tcp_output.c 108.6 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7
/*
 * INET		An implementation of the TCP/IP protocol suite for the LINUX
 *		operating system.  INET is implemented using the  BSD Socket
 *		interface as the means of communication with the user level.
 *
 *		Implementation of the Transmission Control Protocol(TCP).
 *
8
 * Authors:	Ross Biro
L
Linus Torvalds 已提交
9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36
 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
 *		Mark Evans, <evansmp@uhura.aston.ac.uk>
 *		Corey Minyard <wf-rch!minyard@relay.EU.net>
 *		Florian La Roche, <flla@stud.uni-sb.de>
 *		Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
 *		Linus Torvalds, <torvalds@cs.helsinki.fi>
 *		Alan Cox, <gw4pts@gw4pts.ampr.org>
 *		Matthew Dillon, <dillon@apollo.west.oic.com>
 *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
 *		Jorge Cwik, <jorge@laser.satlink.net>
 */

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

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

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

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

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

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

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

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

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

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

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

71 72
/* SND.NXT, if window was not shrunk or the amount of shrunk was less than one
 * window scaling factor due to loss of precision.
L
Linus Torvalds 已提交
73 74 75 76 77
 * If window has been shrunk, what should we make? It is not clear at all.
 * Using SND.UNA we will fail to open window, SND.NXT is out of window. :-(
 * Anything in between SND.UNA...SND.UNA+SND.WND also can be already
 * invalid. OK, let's make this for now:
 */
78
static inline __u32 tcp_acceptable_seq(const struct sock *sk)
L
Linus Torvalds 已提交
79
{
80
	const struct tcp_sock *tp = tcp_sk(sk);
81

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

/* Calculate mss to advertise in SYN segment.
 * RFC1122, RFC1063, draft-ietf-tcpimpl-pmtud-01 state that:
 *
 * 1. It is independent of path mtu.
 * 2. Ideally, it is maximal possible segment size i.e. 65535-40.
 * 3. For IPv4 it is reasonable to calculate it from maximal MTU of
 *    attached devices, because some buggy hosts are confused by
 *    large MSS.
 * 4. We do not make 3, we advertise MSS, calculated from first
 *    hop device mtu, but allow to raise it to ip_rt_min_advmss.
 *    This may be overridden via information stored in routing table.
 * 5. Value 65535 for MSS is valid in IPv6 and means "as large as possible,
 *    probably even Jumbo".
 */
static __u16 tcp_advertise_mss(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
107
	const struct dst_entry *dst = __sk_dst_get(sk);
L
Linus Torvalds 已提交
108 109
	int mss = tp->advmss;

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

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

	return (__u16)mss;
}

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

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

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

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

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

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

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

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

162
/* Account for an ACK we sent. */
S
Stephen Hemminger 已提交
163
static inline void tcp_event_ack_sent(struct sock *sk, unsigned int pkts)
L
Linus Torvalds 已提交
164
{
165 166
	tcp_dec_quickack_mode(sk, pkts);
	inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
L
Linus Torvalds 已提交
167 168
}

169 170 171 172

u32 tcp_default_init_rwnd(u32 mss)
{
	/* Initial receive window should be twice of TCP_INIT_CWND to
W
Weiping Pan 已提交
173
	 * enable proper sending of new unsent data during fast recovery
174 175 176 177 178 179 180 181 182 183
	 * (RFC 3517, Section 4, NextSeg() rule (2)). Further place a
	 * limit when mss is larger than 1460.
	 */
	u32 init_rwnd = TCP_INIT_CWND * 2;

	if (mss > 1460)
		init_rwnd = max((1460 * init_rwnd) / mss, 2U);
	return init_rwnd;
}

L
Linus Torvalds 已提交
184 185 186 187 188 189 190
/* 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.
 */
191
void tcp_select_initial_window(const struct sock *sk, int __space, __u32 mss,
L
Linus Torvalds 已提交
192
			       __u32 *rcv_wnd, __u32 *window_clamp,
193 194
			       int wscale_ok, __u8 *rcv_wscale,
			       __u32 init_rcv_wnd)
L
Linus Torvalds 已提交
195 196 197 198 199
{
	unsigned int space = (__space < 0 ? 0 : __space);

	/* If no clamp set the clamp to the max possible scaled window */
	if (*window_clamp == 0)
200
		(*window_clamp) = (U16_MAX << TCP_MAX_WSCALE);
L
Linus Torvalds 已提交
201 202 203 204
	space = min(*window_clamp, space);

