tcp_output.c 118.2 KB
Newer Older
1
// SPDX-License-Identifier: GPL-2.0-only
L
Linus Torvalds 已提交
2 3 4 5 6 7 8
/*
 * 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).
 *
9
 * Authors:	Ross Biro
L
Linus Torvalds 已提交
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 37
 *		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
 *
 */

38 39
#define pr_fmt(fmt) "TCP: " fmt

L
Linus Torvalds 已提交
40
#include <net/tcp.h>
P
Peter Krystad 已提交
41
#include <net/mptcp.h>
L
Linus Torvalds 已提交
42 43

#include <linux/compiler.h>
44
#include <linux/gfp.h>
L
Linus Torvalds 已提交
45
#include <linux/module.h>
46
#include <linux/static_key.h>
L
Linus Torvalds 已提交
47

48
#include <trace/events/tcp.h>
49

50 51 52 53 54 55 56
/* Refresh clocks of a TCP socket,
 * ensuring monotically increasing values.
 */
void tcp_mstamp_refresh(struct tcp_sock *tp)
{
	u64 val = tcp_clock_ns();

57 58
	tp->tcp_clock_cache = val;
	tp->tcp_mstamp = div_u64(val, NSEC_PER_USEC);
59 60
}

E
Eric Dumazet 已提交
61 62
static bool tcp_write_xmit(struct sock *sk, unsigned int mss_now, int nonagle,
			   int push_one, gfp_t gfp);
63

64
/* Account for new data that has been sent to the network. */
65
static void tcp_event_new_data_sent(struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
66
{
N
Nandita Dukkipati 已提交
67
	struct inet_connection_sock *icsk = inet_csk(sk);
68
	struct tcp_sock *tp = tcp_sk(sk);
69
	unsigned int prior_packets = tp->packets_out;
70

71
	WRITE_ONCE(tp->snd_nxt, TCP_SKB_CB(skb)->end_seq);
72

73 74 75
	__skb_unlink(skb, &sk->sk_write_queue);
	tcp_rbtree_insert(&sk->tcp_rtx_queue, skb);

76 77 78
	if (tp->highest_sack == NULL)
		tp->highest_sack = skb;

79
	tp->packets_out += tcp_skb_pcount(skb);
Y
Yuchung Cheng 已提交
80
	if (!prior_packets || icsk->icsk_pending == ICSK_TIME_LOSS_PROBE)
81
		tcp_rearm_rto(sk);
82

83 84
	NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPORIGDATASENT,
		      tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
85 86
}

87 88
/* 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
89 90 91 92 93
 * If window has been shrunk, what should we make? It is not clear at all.
 * Using SND.UNA we will fail to open window, SND.NXT is out of window. :-(
 * Anything in between SND.UNA...SND.UNA+SND.WND also can be already
 * invalid. OK, let's make this for now:
 */
94
static inline __u32 tcp_acceptable_seq(const struct sock *sk)
L
Linus Torvalds 已提交
95
{
96
	const struct tcp_sock *tp = tcp_sk(sk);
97

98 99 100
	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
101 102
		return tp->snd_nxt;
	else
103
		return tcp_wnd_end(tp);
L
Linus Torvalds 已提交
104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122
}

/* 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);
123
	const struct dst_entry *dst = __sk_dst_get(sk);
L
Linus Torvalds 已提交
124 125
	int mss = tp->advmss;

126 127 128 129 130 131 132
	if (dst) {
		unsigned int metric = dst_metric_advmss(dst);

		if (metric < mss) {
			mss = metric;
			tp->advmss = mss;
		}
L
Linus Torvalds 已提交
133 134 135 136 137 138
	}

	return (__u16)mss;
}

/* RFC2861. Reset CWND after idle period longer RTO to "restart window".
139 140 141
 * This is the first part of cwnd validation mechanism.
 */
void tcp_cwnd_restart(struct sock *sk, s32 delta)
L
Linus Torvalds 已提交
142
{
143
	struct tcp_sock *tp = tcp_sk(sk);
144
	u32 restart_cwnd = tcp_init_cwnd(tp, __sk_dst_get(sk));
L
Linus Torvalds 已提交
145 146
	u32 cwnd = tp->snd_cwnd;

147
	tcp_ca_event(sk, CA_EVENT_CWND_RESTART);
L
Linus Torvalds 已提交
148

149
	tp->snd_ssthresh = tcp_current_ssthresh(sk);
L
Linus Torvalds 已提交
150 151
	restart_cwnd = min(restart_cwnd, cwnd);

152
	while ((delta -= inet_csk(sk)->icsk_rto) > 0 && cwnd > restart_cwnd)
L
Linus Torvalds 已提交
153 154
		cwnd >>= 1;
	tp->snd_cwnd = max(cwnd, restart_cwnd);
155
	tp->snd_cwnd_stamp = tcp_jiffies32;
L
Linus Torvalds 已提交
156 157 158
	tp->snd_cwnd_used = 0;
}

159
/* Congestion state accounting after a packet has been sent. */
S
Stephen Hemminger 已提交
160
static void tcp_event_data_sent(struct tcp_sock *tp,
161
				struct sock *sk)
L
Linus Torvalds 已提交
162
{
163
	struct inet_connection_sock *icsk = inet_csk(sk);
164
	const u32 now = tcp_jiffies32;
L
Linus Torvalds 已提交
165

166 167 168
	if (tcp_packets_in_flight(tp) == 0)
		tcp_ca_event(sk, CA_EVENT_TX_START);

W
Wei Wang 已提交
169 170 171 172
	/* If this is the first data packet sent in response to the
	 * previous received data,
	 * and it is a reply for ato after last received packet,
	 * increase pingpong count.
L
Linus Torvalds 已提交
173
	 */
W
Wei Wang 已提交
174 175 176 177 178
	if (before(tp->lsndtime, icsk->icsk_ack.lrcvtime) &&
	    (u32)(now - icsk->icsk_ack.lrcvtime) < icsk->icsk_ack.ato)
		inet_csk_inc_pingpong_cnt(sk);

	tp->lsndtime = now;
L
Linus Torvalds 已提交
179 180
}

181
/* Account for an ACK we sent. */
182 183
static inline void tcp_event_ack_sent(struct sock *sk, unsigned int pkts,
				      u32 rcv_nxt)
L
Linus Torvalds 已提交
184
{
E
Eric Dumazet 已提交
185 186
	struct tcp_sock *tp = tcp_sk(sk);

E
Eric Dumazet 已提交
187
	if (unlikely(tp->compressed_ack)) {
188
		NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPACKCOMPRESSED,
E
Eric Dumazet 已提交
189 190
			      tp->compressed_ack);
		tp->compressed_ack = 0;
E
Eric Dumazet 已提交
191 192 193
		if (hrtimer_try_to_cancel(&tp->compressed_ack_timer) == 1)
			__sock_put(sk);
	}
194 195 196

	if (unlikely(rcv_nxt != tp->rcv_nxt))
		return;  /* Special ACK sent by DCTCP to reflect ECN */
197 198
	tcp_dec_quickack_mode(sk, pkts);
	inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
L
Linus Torvalds 已提交
199 200 201 202 203 204 205 206 207
}

/* Determine a window scaling and initial window to offer.
 * Based on the assumption that the given amount of space
 * will be offered. Store the results in the tp structure.
 * NOTE: for smooth operation initial space offering should
 * be a multiple of mss if possible. We assume here that mss >= 1.
 * This MUST be enforced by all callers.
 */
208
void tcp_select_initial_window(const struct sock *sk, int __space, __u32 mss,
L
Linus Torvalds 已提交
209
			       __u32 *rcv_wnd, __u32 *window_clamp,
210 211
			       int wscale_ok, __u8 *rcv_wscale,
			       __u32 init_rcv_wnd)
L
Linus Torvalds 已提交
212 213 214 215 216
{
	unsigned int space = (__space < 0 ? 0 : __space);

	/* If no clamp set the clamp to the max possible scaled window */
	if (*window_clamp == 0)
217
		(*window_clamp) = (U16_MAX << TCP_MAX_WSCALE);
L
Linus Torvalds 已提交
218 219 220 221
	space = min(*window_clamp, space);

	/* Quantize space offering to a multiple of mss if possible. */
	if (space > mss)
222
		space = rounddown(space, mss);
L
Linus Torvalds 已提交
223 224

	/* NOTE: offering an initial window larger than 32767
225 226 227 228 229 230
	 * 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
231
	 */
232
	if (sock_net(sk)->ipv4.sysctl_tcp_workaround_signed_windows)
233 234
		(*rcv_wnd) = min(space, MAX_TCP_WINDOW);
	else
235 236 237 238
		(*rcv_wnd) = min_t(u32, space, U16_MAX);

	if (init_rcv_wnd)
		*rcv_wnd = min(*rcv_wnd, init_rcv_wnd * mss);
239

E
Eric Dumazet 已提交
240
	*rcv_wscale = 0;
L
Linus Torvalds 已提交
241
	if (wscale_ok) {
242
		/* Set window scaling on max possible window */
243
		space = max_t(u32, space, sock_net(sk)->ipv4.sysctl_tcp_rmem[2]);
244
		space = max_t(u32, space, sysctl_rmem_max);
245
		space = min_t(u32, space, *window_clamp);
E
Eric Dumazet 已提交
246 247
		*rcv_wscale = clamp_t(int, ilog2(space) - 15,
				      0, TCP_MAX_WSCALE);
L
Linus Torvalds 已提交
248 249
	}
	/* Set the clamp no higher than max representable value */
250
	(*window_clamp) = min_t(__u32, U16_MAX << (*rcv_wscale), *window_clamp);
L
Linus Torvalds 已提交
251
}
E
Eric Dumazet 已提交
252
EXPORT_SYMBOL(tcp_select_initial_window);
L
Linus Torvalds 已提交
253 254 255 256 257 258

/* 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
259
static u16 tcp_select_window(struct sock *sk)
L
Linus Torvalds 已提交
260 261
{
	struct tcp_sock *tp = tcp_sk(sk);
262
	u32 old_win = tp->rcv_wnd;
L
Linus Torvalds 已提交
263 264 265 266
	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
267
	if (new_win < cur_win) {
L
Linus Torvalds 已提交
268 269 270 271 272 273 274
		/* 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.
		 */
275 276 277
		if (new_win == 0)
			NET_INC_STATS(sock_net(sk),
				      LINUX_MIB_TCPWANTZEROWINDOWADV);
278
		new_win = ALIGN(cur_win, 1 << tp->rx_opt.rcv_wscale);
L
Linus Torvalds 已提交
279 280 281 282 283 284 285
	}
	tp->rcv_wnd = new_win;
	tp->rcv_wup = tp->rcv_nxt;

	/* Make sure we do not exceed the maximum possible
	 * scaled window.
	 */
286 287
	if (!tp->rx_opt.rcv_wscale &&
	    sock_net(sk)->ipv4.sysctl_tcp_workaround_signed_windows)
L
Linus Torvalds 已提交
288 289 290 291 292 293 294
		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;

295
	/* If we advertise zero window, disable fast path. */
296
	if (new_win == 0) {
297
		tp->pred_flags = 0;
298 299 300 301 302 303
		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
304 305 306 307

	return new_win;
}

308
/* Packet ECN state for a SYN-ACK */
309
static void tcp_ecn_send_synack(struct sock *sk, struct sk_buff *skb)
310
{
311 312
	const struct tcp_sock *tp = tcp_sk(sk);

E
Eric Dumazet 已提交
313
	TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_CWR;
314
	if (!(tp->ecn_flags & TCP_ECN_OK))
E
Eric Dumazet 已提交
315
		TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_ECE;
316 317
	else if (tcp_ca_needs_ecn(sk) ||
		 tcp_bpf_ca_needs_ecn(sk))
318
		INET_ECN_xmit(sk);
319 320
}

321
/* Packet ECN state for a SYN.  */
322
static void tcp_ecn_send_syn(struct sock *sk, struct sk_buff *skb)
323 324
{
	struct tcp_sock *tp = tcp_sk(sk);
325
	bool bpf_needs_ecn = tcp_bpf_ca_needs_ecn(sk);
326
	bool use_ecn = sock_net(sk)->ipv4.sysctl_tcp_ecn == 1 ||
327
		tcp_ca_needs_ecn(sk) || bpf_needs_ecn;
328 329 330 331 332 333 334

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

		if (dst && dst_feature(dst, RTAX_FEATURE_ECN))
			use_ecn = true;
	}
335 336

	tp->ecn_flags = 0;
337 338

	if (use_ecn) {
E
Eric Dumazet 已提交
339
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ECE | TCPHDR_CWR;
340
		tp->ecn_flags = TCP_ECN_OK;
341
		if (tcp_ca_needs_ecn(sk) || bpf_needs_ecn)
342
			INET_ECN_xmit(sk);
343 344 345
	}
}

346 347 348 349 350 351 352 353 354
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);
}

355
static void
356
tcp_ecn_make_synack(const struct request_sock *req, struct tcphdr *th)
357
{
358
	if (inet_rsk(req)->ecn_ok)
359 360 361
		th->ece = 1;
}

362 363 364
/* Set up ECN state for a packet on a ESTABLISHED socket that is about to
 * be sent.
 */
365
static void tcp_ecn_send(struct sock *sk, struct sk_buff *skb,
366
			 struct tcphdr *th, int tcp_header_len)
367 368 369 370 371 372 373 374
{
	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);
375
			if (tp->ecn_flags & TCP_ECN_QUEUE_CWR) {
376
				tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR;
377
				th->cwr = 1;
378 379
				skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
			}
380
		} else if (!tcp_ca_needs_ecn(sk)) {
381 382 383 384
			/* ACK or retransmitted segment: clear ECT|CE */
			INET_ECN_dontxmit(sk);
		}
		if (tp->ecn_flags & TCP_ECN_DEMAND_CWR)
385
			th->ece = 1;
386 387 388
	}
}

389 390 391 392 393
/* 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)
{
394
	skb->ip_summed = CHECKSUM_PARTIAL;
395

E
Eric Dumazet 已提交
396
	TCP_SKB_CB(skb)->tcp_flags = flags;
397

398
	tcp_skb_pcount_set(skb, 1);
399 400

	TCP_SKB_CB(skb)->seq = seq;
C
Changli Gao 已提交
401
	if (flags & (TCPHDR_SYN | TCPHDR_FIN))
402 403 404 405
		seq++;
	TCP_SKB_CB(skb)->end_seq = seq;
}

E
Eric Dumazet 已提交
406
static inline bool tcp_urg_mode(const struct tcp_sock *tp)
I
Ilpo Järvinen 已提交
407 408 409 410
{
	return tp->snd_una != tp->snd_up;
}

A
Adam Langley 已提交
411 412 413
#define OPTION_SACK_ADVERTISE	(1 << 0)
#define OPTION_TS		(1 << 1)
#define OPTION_MD5		(1 << 2)
414
#define OPTION_WSCALE		(1 << 3)
Y
Yuchung Cheng 已提交
415
#define OPTION_FAST_OPEN_COOKIE	(1 << 8)
416
#define OPTION_SMC		(1 << 9)
P
Peter Krystad 已提交
417
#define OPTION_MPTCP		(1 << 10)
418 419 420 421 422 423 424 425 426 427 428 429 430 431 432

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
433 434

struct tcp_out_options {
Y
Yuchung Cheng 已提交
435 436
	u16 options;		/* bit field of OPTION_* */
	u16 mss;		/* 0 to disable */
A
Adam Langley 已提交
437 438
	u8 ws;			/* window scale, 0 to disable */
	u8 num_sack_blocks;	/* number of SACK blocks to include */
439
	u8 hash_size;		/* bytes in hash_location */
440
	u8 bpf_opt_len;		/* length of BPF hdr option */
441
	__u8 *hash_location;	/* temporary pointer, overloaded */
Y
Yuchung Cheng 已提交
442 443
	__u32 tsval, tsecr;	/* need to include OPTION_TS */
	struct tcp_fastopen_cookie *fastopen_cookie;	/* Fast open cookie */
P
Peter Krystad 已提交
444
	struct mptcp_out_options mptcp;
A
Adam Langley 已提交
445 446
};

447 448
static void mptcp_options_write(__be32 *ptr, const struct tcp_sock *tp,
				struct tcp_out_options *opts)
P
Peter Krystad 已提交
449 450 451
{
#if IS_ENABLED(CONFIG_MPTCP)
	if (unlikely(OPTION_MPTCP & opts->options))
452
		mptcp_write_options(ptr, tp, &opts->mptcp);
P
Peter Krystad 已提交
453 454 455
#endif
}

456
#ifdef CONFIG_CGROUP_BPF
457 458 459 460 461 462 463 464 465 466 467 468
static int bpf_skops_write_hdr_opt_arg0(struct sk_buff *skb,
					enum tcp_synack_type synack_type)
{
	if (unlikely(!skb))
		return BPF_WRITE_HDR_TCP_CURRENT_MSS;

	if (unlikely(synack_type == TCP_SYNACK_COOKIE))
		return BPF_WRITE_HDR_TCP_SYNACK_COOKIE;

	return 0;
}

469 470 471 472 473 474 475 476
/* req, syn_skb and synack_type are used when writing synack */
static void bpf_skops_hdr_opt_len(struct sock *sk, struct sk_buff *skb,
				  struct request_sock *req,
				  struct sk_buff *syn_skb,
				  enum tcp_synack_type synack_type,
				  struct tcp_out_options *opts,
				  unsigned int *remaining)
{
477 478 479
	struct bpf_sock_ops_kern sock_ops;
	int err;

480 481 482 483 484
	if (likely(!BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk),
					   BPF_SOCK_OPS_WRITE_HDR_OPT_CB_FLAG)) ||
	    !*remaining)
		return;

485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530
	/* *remaining has already been aligned to 4 bytes, so *remaining >= 4 */

