3447 3448
	tp->packets_out -= tcp_skb_pcount(syn_data);

3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460
fallback:
	/* Send a regular SYN with Fast Open cookie request option */
	if (fo->cookie.len > 0)
		fo->cookie.len = 0;
	err = tcp_transmit_skb(sk, syn, 1, sk->sk_allocation);
	if (err)
		tp->syn_fastopen = 0;
done:
	fo->cookie.len = -1;  /* Exclude Fast Open option for SYN retries */
	return err;
}

3461
/* Build a SYN and send it off. */
L
Linus Torvalds 已提交
3462 3463 3464 3465
int tcp_connect(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *buff;
3466
	int err;
L
Linus Torvalds 已提交
3467

3468
	tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_CONNECT_CB, 0, NULL);
3469 3470 3471 3472

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

L
Linus Torvalds 已提交
3473 3474
	tcp_connect_init(sk);

A
Andrey Vagin 已提交
3475 3476 3477 3478 3479
	if (unlikely(tp->repair)) {
		tcp_finish_connect(sk, NULL);
		return 0;
	}

3480
	buff = sk_stream_alloc_skb(sk, 0, sk->sk_allocation, true);
3481
	if (unlikely(!buff))
L
Linus Torvalds 已提交
3482 3483
		return -ENOBUFS;

C
Changli Gao 已提交
3484
	tcp_init_nondata_skb(buff, tp->write_seq++, TCPHDR_SYN);
3485 3486
	tcp_mstamp_refresh(tp);
	tp->retrans_stamp = tcp_time_stamp(tp);
3487
	tcp_connect_queue_skb(sk, buff);
3488
	tcp_ecn_send_syn(sk, buff);
3489
	tcp_rbtree_insert(&sk->tcp_rtx_queue, buff);
L
Linus Torvalds 已提交
3490

3491 3492 3493
	/* 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);
3494 3495
	if (err == -ECONNREFUSED)
		return err;
3496 3497 3498 3499 3500 3501

	/* 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;
3502 3503 3504 3505 3506
	buff = tcp_send_head(sk);
	if (unlikely(buff)) {
		tp->snd_nxt	= TCP_SKB_CB(buff)->seq;
		tp->pushed_seq	= TCP_SKB_CB(buff)->seq;
	}
P
Pavel Emelyanov 已提交
3507
	TCP_INC_STATS(sock_net(sk), TCP_MIB_ACTIVEOPENS);
L
Linus Torvalds 已提交
3508 3509

	/* Timer for repeating the SYN until an answer. */
3510 3511
	inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
				  inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
L
Linus Torvalds 已提交
3512 3513
	return 0;
}
E
Eric Dumazet 已提交
3514
EXPORT_SYMBOL(tcp_connect);
L
Linus Torvalds 已提交
3515 3516 3517 3518 3519 3520 3521

/* 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)
{
3522 3523
	struct inet_connection_sock *icsk = inet_csk(sk);
	int ato = icsk->icsk_ack.ato;
L
Linus Torvalds 已提交
3524 3525 3526
	unsigned long timeout;

	if (ato > TCP_DELACK_MIN) {
3527
		const struct tcp_sock *tp = tcp_sk(sk);
3528
		int max_ato = HZ / 2;
L
Linus Torvalds 已提交
3529

3530 3531
		if (icsk->icsk_ack.pingpong ||
		    (icsk->icsk_ack.pending & ICSK_ACK_PUSHED))
L
Linus Torvalds 已提交
3532 3533 3534 3535 3536
			max_ato = TCP_DELACK_MAX;

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

		/* If some rtt estimate is known, use it to bound delayed ack.
3537
		 * Do not use inet_csk(sk)->icsk_rto here, use results of rtt measurements
L
Linus Torvalds 已提交
3538 3539
		 * directly.
		 */
3540 3541 3542
		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
3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554