	/* Quantize space offering to a multiple of mss if possible. */
	if (space > mss)
205
		space = rounddown(space, mss);
L
Linus Torvalds 已提交
206 207

	/* NOTE: offering an initial window larger than 32767
208 209 210 211 212 213
	 * 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
214
	 */
215
	if (sock_net(sk)->ipv4.sysctl_tcp_workaround_signed_windows)
216 217 218 219
		(*rcv_wnd) = min(space, MAX_TCP_WINDOW);
	else
		(*rcv_wnd) = space;

L
Linus Torvalds 已提交
220 221
	(*rcv_wscale) = 0;
	if (wscale_ok) {
222
		/* Set window scaling on max possible window */
223 224
		space = max_t(u32, space, sysctl_tcp_rmem[2]);
		space = max_t(u32, space, sysctl_rmem_max);
225
		space = min_t(u32, space, *window_clamp);
226
		while (space > U16_MAX && (*rcv_wscale) < TCP_MAX_WSCALE) {
L
Linus Torvalds 已提交
227 228 229 230 231
			space >>= 1;
			(*rcv_wscale)++;
		}
	}

232
	if (mss > (1 << *rcv_wscale)) {
233 234 235
		if (!init_rcv_wnd) /* Use default unless specified otherwise */
			init_rcv_wnd = tcp_default_init_rwnd(mss);
		*rcv_wnd = min(*rcv_wnd, init_rcv_wnd * mss);
L
Linus Torvalds 已提交
236 237 238
	}

	/* Set the clamp no higher than max representable value */
239
	(*window_clamp) = min_t(__u32, U16_MAX << (*rcv_wscale), *window_clamp);
L
Linus Torvalds 已提交
240
}
E
Eric Dumazet 已提交
241
EXPORT_SYMBOL(tcp_select_initial_window);
L
Linus Torvalds 已提交
242 243 244 245 246 247

/* 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
248
static u16 tcp_select_window(struct sock *sk)
L
Linus Torvalds 已提交
249 250
{
	struct tcp_sock *tp = tcp_sk(sk);
251
	u32 old_win = tp->rcv_wnd;
L
Linus Torvalds 已提交
252 253 254 255
	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
256
	if (new_win < cur_win) {
L
Linus Torvalds 已提交
257 258 259 260 261 262 263
		/* 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.
		 */
264 265 266
		if (new_win == 0)
			NET_INC_STATS(sock_net(sk),
				      LINUX_MIB_TCPWANTZEROWINDOWADV);
267
		new_win = ALIGN(cur_win, 1 << tp->rx_opt.rcv_wscale);
L
Linus Torvalds 已提交
268 269 270 271 272 273 274
	}
	tp->rcv_wnd = new_win;
	tp->rcv_wup = tp->rcv_nxt;

	/* Make sure we do not exceed the maximum possible
	 * scaled window.
	 */
275 276
	if (!tp->rx_opt.rcv_wscale &&
	    sock_net(sk)->ipv4.sysctl_tcp_workaround_signed_windows)
L
Linus Torvalds 已提交
277 278 279 280 281 282 283
		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;

284
	/* If we advertise zero window, disable fast path. */
285
	if (new_win == 0) {
286
		tp->pred_flags = 0;
287 288 289 290 291 292
		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
293 294 295 296

	return new_win;
}

297
/* Packet ECN state for a SYN-ACK */
298
static void tcp_ecn_send_synack(struct sock *sk, struct sk_buff *skb)
299
{
300 301
	const struct tcp_sock *tp = tcp_sk(sk);

E
Eric Dumazet 已提交
302
	TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_CWR;
303
	if (!(tp->ecn_flags & TCP_ECN_OK))
E
Eric Dumazet 已提交
304
		TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_ECE;
305 306
	else if (tcp_ca_needs_ecn(sk) ||
		 tcp_bpf_ca_needs_ecn(sk))
307
		INET_ECN_xmit(sk);
308 309
}

310
/* Packet ECN state for a SYN.  */
311
static void tcp_ecn_send_syn(struct sock *sk, struct sk_buff *skb)
312 313
{
	struct tcp_sock *tp = tcp_sk(sk);
314
	bool bpf_needs_ecn = tcp_bpf_ca_needs_ecn(sk);
315
	bool use_ecn = sock_net(sk)->ipv4.sysctl_tcp_ecn == 1 ||
316
		tcp_ca_needs_ecn(sk) || bpf_needs_ecn;
317 318 319 320 321 322 323