	/* init sock_ops */
	memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));

	sock_ops.op = BPF_SOCK_OPS_HDR_OPT_LEN_CB;

	if (req) {
		/* The listen "sk" cannot be passed here because
		 * it is not locked.  It would not make too much
		 * sense to do bpf_setsockopt(listen_sk) based
		 * on individual connection request also.
		 *
		 * Thus, "req" is passed here and the cgroup-bpf-progs
		 * of the listen "sk" will be run.
		 *
		 * "req" is also used here for fastopen even the "sk" here is
		 * a fullsock "child" sk.  It is to keep the behavior
		 * consistent between fastopen and non-fastopen on
		 * the bpf programming side.
		 */
		sock_ops.sk = (struct sock *)req;
		sock_ops.syn_skb = syn_skb;
	} else {
		sock_owned_by_me(sk);

		sock_ops.is_fullsock = 1;
		sock_ops.sk = sk;
	}

	sock_ops.args[0] = bpf_skops_write_hdr_opt_arg0(skb, synack_type);
	sock_ops.remaining_opt_len = *remaining;
	/* tcp_current_mss() does not pass a skb */
	if (skb)
		bpf_skops_init_skb(&sock_ops, skb, 0);

	err = BPF_CGROUP_RUN_PROG_SOCK_OPS_SK(&sock_ops, sk);

	if (err || sock_ops.remaining_opt_len == *remaining)
		return;

	opts->bpf_opt_len = *remaining - sock_ops.remaining_opt_len;
	/* round up to 4 bytes */
	opts->bpf_opt_len = (opts->bpf_opt_len + 3) & ~3;

	*remaining -= opts->bpf_opt_len;
531 532 533 534 535 536 537 538
}

static void bpf_skops_write_hdr_opt(struct sock *sk, struct sk_buff *skb,
				    struct request_sock *req,
				    struct sk_buff *syn_skb,
				    enum tcp_synack_type synack_type,
				    struct tcp_out_options *opts)
{
539 540 541 542 543
	u8 first_opt_off, nr_written, max_opt_len = opts->bpf_opt_len;
	struct bpf_sock_ops_kern sock_ops;
	int err;

	if (likely(!max_opt_len))
544 545
		return;

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
	memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));

	sock_ops.op = BPF_SOCK_OPS_WRITE_HDR_OPT_CB;

	if (req) {
		sock_ops.sk = (struct sock *)req;
		sock_ops.syn_skb = syn_skb;
	} else {
		sock_owned_by_me(sk);

		sock_ops.is_fullsock = 1;
		sock_ops.sk = sk;
	}

	sock_ops.args[0] = bpf_skops_write_hdr_opt_arg0(skb, synack_type);
	sock_ops.remaining_opt_len = max_opt_len;
	first_opt_off = tcp_hdrlen(skb) - max_opt_len;
	bpf_skops_init_skb(&sock_ops, skb, first_opt_off);

	err = BPF_CGROUP_RUN_PROG_SOCK_OPS_SK(&sock_ops, sk);

	if (err)
		nr_written = 0;
	else
		nr_written = max_opt_len - sock_ops.remaining_opt_len;

	if (nr_written < max_opt_len)
		memset(skb->data + first_opt_off + nr_written, TCPOPT_NOP,
		       max_opt_len - nr_written);
575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594
}
#else
static void bpf_skops_hdr_opt_len(struct sock *sk, struct sk_buff *skb,
				  struct request_sock *req,
				  struct sk_buff *syn_skb,
				  enum tcp_synack_type synack_type,
				  struct tcp_out_options *opts,
				  unsigned int *remaining)
{
}

static void bpf_skops_write_hdr_opt(struct sock *sk, struct sk_buff *skb,
				    struct request_sock *req,
				    struct sk_buff *syn_skb,
				    enum tcp_synack_type synack_type,
				    struct tcp_out_options *opts)
{
}
#endif

595 596 597
/* Write previously computed TCP options to the packet.
 *
 * Beware: Something in the Internet is very sensitive to the ordering of
598 599
 * 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
600
 * inter-operability perspective it seems that we're somewhat stuck with
601 602 603 604 605 606 607
 * 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
608
static void tcp_options_write(__be32 *ptr, struct tcp_sock *tp,
609 610
			      struct tcp_out_options *opts)
{
Y
Yuchung Cheng 已提交
611
	u16 options = opts->options;	/* mungable copy */
612 613

	if (unlikely(OPTION_MD5 & options)) {
C
Christoph Paasch 已提交
614 615
		*ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
			       (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
616 617
		/* overload cookie hash location */
		opts->hash_location = (__u8 *)ptr;
A
Adam Langley 已提交
618
		ptr += 4;
S
Stephen Hemminger 已提交
619
	}
A
Adam Langley 已提交
620

621 622 623 624 625 626
	if (unlikely(opts->mss)) {
		*ptr++ = htonl((TCPOPT_MSS << 24) |
			       (TCPOLEN_MSS << 16) |
			       opts->mss);
	}

627 628
	if (likely(OPTION_TS & options)) {
		if (unlikely(OPTION_SACK_ADVERTISE & options)) {
A
Adam Langley 已提交
629 630 631 632
			*ptr++ = htonl((TCPOPT_SACK_PERM << 24) |
				       (TCPOLEN_SACK_PERM << 16) |
				       (TCPOPT_TIMESTAMP << 8) |
				       TCPOLEN_TIMESTAMP);
633
			options &= ~OPTION_SACK_ADVERTISE;
A
Adam Langley 已提交
634 635 636 637 638 639 640 641 642 643
		} else {
			*ptr++ = htonl((TCPOPT_NOP << 24) |
				       (TCPOPT_NOP << 16) |
				       (TCPOPT_TIMESTAMP << 8) |
				       TCPOLEN_TIMESTAMP);
		}
		*ptr++ = htonl(opts->tsval);
		*ptr++ = htonl(opts->tsecr);
	}

644
	if (unlikely(OPTION_SACK_ADVERTISE & options)) {
A
Adam Langley 已提交
645 646 647 648 649 650
		*ptr++ = htonl((TCPOPT_NOP << 24) |
			       (TCPOPT_NOP << 16) |
			       (TCPOPT_SACK_PERM << 8) |
			       TCPOLEN_SACK_PERM);
	}

651
	if (unlikely(OPTION_WSCALE & options)) {
A
Adam Langley 已提交
652 653 654 655 656 657 658 659 660
		*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
661 662 663 664 665
		int this_sack;

		*ptr++ = htonl((TCPOPT_NOP  << 24) |
			       (TCPOPT_NOP  << 16) |
			       (TCPOPT_SACK <<  8) |
A
Adam Langley 已提交
666
			       (TCPOLEN_SACK_BASE + (opts->num_sack_blocks *
S
Stephen Hemminger 已提交
667
						     TCPOLEN_SACK_PERBLOCK)));
S
Stephen Hemminger 已提交
668

A
Adam Langley 已提交
669 670
		for (this_sack = 0; this_sack < opts->num_sack_blocks;
		     ++this_sack) {
S
Stephen Hemminger 已提交
671 672 673
			*ptr++ = htonl(sp[this_sack].start_seq);
			*ptr++ = htonl(sp[this_sack].end_seq);
		}
S
Stephen Hemminger 已提交
674

675
		tp->rx_opt.dsack = 0;
S
Stephen Hemminger 已提交
676
	}
Y
Yuchung Cheng 已提交
677 678 679

	if (unlikely(OPTION_FAST_OPEN_COOKIE & options)) {
		struct tcp_fastopen_cookie *foc = opts->fastopen_cookie;
680 681 682 683 684 685 686 687 688 689 690 691 692
		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
693

694 695 696 697
		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
698
		}
699
		ptr += (len + 3) >> 2;
Y
Yuchung Cheng 已提交
700
	}
701 702

	smc_options_write(ptr, &options);
P
Peter Krystad 已提交
703

704
	mptcp_options_write(ptr, tp, opts);
705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737
}

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
738 739
}

740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755
static void mptcp_set_option_cond(const struct request_sock *req,
				  struct tcp_out_options *opts,
				  unsigned int *remaining)
{
	if (rsk_is_mptcp(req)) {
		unsigned int size;

		if (mptcp_synack_options(req, &size, &opts->mptcp)) {
			if (*remaining >= size) {
				opts->options |= OPTION_MPTCP;
				*remaining -= size;
			}
		}
	}
}

756 757 758
/* Compute TCP options for SYN packets. This is not the final
 * network wire format yet.
 */
759
static unsigned int tcp_syn_options(struct sock *sk, struct sk_buff *skb,
A
Adam Langley 已提交
760
				struct tcp_out_options *opts,
761 762
				struct tcp_md5sig_key **md5)
{
A
Adam Langley 已提交
763
	struct tcp_sock *tp = tcp_sk(sk);
764
	unsigned int remaining = MAX_TCP_OPTION_SPACE;
765
	struct tcp_fastopen_request *fastopen = tp->fastopen_req;
A
Adam Langley 已提交
766

767
	*md5 = NULL;
768
#ifdef CONFIG_TCP_MD5SIG
769
	if (static_branch_unlikely(&tcp_md5_needed) &&
770
	    rcu_access_pointer(tp->md5sig_info)) {
771 772 773 774 775
		*md5 = tp->af_specific->md5_lookup(sk, sk);
		if (*md5) {
			opts->options |= OPTION_MD5;
			remaining -= TCPOLEN_MD5SIG_ALIGNED;
		}
776 777
	}
#endif
A
Adam Langley 已提交
778 779 780 781 782 783 784 785 786 787 788

	/* 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);
789
	remaining -= TCPOLEN_MSS_ALIGNED;
A
Adam Langley 已提交
790

791
	if (likely(sock_net(sk)->ipv4.sysctl_tcp_timestamps && !*md5)) {
A
Adam Langley 已提交
792
		opts->options |= OPTION_TS;
E
Eric Dumazet 已提交
793
		opts->tsval = tcp_skb_timestamp(skb) + tp->tsoffset;
A
Adam Langley 已提交
794
		opts->tsecr = tp->rx_opt.ts_recent;
795
		remaining -= TCPOLEN_TSTAMP_ALIGNED;
A
Adam Langley 已提交
796
	}
797
	if (likely(sock_net(sk)->ipv4.sysctl_tcp_window_scaling)) {
A
Adam Langley 已提交
798
		opts->ws = tp->rx_opt.rcv_wscale;
799
		opts->options |= OPTION_WSCALE;
800
		remaining -= TCPOLEN_WSCALE_ALIGNED;
A
Adam Langley 已提交
801
	}
E
Eric Dumazet 已提交
802
	if (likely(sock_net(sk)->ipv4.sysctl_tcp_sack)) {
A
Adam Langley 已提交
803
		opts->options |= OPTION_SACK_ADVERTISE;
804
		if (unlikely(!(OPTION_TS & opts->options)))
805
			remaining -= TCPOLEN_SACKPERM_ALIGNED;
A
Adam Langley 已提交
806 807
	}

808
	if (fastopen && fastopen->cookie.len >= 0) {
809 810 811 812
		u32 need = fastopen->cookie.len;

		need += fastopen->cookie.exp ? TCPOLEN_EXP_FASTOPEN_BASE :
					       TCPOLEN_FASTOPEN_BASE;
813 814 815 816 817 818
		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;
819
			tp->syn_fastopen_exp = fastopen->cookie.exp ? 1 : 0;
820 821
		}
	}
822

823 824
	smc_set_option(tp, opts, &remaining);

825 826 827
	if (sk_is_mptcp(sk)) {
		unsigned int size;

828
		if (mptcp_syn_options(sk, skb, &size, &opts->mptcp)) {
829 830 831 832 833
			opts->options |= OPTION_MPTCP;
			remaining -= size;
		}
	}

834 835
	bpf_skops_hdr_opt_len(sk, skb, NULL, NULL, 0, opts, &remaining);

836
	return MAX_TCP_OPTION_SPACE - remaining;
S
Stephen Hemminger 已提交
837 838
}

839
/* Set up TCP options for SYN-ACKs. */
840 841
static unsigned int tcp_synack_options(const struct sock *sk,
				       struct request_sock *req,
842 843 844
				       unsigned int mss, struct sk_buff *skb,
				       struct tcp_out_options *opts,
				       const struct tcp_md5sig_key *md5,
845
				       struct tcp_fastopen_cookie *foc,
846 847
				       enum tcp_synack_type synack_type,
				       struct sk_buff *syn_skb)
848
{
A
Adam Langley 已提交
849
	struct inet_request_sock *ireq = inet_rsk(req);
850
	unsigned int remaining = MAX_TCP_OPTION_SPACE;
A
Adam Langley 已提交
851

852
#ifdef CONFIG_TCP_MD5SIG
E
Eric Dumazet 已提交
853
	if (md5) {
A
Adam Langley 已提交
854
		opts->options |= OPTION_MD5;
855 856 857 858 859 860 861
		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.
		 */
862 863
		if (synack_type != TCP_SYNACK_COOKIE)
			ireq->tstamp_ok &= !ireq->sack_ok;
864 865
	}
#endif
A
Adam Langley 已提交
866

867
	/* We always send an MSS option. */
A
Adam Langley 已提交
868
	opts->mss = mss;
869
	remaining -= TCPOLEN_MSS_ALIGNED;
A
Adam Langley 已提交
870 871 872

	if (likely(ireq->wscale_ok)) {
		opts->ws = ireq->rcv_wscale;
873
		opts->options |= OPTION_WSCALE;
874
		remaining -= TCPOLEN_WSCALE_ALIGNED;
A
Adam Langley 已提交
875
	}
E
Eric Dumazet 已提交
876
	if (likely(ireq->tstamp_ok)) {
A
Adam Langley 已提交
877
		opts->options |= OPTION_TS;
878
		opts->tsval = tcp_skb_timestamp(skb) + tcp_rsk(req)->ts_off;
A
Adam Langley 已提交
879
		opts->tsecr = req->ts_recent;
880
		remaining -= TCPOLEN_TSTAMP_ALIGNED;
A
Adam Langley 已提交
881 882 883
	}
	if (likely(ireq->sack_ok)) {
		opts->options |= OPTION_SACK_ADVERTISE;
E
Eric Dumazet 已提交
884
		if (unlikely(!ireq->tstamp_ok))
885
			remaining -= TCPOLEN_SACKPERM_ALIGNED;
A
Adam Langley 已提交
886
	}
887 888 889 890 891
	if (foc != NULL && foc->len >= 0) {
		u32 need = foc->len;

		need += foc->exp ? TCPOLEN_EXP_FASTOPEN_BASE :
				   TCPOLEN_FASTOPEN_BASE;
892 893 894 895 896 897 898
		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
899

900 901
	mptcp_set_option_cond(req, opts, &remaining);

902 903
	smc_set_option_cond(tcp_sk(sk), ireq, opts, &remaining);

904 905 906
	bpf_skops_hdr_opt_len((struct sock *)sk, skb, req, syn_skb,
			      synack_type, opts, &remaining);

907
	return MAX_TCP_OPTION_SPACE - remaining;
A
Adam Langley 已提交
908 909
}

910 911 912
/* Compute TCP options for ESTABLISHED sockets. This is not the
 * final wire format yet.
 */
913
static unsigned int tcp_established_options(struct sock *sk, struct sk_buff *skb,
A
Adam Langley 已提交
914
					struct tcp_out_options *opts,
915 916
					struct tcp_md5sig_key **md5)
{
A
Adam Langley 已提交
917
	struct tcp_sock *tp = tcp_sk(sk);
918
	unsigned int size = 0;
919
	unsigned int eff_sacks;
A
Adam Langley 已提交
920

921 922
	opts->options = 0;

923
	*md5 = NULL;
A
Adam Langley 已提交
924
#ifdef CONFIG_TCP_MD5SIG
925
	if (static_branch_unlikely(&tcp_md5_needed) &&
926
	    rcu_access_pointer(tp->md5sig_info)) {
927 928 929 930 931
		*md5 = tp->af_specific->md5_lookup(sk, sk);
		if (*md5) {
			opts->options |= OPTION_MD5;
			size += TCPOLEN_MD5SIG_ALIGNED;
		}
A
Adam Langley 已提交
932 933 934 935 936
	}
#endif

	if (likely(tp->rx_opt.tstamp_ok)) {
		opts->options |= OPTION_TS;
E
Eric Dumazet 已提交
937
		opts->tsval = skb ? tcp_skb_timestamp(skb) + tp->tsoffset : 0;
A
Adam Langley 已提交
938 939 940 941
		opts->tsecr = tp->rx_opt.ts_recent;
		size += TCPOLEN_TSTAMP_ALIGNED;
	}

942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958
	/* MPTCP options have precedence over SACK for the limited TCP
	 * option space because a MPTCP connection would be forced to
	 * fall back to regular TCP if a required multipath option is
	 * missing. SACK still gets a chance to use whatever space is
	 * left.
	 */
	if (sk_is_mptcp(sk)) {
		unsigned int remaining = MAX_TCP_OPTION_SPACE - size;
		unsigned int opt_size = 0;

		if (mptcp_established_options(sk, skb, &opt_size, remaining,
					      &opts->mptcp)) {
			opts->options |= OPTION_MPTCP;
			size += opt_size;
		}
	}

959 960
	eff_sacks = tp->rx_opt.num_sacks + tp->rx_opt.dsack;
	if (unlikely(eff_sacks)) {
961
		const unsigned int remaining = MAX_TCP_OPTION_SPACE - size;
962 963 964 965
		if (unlikely(remaining < TCPOLEN_SACK_BASE_ALIGNED +
					 TCPOLEN_SACK_PERBLOCK))
			return size;

A
Adam Langley 已提交
966
		opts->num_sack_blocks =
967
			min_t(unsigned int, eff_sacks,
A
Adam Langley 已提交
968 969
			      (remaining - TCPOLEN_SACK_BASE_ALIGNED) /
			      TCPOLEN_SACK_PERBLOCK);
970 971 972

		size += TCPOLEN_SACK_BASE_ALIGNED +
			opts->num_sack_blocks * TCPOLEN_SACK_PERBLOCK;
A
Adam Langley 已提交
973 974
	}

975 976 977 978 979 980 981 982 983
	if (unlikely(BPF_SOCK_OPS_TEST_FLAG(tp,
					    BPF_SOCK_OPS_WRITE_HDR_OPT_CB_FLAG))) {
		unsigned int remaining = MAX_TCP_OPTION_SPACE - size;

		bpf_skops_hdr_opt_len(sk, skb, NULL, NULL, 0, opts, &remaining);

		size = MAX_TCP_OPTION_SPACE - remaining;
	}

A
Adam Langley 已提交
984
	return size;
S
Stephen Hemminger 已提交
985
}
L
Linus Torvalds 已提交
986

E
Eric Dumazet 已提交
987 988 989 990 991 992 993 994 995

/* 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
996
 * The invariant being skb->truesize subtracted from sk->sk_wmem_alloc
E
Eric Dumazet 已提交
997 998 999 1000 1001 1002 1003 1004 1005 1006 1007
 *
 * 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);