			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. */
3555
	if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
L
Linus Torvalds 已提交
3556 3557 3558
		/* If delack timer was blocked or is about to expire,
		 * send ACK now.
		 */
3559 3560
		if (icsk->icsk_ack.blocked ||
		    time_before_eq(icsk->icsk_ack.timeout, jiffies + (ato >> 2))) {
L
Linus Torvalds 已提交
3561 3562 3563 3564
			tcp_send_ack(sk);
			return;
		}

3565 3566
		if (!time_before(timeout, icsk->icsk_ack.timeout))
			timeout = icsk->icsk_ack.timeout;
L
Linus Torvalds 已提交
3567
	}
3568 3569 3570
	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
3571 3572 3573 3574 3575
}

/* This routine sends an ack and also updates the window. */
void tcp_send_ack(struct sock *sk)
{
3576
	struct sk_buff *buff;
L
Linus Torvalds 已提交
3577

3578 3579 3580
	/* If we have been reset, we may not send again. */
	if (sk->sk_state == TCP_CLOSE)
		return;
L
Linus Torvalds 已提交
3581

3582 3583 3584 3585
	/* We are not putting this on the write queue, so
	 * tcp_transmit_skb() will set the ownership to this
	 * sock.
	 */
3586 3587 3588
	buff = alloc_skb(MAX_TCP_HEADER,
			 sk_gfp_mask(sk, GFP_ATOMIC | __GFP_NOWARN));
	if (unlikely(!buff)) {
3589 3590 3591 3592 3593
		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
3594
	}
3595 3596 3597

	/* Reserve space for headers and prepare control bits. */
	skb_reserve(buff, MAX_TCP_HEADER);
C
Changli Gao 已提交
3598
	tcp_init_nondata_skb(buff, tcp_acceptable_seq(sk), TCPHDR_ACK);
3599

E
Eric Dumazet 已提交
3600 3601 3602 3603 3604 3605
	/* We do not want pure acks influencing TCP Small Queues or fq/pacing
	 * too much.
	 * SKB_TRUESIZE(max(1 .. 66, MAX_TCP_HEADER)) is unfortunately ~784
	 */
	skb_set_tcp_pure_ack(buff);

3606
	/* Send it off, this clears delayed acks for us. */
3607
	tcp_transmit_skb(sk, buff, 0, (__force gfp_t)0);
L
Linus Torvalds 已提交
3608
}
3609
EXPORT_SYMBOL_GPL(tcp_send_ack);
L
Linus Torvalds 已提交
3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621

/* 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.
 */
3622
static int tcp_xmit_probe_skb(struct sock *sk, int urgent, int mib)
L
Linus Torvalds 已提交
3623 3624 3625 3626 3627
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb;

	/* We don't queue it, tcp_transmit_skb() sets ownership. */
3628 3629
	skb = alloc_skb(MAX_TCP_HEADER,
			sk_gfp_mask(sk, GFP_ATOMIC | __GFP_NOWARN));
3630
	if (!skb)
L
Linus Torvalds 已提交
3631 3632 3633 3634 3635 3636 3637 3638
		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
3639
	tcp_init_nondata_skb(skb, tp->snd_una - !urgent, TCPHDR_ACK);
3640
	NET_INC_STATS(sock_net(sk), mib);
3641
	return tcp_transmit_skb(sk, skb, 0, (__force gfp_t)0);
L
Linus Torvalds 已提交
3642 3643
}

3644
/* Called from setsockopt( ... TCP_REPAIR ) */
P
Pavel Emelyanov 已提交
3645 3646 3647 3648
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;
3649
		tcp_mstamp_refresh(tcp_sk(sk));
3650
		tcp_xmit_probe_skb(sk, 0, LINUX_MIB_TCPWINPROBE);
P
Pavel Emelyanov 已提交
3651 3652 3653
	}
}

3654
/* Initiate keepalive or window probe from timer. */
3655
int tcp_write_wakeup(struct sock *sk, int mib)
L
Linus Torvalds 已提交
3656
{
3657 3658
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb;
L
Linus Torvalds 已提交
3659