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

		if (dst && dst_feature(dst, RTAX_FEATURE_ECN))
			use_ecn = true;
	}
324 325

	tp->ecn_flags = 0;
326 327

	if (use_ecn) {
E
Eric Dumazet 已提交
328
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ECE | TCPHDR_CWR;
329
		tp->ecn_flags = TCP_ECN_OK;
330
		if (tcp_ca_needs_ecn(sk) || bpf_needs_ecn)
331
			INET_ECN_xmit(sk);
332 333 334
	}
}

335 336 337 338 339 340 341 342 343
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);
}

344
static void
345
tcp_ecn_make_synack(const struct request_sock *req, struct tcphdr *th)
346
{
347
	if (inet_rsk(req)->ecn_ok)
348 349 350
		th->ece = 1;
}

351 352 353
/* Set up ECN state for a packet on a ESTABLISHED socket that is about to
 * be sent.
 */
354
static void tcp_ecn_send(struct sock *sk, struct sk_buff *skb,
355
			 struct tcphdr *th, int tcp_header_len)
356 357 358 359 360 361 362 363
{
	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);
364
			if (tp->ecn_flags & TCP_ECN_QUEUE_CWR) {
365
				tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR;
366
				th->cwr = 1;
367 368
				skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
			}
369
		} else if (!tcp_ca_needs_ecn(sk)) {
370 371 372 373
			/* ACK or retransmitted segment: clear ECT|CE */
			INET_ECN_dontxmit(sk);
		}
		if (tp->ecn_flags & TCP_ECN_DEMAND_CWR)
374
			th->ece = 1;
375 376 377
	}
}

378 379 380 381 382
/* 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)
{
383
	skb->ip_summed = CHECKSUM_PARTIAL;
384

E
Eric Dumazet 已提交
385
	TCP_SKB_CB(skb)->tcp_flags = flags;
386 387
	TCP_SKB_CB(skb)->sacked = 0;

388
	tcp_skb_pcount_set(skb, 1);
389 390

	TCP_SKB_CB(skb)->seq = seq;
C
Changli Gao 已提交
391
	if (flags & (TCPHDR_SYN | TCPHDR_FIN))
392 393 394 395
		seq++;
	TCP_SKB_CB(skb)->end_seq = seq;
}

E
Eric Dumazet 已提交
396
static inline bool tcp_urg_mode(const struct tcp_sock *tp)
I
Ilpo Järvinen 已提交
397 398 399 400
{
	return tp->snd_una != tp->snd_up;
}

A
Adam Langley 已提交
401 402 403
#define OPTION_SACK_ADVERTISE	(1 << 0)
#define OPTION_TS		(1 << 1)
#define OPTION_MD5		(1 << 2)
404
#define OPTION_WSCALE		(1 << 3)
Y
Yuchung Cheng 已提交
405
#define OPTION_FAST_OPEN_COOKIE	(1 << 8)
406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421
#define OPTION_SMC		(1 << 9)

static void smc_options_write(__be32 *ptr, u16 *options)
{
#if IS_ENABLED(CONFIG_SMC)
	if (static_branch_unlikely(&tcp_have_smc)) {
		if (unlikely(OPTION_SMC & *options)) {
			*ptr++ = htonl((TCPOPT_NOP  << 24) |
				       (TCPOPT_NOP  << 16) |
				       (TCPOPT_EXP <<  8) |
				       (TCPOLEN_EXP_SMC_BASE));
			*ptr++ = htonl(TCPOPT_SMC_MAGIC);
		}
	}
#endif
}
A
Adam Langley 已提交
422 423

struct tcp_out_options {
Y
Yuchung Cheng 已提交
424 425
	u16 options;		/* bit field of OPTION_* */
	u16 mss;		/* 0 to disable */
A
Adam Langley 已提交
426 427
	u8 ws;			/* window scale, 0 to disable */
	u8 num_sack_blocks;	/* number of SACK blocks to include */
428 429
	u8 hash_size;		/* bytes in hash_location */
	__u8 *hash_location;	/* temporary pointer, overloaded */
Y
Yuchung Cheng 已提交
430 431
	__u32 tsval, tsecr;	/* need to include OPTION_TS */
	struct tcp_fastopen_cookie *fastopen_cookie;	/* Fast open cookie */
A
Adam Langley 已提交
432 433
};