1008
static void tcp_tsq_write(struct sock *sk)
1009 1010 1011
{
	if ((1 << sk->sk_state) &
	    (TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_CLOSING |
1012 1013 1014 1015
	     TCPF_CLOSE_WAIT  | TCPF_LAST_ACK)) {
		struct tcp_sock *tp = tcp_sk(sk);

		if (tp->lost_out > tp->retrans_out &&
1016 1017
		    tp->snd_cwnd > tcp_packets_in_flight(tp)) {
			tcp_mstamp_refresh(tp);
1018
			tcp_xmit_retransmit_queue(sk);
1019
		}
1020 1021

		tcp_write_xmit(sk, tcp_current_mss(sk), tp->nonagle,
J
John Ogness 已提交
1022
			       0, GFP_ATOMIC);
1023
	}
1024
}
1025 1026 1027 1028 1029 1030 1031 1032 1033 1034

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
1035
/*
S
stephen hemminger 已提交
1036
 * One tasklet per cpu tries to send more skbs.
E
Eric Dumazet 已提交
1037
 * We run in tasklet context but need to disable irqs when
S
stephen hemminger 已提交
1038
 * transferring tsq->head because tcp_wfree() might
E
Eric Dumazet 已提交
1039 1040
 * interrupt us (non NAPI drivers)
 */
1041
static void tcp_tasklet_func(struct tasklet_struct *t)
E
Eric Dumazet 已提交
1042
{
1043
	struct tsq_tasklet *tsq = from_tasklet(tsq,  t, tasklet);
E
Eric Dumazet 已提交
1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058
	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;
1059
		smp_mb__before_atomic();
1060 1061
		clear_bit(TSQ_QUEUED, &sk->sk_tsq_flags);

1062
		tcp_tsq_handler(sk);
E
Eric Dumazet 已提交
1063 1064 1065 1066
		sk_free(sk);
	}
}

E
Eric Dumazet 已提交
1067 1068 1069 1070
#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
1071 1072 1073 1074 1075 1076 1077 1078 1079
/**
 * 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)
{
1080
	unsigned long flags, nflags;
E
Eric Dumazet 已提交
1081

1082 1083
	/* perform an atomic operation only if at least one flag is set */
	do {
1084
		flags = sk->sk_tsq_flags;
1085 1086 1087
		if (!(flags & TCP_DEFERRED_ALL))
			return;
		nflags = flags & ~TCP_DEFERRED_ALL;
1088
	} while (cmpxchg(&sk->sk_tsq_flags, flags, nflags) != flags);
1089

1090 1091 1092 1093
	if (flags & TCPF_TSQ_DEFERRED) {
		tcp_tsq_write(sk);
		__sock_put(sk);
	}
1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104
	/* 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
1105
	if (flags & TCPF_WRITE_TIMER_DEFERRED) {
1106
		tcp_write_timer_handler(sk);
1107 1108
		__sock_put(sk);
	}
E
Eric Dumazet 已提交
1109
	if (flags & TCPF_DELACK_TIMER_DEFERRED) {
1110
		tcp_delack_timer_handler(sk);
1111 1112
		__sock_put(sk);
	}
E
Eric Dumazet 已提交
1113
	if (flags & TCPF_MTU_REDUCED_DEFERRED) {
1114
		inet_csk(sk)->icsk_af_ops->mtu_reduced(sk);
1115 1116
		__sock_put(sk);
	}
E
Eric Dumazet 已提交
1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127
}
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);
1128
		tasklet_setup(&tsq->tasklet, tcp_tasklet_func);
E
Eric Dumazet 已提交
1129 1130 1131 1132 1133
	}
}

/*
 * Write buffer destructor automatically called from kfree_skb.
S
stephen hemminger 已提交
1134
 * We can't xmit new skbs from this context, as we might already
E
Eric Dumazet 已提交
1135 1136
 * hold qdisc lock.
 */
E
Eric Dumazet 已提交
1137
void tcp_wfree(struct sk_buff *skb)
E
Eric Dumazet 已提交
1138 1139 1140
{
	struct sock *sk = skb->sk;
	struct tcp_sock *tp = tcp_sk(sk);
1141
	unsigned long flags, nval, oval;
1142 1143 1144 1145

	/* Keep one reference on sk_wmem_alloc.
	 * Will be released by sk_free() from here or tcp_tasklet_func()
	 */
1146
	WARN_ON(refcount_sub_and_test(skb->truesize - 1, &sk->sk_wmem_alloc));
1147 1148 1149 1150 1151 1152 1153 1154

	/* 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)
	 */
1155
	if (refcount_read(&sk->sk_wmem_alloc) >= SKB_TRUESIZE(1) && this_cpu_ksoftirqd() == current)
1156
		goto out;
E
Eric Dumazet 已提交
1157

1158
	for (oval = READ_ONCE(sk->sk_tsq_flags);; oval = nval) {
E
Eric Dumazet 已提交
1159
		struct tsq_tasklet *tsq;
1160
		bool empty;
E
Eric Dumazet 已提交
1161

1162 1163 1164
		if (!(oval & TSQF_THROTTLED) || (oval & TSQF_QUEUED))
			goto out;

1165
		nval = (oval & ~TSQF_THROTTLED) | TSQF_QUEUED;
1166
		nval = cmpxchg(&sk->sk_tsq_flags, oval, nval);
1167 1168 1169
		if (nval != oval)
			continue;

E
Eric Dumazet 已提交
1170 1171
		/* queue this socket to tasklet queue */
		local_irq_save(flags);
1172
		tsq = this_cpu_ptr(&tsq_tasklet);
1173
		empty = list_empty(&tsq->head);
E
Eric Dumazet 已提交
1174
		list_add(&tp->tsq_node, &tsq->head);
1175 1176
		if (empty)
			tasklet_schedule(&tsq->tasklet);
E
Eric Dumazet 已提交
1177
		local_irq_restore(flags);
1178
		return;
E
Eric Dumazet 已提交
1179
	}
1180 1181
out:
	sk_free(sk);
E
Eric Dumazet 已提交
1182 1183
}

1184 1185
/* Note: Called under soft irq.
 * We can call TCP stack right away, unless socket is owned by user.
1186 1187 1188 1189 1190 1191
 */
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;

1192 1193
	tcp_tsq_handler(sk);
	sock_put(sk);
1194 1195 1196 1197

	return HRTIMER_NORESTART;
}

1198 1199
static void tcp_update_skb_after_send(struct sock *sk, struct sk_buff *skb,
				      u64 prior_wstamp)
1200
{
1201 1202 1203
	struct tcp_sock *tp = tcp_sk(sk);

	if (sk->sk_pacing_status != SK_PACING_NONE) {
1204
		unsigned long rate = sk->sk_pacing_rate;
1205 1206 1207 1208 1209

		/* Original sch_fq does not pace first 10 MSS
		 * Note that tp->data_segs_out overflows after 2^32 packets,
		 * this is a minor annoyance.
		 */
1210
		if (rate != ~0UL && rate && tp->data_segs_out >= 10) {
1211 1212 1213 1214 1215 1216
			u64 len_ns = div64_ul((u64)skb->len * NSEC_PER_SEC, rate);
			u64 credit = tp->tcp_wstamp_ns - prior_wstamp;

			/* take into account OS jitter */
			len_ns -= min_t(u64, len_ns / 2, credit);
			tp->tcp_wstamp_ns += len_ns;
1217 1218
		}
	}
1219 1220 1221
	list_move_tail(&skb->tcp_tsorted_anchor, &tp->tsorted_sent_queue);
}

1222 1223
INDIRECT_CALLABLE_DECLARE(int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl));
INDIRECT_CALLABLE_DECLARE(int inet6_csk_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl));
1224
INDIRECT_CALLABLE_DECLARE(void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb));
1225

L
Linus Torvalds 已提交
1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236
/* 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.
 */
1237 1238
static int __tcp_transmit_skb(struct sock *sk, struct sk_buff *skb,
			      int clone_it, gfp_t gfp_mask, u32 rcv_nxt)
L
Linus Torvalds 已提交
1239
{
1240 1241 1242 1243
	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
1244
	struct tcp_out_options opts;
1245
	unsigned int tcp_options_size, tcp_header_size;
1246
	struct sk_buff *oskb = NULL;
1247
	struct tcp_md5sig_key *md5;
1248
	struct tcphdr *th;
1249
	u64 prior_wstamp;
1250 1251 1252
	int err;

	BUG_ON(!skb || !tcp_skb_pcount(skb));
1253
	tp = tcp_sk(sk);
1254 1255 1256
	prior_wstamp = tp->tcp_wstamp_ns;
	tp->tcp_wstamp_ns = max(tp->tcp_wstamp_ns, tp->tcp_clock_cache);
	skb->skb_mstamp_ns = tp->tcp_wstamp_ns;
1257
	if (clone_it) {
1258
		oskb = skb;
1259 1260 1261 1262 1263 1264 1265 1266

		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);

1267 1268
		if (unlikely(!skb))
			return -ENOBUFS;
1269 1270 1271 1272
		/* retransmit skbs might have a non zero value in skb->dev
		 * because skb->dev is aliased with skb->rbnode.rb_left
		 */
		skb->dev = NULL;
1273
	}
1274

1275 1276
	inet = inet_sk(sk);
	tcb = TCP_SKB_CB(skb);
A
Adam Langley 已提交
1277
	memset(&opts, 0, sizeof(opts));
L
Linus Torvalds 已提交
1278

1279
	if (unlikely(tcb->tcp_flags & TCPHDR_SYN)) {
A
Adam Langley 已提交
1280
		tcp_options_size = tcp_syn_options(sk, skb, &opts, &md5);
1281
	} else {
A
Adam Langley 已提交
1282 1283
		tcp_options_size = tcp_established_options(sk, skb, &opts,
							   &md5);
1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294
		/* Force a PSH flag on all (GSO) packets to expedite GRO flush
		 * at receiver : This slightly improve GRO performance.
		 * Note that we do not force the PSH flag for non GSO packets,
		 * because they might be sent under high congestion events,
		 * and in this case it is better to delay the delivery of 1-MSS
		 * packets and thus the corresponding ACK packet that would
		 * release the following packet.
		 */
		if (tcp_skb_pcount(skb) > 1)
			tcb->tcp_flags |= TCPHDR_PSH;
	}
A
Adam Langley 已提交
1295
	tcp_header_size = tcp_options_size + sizeof(struct tcphdr);
1296

E
Eric Dumazet 已提交
1297
	/* if no packet is in qdisc/device queue, then allow XPS to select
1298
	 * another queue. We can be called from tcp_tsq_handler()
1299
	 * which holds one reference to sk.
1300 1301 1302
	 *
	 * 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
1303
	 */
1304
	skb->ooo_okay = sk_wmem_alloc_get(sk) < SKB_TRUESIZE(1);
1305

1306 1307 1308 1309 1310 1311 1312
	/* 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;

1313 1314
	skb_push(skb, tcp_header_size);
	skb_reset_transport_header(skb);
E
Eric Dumazet 已提交
1315 1316 1317

	skb_orphan(skb);
	skb->sk = sk;
E
Eric Dumazet 已提交
1318
	skb->destructor = skb_is_tcp_pure_ack(skb) ? __sock_wfree : tcp_wfree;
1319
	refcount_add(skb->truesize, &sk->sk_wmem_alloc);
1320

1321 1322
	skb_set_dst_pending_confirm(skb, sk->sk_dst_pending_confirm);

1323
	/* Build TCP header and checksum it. */
1324
	th = (struct tcphdr *)skb->data;
E
Eric Dumazet 已提交
1325 1326
	th->source		= inet->inet_sport;
	th->dest		= inet->inet_dport;
1327
	th->seq			= htonl(tcb->seq);
1328
	th->ack_seq		= htonl(rcv_nxt);
1329
	*(((__be16 *)th) + 6)	= htons(((tcp_header_size >> 2) << 12) |
E
Eric Dumazet 已提交
1330
					tcb->tcp_flags);
1331 1332 1333

	th->check		= 0;
	th->urg_ptr		= 0;
L
Linus Torvalds 已提交
1334

I
Ilpo Järvinen 已提交
1335
	/* The urg_mode check is necessary during a below snd_una win probe */
1336 1337 1338 1339 1340
	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)) {
1341
			th->urg_ptr = htons(0xFFFF);
1342 1343
			th->urg = 1;
		}
1344
	}
L
Linus Torvalds 已提交
1345

1346
	skb_shinfo(skb)->gso_type = sk->sk_gso_type;
1347 1348 1349 1350 1351 1352 1353 1354 1355
	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));
	}
1356 1357 1358

	tcp_options_write((__be32 *)(th + 1), tp, &opts);

1359 1360 1361
#ifdef CONFIG_TCP_MD5SIG
	/* Calculate the MD5 hash, as we have all we need now */
	if (md5) {
E
Eric Dumazet 已提交
1362
		sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1363
		tp->af_specific->calc_md5_hash(opts.hash_location,
1364
					       md5, sk, skb);
1365 1366 1367
	}
#endif

1368 1369 1370
	/* BPF prog is the last one writing header option */
	bpf_skops_write_hdr_opt(sk, skb, NULL, NULL, 0, &opts);

1371 1372 1373
	INDIRECT_CALL_INET(icsk->icsk_af_ops->send_check,
			   tcp_v6_send_check, tcp_v4_send_check,
			   sk, skb);
L
Linus Torvalds 已提交
1374

E
Eric Dumazet 已提交
1375
	if (likely(tcb->tcp_flags & TCPHDR_ACK))
1376
		tcp_event_ack_sent(sk, tcp_skb_pcount(skb), rcv_nxt);
L
Linus Torvalds 已提交
1377

1378
	if (skb->len != tcp_header_size) {
1379
		tcp_event_data_sent(tp, sk);
1380
		tp->data_segs_out += tcp_skb_pcount(skb);
W
Wei Wang 已提交
1381
		tp->bytes_sent += skb->len - tcp_header_size;
1382
	}
L
Linus Torvalds 已提交
1383

1384
	if (after(tcb->end_seq, tp->snd_nxt) || tcb->seq == tcb->end_seq)
1385 1386
		TCP_ADD_STATS(sock_net(sk), TCP_MIB_OUTSEGS,
			      tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
1387

1388
	tp->segs_out += tcp_skb_pcount(skb);
1389
	skb_set_hash_from_sk(skb, sk);
1390
	/* OK, its time to fill skb_shinfo(skb)->gso_{segs|size} */
1391
	skb_shinfo(skb)->gso_segs = tcp_skb_pcount(skb);
1392
	skb_shinfo(skb)->gso_size = tcp_skb_mss(skb);
1393

1394
	/* Leave earliest departure time in skb->tstamp (skb->skb_mstamp_ns) */
1395 1396 1397 1398 1399

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

1400 1401
	tcp_add_tx_delay(skb, tp);

1402 1403 1404
	err = INDIRECT_CALL_INET(icsk->icsk_af_ops->queue_xmit,
				 inet6_csk_xmit, ip_queue_xmit,
				 sk, skb, &inet->cork.fl);
E
Eric Dumazet 已提交
1405

1406 1407 1408 1409
	if (unlikely(err > 0)) {
		tcp_enter_cwr(sk);
		err = net_xmit_eval(err);
	}
E
Eric Dumazet 已提交
1410
	if (!err && oskb) {
1411
		tcp_update_skb_after_send(sk, oskb, prior_wstamp);
E
Eric Dumazet 已提交
1412 1413
		tcp_rate_skb_sent(sk, oskb);
	}
1414
	return err;
L
Linus Torvalds 已提交
1415 1416
}

1417 1418 1419 1420 1421 1422 1423
static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it,
			    gfp_t gfp_mask)
{
	return __tcp_transmit_skb(sk, skb, clone_it, gfp_mask,
				  tcp_sk(sk)->rcv_nxt);
}

1424
/* This routine just queues the buffer for sending.
L
Linus Torvalds 已提交
1425 1426 1427 1428 1429 1430 1431 1432 1433
 *
 * 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. */
1434
	WRITE_ONCE(tp->write_seq, TCP_SKB_CB(skb)->end_seq);
1435
	__skb_header_release(skb);
1436
	tcp_add_write_queue_tail(sk, skb);
1437
	sk_wmem_queued_add(sk, skb->truesize);
1438
	sk_mem_charge(sk, skb->truesize);
L
Linus Torvalds 已提交
1439 1440
}

1441
/* Initialize TSO segments for a packet. */
1442
static void tcp_set_skb_tso_segs(struct sk_buff *skb, unsigned int mss_now)
1443
{
1444
	if (skb->len <= mss_now) {
1445 1446 1447
		/* Avoid the costly divide in the normal
		 * non-TSO case.
		 */
1448
		tcp_skb_pcount_set(skb, 1);
1449
		TCP_SKB_CB(skb)->tcp_gso_size = 0;
1450
	} else {
1451
		tcp_skb_pcount_set(skb, DIV_ROUND_UP(skb->len, mss_now));
1452
		TCP_SKB_CB(skb)->tcp_gso_size = mss_now;
L
Linus Torvalds 已提交
1453 1454 1455
	}
}

1456 1457 1458
/* Pcount in the middle of the write queue got changed, we need to do various
 * tweaks to fix counters
 */
1459
static void tcp_adjust_pcount(struct sock *sk, const struct sk_buff *skb, int decr)
1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477
{
	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
1478
	    (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED))
1479 1480 1481 1482 1483
		tp->lost_cnt_hint -= decr;

	tcp_verify_left_out(tp);
}

1484 1485 1486 1487 1488 1489
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);
}

1490 1491 1492 1493
static void tcp_fragment_tstamp(struct sk_buff *skb, struct sk_buff *skb2)
{
	struct skb_shared_info *shinfo = skb_shinfo(skb);

1494
	if (unlikely(tcp_has_tx_tstamp(skb)) &&
1495 1496 1497 1498 1499 1500 1501
	    !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);
1502 1503
		TCP_SKB_CB(skb2)->txstamp_ack = TCP_SKB_CB(skb)->txstamp_ack;
		TCP_SKB_CB(skb)->txstamp_ack = 0;
1504 1505 1506
	}
}

1507 1508 1509 1510 1511 1512
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;
}

1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524
/* 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
1525 1526
/* 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
1527
 * packet to the list.  This won't be called frequently, I hope.
L
Linus Torvalds 已提交
1528 1529
 * Remember, these are still headerless SKBs at this point.
 */
1530 1531
int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
		 struct sk_buff *skb, u32 len,
1532
		 unsigned int mss_now, gfp_t gfp)
L
Linus Torvalds 已提交
1533 1534 1535
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *buff;
1536
	int nsize, old_factor;
1537
	long limit;
1538
	int nlen;
1539
	u8 flags;
L
Linus Torvalds 已提交
1540