3660 3661 3662
	if (sk->sk_state == TCP_CLOSE)
		return -1;

3663 3664
	skb = tcp_send_head(sk);
	if (skb && before(TCP_SKB_CB(skb)->seq, tcp_wnd_end(tp))) {
3665
		int err;
I
Ilpo Järvinen 已提交
3666
		unsigned int mss = tcp_current_mss(sk);
3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678
		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
3679
			TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
3680 3681
			if (tcp_fragment(sk, TCP_FRAG_IN_WRITE_QUEUE,
					 skb, seg_size, mss, GFP_ATOMIC))
3682 3683
				return -1;
		} else if (!tcp_skb_pcount(skb))
3684
			tcp_set_skb_tso_segs(skb, mss);
3685

E
Eric Dumazet 已提交
3686
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
3687 3688 3689 3690 3691
		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
3692
		if (between(tp->snd_up, tp->snd_una + 1, tp->snd_una + 0xFFFF))
3693 3694
			tcp_xmit_probe_skb(sk, 1, mib);
		return tcp_xmit_probe_skb(sk, 0, mib);
L
Linus Torvalds 已提交
3695 3696 3697 3698 3699 3700 3701 3702
	}
}

/* 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)
{
3703
	struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
3704
	struct tcp_sock *tp = tcp_sk(sk);
3705
	struct net *net = sock_net(sk);
3706
	unsigned long probe_max;
L
Linus Torvalds 已提交
3707 3708
	int err;

3709
	err = tcp_write_wakeup(sk, LINUX_MIB_TCPWINPROBE);
L
Linus Torvalds 已提交
3710

3711
	if (tp->packets_out || tcp_write_queue_empty(sk)) {
L
Linus Torvalds 已提交
3712
		/* Cancel probe timer, if it is not required. */
3713
		icsk->icsk_probes_out = 0;
3714
		icsk->icsk_backoff = 0;
L
Linus Torvalds 已提交
3715 3716 3717 3718
		return;
	}

	if (err <= 0) {
3719
		if (icsk->icsk_backoff < net->ipv4.sysctl_tcp_retries2)
3720
			icsk->icsk_backoff++;
3721
		icsk->icsk_probes_out++;
3722
		probe_max = TCP_RTO_MAX;
L
Linus Torvalds 已提交
3723 3724
	} else {
		/* If packet was not sent due to local congestion,
3725
		 * do not backoff and do not remember icsk_probes_out.
L
Linus Torvalds 已提交
3726 3727 3728 3729
		 * Let local senders to fight for local resources.
		 *
		 * Use accumulated backoff yet.
		 */
3730 3731
		if (!icsk->icsk_probes_out)
			icsk->icsk_probes_out = 1;
3732
		probe_max = TCP_RESOURCE_PROBE_INTERVAL;
L
Linus Torvalds 已提交
3733
	}
3734
	inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
3735
				  tcp_probe0_when(sk, probe_max),
3736
				  TCP_RTO_MAX);
L
Linus Torvalds 已提交
3737
}
3738

3739
int tcp_rtx_synack(const struct sock *sk, struct request_sock *req)
3740 3741 3742 3743 3744
{
	const struct tcp_request_sock_ops *af_ops = tcp_rsk(req)->af_specific;
	struct flowi fl;
	int res;

3745
	tcp_rsk(req)->txhash = net_tx_rndhash();
3746
	res = af_ops->send_synack(sk, NULL, &fl, req, NULL, TCP_SYNACK_NORMAL);
3747
	if (!res) {
E
Eric Dumazet 已提交
3748
		__TCP_INC_STATS(sock_net(sk), TCP_MIB_RETRANSSEGS);
3749
		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPSYNRETRANS);
3750 3751
		if (unlikely(tcp_passive_fastopen(sk)))
			tcp_sk(sk)->total_retrans++;
3752
		trace_tcp_retransmit_synack(sk, req);
3753 3754 3755 3756
	}
	return res;
}
EXPORT_SYMBOL(tcp_rtx_synack);