434 435 436
/* Write previously computed TCP options to the packet.
 *
 * Beware: Something in the Internet is very sensitive to the ordering of
437 438
 * 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
439
 * inter-operability perspective it seems that we're somewhat stuck with
440 441 442 443 444 445 446
 * 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
447
static void tcp_options_write(__be32 *ptr, struct tcp_sock *tp,
448 449
			      struct tcp_out_options *opts)
{
Y
Yuchung Cheng 已提交
450
	u16 options = opts->options;	/* mungable copy */
451 452

	if (unlikely(OPTION_MD5 & options)) {
C
Christoph Paasch 已提交
453 454
		*ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
			       (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
455 456
		/* overload cookie hash location */
		opts->hash_location = (__u8 *)ptr;
A
Adam Langley 已提交
457
		ptr += 4;
S
Stephen Hemminger 已提交
458
	}
A
Adam Langley 已提交
459

460 461 462 463 464 465
	if (unlikely(opts->mss)) {
		*ptr++ = htonl((TCPOPT_MSS << 24) |
			       (TCPOLEN_MSS << 16) |
			       opts->mss);
	}

466 467
	if (likely(OPTION_TS & options)) {
		if (unlikely(OPTION_SACK_ADVERTISE & options)) {
A
Adam Langley 已提交
468 469 470 471
			*ptr++ = htonl((TCPOPT_SACK_PERM << 24) |
				       (TCPOLEN_SACK_PERM << 16) |
				       (TCPOPT_TIMESTAMP << 8) |
				       TCPOLEN_TIMESTAMP);
472
			options &= ~OPTION_SACK_ADVERTISE;
A
Adam Langley 已提交
473 474 475 476 477 478 479 480 481 482
		} else {
			*ptr++ = htonl((TCPOPT_NOP << 24) |
				       (TCPOPT_NOP << 16) |
				       (TCPOPT_TIMESTAMP << 8) |
				       TCPOLEN_TIMESTAMP);
		}
		*ptr++ = htonl(opts->tsval);
		*ptr++ = htonl(opts->tsecr);
	}

483
	if (unlikely(OPTION_SACK_ADVERTISE & options)) {
A
Adam Langley 已提交
484 485 486 487 488 489
		*ptr++ = htonl((TCPOPT_NOP << 24) |
			       (TCPOPT_NOP << 16) |
			       (TCPOPT_SACK_PERM << 8) |
			       TCPOLEN_SACK_PERM);
	}

490
	if (unlikely(OPTION_WSCALE & options)) {
A
Adam Langley 已提交
491 492 493 494 495 496 497 498 499
		*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
500 501 502 503 504
		int this_sack;

		*ptr++ = htonl((TCPOPT_NOP  << 24) |
			       (TCPOPT_NOP  << 16) |
			       (TCPOPT_SACK <<  8) |
A
Adam Langley 已提交
505
			       (TCPOLEN_SACK_BASE + (opts->num_sack_blocks *
S
Stephen Hemminger 已提交
506
						     TCPOLEN_SACK_PERBLOCK)));
S
Stephen Hemminger 已提交
507

A
Adam Langley 已提交
508 509
		for (this_sack = 0; this_sack < opts->num_sack_blocks;
		     ++this_sack) {
S
Stephen Hemminger 已提交
510 511 512
			*ptr++ = htonl(sp[this_sack].start_seq);
			*ptr++ = htonl(sp[this_sack].end_seq);
		}
S
Stephen Hemminger 已提交
513

514
		tp->rx_opt.dsack = 0;
S
Stephen Hemminger 已提交
515
	}
Y
Yuchung Cheng 已提交
516 517 518

	if (unlikely(OPTION_FAST_OPEN_COOKIE & options)) {
		struct tcp_fastopen_cookie *foc = opts->fastopen_cookie;
519 520 521 522 523 524 525 526 527 528 529 530 531
		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
532

533 534 535 536
		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
537
		}
538
		ptr += (len + 3) >> 2;
Y
Yuchung Cheng 已提交
539
	}
540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574

	smc_options_write(ptr, &options);
}

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

static void smc_set_option_cond(const struct tcp_sock *tp,
				const struct inet_request_sock *ireq,
				struct tcp_out_options *opts,
				unsigned int *remaining)
{
#if IS_ENABLED(CONFIG_SMC)
	if (static_branch_unlikely(&tcp_have_smc)) {
		if (tp->syn_smc && ireq->smc_ok) {
			if (*remaining >= TCPOLEN_EXP_SMC_BASE_ALIGNED) {
				opts->options |= OPTION_SMC;
				*remaining -= TCPOLEN_EXP_SMC_BASE_ALIGNED;
			}
		}
	}
#endif
A
Adam Langley 已提交
575 576
}