1541 1542
	if (WARN_ON(len > skb->len))
		return -EINVAL;
1543

L
Linus Torvalds 已提交
1544 1545 1546 1547
	nsize = skb_headlen(skb) - len;
	if (nsize < 0)
		nsize = 0;

1548 1549 1550 1551 1552 1553 1554 1555 1556 1557
	/* tcp_sendmsg() can overshoot sk_wmem_queued by one full size skb.
	 * We need some allowance to not penalize applications setting small
	 * SO_SNDBUF values.
	 * Also allow first and last skb in retransmit queue to be split.
	 */
	limit = sk->sk_sndbuf + 2 * SKB_TRUESIZE(GSO_MAX_SIZE);
	if (unlikely((sk->sk_wmem_queued >> 1) > limit &&
		     tcp_queue != TCP_FRAG_IN_WRITE_QUEUE &&
		     skb != tcp_rtx_queue_head(sk) &&
		     skb != tcp_rtx_queue_tail(sk))) {
1558 1559 1560 1561
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPWQUEUETOOBIG);
		return -ENOMEM;
	}

1562
	if (skb_unclone(skb, gfp))
L
Linus Torvalds 已提交
1563 1564 1565
		return -ENOMEM;

	/* Get a new skb... force flag on. */
E
Eric Dumazet 已提交
1566
	buff = tcp_stream_alloc_skb(sk, nsize, gfp, true);
1567
	if (!buff)
L
Linus Torvalds 已提交
1568
		return -ENOMEM; /* We'll just try again later. */
1569
	skb_copy_decrypted(buff, skb);
1570
	mptcp_skb_ext_copy(buff, skb);
H
Herbert Xu 已提交
1571

1572
	sk_wmem_queued_add(sk, buff->truesize);
1573
	sk_mem_charge(sk, buff->truesize);
1574 1575 1576
	nlen = skb->len - len - nsize;
	buff->truesize += nlen;
	skb->truesize -= nlen;
L
Linus Torvalds 已提交
1577 1578 1579 1580 1581 1582 1583

	/* 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
1584 1585 1586
	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;
1587
	TCP_SKB_CB(buff)->sacked = TCP_SKB_CB(skb)->sacked;
1588
	tcp_skb_fragment_eor(skb, buff);
L
Linus Torvalds 已提交
1589

1590
	skb_split(skb, buff, len);
L
Linus Torvalds 已提交
1591

1592
	buff->tstamp = skb->tstamp;
1593
	tcp_fragment_tstamp(skb, buff);
L
Linus Torvalds 已提交
1594

1595 1596
	old_factor = tcp_skb_pcount(skb);

L
Linus Torvalds 已提交
1597
	/* Fix up tso_factor for both original and new SKB.  */
1598 1599
	tcp_set_skb_tso_segs(skb, mss_now);
	tcp_set_skb_tso_segs(buff, mss_now);
L
Linus Torvalds 已提交
1600

1601 1602 1603
	/* Update delivered info for the new segment */
	TCP_SKB_CB(buff)->tx = TCP_SKB_CB(skb)->tx;

1604 1605 1606
	/* If this packet has been sent out already, we must
	 * adjust the various packet counters.
	 */
1607
	if (!before(tp->snd_nxt, TCP_SKB_CB(buff)->end_seq)) {
1608 1609
		int diff = old_factor - tcp_skb_pcount(skb) -
			tcp_skb_pcount(buff);
L
Linus Torvalds 已提交
1610

1611 1612
		if (diff)
			tcp_adjust_pcount(sk, skb, diff);
L
Linus Torvalds 已提交
1613 1614 1615
	}

	/* Link BUFF into the send queue. */
1616
	__skb_header_release(buff);
1617
	tcp_insert_write_queue_after(skb, buff, sk, tcp_queue);
1618 1619
	if (tcp_queue == TCP_FRAG_IN_RTX_QUEUE)
		list_add(&buff->tcp_tsorted_anchor, &skb->tcp_tsorted_anchor);
L
Linus Torvalds 已提交
1620 1621 1622 1623

	return 0;
}

1624 1625
/* 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
1626
 */
1627
static int __pskb_trim_head(struct sk_buff *skb, int len)
L
Linus Torvalds 已提交
1628
{
1629
	struct skb_shared_info *shinfo;
L
Linus Torvalds 已提交
1630 1631
	int i, k, eat;

E
Eric Dumazet 已提交
1632 1633 1634 1635 1636
	eat = min_t(int, len, skb_headlen(skb));
	if (eat) {
		__skb_pull(skb, eat);
		len -= eat;
		if (!len)
1637
			return 0;
E
Eric Dumazet 已提交
1638
	}
L
Linus Torvalds 已提交
1639 1640
	eat = len;
	k = 0;
1641 1642 1643
	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
1644 1645

		if (size <= eat) {
1646
			skb_frag_unref(skb, i);
E
Eric Dumazet 已提交
1647
			eat -= size;
L
Linus Torvalds 已提交
1648
		} else {
1649
			shinfo->frags[k] = shinfo->frags[i];
L
Linus Torvalds 已提交
1650
			if (eat) {
J
Jonathan Lemon 已提交
1651
				skb_frag_off_add(&shinfo->frags[k], eat);
1652
				skb_frag_size_sub(&shinfo->frags[k], eat);
L
Linus Torvalds 已提交
1653 1654 1655 1656 1657
				eat = 0;
			}
			k++;
		}
	}
1658
	shinfo->nr_frags = k;
L
Linus Torvalds 已提交
1659 1660 1661

	skb->data_len -= len;
	skb->len = skb->data_len;
1662
	return len;
L
Linus Torvalds 已提交
1663 1664
}

1665
/* Remove acked data from a packet in the transmit queue. */
L
Linus Torvalds 已提交
1666 1667
int tcp_trim_head(struct sock *sk, struct sk_buff *skb, u32 len)
{
1668 1669
	u32 delta_truesize;

1670
	if (skb_unclone(skb, GFP_ATOMIC))
L
Linus Torvalds 已提交
1671 1672
		return -ENOMEM;

1673
	delta_truesize = __pskb_trim_head(skb, len);
L
Linus Torvalds 已提交
1674 1675 1676

	TCP_SKB_CB(skb)->seq += len;

1677 1678
	if (delta_truesize) {
		skb->truesize	   -= delta_truesize;
1679
		sk_wmem_queued_add(sk, -delta_truesize);
1680 1681
		if (!skb_zcopy_pure(skb))
			sk_mem_uncharge(sk, delta_truesize);
1682
	}
L
Linus Torvalds 已提交
1683

1684
	/* Any change of skb->len requires recalculation of tso factor. */
L
Linus Torvalds 已提交
1685
	if (tcp_skb_pcount(skb) > 1)
1686
		tcp_set_skb_tso_segs(skb, tcp_skb_mss(skb));
L
Linus Torvalds 已提交
1687 1688 1689 1690

	return 0;
}

1691 1692
/* 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
1693
{
1694 1695
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1696 1697 1698 1699 1700 1701 1702
	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);

1703 1704 1705 1706 1707 1708 1709 1710
	/* 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
1711 1712 1713 1714 1715 1716 1717 1718
	/* 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 */
E
Eric Dumazet 已提交
1719
	mss_now = max(mss_now, sock_net(sk)->ipv4.sysctl_tcp_min_snd_mss);
J
John Heffner 已提交
1720 1721 1722
	return mss_now;
}

1723 1724 1725 1726 1727 1728 1729
/* 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));
}
1730
EXPORT_SYMBOL(tcp_mtu_to_mss);
1731

J
John Heffner 已提交
1732
/* Inverse of above */
1733
int tcp_mss_to_mtu(struct sock *sk, int mss)
J
John Heffner 已提交
1734
{
1735 1736
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1737 1738 1739 1740 1741 1742 1743
	int mtu;

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

1744 1745 1746 1747 1748 1749 1750
	/* 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
1751 1752
	return mtu;
}
1753
EXPORT_SYMBOL(tcp_mss_to_mtu);
J
John Heffner 已提交
1754

1755
/* MTU probing init per socket */
J
John Heffner 已提交
1756 1757 1758 1759
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
1760
	struct net *net = sock_net(sk);
J
John Heffner 已提交
1761

F
Fan Du 已提交
1762
	icsk->icsk_mtup.enabled = net->ipv4.sysctl_tcp_mtu_probing > 1;
J
John Heffner 已提交
1763
	icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp + sizeof(struct tcphdr) +
1764
			       icsk->icsk_af_ops->net_header_len;
F
Fan Du 已提交
1765
	icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, net->ipv4.sysctl_tcp_base_mss);
J
John Heffner 已提交
1766
	icsk->icsk_mtup.probe_size = 0;
1767
	if (icsk->icsk_mtup.enabled)
1768
		icsk->icsk_mtup.probe_timestamp = tcp_jiffies32;
J
John Heffner 已提交
1769
}
E
Eric Dumazet 已提交
1770
EXPORT_SYMBOL(tcp_mtup_init);
J
John Heffner 已提交
1771

L
Linus Torvalds 已提交
1772 1773 1774 1775 1776 1777
/* 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
1778
   It is minimum of user_mss and mss received with SYN.
L
Linus Torvalds 已提交
1779 1780
   It also does not include TCP options.

1781
   inet_csk(sk)->icsk_pmtu_cookie is last pmtu, seen by this function.
L
Linus Torvalds 已提交
1782 1783 1784 1785 1786 1787 1788 1789 1790

   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.

1791 1792
   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
1793 1794 1795 1796
 */
unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu)
{
	struct tcp_sock *tp = tcp_sk(sk);
1797
	struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1798
	int mss_now;
L
Linus Torvalds 已提交
1799

J
John Heffner 已提交
1800 1801
	if (icsk->icsk_mtup.search_high > pmtu)
		icsk->icsk_mtup.search_high = pmtu;
L
Linus Torvalds 已提交
1802

J
John Heffner 已提交
1803
	mss_now = tcp_mtu_to_mss(sk, pmtu);
1804
	mss_now = tcp_bound_to_half_wnd(tp, mss_now);
L
Linus Torvalds 已提交
1805 1806

	/* And store cached results */
1807
	icsk->icsk_pmtu_cookie = pmtu;
J
John Heffner 已提交
1808 1809
	if (icsk->icsk_mtup.enabled)
		mss_now = min(mss_now, tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low));
1810
	tp->mss_cache = mss_now;
L
Linus Torvalds 已提交
1811 1812 1813

	return mss_now;
}
E
Eric Dumazet 已提交
1814
EXPORT_SYMBOL(tcp_sync_mss);
L
Linus Torvalds 已提交
1815 1816 1817 1818

/* Compute the current effective MSS, taking SACKs and IP options,
 * and even PMTU discovery events into account.
 */
I
Ilpo Järvinen 已提交
1819
unsigned int tcp_current_mss(struct sock *sk)
L
Linus Torvalds 已提交
1820
{
1821 1822
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct dst_entry *dst = __sk_dst_get(sk);
1823
	u32 mss_now;
1824
	unsigned int header_len;
A
Adam Langley 已提交
1825 1826
	struct tcp_out_options opts;
	struct tcp_md5sig_key *md5;
1827 1828 1829

	mss_now = tp->mss_cache;

L
Linus Torvalds 已提交
1830 1831
	if (dst) {
		u32 mtu = dst_mtu(dst);
1832
		if (mtu != inet_csk(sk)->icsk_pmtu_cookie)
L
Linus Torvalds 已提交
1833 1834 1835
			mss_now = tcp_sync_mss(sk, mtu);
	}

A
Adam Langley 已提交
1836 1837 1838 1839 1840 1841 1842 1843 1844 1845
	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;
	}
1846

L
Linus Torvalds 已提交
1847 1848 1849
	return mss_now;
}

1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868
/* 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;
	}
1869
	tp->snd_cwnd_stamp = tcp_jiffies32;
1870 1871
}

1872
static void tcp_cwnd_validate(struct sock *sk, bool is_cwnd_limited)
1873
{
1874
	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1875
	struct tcp_sock *tp = tcp_sk(sk);
1876

1877 1878 1879 1880
	/* 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) ||
1881 1882
	    tp->packets_out > tp->max_packets_out ||
	    is_cwnd_limited) {
1883 1884 1885 1886
		tp->max_packets_out = tp->packets_out;
		tp->max_packets_seq = tp->snd_nxt;
		tp->is_cwnd_limited = is_cwnd_limited;
	}
1887

1888
	if (tcp_is_cwnd_limited(sk)) {
1889 1890
		/* Network is feed fully. */
		tp->snd_cwnd_used = 0;
1891
		tp->snd_cwnd_stamp = tcp_jiffies32;
1892 1893 1894 1895 1896
	} else {
		/* Network starves. */
		if (tp->packets_out > tp->snd_cwnd_used)
			tp->snd_cwnd_used = tp->packets_out;

1897
		if (sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle &&
1898
		    (s32)(tcp_jiffies32 - tp->snd_cwnd_stamp) >= inet_csk(sk)->icsk_rto &&
1899
		    !ca_ops->cong_control)
1900
			tcp_cwnd_application_limited(sk);
1901 1902 1903 1904 1905

		/* 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)
1906
		 * 3) no more data to send (tcp_write_queue_empty())
1907 1908
		 * 4) application is hitting buffer limit (SOCK_NOSPACE)
		 */
1909
		if (tcp_write_queue_empty(sk) && sk->sk_socket &&
1910 1911 1912
		    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);
1913 1914 1915
	}
}

E
Eric Dumazet 已提交
1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929
/* 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
1930
 */
E
Eric Dumazet 已提交
1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945
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,
1946
			    int nonagle)
E
Eric Dumazet 已提交
1947 1948 1949 1950 1951
{
	return partial &&
		((nonagle & TCP_NAGLE_CORK) ||
		 (!nonagle && tp->packets_out && tcp_minshall_check(tp)));
}
E
Eric Dumazet 已提交
1952 1953 1954 1955

/* Return how many segs we'd like on a TSO packet,
 * to send one TSO packet per ms
 */
1956 1957
static u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
			    int min_tso_segs)
E
Eric Dumazet 已提交
1958 1959 1960
{
	u32 bytes, segs;

1961
	bytes = min_t(unsigned long,
1962
		      sk->sk_pacing_rate >> READ_ONCE(sk->sk_pacing_shift),
1963
		      sk->sk_gso_max_size - 1 - MAX_TCP_HEADER);
E
Eric Dumazet 已提交
1964 1965 1966 1967 1968 1969

	/* 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.
	 */
1970
	segs = max_t(u32, bytes / mss_now, min_tso_segs);
E
Eric Dumazet 已提交
1971

1972
	return segs;
E
Eric Dumazet 已提交
1973 1974
}

1975 1976 1977 1978 1979 1980
/* 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;
1981
	u32 min_tso, tso_segs;
1982

1983 1984 1985 1986 1987
	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);
1988
	return min_t(u32, tso_segs, sk->sk_gso_max_segs);
1989 1990
}

E
Eric Dumazet 已提交
1991 1992 1993 1994 1995 1996
/* 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)
1997
{
1998
	const struct tcp_sock *tp = tcp_sk(sk);
E
Eric Dumazet 已提交
1999
	u32 partial, needed, window, max_len;
2000

2001
	window = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
2002
	max_len = mss_now * max_segs;
2003

2004 2005
	if (likely(max_len <= window && skb != tcp_write_queue_tail(sk)))
		return max_len;
2006

2007 2008
	needed = min(skb->len, window);

2009 2010
	if (max_len <= needed)
		return max_len;
2011

E
Eric Dumazet 已提交
2012 2013 2014 2015 2016
	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
	 */
2017
	if (tcp_nagle_check(partial != 0, tp, nonagle))
E
Eric Dumazet 已提交
2018 2019 2020
		return needed - partial;

	return needed;
2021 2022 2023 2024 2025
}

/* 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.
 */
2026 2027
static inline unsigned int tcp_cwnd_test(const struct tcp_sock *tp,
					 const struct sk_buff *skb)
2028
{
E
Eric Dumazet 已提交
2029
	u32 in_flight, cwnd, halfcwnd;
2030 2031

	/* Don't be strict about the congestion window for the final FIN.  */
E
Eric Dumazet 已提交
2032 2033
	if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) &&
	    tcp_skb_pcount(skb) == 1)
2034 2035 2036 2037
		return 1;

	in_flight = tcp_packets_in_flight(tp);
	cwnd = tp->snd_cwnd;
E
Eric Dumazet 已提交
2038 2039
	if (in_flight >= cwnd)
		return 0;
2040

E
Eric Dumazet 已提交
2041 2042 2043 2044 2045
	/* For better scheduling, ensure we have at least
	 * 2 GSO packets in flight.
	 */
	halfcwnd = max(cwnd >> 1, 1U);
	return min(halfcwnd, cwnd - in_flight);
2046 2047
}

2048
/* Initialize TSO state of a skb.
2049
 * This must be invoked the first time we consider transmitting
2050 2051
 * SKB onto the wire.
 */
2052
static int tcp_init_tso_segs(struct sk_buff *skb, unsigned int mss_now)
2053 2054 2055
{
	int tso_segs = tcp_skb_pcount(skb);

2056
	if (!tso_segs || (tso_segs > 1 && tcp_skb_mss(skb) != mss_now)) {
2057
		tcp_set_skb_tso_segs(skb, mss_now);
2058 2059 2060 2061 2062 2063
		tso_segs = tcp_skb_pcount(skb);
	}
	return tso_segs;
}


E
Eric Dumazet 已提交
2064
/* Return true if the Nagle test allows this packet to be
2065 2066
 * sent now.
 */
E
Eric Dumazet 已提交
2067 2068
static inline bool tcp_nagle_test(const struct tcp_sock *tp, const struct sk_buff *skb,
				  unsigned int cur_mss, int nonagle)
2069 2070 2071 2072 2073 2074 2075 2076
{
	/* 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
2077
		return true;
2078

Y
Yuchung Cheng 已提交
2079 2080
	/* 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
2081
		return true;
2082

2083
	if (!tcp_nagle_check(skb->len < cur_mss, tp, nonagle))
E
Eric Dumazet 已提交
2084
		return true;
2085

E
Eric Dumazet 已提交
2086
	return false;
2087 2088 2089
}

/* Does at least the first segment of SKB fit into the send window? */
E
Eric Dumazet 已提交
2090 2091 2092
static bool tcp_snd_wnd_test(const struct tcp_sock *tp,
			     const struct sk_buff *skb,
			     unsigned int cur_mss)
2093 2094 2095 2096 2097 2098
{
	u32 end_seq = TCP_SKB_CB(skb)->end_seq;

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

2099
	return !after(end_seq, tcp_wnd_end(tp));
2100 2101 2102 2103 2104 2105 2106 2107 2108
}

/* 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).
 */
2109
static int tso_fragment(struct sock *sk, struct sk_buff *skb, unsigned int len,
2110
			unsigned int mss_now, gfp_t gfp)
2111 2112
{
	int nlen = skb->len - len;
2113
	struct sk_buff *buff;
2114
	u8 flags;
2115 2116

	/* All of a TSO frame must be composed of paged data.  */
2117
	if (skb->len != skb->data_len)
2118 2119
		return tcp_fragment(sk, TCP_FRAG_IN_WRITE_QUEUE,
				    skb, len, mss_now, gfp);
2120

E
Eric Dumazet 已提交
2121
	buff = tcp_stream_alloc_skb(sk, 0, gfp, true);
2122
	if (unlikely(!buff))
2123
		return -ENOMEM;
2124
	skb_copy_decrypted(buff, skb);
2125
	mptcp_skb_ext_copy(buff, skb);
2126

2127
	sk_wmem_queued_add(sk, buff->truesize);
2128
	sk_mem_charge(sk, buff->truesize);
2129
	buff->truesize += nlen;
2130 2131 2132 2133 2134 2135 2136 2137
	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
2138 2139 2140
	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;
2141

2142 2143
	tcp_skb_fragment_eor(skb, buff);

2144
	skb_split(skb, buff, len);
2145
	tcp_fragment_tstamp(skb, buff);
2146 2147

	/* Fix up tso_factor for both original and new SKB.  */
2148 2149
	tcp_set_skb_tso_segs(skb, mss_now);
	tcp_set_skb_tso_segs(buff, mss_now);
2150 2151

	/* Link BUFF into the send queue. */
2152
	__skb_header_release(buff);
2153
	tcp_insert_write_queue_after(skb, buff, sk, TCP_FRAG_IN_WRITE_QUEUE);
2154 2155 2156 2157 2158 2159 2160 2161 2162

	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.
 */
2163
static bool tcp_tso_should_defer(struct sock *sk, struct sk_buff *skb,
2164 2165 2166
				 bool *is_cwnd_limited,
				 bool *is_rwnd_limited,
				 u32 max_segs)
2167
{
2168
	const struct inet_connection_sock *icsk = inet_csk(sk);
2169
	u32 send_win, cong_win, limit, in_flight;
2170 2171
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *head;
2172
	int win_divisor;
2173
	s64 delta;
2174

2175
	if (icsk->icsk_ca_state >= TCP_CA_Recovery)
J
John Heffner 已提交
2176 2177
		goto send_now;

E
Eric Dumazet 已提交
2178
	/* Avoid bursty behavior by allowing defer
2179 2180 2181
	 * only if the last write was recent (1 ms).
	 * Note that tp->tcp_wstamp_ns can be in the future if we have
	 * packets waiting in a qdisc or device for EDT delivery.
E
Eric Dumazet 已提交
2182
	 */
2183 2184
	delta = tp->tcp_clock_cache - tp->tcp_wstamp_ns - NSEC_PER_MSEC;
	if (delta > 0)
J
John Heffner 已提交
2185
		goto send_now;
2186

2187 2188
	in_flight = tcp_packets_in_flight(tp);

2189 2190
	BUG_ON(tcp_skb_pcount(skb) <= 1);
	BUG_ON(tp->snd_cwnd <= in_flight);
2191

2192
	send_win = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
2193 2194 2195 2196 2197 2198

	/* 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);

2199
	/* If a full-sized TSO skb can be sent, do it. */
E
Eric Dumazet 已提交
2200
	if (limit >= max_segs * tp->mss_cache)
J
John Heffner 已提交
2201
		goto send_now;
2202

2203 2204 2205 2206
	/* 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;

2207
	win_divisor = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_tso_win_divisor);
2208
	if (win_divisor) {
2209 2210 2211 2212 2213
		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.
		 */
2214
		chunk /= win_divisor;
2215
		if (limit >= chunk)
J
John Heffner 已提交
2216
			goto send_now;
2217 2218 2219 2220 2221 2222
	} 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.
		 */
2223
		if (limit > tcp_max_tso_deferred_mss(tp) * tp->mss_cache)
J
John Heffner 已提交
2224
			goto send_now;
2225 2226
	}

2227 2228 2229 2230
	/* TODO : use tsorted_sent_queue ? */
	head = tcp_rtx_queue_head(sk);
	if (!head)
		goto send_now;
2231
	delta = tp->tcp_clock_cache - head->tstamp;
2232
	/* If next ACK is likely to come too late (half srtt), do not defer */
2233
	if ((s64)(delta - (u64)NSEC_PER_USEC * (tp->srtt_us >> 4)) < 0)
2234 2235
		goto send_now;

2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252
	/* Ok, it looks like it is advisable to defer.
	 * Three cases are tracked :
	 * 1) We are cwnd-limited
	 * 2) We are rwnd-limited
	 * 3) We are application limited.
	 */
	if (cong_win < send_win) {
		if (cong_win <= skb->len) {
			*is_cwnd_limited = true;
			return true;
		}
	} else {
		if (send_win <= skb->len) {
			*is_rwnd_limited = true;
			return true;
		}
	}
J
John Heffner 已提交
2253

2254
	/* If this packet won't get more data, do not wait. */
2255 2256
	if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) ||
	    TCP_SKB_CB(skb)->eor)
2257
		goto send_now;
2258

E
Eric Dumazet 已提交
2259
	return true;
J
John Heffner 已提交
2260 2261

send_now:
E
Eric Dumazet 已提交
2262
	return false;
2263 2264
}

2265 2266 2267 2268 2269 2270 2271 2272 2273
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;
2274
	delta = tcp_jiffies32 - icsk->icsk_mtup.probe_timestamp;
2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285
	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 */
2286
		icsk->icsk_mtup.probe_timestamp = tcp_jiffies32;
2287 2288 2289
	}
}

2290 2291 2292 2293 2294 2295 2296 2297 2298
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;

2299 2300 2301
		if (unlikely(TCP_SKB_CB(skb)->eor) ||
		    tcp_has_tx_tstamp(skb) ||
		    !skb_pure_zcopy_same(skb, next))
2302 2303 2304 2305 2306 2307 2308 2309
			return false;

		len -= skb->len;
	}

	return true;
}

J
John Heffner 已提交
2310
/* Create a new MTU probe if we are ready.
2311 2312 2313 2314
 * 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
2315 2316
 * Returns 0 if we should wait to probe (no cwnd available),
 *         1 if a probe was sent,
2317 2318
 *         -1 otherwise
 */
J
John Heffner 已提交
2319 2320 2321
static int tcp_mtu_probe(struct sock *sk)
{
	struct inet_connection_sock *icsk = inet_csk(sk);
2322
	struct tcp_sock *tp = tcp_sk(sk);
J
John Heffner 已提交
2323
	struct sk_buff *skb, *nskb, *next;
2324
	struct net *net = sock_net(sk);
J
John Heffner 已提交
2325
	int probe_size;
2326
	int size_needed;
2327
	int copy, len;
J
John Heffner 已提交
2328
	int mss_now;
2329
	int interval;
J
John Heffner 已提交
2330 2331 2332 2333

	/* Not currently probing/verifying,
	 * not in recovery,
	 * have enough cwnd, and
2334 2335 2336 2337 2338 2339 2340
	 * 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
2341 2342
		return -1;

2343 2344 2345 2346
	/* 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
2347
	mss_now = tcp_current_mss(sk);
2348 2349
	probe_size = tcp_mtu_to_mss(sk, (icsk->icsk_mtup.search_high +
				    icsk->icsk_mtup.search_low) >> 1);
2350
	size_needed = probe_size + (tp->reordering + 1) * tp->mss_cache;
2351
	interval = icsk->icsk_mtup.search_high - icsk->icsk_mtup.search_low;
2352 2353 2354 2355
	/* 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.
	 */
2356
	if (probe_size > tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_high) ||
2357 2358 2359 2360 2361
		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
2362 2363 2364 2365
		return -1;
	}

	/* Have enough data in the send queue to probe? */
2366
	if (tp->write_seq - tp->snd_nxt < size_needed)
J
John Heffner 已提交
2367 2368
		return -1;

2369 2370
	if (tp->snd_wnd < size_needed)
		return -1;
2371
	if (after(tp->snd_nxt + size_needed, tcp_wnd_end(tp)))
2372
		return 0;
J
John Heffner 已提交
2373

2374 2375 2376
	/* 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
2377 2378 2379 2380 2381
			return -1;
		else
			return 0;
	}

2382 2383 2384
	if (!tcp_can_coalesce_send_queue_head(sk, probe_size))
		return -1;

J
John Heffner 已提交
2385
	/* We're allowed to probe.  Build it now. */
E
Eric Dumazet 已提交
2386
	nskb = tcp_stream_alloc_skb(sk, probe_size, GFP_ATOMIC, false);
2387
	if (!nskb)
J
John Heffner 已提交
2388
		return -1;
2389
	sk_wmem_queued_add(sk, nskb->truesize);
2390
	sk_mem_charge(sk, nskb->truesize);
J
John Heffner 已提交
2391

2392
	skb = tcp_send_head(sk);
2393
	skb_copy_decrypted(nskb, skb);
2394
	mptcp_skb_ext_copy(nskb, skb);
J
John Heffner 已提交
2395 2396 2397

	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
2398
	TCP_SKB_CB(nskb)->tcp_flags = TCPHDR_ACK;
J
John Heffner 已提交
2399

2400
	tcp_insert_write_queue_before(nskb, skb, sk);
2401
	tcp_highest_sack_replace(sk, skb, nskb);
2402

J
John Heffner 已提交
2403
	len = 0;
2404
	tcp_for_write_queue_from_safe(skb, next, sk) {
J
John Heffner 已提交
2405
		copy = min_t(int, skb->len, probe_size - len);
2406
		skb_copy_bits(skb, 0, skb_put(nskb, copy), copy);
J
John Heffner 已提交
2407 2408 2409 2410

		if (skb->len <= copy) {
			/* We've eaten all the data from this skb.
			 * Throw it away. */
E
Eric Dumazet 已提交
2411
			TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;
2412 2413 2414 2415
			/* 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;
2416
			tcp_skb_collapse_tstamp(nskb, skb);
2417
			tcp_unlink_write_queue(skb, sk);
T
Talal Ahmad 已提交
2418
			tcp_wmem_free_skb(sk, skb);
J
John Heffner 已提交
2419
		} else {
E
Eric Dumazet 已提交
2420
			TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags &
C
Changli Gao 已提交
2421
						   ~(TCPHDR_FIN|TCPHDR_PSH);
J
John Heffner 已提交
2422 2423 2424 2425
			if (!skb_shinfo(skb)->nr_frags) {
				skb_pull(skb, copy);
			} else {
				__pskb_trim_head(skb, copy);
2426
				tcp_set_skb_tso_segs(skb, mss_now);
J
John Heffner 已提交
2427 2428 2429 2430 2431
			}
			TCP_SKB_CB(skb)->seq += copy;
		}

		len += copy;
2432 2433 2434

		if (len >= probe_size)
			break;
J
John Heffner 已提交
2435
	}
2436
	tcp_init_tso_segs(nskb, nskb->len);
J
John Heffner 已提交
2437 2438

	/* We're ready to send.  If this fails, the probe will
E
Eric Dumazet 已提交
2439 2440
	 * be resegmented into mss-sized pieces by tcp_write_xmit().
	 */
J
John Heffner 已提交
2441 2442
	if (!tcp_transmit_skb(sk, nskb, 1, GFP_ATOMIC)) {
		/* Decrement cwnd here because we are sending
2443
		 * effectively two packets. */
J
John Heffner 已提交
2444
		tp->snd_cwnd--;
2445
		tcp_event_new_data_sent(sk, nskb);
J
John Heffner 已提交
2446 2447

		icsk->icsk_mtup.probe_size = tcp_mss_to_mtu(sk, nskb->len);
2448 2449
		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
2450 2451 2452 2453 2454 2455 2456

		return 1;
	}

	return -1;
}

E
Eric Dumazet 已提交
2457
static bool tcp_pacing_check(struct sock *sk)
2458
{
E
Eric Dumazet 已提交
2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473
	struct tcp_sock *tp = tcp_sk(sk);

	if (!tcp_needs_internal_pacing(sk))
		return false;

	if (tp->tcp_wstamp_ns <= tp->tcp_clock_cache)
		return false;

	if (!hrtimer_is_queued(&tp->pacing_timer)) {
		hrtimer_start(&tp->pacing_timer,
			      ns_to_ktime(tp->tcp_wstamp_ns),
			      HRTIMER_MODE_ABS_PINNED_SOFT);
		sock_hold(sk);
	}
	return true;
2474 2475
}

2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489
/* 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)
{
2490
	unsigned long limit;
2491

2492 2493
	limit = max_t(unsigned long,
		      2 * skb->truesize,
2494
		      sk->sk_pacing_rate >> READ_ONCE(sk->sk_pacing_shift));
2495 2496 2497
	if (sk->sk_pacing_status == SK_PACING_NONE)
		limit = min_t(unsigned long, limit,
			      sock_net(sk)->ipv4.sysctl_tcp_limit_output_bytes);
2498 2499
	limit <<= factor;

2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511
	if (static_branch_unlikely(&tcp_tx_delay_enabled) &&
	    tcp_sk(sk)->tcp_tx_delay) {
		u64 extra_bytes = (u64)sk->sk_pacing_rate * tcp_sk(sk)->tcp_tx_delay;

		/* TSQ is based on skb truesize sum (sk_wmem_alloc), so we
		 * approximate our needs assuming an ~100% skb->truesize overhead.
		 * USEC_PER_SEC is approximated by 2^20.
		 * do_div(extra_bytes, USEC_PER_SEC/2) is replaced by a right shift.
		 */
		extra_bytes >>= (20 - 1);
		limit += extra_bytes;
	}
2512
	if (refcount_read(&sk->sk_wmem_alloc) > limit) {
2513
		/* Always send skb if rtx queue is empty.
2514 2515 2516 2517
		 * No need to wait for TX completion to call us back,
		 * after softirq/tasklet schedule.
		 * This helps when TX completions are delayed too much.
		 */
2518
		if (tcp_rtx_queue_empty(sk))
2519 2520
			return false;

2521
		set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags);
2522 2523 2524 2525 2526
		/* It is possible TX completion already happened
		 * before we set TSQ_THROTTLED, so we must
		 * test again the condition.
		 */
		smp_mb__after_atomic();
2527
		if (refcount_read(&sk->sk_wmem_alloc) > limit)
2528 2529 2530 2531 2532
			return true;
	}
	return false;
}

2533 2534
static void tcp_chrono_set(struct tcp_sock *tp, const enum tcp_chrono new)
{
2535
	const u32 now = tcp_jiffies32;
2536
	enum tcp_chrono old = tp->chrono_type;
2537

2538 2539
	if (old > TCP_CHRONO_UNSPEC)
		tp->chrono_stat[old - 1] += now - tp->chrono_start;
2540 2541 2542 2543 2544 2545 2546 2547 2548
	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
2549 2550
	 * chronograph then the highest priority enum takes precedence
	 * over the other conditions. So that if something "more interesting"
2551 2552 2553 2554 2555 2556 2557 2558 2559 2560
	 * 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);

2561 2562 2563 2564 2565 2566 2567 2568

	/* 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.
	 */
2569
	if (tcp_rtx_and_write_queues_empty(sk))
2570 2571 2572
		tcp_chrono_set(tp, TCP_CHRONO_UNSPEC);
	else if (type == tp->chrono_type)
		tcp_chrono_set(tp, TCP_CHRONO_BUSY);
2573 2574
}

L
Linus Torvalds 已提交
2575 2576 2577 2578
/* This routine writes packets to the network.  It advances the
 * send_head.  This happens as incoming acks open up the remote
 * window for us.
 *
2579 2580 2581 2582
 * 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
2583 2584 2585
 * 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
2586 2587
 * 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
2588
 */
E
Eric Dumazet 已提交
2589 2590
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
2591 2592
{
	struct tcp_sock *tp = tcp_sk(sk);
2593
	struct sk_buff *skb;
2594 2595
	unsigned int tso_segs, sent_pkts;
	int cwnd_quota;
J
John Heffner 已提交
2596
	int result;
2597
	bool is_cwnd_limited = false, is_rwnd_limited = false;
E
Eric Dumazet 已提交
2598
	u32 max_segs;
L
Linus Torvalds 已提交
2599

2600
	sent_pkts = 0;
J
John Heffner 已提交
2601

2602
	tcp_mstamp_refresh(tp);
2603 2604 2605 2606
	if (!push_one) {
		/* Do MTU probing. */
		result = tcp_mtu_probe(sk);
		if (!result) {
E
Eric Dumazet 已提交
2607
			return false;
2608 2609 2610
		} else if (result > 0) {
			sent_pkts = 1;
		}
J
John Heffner 已提交
2611 2612
	}

2613
	max_segs = tcp_tso_segs(sk, mss_now);
2614
	while ((skb = tcp_send_head(sk))) {
2615 2616
		unsigned int limit;

2617 2618 2619 2620
		if (unlikely(tp->repair) && tp->repair_queue == TCP_SEND_QUEUE) {
			/* "skb_mstamp_ns" is used as a start point for the retransmit timer */
			skb->skb_mstamp_ns = tp->tcp_wstamp_ns = tp->tcp_clock_cache;
			list_move_tail(&skb->tcp_tsorted_anchor, &tp->tsorted_sent_queue);
2621
			tcp_init_tso_segs(skb, mss_now);
2622 2623 2624
			goto repair; /* Skip network transmission */
		}

2625 2626 2627
		if (tcp_pacing_check(sk))
			break;

2628
		tso_segs = tcp_init_tso_segs(skb, mss_now);
2629
		BUG_ON(!tso_segs);
2630

2631
		cwnd_quota = tcp_cwnd_test(tp, skb);
N
Nandita Dukkipati 已提交
2632 2633 2634 2635 2636 2637 2638
		if (!cwnd_quota) {
			if (push_one == 2)
				/* Force out a loss probe pkt. */
				cwnd_quota = 1;
			else
				break;
		}
2639

2640 2641
		if (unlikely(!tcp_snd_wnd_test(tp, skb, mss_now))) {
			is_rwnd_limited = true;
2642
			break;
2643
		}
2644

2645
		if (tso_segs == 1) {
2646 2647 2648 2649 2650
			if (unlikely(!tcp_nagle_test(tp, skb, mss_now,
						     (tcp_skb_is_last(sk, skb) ?
						      nonagle : TCP_NAGLE_PUSH))))
				break;
		} else {
2651
			if (!push_one &&
E
Eric Dumazet 已提交
2652
			    tcp_tso_should_defer(sk, skb, &is_cwnd_limited,
2653
						 &is_rwnd_limited, max_segs))
2654 2655
				break;
		}
2656

E
Eric Dumazet 已提交
2657
		limit = mss_now;
2658
		if (tso_segs > 1 && !tcp_urg_mode(tp))
E
Eric Dumazet 已提交
2659 2660 2661 2662 2663 2664 2665
			limit = tcp_mss_split_point(sk, skb, mss_now,
						    min_t(unsigned int,
							  cwnd_quota,
							  max_segs),
						    nonagle);

		if (skb->len > limit &&
2666
		    unlikely(tso_fragment(sk, skb, limit, mss_now, gfp)))
E
Eric Dumazet 已提交
2667 2668
			break;

2669 2670
		if (tcp_small_queue_check(sk, skb, 0))
			break;
E
Eric Dumazet 已提交
2671

2672 2673 2674 2675 2676 2677 2678 2679
		/* Argh, we hit an empty skb(), presumably a thread
		 * is sleeping in sendmsg()/sk_stream_wait_memory().
		 * We do not want to send a pure-ack packet and have
		 * a strange looking rtx queue with empty packet(s).
		 */
		if (TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq)
			break;

2680
		if (unlikely(tcp_transmit_skb(sk, skb, 1, gfp)))
2681
			break;
L
Linus Torvalds 已提交
2682

2683
repair:
2684 2685 2686
		/* Advance the send_head.  This one is sent out.
		 * This call will increment packets_out.
		 */
2687
		tcp_event_new_data_sent(sk, skb);
L
Linus Torvalds 已提交
2688

2689
		tcp_minshall_update(tp, mss_now, skb);
2690
		sent_pkts += tcp_skb_pcount(skb);
2691 2692 2693

		if (push_one)
			break;
2694
	}
L
Linus Torvalds 已提交
2695

2696 2697 2698 2699 2700
	if (is_rwnd_limited)
		tcp_chrono_start(sk, TCP_CHRONO_RWND_LIMITED);
	else
		tcp_chrono_stop(sk, TCP_CHRONO_RWND_LIMITED);

2701 2702 2703 2704
	is_cwnd_limited |= (tcp_packets_in_flight(tp) >= tp->snd_cwnd);
	if (likely(sent_pkts || is_cwnd_limited))
		tcp_cwnd_validate(sk, is_cwnd_limited);

2705
	if (likely(sent_pkts)) {
2706 2707
		if (tcp_in_cwnd_reduction(sk))
			tp->prr_out += sent_pkts;
N
Nandita Dukkipati 已提交
2708 2709 2710

		/* Send one loss probe per tail loss episode. */
		if (push_one != 2)
2711
			tcp_schedule_loss_probe(sk, false);
E
Eric Dumazet 已提交
2712
		return false;
L
Linus Torvalds 已提交
2713
	}
2714
	return !tp->packets_out && !tcp_write_queue_empty(sk);
N
Nandita Dukkipati 已提交
2715 2716
}

2717
bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto)
N
Nandita Dukkipati 已提交
2718 2719 2720
{
	struct inet_connection_sock *icsk = inet_csk(sk);
	struct tcp_sock *tp = tcp_sk(sk);
2721
	u32 timeout, rto_delta_us;
2722
	int early_retrans;
N
Nandita Dukkipati 已提交
2723 2724 2725 2726

	/* Don't do any loss probe on a Fast Open connection before 3WHS
	 * finishes.
	 */
2727
	if (rcu_access_pointer(tp->fastopen_rsk))
N
Nandita Dukkipati 已提交
2728 2729
		return false;

2730
	early_retrans = sock_net(sk)->ipv4.sysctl_tcp_early_retrans;
N
Nandita Dukkipati 已提交
2731
	/* Schedule a loss probe in 2*RTT for SACK capable connections
N
Neal Cardwell 已提交
2732
	 * not in loss recovery, that are either limited by cwnd or application.
N
Nandita Dukkipati 已提交
2733
	 */
2734
	if ((early_retrans != 3 && early_retrans != 4) ||
Y
Yuchung Cheng 已提交
2735
	    !tp->packets_out || !tcp_is_sack(tp) ||
N
Neal Cardwell 已提交
2736 2737
	    (icsk->icsk_ca_state != TCP_CA_Open &&
	     icsk->icsk_ca_state != TCP_CA_CWR))
N
Nandita Dukkipati 已提交
2738 2739
		return false;

2740
	/* Probe timeout is 2*rtt. Add minimum RTO to account
2741 2742
	 * 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
2743
	 */
2744 2745 2746 2747 2748 2749 2750 2751 2752
	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
2753

2754
	/* If the RTO formula yields an earlier time, then use that time. */
2755 2756 2757
	rto_delta_us = advancing_rto ?
			jiffies_to_usecs(inet_csk(sk)->icsk_rto) :
			tcp_rto_delta_us(sk);  /* How far in future is RTO? */
2758 2759
	if (rto_delta_us > 0)
		timeout = min_t(u32, timeout, usecs_to_jiffies(rto_delta_us));
N
Nandita Dukkipati 已提交
2760

2761
	tcp_reset_xmit_timer(sk, ICSK_TIME_LOSS_PROBE, timeout, TCP_RTO_MAX);
N
Nandita Dukkipati 已提交
2762 2763 2764
	return true;
}

2765 2766 2767 2768
/* 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 !
 */
2769
static bool skb_still_in_host_queue(struct sock *sk,
2770 2771
				    const struct sk_buff *skb)
{
2772
	if (unlikely(skb_fclone_busy(sk, skb))) {
2773 2774 2775 2776 2777 2778 2779
		set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags);
		smp_mb__after_atomic();
		if (skb_fclone_busy(sk, skb)) {
			NET_INC_STATS(sock_net(sk),
				      LINUX_MIB_TCPSPURIOUS_RTX_HOSTQUEUES);
			return true;
		}
2780 2781 2782 2783
	}
	return false;
}

2784
/* When probe timeout (PTO) fires, try send a new segment if possible, else
N
Nandita Dukkipati 已提交
2785 2786 2787 2788
 * retransmit the last segment.
 */
void tcp_send_loss_probe(struct sock *sk)
{
N
Nandita Dukkipati 已提交
2789
	struct tcp_sock *tp = tcp_sk(sk);
N
Nandita Dukkipati 已提交
2790 2791 2792 2793
	struct sk_buff *skb;
	int pcount;
	int mss = tcp_current_mss(sk);

2794 2795 2796 2797 2798
	/* At most one outstanding TLP */
	if (tp->tlp_high_seq)
		goto rearm_timer;

	tp->tlp_retrans = 0;
2799
	skb = tcp_send_head(sk);
2800 2801 2802 2803 2804 2805
	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
2806
	}
2807
	skb = skb_rb_last(&sk->tcp_rtx_queue);
2808 2809 2810 2811 2812 2813 2814
	if (unlikely(!skb)) {
		WARN_ONCE(tp->packets_out,
			  "invalid inflight: %u state %u cwnd %u mss %d\n",
			  tp->packets_out, sk->sk_state, tp->snd_cwnd, mss);
		inet_csk(sk)->icsk_pending = 0;
		return;
	}
N
Nandita Dukkipati 已提交
2815

2816 2817 2818
	if (skb_still_in_host_queue(sk, skb))
		goto rearm_timer;

N
Nandita Dukkipati 已提交
2819 2820 2821 2822 2823
	pcount = tcp_skb_pcount(skb);
	if (WARN_ON(!pcount))
		goto rearm_timer;

	if ((pcount > 1) && (skb->len > (pcount - 1) * mss)) {
2824 2825
		if (unlikely(tcp_fragment(sk, TCP_FRAG_IN_RTX_QUEUE, skb,
					  (pcount - 1) * mss, mss,
2826
					  GFP_ATOMIC)))
N
Nandita Dukkipati 已提交
2827
			goto rearm_timer;
2828
		skb = skb_rb_next(skb);
N
Nandita Dukkipati 已提交
2829 2830 2831 2832 2833
	}

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

2834
	if (__tcp_retransmit_skb(sk, skb, 1))
2835
		goto rearm_timer;
N
Nandita Dukkipati 已提交
2836

2837 2838 2839
	tp->tlp_retrans = 1;

probe_sent:
N
Nandita Dukkipati 已提交
2840
	/* Record snd_nxt for loss detection. */
2841
	tp->tlp_high_seq = tp->snd_nxt;
N
Nandita Dukkipati 已提交
2842

2843
	NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPLOSSPROBES);
2844 2845
	/* Reset s.t. tcp_rearm_rto will restart timer from now */
	inet_csk(sk)->icsk_pending = 0;
N
Nandita Dukkipati 已提交
2846
rearm_timer:
2847
	tcp_rearm_rto(sk);
L
Linus Torvalds 已提交
2848 2849
}

2850 2851 2852 2853
/* 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.
 */
2854 2855
void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
			       int nonagle)
2856
{
2857 2858 2859 2860 2861 2862 2863
	/* 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;

2864
	if (tcp_write_xmit(sk, cur_mss, nonagle, 0,
2865
			   sk_gfp_mask(sk, GFP_ATOMIC)))
2866
		tcp_check_probe_timer(sk);
2867 2868
}

2869 2870 2871 2872 2873
/* 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)
{
2874
	struct sk_buff *skb = tcp_send_head(sk);
2875 2876 2877

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

2878
	tcp_write_xmit(sk, mss_now, TCP_NAGLE_PUSH, 1, sk->sk_allocation);
2879 2880
}

L
Linus Torvalds 已提交
2881 2882
/* This function returns the amount that we can raise the
 * usable window based on the following constraints
2883
 *
L
Linus Torvalds 已提交
2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901
 * 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.
2902
 *
L
Linus Torvalds 已提交
2903 2904
 * Of course, if the sender implements sender side SWS prevention
 * then this will not be a problem.
2905
 *
L
Linus Torvalds 已提交
2906
 * BSD seems to make the following compromise:
2907
 *
L
Linus Torvalds 已提交
2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934
 *	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)
{
2935
	struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
2936
	struct tcp_sock *tp = tcp_sk(sk);
S
Stephen Hemminger 已提交
2937
	/* MSS for the peer's data.  Previous versions used mss_clamp
L
Linus Torvalds 已提交
2938 2939 2940 2941 2942
	 * 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
	 */
2943
	int mss = icsk->icsk_ack.rcv_mss;
L
Linus Torvalds 已提交
2944
	int free_space = tcp_space(sk);
2945
	int allowed_space = tcp_full_space(sk);
2946 2947 2948 2949 2950 2951
	int full_space, window;

	if (sk_is_mptcp(sk))
		mptcp_space(sk, &free_space, &allowed_space);

	full_space = min_t(int, tp->window_clamp, allowed_space);
L
Linus Torvalds 已提交
2952

2953
	if (unlikely(mss > full_space)) {
2954
		mss = full_space;
2955 2956 2957
		if (mss <= 0)
			return 0;
	}
2958
	if (free_space < (full_space >> 1)) {
2959
		icsk->icsk_ack.quick = 0;
L
Linus Torvalds 已提交
2960

2961
		if (tcp_under_memory_pressure(sk))
2962
			tcp_adjust_rcv_ssthresh(sk);
L
Linus Torvalds 已提交
2963

2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976
		/* 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
2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992
			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.
		 */
2993
		window = ALIGN(window, (1 << tp->rx_opt.rcv_wscale));
L
Linus Torvalds 已提交
2994
	} else {
2995
		window = tp->rcv_wnd;
L
Linus Torvalds 已提交
2996 2997 2998 2999 3000 3001 3002 3003 3004
		/* 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)
3005
			window = rounddown(free_space, mss);
3006
		else if (mss == full_space &&
3007
			 free_space > window + (full_space >> 1))
3008
			window = free_space;
L
Linus Torvalds 已提交
3009 3010 3011 3012 3013
	}

	return window;
}

3014 3015
void tcp_skb_collapse_tstamp(struct sk_buff *skb,
			     const struct sk_buff *next_skb)
3016
{
3017 3018 3019
	if (unlikely(tcp_has_tx_tstamp(next_skb))) {
		const struct skb_shared_info *next_shinfo =
			skb_shinfo(next_skb);
3020 3021
		struct skb_shared_info *shinfo = skb_shinfo(skb);

3022
		shinfo->tx_flags |= next_shinfo->tx_flags & SKBTX_ANY_TSTAMP;
3023
		shinfo->tskey = next_shinfo->tskey;
3024 3025
		TCP_SKB_CB(skb)->txstamp_ack |=
			TCP_SKB_CB(next_skb)->txstamp_ack;
3026 3027 3028
	}
}

3029
/* Collapses two adjacent SKB's during retransmission. */
3030
static bool tcp_collapse_retrans(struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
3031 3032
{
	struct tcp_sock *tp = tcp_sk(sk);
3033
	struct sk_buff *next_skb = skb_rb_next(skb);
3034
	int next_skb_size;
L
Linus Torvalds 已提交
3035

3036
	next_skb_size = next_skb->len;
L
Linus Torvalds 已提交
3037

3038
	BUG_ON(tcp_skb_pcount(skb) != 1 || tcp_skb_pcount(next_skb) != 1);
3039

3040 3041 3042
	if (next_skb_size && !tcp_skb_shift(skb, next_skb, 1, next_skb_size))
		return false;

3043
	tcp_highest_sack_replace(sk, next_skb, skb);
L
Linus Torvalds 已提交
3044

3045 3046
	/* 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
3047

3048
	/* Merge over control information. This moves PSH/FIN etc. over */
E
Eric Dumazet 已提交
3049
	TCP_SKB_CB(skb)->tcp_flags |= TCP_SKB_CB(next_skb)->tcp_flags;
3050 3051 3052 3053 3054

	/* 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;
3055
	TCP_SKB_CB(skb)->eor = TCP_SKB_CB(next_skb)->eor;
3056 3057

	/* changed transmit queue under us so clear hints */
3058 3059 3060
	tcp_clear_retrans_hints_partial(tp);
	if (next_skb == tp->retransmit_skb_hint)
		tp->retransmit_skb_hint = skb;
3061

3062 3063
	tcp_adjust_pcount(sk, next_skb, tcp_skb_pcount(next_skb));

3064 3065
	tcp_skb_collapse_tstamp(skb, next_skb);

3066
	tcp_rtx_queue_unlink_and_free(next_skb, sk);
3067
	return true;
L
Linus Torvalds 已提交
3068 3069
}

3070
/* Check if coalescing SKBs is legal. */
E
Eric Dumazet 已提交
3071
static bool tcp_can_collapse(const struct sock *sk, const struct sk_buff *skb)
3072 3073
{
	if (tcp_skb_pcount(skb) > 1)
E
Eric Dumazet 已提交
3074
		return false;
3075
	if (skb_cloned(skb))
E
Eric Dumazet 已提交
3076
		return false;
E
Eric Dumazet 已提交
3077
	/* Some heuristics for collapsing over SACK'd could be invented */
3078
	if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)
E
Eric Dumazet 已提交
3079
		return false;
3080

E
Eric Dumazet 已提交
3081
	return true;
3082 3083
}

3084 3085 3086
/* Collapse packets in the retransmit queue to make to create
 * less packets on the wire. This is only done on retransmission.
 */
3087 3088 3089 3090 3091
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
3092
	bool first = true;
3093

3094
	if (!sock_net(sk)->ipv4.sysctl_tcp_retrans_collapse)
3095
		return;
E
Eric Dumazet 已提交
3096
	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)
3097 3098
		return;

3099
	skb_rbtree_walk_from_safe(skb, tmp) {
3100 3101 3102
		if (!tcp_can_collapse(sk, skb))
			break;

3103
		if (!tcp_skb_can_collapse(to, skb))
3104 3105
			break;

3106 3107 3108
		space -= skb->len;

		if (first) {
E
Eric Dumazet 已提交
3109
			first = false;
3110 3111 3112 3113 3114 3115 3116 3117 3118
			continue;
		}

		if (space < 0)
			break;

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

3119 3120
		if (!tcp_collapse_retrans(sk, to))
			break;
3121 3122 3123
	}
}

L
Linus Torvalds 已提交
3124 3125 3126 3127
/* 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.
 */
3128
int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs)
L
Linus Torvalds 已提交
3129
{
J
John Heffner 已提交
3130
	struct inet_connection_sock *icsk = inet_csk(sk);
3131
	struct tcp_sock *tp = tcp_sk(sk);
3132
	unsigned int cur_mss;
3133 3134
	int diff, len, err;

L
Linus Torvalds 已提交
3135

3136 3137
	/* Inconclusive MTU probe */
	if (icsk->icsk_mtup.probe_size)
J
John Heffner 已提交
3138
		icsk->icsk_mtup.probe_size = 0;
L
Linus Torvalds 已提交
3139

3140 3141 3142
	if (skb_still_in_host_queue(sk, skb))
		return -EBUSY;

L
Linus Torvalds 已提交
3143
	if (before(TCP_SKB_CB(skb)->seq, tp->snd_una)) {
3144 3145 3146 3147
		if (unlikely(before(TCP_SKB_CB(skb)->end_seq, tp->snd_una))) {
			WARN_ON_ONCE(1);
			return -EINVAL;
		}
L
Linus Torvalds 已提交
3148 3149 3150 3151
		if (tcp_trim_head(sk, skb, tp->snd_una - TCP_SKB_CB(skb)->seq))
			return -ENOMEM;
	}

3152 3153 3154
	if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk))
		return -EHOSTUNREACH; /* Routing failure or similar. */

I
Ilpo Järvinen 已提交
3155
	cur_mss = tcp_current_mss(sk);
3156

L
Linus Torvalds 已提交
3157 3158 3159 3160 3161
	/* 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.
	 */
3162 3163
	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
3164 3165
		return -EAGAIN;

3166 3167
	len = cur_mss * segs;
	if (skb->len > len) {
3168 3169
		if (tcp_fragment(sk, TCP_FRAG_IN_RTX_QUEUE, skb, len,
				 cur_mss, GFP_ATOMIC))
L
Linus Torvalds 已提交
3170
			return -ENOMEM; /* We'll try again later. */
3171
	} else {
3172 3173
		if (skb_unclone(skb, GFP_ATOMIC))
			return -ENOMEM;
3174

3175 3176 3177 3178 3179 3180 3181
		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
3182 3183
	}

3184 3185 3186 3187
	/* 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);

3188 3189 3190 3191 3192 3193
	/* 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;
3194
	tp->bytes_retrans += skb->len;
3195

3196 3197 3198 3199 3200 3201
	/* 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)) {
3202 3203
		struct sk_buff *nskb;

3204 3205
		tcp_skb_tsorted_save(skb) {
			nskb = __pskb_copy(skb, MAX_TCP_HEADER, GFP_ATOMIC);
3206 3207 3208 3209 3210 3211
			if (nskb) {
				nskb->dev = NULL;
				err = tcp_transmit_skb(sk, nskb, 0, GFP_ATOMIC);
			} else {
				err = -ENOBUFS;
			}
3212 3213
		} tcp_skb_tsorted_restore(skb);

3214
		if (!err) {
3215
			tcp_update_skb_after_send(sk, skb, tp->tcp_wstamp_ns);
3216 3217
			tcp_rate_skb_sent(sk, skb);
		}
E
Eric Dumazet 已提交
3218
	} else {
3219
		err = tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
E
Eric Dumazet 已提交
3220
	}
3221

3222 3223 3224 3225 3226
	/* To avoid taking spuriously low RTT samples based on a timestamp
	 * for a transmit that never happened, always mark EVER_RETRANS
	 */
	TCP_SKB_CB(skb)->sacked |= TCPCB_EVER_RETRANS;

3227 3228 3229 3230
	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);

3231
	if (likely(!err)) {
3232
		trace_tcp_retransmit_skb(sk, skb);
3233
	} else if (err != -EBUSY) {
3234
		NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPRETRANSFAIL, segs);
3235
	}
3236
	return err;
3237 3238
}

3239
int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs)
3240 3241
{
	struct tcp_sock *tp = tcp_sk(sk);
3242
	int err = __tcp_retransmit_skb(sk, skb, segs);
L
Linus Torvalds 已提交
3243 3244 3245

	if (err == 0) {
#if FASTRETRANS_DEBUG > 0
3246
		if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) {
3247
			net_dbg_ratelimited("retrans_out leaked\n");
L
Linus Torvalds 已提交
3248 3249 3250 3251 3252
		}
#endif
		TCP_SKB_CB(skb)->sacked |= TCPCB_RETRANS;
		tp->retrans_out += tcp_skb_pcount(skb);
	}
Y
Yuchung Cheng 已提交
3253

3254 3255 3256 3257
	/* Save stamp of the first (attempted) retransmit. */
	if (!tp->retrans_stamp)
		tp->retrans_stamp = tcp_skb_timestamp(skb);

Y
Yuchung Cheng 已提交
3258 3259 3260
	if (tp->undo_retrans < 0)
		tp->undo_retrans = 0;
	tp->undo_retrans += tcp_skb_pcount(skb);
L
Linus Torvalds 已提交
3261 3262 3263 3264 3265 3266 3267 3268 3269 3270
	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)
{
3271
	const struct inet_connection_sock *icsk = inet_csk(sk);
3272
	struct sk_buff *skb, *rtx_head, *hole = NULL;
L
Linus Torvalds 已提交
3273
	struct tcp_sock *tp = tcp_sk(sk);
3274
	bool rearm_timer = false;
3275
	u32 max_segs;
3276
	int mib_idx;
3277

3278 3279 3280
	if (!tp->packets_out)
		return;

3281 3282
	rtx_head = tcp_rtx_queue_head(sk);
	skb = tp->retransmit_skb_hint ?: rtx_head;
3283
	max_segs = tcp_tso_segs(sk, tcp_current_mss(sk));
3284
	skb_rbtree_walk_from(skb) {
E
Eric Dumazet 已提交
3285
		__u8 sacked;
3286
		int segs;
L
Linus Torvalds 已提交
3287

3288 3289 3290
		if (tcp_pacing_check(sk))
			break;

3291
		/* we could do better than to assign each time */
3292
		if (!hole)
3293
			tp->retransmit_skb_hint = skb;
3294

3295 3296
		segs = tp->snd_cwnd - tcp_packets_in_flight(tp);
		if (segs <= 0)
3297
			break;
E
Eric Dumazet 已提交
3298
		sacked = TCP_SKB_CB(skb)->sacked;
3299 3300 3301 3302
		/* 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
3303

3304 3305
		if (tp->retrans_out >= tp->lost_out) {
			break;
3306
		} else if (!(sacked & TCPCB_LOST)) {
3307
			if (!hole && !(sacked & (TCPCB_SACKED_RETRANS|TCPCB_SACKED_ACKED)))
3308 3309
				hole = skb;
			continue;
L
Linus Torvalds 已提交
3310

3311 3312 3313 3314 3315 3316
		} 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
3317

3318
		if (sacked & (TCPCB_SACKED_ACKED|TCPCB_SACKED_RETRANS))
L
Linus Torvalds 已提交
3319 3320
			continue;

3321
		if (tcp_small_queue_check(sk, skb, 1))
3322
			break;
3323

3324
		if (tcp_retransmit_skb(sk, skb, segs))
3325
			break;
3326

3327
		NET_ADD_STATS(sock_net(sk), mib_idx, tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
3328

3329
		if (tcp_in_cwnd_reduction(sk))
3330 3331
			tp->prr_out += tcp_skb_pcount(skb);

3332
		if (skb == rtx_head &&
3333
		    icsk->icsk_pending != ICSK_TIME_REO_TIMEOUT)
3334 3335
			rearm_timer = true;

L
Linus Torvalds 已提交
3336
	}
3337 3338 3339 3340
	if (rearm_timer)
		tcp_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
				     inet_csk(sk)->icsk_rto,
				     TCP_RTO_MAX);
L
Linus Torvalds 已提交
3341 3342
}

3343 3344
/* We allow to exceed memory limits for FIN packets to expedite
 * connection tear down and (memory) recovery.
3345 3346
 * Otherwise tcp_send_fin() could be tempted to either delay FIN
 * or even be forced to close flow without any FIN.
3347 3348
 * In general, we want to allow one skb per socket to avoid hangs
 * with edge trigger epoll()
3349
 */
3350
void sk_forced_mem_schedule(struct sock *sk, int size)
3351
{
3352
	int amt;
3353 3354 3355 3356 3357

	if (size <= sk->sk_forward_alloc)
		return;
	amt = sk_mem_pages(size);
	sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
3358 3359
	sk_memory_allocated_add(sk, amt);

3360
	if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3361 3362
		mem_cgroup_charge_skmem(sk->sk_memcg, amt,
					gfp_memcg_charge() | __GFP_NOFAIL);
3363 3364
}

3365 3366
/* 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
3367 3368 3369
 */
void tcp_send_fin(struct sock *sk)
{
3370
	struct sk_buff *skb, *tskb, *tail = tcp_write_queue_tail(sk);
3371 3372
	struct tcp_sock *tp = tcp_sk(sk);

3373 3374 3375 3376
	/* 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
3377
	 */
3378
	tskb = tail;
3379 3380 3381 3382
	if (!tskb && tcp_under_memory_pressure(sk))
		tskb = skb_rb_last(&sk->tcp_rtx_queue);

	if (tskb) {
3383 3384
		TCP_SKB_CB(tskb)->tcp_flags |= TCPHDR_FIN;
		TCP_SKB_CB(tskb)->end_seq++;
L
Linus Torvalds 已提交
3385
		tp->write_seq++;
3386
		if (!tail) {
3387 3388 3389 3390 3391 3392
			/* 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.
			 */
3393
			WRITE_ONCE(tp->snd_nxt, tp->snd_nxt + 1);
3394 3395
			return;
		}
L
Linus Torvalds 已提交
3396
	} else {
3397
		skb = alloc_skb_fclone(MAX_TCP_HEADER, sk->sk_allocation);
3398
		if (unlikely(!skb))
3399
			return;
3400

3401
		INIT_LIST_HEAD(&skb->tcp_tsorted_anchor);
3402
		skb_reserve(skb, MAX_TCP_HEADER);
3403
		sk_forced_mem_schedule(sk, skb->truesize);
L
Linus Torvalds 已提交
3404
		/* FIN eats a sequence byte, write_seq advanced by tcp_queue_skb(). */
3405
		tcp_init_nondata_skb(skb, tp->write_seq,
C
Changli Gao 已提交
3406
				     TCPHDR_ACK | TCPHDR_FIN);
L
Linus Torvalds 已提交
3407 3408
		tcp_queue_skb(sk, skb);
	}
3409
	__tcp_push_pending_frames(sk, tcp_current_mss(sk), TCP_NAGLE_OFF);
L
Linus Torvalds 已提交
3410 3411 3412 3413 3414
}

/* 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
3415
 * by RFC 2525, section 2.17.  -DaveM
L
Linus Torvalds 已提交
3416
 */
A
Al Viro 已提交
3417
void tcp_send_active_reset(struct sock *sk, gfp_t priority)
L
Linus Torvalds 已提交
3418 3419 3420
{
	struct sk_buff *skb;

3421 3422
	TCP_INC_STATS(sock_net(sk), TCP_MIB_OUTRSTS);

L
Linus Torvalds 已提交
3423 3424 3425
	/* 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
3426
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTFAILED);
L
Linus Torvalds 已提交
3427 3428 3429 3430 3431
		return;
	}

	/* Reserve space for headers and prepare control bits. */
	skb_reserve(skb, MAX_TCP_HEADER);
3432
	tcp_init_nondata_skb(skb, tcp_acceptable_seq(sk),
C
Changli Gao 已提交
3433
			     TCPHDR_ACK | TCPHDR_RST);
3434
	tcp_mstamp_refresh(tcp_sk(sk));
L
Linus Torvalds 已提交
3435
	/* Send it off. */
3436
	if (tcp_transmit_skb(sk, skb, 0, priority))
P
Pavel Emelyanov 已提交
3437
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTFAILED);
3438 3439 3440 3441 3442

	/* 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
3443 3444
}

3445 3446
/* 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
3447 3448 3449 3450 3451 3452
 * 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)
{
3453
	struct sk_buff *skb;
L
Linus Torvalds 已提交
3454

3455
	skb = tcp_rtx_queue_head(sk);
3456
	if (!skb || !(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
3457
		pr_err("%s: wrong queue state\n", __func__);
L
Linus Torvalds 已提交
3458 3459
		return -EFAULT;
	}
E
Eric Dumazet 已提交
3460
	if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)) {
L
Linus Torvalds 已提交
3461
		if (skb_cloned(skb)) {
3462 3463 3464 3465 3466
			struct sk_buff *nskb;

			tcp_skb_tsorted_save(skb) {
				nskb = skb_copy(skb, GFP_ATOMIC);
			} tcp_skb_tsorted_restore(skb);
3467
			if (!nskb)
L
Linus Torvalds 已提交
3468
				return -ENOMEM;
3469
			INIT_LIST_HEAD(&nskb->tcp_tsorted_anchor);
3470
			tcp_highest_sack_replace(sk, skb, nskb);
3471
			tcp_rtx_queue_unlink_and_free(skb, sk);
3472
			__skb_header_release(nskb);
3473
			tcp_rbtree_insert(&sk->tcp_rtx_queue, nskb);
3474
			sk_wmem_queued_add(sk, nskb->truesize);
3475
			sk_mem_charge(sk, nskb->truesize);
L
Linus Torvalds 已提交
3476 3477 3478
			skb = nskb;
		}

E
Eric Dumazet 已提交
3479
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ACK;
3480
		tcp_ecn_send_synack(sk, skb);
L
Linus Torvalds 已提交
3481
	}
3482
	return tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
L
Linus Torvalds 已提交
3483 3484
}

3485
/**
3486 3487 3488 3489 3490 3491 3492 3493
 * tcp_make_synack - Allocate one skb and build a SYNACK packet.
 * @sk: listener socket
 * @dst: dst entry attached to the SYNACK. It is consumed and caller
 *       should not use it again.
 * @req: request_sock pointer
 * @foc: cookie for tcp fast open
 * @synack_type: Type of synack to prepare
 * @syn_skb: SYN packet just received.  It could be NULL for rtx case.
3494
 */
3495
struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
3496
				struct request_sock *req,
3497
				struct tcp_fastopen_cookie *foc,
3498 3499
				enum tcp_synack_type synack_type,
				struct sk_buff *syn_skb)
L
Linus Torvalds 已提交
3500
{
3501
	struct inet_request_sock *ireq = inet_rsk(req);
3502
	const struct tcp_sock *tp = tcp_sk(sk);
E
Eric Dumazet 已提交
3503
	struct tcp_md5sig_key *md5 = NULL;
3504 3505
	struct tcp_out_options opts;
	struct sk_buff *skb;
3506
	int tcp_header_size;
3507
	struct tcphdr *th;
3508
	int mss;
3509
	u64 now;
L
Linus Torvalds 已提交
3510

3511
	skb = alloc_skb(MAX_TCP_HEADER, GFP_ATOMIC);
3512 3513
	if (unlikely(!skb)) {
		dst_release(dst);
L
Linus Torvalds 已提交
3514
		return NULL;
3515
	}
L
Linus Torvalds 已提交
3516 3517 3518
	/* Reserve space for headers. */
	skb_reserve(skb, MAX_TCP_HEADER);

3519 3520
	switch (synack_type) {
	case TCP_SYNACK_NORMAL:
3521
		skb_set_owner_w(skb, req_to_sk(req));
3522 3523 3524 3525 3526 3527 3528
		break;
	case TCP_SYNACK_COOKIE:
		/* Under synflood, we do not attach skb to a socket,
		 * to avoid false sharing.
		 */
		break;
	case TCP_SYNACK_FASTOPEN:
3529 3530 3531 3532 3533
		/* 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);
3534
		break;
3535
	}
3536
	skb_dst_set(skb, dst);
L
Linus Torvalds 已提交
3537

E
Eric Dumazet 已提交
3538
	mss = tcp_mss_clamp(tp, dst_metric_advmss(dst));
3539

A
Adam Langley 已提交
3540
	memset(&opts, 0, sizeof(opts));
3541
	now = tcp_clock_ns();
3542
#ifdef CONFIG_SYN_COOKIES
3543
	if (unlikely(synack_type == TCP_SYNACK_COOKIE && ireq->tstamp_ok))
3544
		skb->skb_mstamp_ns = cookie_init_timestamp(req, now);
3545 3546
	else
#endif
3547
	{
3548
		skb->skb_mstamp_ns = now;
3549 3550 3551
		if (!tcp_rsk(req)->snt_synack) /* Timestamp first SYNACK */
			tcp_rsk(req)->snt_synack = tcp_skb_timestamp_us(skb);
	}
E
Eric Dumazet 已提交
3552 3553 3554

#ifdef CONFIG_TCP_MD5SIG
	rcu_read_lock();
3555
	md5 = tcp_rsk(req)->af_specific->req_md5_lookup(sk, req_to_sk(req));
E
Eric Dumazet 已提交
3556
#endif
3557
	skb_set_hash(skb, tcp_rsk(req)->txhash, PKT_HASH_TYPE_L4);
3558 3559
	/* bpf program will be interested in the tcp_flags */
	TCP_SKB_CB(skb)->tcp_flags = TCPHDR_SYN | TCPHDR_ACK;
3560
	tcp_header_size = tcp_synack_options(sk, req, mss, skb, &opts, md5,
3561 3562
					     foc, synack_type,
					     syn_skb) + sizeof(*th);
3563

3564 3565
	skb_push(skb, tcp_header_size);
	skb_reset_transport_header(skb);
L
Linus Torvalds 已提交
3566

3567
	th = (struct tcphdr *)skb->data;
L
Linus Torvalds 已提交
3568 3569 3570
	memset(th, 0, sizeof(struct tcphdr));
	th->syn = 1;
	th->ack = 1;
3571
	tcp_ecn_make_synack(req, th);
E
Eric Dumazet 已提交
3572
	th->source = htons(ireq->ir_num);
3573
	th->dest = ireq->ir_rmt_port;
3574
	skb->mark = ireq->ir_mark;
3575 3576
	skb->ip_summed = CHECKSUM_PARTIAL;
	th->seq = htonl(tcp_rsk(req)->snt_isn);
3577 3578
	/* 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
3579 3580

	/* RFC1323: The window in SYN & SYN/ACK segments is never scaled. */
3581
	th->window = htons(min(req->rsk_rcv_wnd, 65535U));
3582
	tcp_options_write((__be32 *)(th + 1), NULL, &opts);
L
Linus Torvalds 已提交
3583
	th->doff = (tcp_header_size >> 2);
E
Eric Dumazet 已提交
3584
	__TCP_INC_STATS(sock_net(sk), TCP_MIB_OUTSEGS);
3585 3586 3587

#ifdef CONFIG_TCP_MD5SIG
	/* Okay, we have all we need - do the md5 hash if needed */
E
Eric Dumazet 已提交
3588
	if (md5)
3589
		tcp_rsk(req)->af_specific->calc_md5_hash(opts.hash_location,
3590
					       md5, req_to_sk(req), skb);
E
Eric Dumazet 已提交
3591
	rcu_read_unlock();
3592 3593
#endif

3594 3595 3596
	bpf_skops_write_hdr_opt((struct sock *)sk, skb, req, syn_skb,
				synack_type, &opts);

3597 3598 3599
	skb->skb_mstamp_ns = now;
	tcp_add_tx_delay(skb, tp);

L
Linus Torvalds 已提交
3600 3601
	return skb;
}
E
Eric Dumazet 已提交
3602
EXPORT_SYMBOL(tcp_make_synack);
L
Linus Torvalds 已提交
3603

3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614
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);
3615 3616
	if (likely(ca && bpf_try_module_get(ca, ca->owner))) {
		bpf_module_put(icsk->icsk_ca_ops, icsk->icsk_ca_ops->owner);
3617 3618 3619 3620 3621 3622
		icsk->icsk_ca_dst_locked = tcp_ca_dst_locked(dst);
		icsk->icsk_ca_ops = ca;
	}
	rcu_read_unlock();
}

3623
/* Do all connect socket setups that can be done AF independent. */
3624
static void tcp_connect_init(struct sock *sk)
L
Linus Torvalds 已提交
3625
{
3626
	const struct dst_entry *dst = __sk_dst_get(sk);
L
Linus Torvalds 已提交
3627 3628
	struct tcp_sock *tp = tcp_sk(sk);
	__u8 rcv_wscale;
3629
	u32 rcv_wnd;
L
Linus Torvalds 已提交
3630 3631 3632 3633

	/* 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.
	 */
3634 3635 3636
	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
3637

3638
#ifdef CONFIG_TCP_MD5SIG
3639
	if (tp->af_specific->md5_lookup(sk, sk))
3640 3641 3642
		tp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
#endif

L
Linus Torvalds 已提交
3643 3644 3645 3646
	/* 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
3647
	tcp_mtup_init(sk);
L
Linus Torvalds 已提交
3648 3649
	tcp_sync_mss(sk, dst_mtu(dst));

3650 3651
	tcp_ca_dst_init(sk, dst);

L
Linus Torvalds 已提交
3652 3653
	if (!tp->window_clamp)
		tp->window_clamp = dst_metric(dst, RTAX_WINDOW);
E
Eric Dumazet 已提交
3654
	tp->advmss = tcp_mss_clamp(tp, dst_metric_advmss(dst));
3655

L
Linus Torvalds 已提交
3656 3657
	tcp_initialize_rcv_mss(sk);

3658 3659 3660 3661 3662
	/* 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);

3663 3664 3665 3666
	rcv_wnd = tcp_rwnd_init_bpf(sk);
	if (rcv_wnd == 0)
		rcv_wnd = dst_metric(dst, RTAX_INITRWND);

3667
	tcp_select_initial_window(sk, tcp_full_space(sk),
L
Linus Torvalds 已提交
3668 3669 3670
				  tp->advmss - (tp->rx_opt.ts_recent_stamp ? tp->tcp_header_len - sizeof(struct tcphdr) : 0),
				  &tp->rcv_wnd,
				  &tp->window_clamp,
3671
				  sock_net(sk)->ipv4.sysctl_tcp_window_scaling,
3672
				  &rcv_wscale,
3673
				  rcv_wnd);
L
Linus Torvalds 已提交
3674 3675 3676 3677 3678 3679 3680

	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;
3681
	tcp_init_wl(tp, 0);
3682
	tcp_write_queue_purge(sk);
L
Linus Torvalds 已提交
3683 3684
	tp->snd_una = tp->write_seq;
	tp->snd_sml = tp->write_seq;
I
Ilpo Järvinen 已提交
3685
	tp->snd_up = tp->write_seq;
3686
	WRITE_ONCE(tp->snd_nxt, tp->write_seq);
P
Pavel Emelyanov 已提交
3687 3688 3689

	if (likely(!tp->repair))
		tp->rcv_nxt = 0;
3690
	else
3691
		tp->rcv_tstamp = tcp_jiffies32;
P
Pavel Emelyanov 已提交
3692
	tp->rcv_wup = tp->rcv_nxt;
3693
	WRITE_ONCE(tp->copied_seq, tp->rcv_nxt);
L
Linus Torvalds 已提交
3694

3695
	inet_csk(sk)->icsk_rto = tcp_timeout_init(sk);
3696
	inet_csk(sk)->icsk_retransmits = 0;
L
Linus Torvalds 已提交
3697 3698 3699
	tcp_clear_retrans(tp);
}

3700 3701 3702 3703 3704 3705
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;
3706
	__skb_header_release(skb);
3707
	sk_wmem_queued_add(sk, skb->truesize);
3708
	sk_mem_charge(sk, skb->truesize);
3709
	WRITE_ONCE(tp->write_seq, tcb->end_seq);
3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723
	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;
3724
	int space, err = 0;
3725
	struct sk_buff *syn_data;
3726

3727
	tp->rx_opt.mss_clamp = tp->advmss;  /* If MSS is not cached */
3728
	if (!tcp_fastopen_cookie_check(sk, &tp->rx_opt.mss_clamp, &fo->cookie))
3729 3730 3731 3732 3733 3734
		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
3735 3736
	tp->rx_opt.mss_clamp = tcp_mss_clamp(tp, tp->rx_opt.mss_clamp);

3737
	space = __tcp_mtu_to_mss(sk, inet_csk(sk)->icsk_pmtu_cookie) -
3738 3739
		MAX_TCP_OPTION_SPACE;

3740 3741 3742 3743 3744
	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));

E
Eric Dumazet 已提交
3745
	syn_data = tcp_stream_alloc_skb(sk, space, sk->sk_allocation, false);
3746
	if (!syn_data)
3747
		goto fallback;
3748
	memcpy(syn_data->cb, syn->cb, sizeof(syn->cb));
3749 3750 3751 3752
	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
3753
			tcp_skb_tsorted_anchor_cleanup(syn_data);
3754 3755 3756 3757 3758 3759 3760
			kfree_skb(syn_data);
			goto fallback;
		}
		if (copied != space) {
			skb_trim(syn_data, copied);
			space = copied;
		}
3761
		skb_zcopy_set(syn_data, fo->uarg, NULL);
3762
	}
3763 3764 3765 3766
	/* No more data pending in inet_wait_for_connect() */
	if (space == fo->size)
		fo->data = NULL;
	fo->copied = space;
3767

3768
	tcp_connect_queue_skb(sk, syn_data);
3769 3770
	if (syn_data->len)
		tcp_chrono_start(sk, TCP_CHRONO_BUSY);
3771

3772
	err = tcp_transmit_skb(sk, syn_data, 1, sk->sk_allocation);
3773

3774
	syn->skb_mstamp_ns = syn_data->skb_mstamp_ns;
E
Eric Dumazet 已提交
3775

3776 3777 3778 3779 3780 3781 3782 3783
	/* 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) {
3784
		tp->syn_data = (fo->copied > 0);
3785
		tcp_rbtree_insert(&sk->tcp_rtx_queue, syn_data);
3786
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPORIGDATASENT);
3787 3788 3789
		goto done;
	}

3790 3791
	/* data was not sent, put it in write_queue */
	__skb_queue_tail(&sk->sk_write_queue, syn_data);
3792 3793
	tp->packets_out -= tcp_skb_pcount(syn_data);

3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805
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;
}

3806
/* Build a SYN and send it off. */
L
Linus Torvalds 已提交
3807 3808 3809 3810
int tcp_connect(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *buff;
3811
	int err;
L
Linus Torvalds 已提交
3812

3813
	tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_CONNECT_CB, 0, NULL);
3814 3815 3816 3817

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

L
Linus Torvalds 已提交
3818 3819
	tcp_connect_init(sk);

A
Andrey Vagin 已提交
3820 3821 3822 3823 3824
	if (unlikely(tp->repair)) {
		tcp_finish_connect(sk, NULL);
		return 0;
	}

E
Eric Dumazet 已提交
3825
	buff = tcp_stream_alloc_skb(sk, 0, sk->sk_allocation, true);
3826
	if (unlikely(!buff))
L
Linus Torvalds 已提交
3827 3828
		return -ENOBUFS;

C
Changli Gao 已提交
3829
	tcp_init_nondata_skb(buff, tp->write_seq++, TCPHDR_SYN);
3830 3831
	tcp_mstamp_refresh(tp);
	tp->retrans_stamp = tcp_time_stamp(tp);
3832
	tcp_connect_queue_skb(sk, buff);
3833
	tcp_ecn_send_syn(sk, buff);
3834
	tcp_rbtree_insert(&sk->tcp_rtx_queue, buff);
L
Linus Torvalds 已提交
3835

3836 3837 3838
	/* 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);
3839 3840
	if (err == -ECONNREFUSED)
		return err;
3841 3842 3843 3844

	/* We change tp->snd_nxt after the tcp_transmit_skb() call
	 * in order to make this packet get counted in tcpOutSegs.
	 */
3845
	WRITE_ONCE(tp->snd_nxt, tp->write_seq);
3846
	tp->pushed_seq = tp->write_seq;
3847 3848
	buff = tcp_send_head(sk);
	if (unlikely(buff)) {
3849
		WRITE_ONCE(tp->snd_nxt, TCP_SKB_CB(buff)->seq);
3850 3851
		tp->pushed_seq	= TCP_SKB_CB(buff)->seq;
	}
P
Pavel Emelyanov 已提交
3852
	TCP_INC_STATS(sock_net(sk), TCP_MIB_ACTIVEOPENS);
L
Linus Torvalds 已提交
3853 3854

	/* Timer for repeating the SYN until an answer. */
3855 3856
	inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
				  inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
L
Linus Torvalds 已提交
3857 3858
	return 0;
}
E
Eric Dumazet 已提交
3859
EXPORT_SYMBOL(tcp_connect);
L
Linus Torvalds 已提交
3860 3861 3862 3863 3864 3865 3866

/* 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)
{
3867 3868
	struct inet_connection_sock *icsk = inet_csk(sk);
	int ato = icsk->icsk_ack.ato;
L
Linus Torvalds 已提交
3869 3870 3871
	unsigned long timeout;

	if (ato > TCP_DELACK_MIN) {
3872
		const struct tcp_sock *tp = tcp_sk(sk);
3873
		int max_ato = HZ / 2;
L
Linus Torvalds 已提交
3874

W
Wei Wang 已提交
3875
		if (inet_csk_in_pingpong_mode(sk) ||
3876
		    (icsk->icsk_ack.pending & ICSK_ACK_PUSHED))
L
Linus Torvalds 已提交
3877 3878 3879 3880 3881
			max_ato = TCP_DELACK_MAX;

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

		/* If some rtt estimate is known, use it to bound delayed ack.
3882
		 * Do not use inet_csk(sk)->icsk_rto here, use results of rtt measurements
L
Linus Torvalds 已提交
3883 3884
		 * directly.
		 */
3885 3886 3887
		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
3888 3889 3890 3891 3892 3893 3894 3895

			if (rtt < max_ato)
				max_ato = rtt;
		}

		ato = min(ato, max_ato);
	}

3896 3897
	ato = min_t(u32, ato, inet_csk(sk)->icsk_delack_max);

L
Linus Torvalds 已提交
3898 3899 3900 3901
	/* Stay within the limit we were given */
	timeout = jiffies + ato;

	/* Use new timeout only if there wasn't a older one earlier. */
3902
	if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
E
Eric Dumazet 已提交
3903 3904
		/* If delack timer is about to expire, send ACK now. */
		if (time_before_eq(icsk->icsk_ack.timeout, jiffies + (ato >> 2))) {
L
Linus Torvalds 已提交
3905 3906 3907 3908
			tcp_send_ack(sk);
			return;
		}

3909 3910
		if (!time_before(timeout, icsk->icsk_ack.timeout))
			timeout = icsk->icsk_ack.timeout;
L
Linus Torvalds 已提交
3911
	}
3912 3913 3914
	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
3915 3916 3917
}

/* This routine sends an ack and also updates the window. */
3918
void __tcp_send_ack(struct sock *sk, u32 rcv_nxt)
L
Linus Torvalds 已提交
3919
{
3920
	struct sk_buff *buff;
L
Linus Torvalds 已提交
3921

3922 3923 3924
	/* If we have been reset, we may not send again. */
	if (sk->sk_state == TCP_CLOSE)
		return;
L
Linus Torvalds 已提交
3925

3926 3927 3928 3929
	/* We are not putting this on the write queue, so
	 * tcp_transmit_skb() will set the ownership to this
	 * sock.
	 */
3930 3931 3932
	buff = alloc_skb(MAX_TCP_HEADER,
			 sk_gfp_mask(sk, GFP_ATOMIC | __GFP_NOWARN));
	if (unlikely(!buff)) {
3933 3934 3935 3936 3937 3938
		struct inet_connection_sock *icsk = inet_csk(sk);
		unsigned long delay;

		delay = TCP_DELACK_MAX << icsk->icsk_ack.retry;
		if (delay < TCP_RTO_MAX)
			icsk->icsk_ack.retry++;
3939
		inet_csk_schedule_ack(sk);
3940 3941
		icsk->icsk_ack.ato = TCP_ATO_MIN;
		inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, delay, TCP_RTO_MAX);
3942
		return;
L
Linus Torvalds 已提交
3943
	}
3944 3945 3946

	/* Reserve space for headers and prepare control bits. */
	skb_reserve(buff, MAX_TCP_HEADER);
C
Changli Gao 已提交
3947
	tcp_init_nondata_skb(buff, tcp_acceptable_seq(sk), TCPHDR_ACK);
3948

E
Eric Dumazet 已提交
3949 3950 3951 3952 3953 3954
	/* 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);

3955
	/* Send it off, this clears delayed acks for us. */
3956 3957
	__tcp_transmit_skb(sk, buff, 0, (__force gfp_t)0, rcv_nxt);
}
3958
EXPORT_SYMBOL_GPL(__tcp_send_ack);
3959 3960 3961 3962

void tcp_send_ack(struct sock *sk)
{
	__tcp_send_ack(sk, tcp_sk(sk)->rcv_nxt);
L
Linus Torvalds 已提交
3963 3964 3965 3966 3967 3968 3969 3970 3971 3972 3973 3974 3975
}

/* 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.
 */
3976
static int tcp_xmit_probe_skb(struct sock *sk, int urgent, int mib)
L
Linus Torvalds 已提交
3977 3978 3979 3980 3981
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb;

	/* We don't queue it, tcp_transmit_skb() sets ownership. */
3982 3983
	skb = alloc_skb(MAX_TCP_HEADER,
			sk_gfp_mask(sk, GFP_ATOMIC | __GFP_NOWARN));
3984
	if (!skb)
L
Linus Torvalds 已提交
3985 3986 3987 3988 3989 3990 3991 3992
		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
3993
	tcp_init_nondata_skb(skb, tp->snd_una - !urgent, TCPHDR_ACK);
3994
	NET_INC_STATS(sock_net(sk), mib);
3995
	return tcp_transmit_skb(sk, skb, 0, (__force gfp_t)0);
L
Linus Torvalds 已提交
3996 3997
}

3998
/* Called from setsockopt( ... TCP_REPAIR ) */
P
Pavel Emelyanov 已提交
3999 4000 4001 4002
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;
4003
		tcp_mstamp_refresh(tcp_sk(sk));
4004
		tcp_xmit_probe_skb(sk, 0, LINUX_MIB_TCPWINPROBE);
P
Pavel Emelyanov 已提交
4005 4006 4007
	}
}

4008
/* Initiate keepalive or window probe from timer. */
4009
int tcp_write_wakeup(struct sock *sk, int mib)
L
Linus Torvalds 已提交
4010
{
4011 4012
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb;
L
Linus Torvalds 已提交
4013

4014 4015 4016
	if (sk->sk_state == TCP_CLOSE)
		return -1;

4017 4018
	skb = tcp_send_head(sk);
	if (skb && before(TCP_SKB_CB(skb)->seq, tcp_wnd_end(tp))) {
4019
		int err;
I
Ilpo Järvinen 已提交
4020
		unsigned int mss = tcp_current_mss(sk);
4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032
		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
4033
			TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
4034 4035
			if (tcp_fragment(sk, TCP_FRAG_IN_WRITE_QUEUE,
					 skb, seg_size, mss, GFP_ATOMIC))
4036 4037
				return -1;
		} else if (!tcp_skb_pcount(skb))
4038
			tcp_set_skb_tso_segs(skb, mss);
4039

E
Eric Dumazet 已提交
4040
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
4041 4042 4043 4044 4045
		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
4046
		if (between(tp->snd_up, tp->snd_una + 1, tp->snd_una + 0xFFFF))
4047 4048
			tcp_xmit_probe_skb(sk, 1, mib);
		return tcp_xmit_probe_skb(sk, 0, mib);
L
Linus Torvalds 已提交
4049 4050 4051 4052 4053 4054 4055 4056
	}
}

/* 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)
{
4057
	struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
4058
	struct tcp_sock *tp = tcp_sk(sk);
4059
	struct net *net = sock_net(sk);
4060
	unsigned long timeout;
L
Linus Torvalds 已提交
4061 4062
	int err;

4063
	err = tcp_write_wakeup(sk, LINUX_MIB_TCPWINPROBE);
L
Linus Torvalds 已提交
4064

4065
	if (tp->packets_out || tcp_write_queue_empty(sk)) {
L
Linus Torvalds 已提交
4066
		/* Cancel probe timer, if it is not required. */
4067
		icsk->icsk_probes_out = 0;
4068
		icsk->icsk_backoff = 0;
4069
		icsk->icsk_probes_tstamp = 0;
L
Linus Torvalds 已提交
4070 4071 4072
		return;
	}

4073
	icsk->icsk_probes_out++;
L
Linus Torvalds 已提交
4074
	if (err <= 0) {
4075
		if (icsk->icsk_backoff < net->ipv4.sysctl_tcp_retries2)
4076
			icsk->icsk_backoff++;
4077
		timeout = tcp_probe0_when(sk, TCP_RTO_MAX);
L
Linus Torvalds 已提交
4078 4079
	} else {
		/* If packet was not sent due to local congestion,
4080
		 * Let senders fight for local resources conservatively.
L
Linus Torvalds 已提交
4081
		 */
4082 4083
		timeout = TCP_RESOURCE_PROBE_INTERVAL;
	}
4084 4085

	timeout = tcp_clamp_probe0_to_user_timeout(sk, timeout);
4086
	tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, timeout, TCP_RTO_MAX);
L
Linus Torvalds 已提交
4087
}
4088

4089
int tcp_rtx_synack(const struct sock *sk, struct request_sock *req)
4090 4091 4092 4093 4094
{
	const struct tcp_request_sock_ops *af_ops = tcp_rsk(req)->af_specific;
	struct flowi fl;
	int res;

4095
	tcp_rsk(req)->txhash = net_tx_rndhash();
4096 4097
	res = af_ops->send_synack(sk, NULL, &fl, req, NULL, TCP_SYNACK_NORMAL,
				  NULL);
4098
	if (!res) {
E
Eric Dumazet 已提交
4099
		__TCP_INC_STATS(sock_net(sk), TCP_MIB_RETRANSSEGS);
4100
		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPSYNRETRANS);
4101 4102
		if (unlikely(tcp_passive_fastopen(sk)))
			tcp_sk(sk)->total_retrans++;
4103
		trace_tcp_retransmit_synack(sk, req);
4104 4105 4106 4107
	}
	return res;
}
EXPORT_SYMBOL(tcp_rtx_synack);