577 578 579
/* Compute TCP options for SYN packets. This is not the final
 * network wire format yet.
 */
580
static unsigned int tcp_syn_options(struct sock *sk, struct sk_buff *skb,
A
Adam Langley 已提交
581
				struct tcp_out_options *opts,
582 583
				struct tcp_md5sig_key **md5)
{
A
Adam Langley 已提交
584
	struct tcp_sock *tp = tcp_sk(sk);
585
	unsigned int remaining = MAX_TCP_OPTION_SPACE;
586
	struct tcp_fastopen_request *fastopen = tp->fastopen_req;
A
Adam Langley 已提交
587

588
#ifdef CONFIG_TCP_MD5SIG
A
Adam Langley 已提交
589 590 591
	*md5 = tp->af_specific->md5_lookup(sk, sk);
	if (*md5) {
		opts->options |= OPTION_MD5;
592
		remaining -= TCPOLEN_MD5SIG_ALIGNED;
593
	}
A
Adam Langley 已提交
594 595
#else
	*md5 = NULL;
596
#endif
A
Adam Langley 已提交
597 598 599 600 601 602 603 604 605 606 607

	/* 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);
608
	remaining -= TCPOLEN_MSS_ALIGNED;
A
Adam Langley 已提交
609

610
	if (likely(sock_net(sk)->ipv4.sysctl_tcp_timestamps && !*md5)) {
A
Adam Langley 已提交
611
		opts->options |= OPTION_TS;
E
Eric Dumazet 已提交
612
		opts->tsval = tcp_skb_timestamp(skb) + tp->tsoffset;
A
Adam Langley 已提交
613
		opts->tsecr = tp->rx_opt.ts_recent;
614
		remaining -= TCPOLEN_TSTAMP_ALIGNED;
A
Adam Langley 已提交
615
	}
616
	if (likely(sock_net(sk)->ipv4.sysctl_tcp_window_scaling)) {
A
Adam Langley 已提交
617
		opts->ws = tp->rx_opt.rcv_wscale;
618
		opts->options |= OPTION_WSCALE;
619
		remaining -= TCPOLEN_WSCALE_ALIGNED;
A
Adam Langley 已提交
620
	}
E
Eric Dumazet 已提交
621
	if (likely(sock_net(sk)->ipv4.sysctl_tcp_sack)) {
A
Adam Langley 已提交
622
		opts->options |= OPTION_SACK_ADVERTISE;
623
		if (unlikely(!(OPTION_TS & opts->options)))
624
			remaining -= TCPOLEN_SACKPERM_ALIGNED;
A
Adam Langley 已提交
625 626
	}

627
	if (fastopen && fastopen->cookie.len >= 0) {
628 629 630 631
		u32 need = fastopen->cookie.len;

		need += fastopen->cookie.exp ? TCPOLEN_EXP_FASTOPEN_BASE :
					       TCPOLEN_FASTOPEN_BASE;
632 633 634 635 636 637
		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;
638
			tp->syn_fastopen_exp = fastopen->cookie.exp ? 1 : 0;
639 640
		}
	}
641

642 643
	smc_set_option(tp, opts, &remaining);

644
	return MAX_TCP_OPTION_SPACE - remaining;
S
Stephen Hemminger 已提交
645 646
}

647
/* Set up TCP options for SYN-ACKs. */
648 649
static unsigned int tcp_synack_options(const struct sock *sk,
				       struct request_sock *req,
650 651 652 653
				       unsigned int mss, struct sk_buff *skb,
				       struct tcp_out_options *opts,
				       const struct tcp_md5sig_key *md5,
				       struct tcp_fastopen_cookie *foc)
654
{
A
Adam Langley 已提交
655
	struct inet_request_sock *ireq = inet_rsk(req);
656
	unsigned int remaining = MAX_TCP_OPTION_SPACE;
A
Adam Langley 已提交
657

658
#ifdef CONFIG_TCP_MD5SIG
E
Eric Dumazet 已提交
659
	if (md5) {
A
Adam Langley 已提交
660
		opts->options |= OPTION_MD5;
661 662 663 664 665 666 667
		remaining -= TCPOLEN_MD5SIG_ALIGNED;

		/* We can't fit any SACK blocks in a packet with MD5 + TS
		 * options. There was discussion about disabling SACK
		 * rather than TS in order to fit in better with old,
		 * buggy kernels, but that was deemed to be unnecessary.
		 */
E
Eric Dumazet 已提交
668
		ireq->tstamp_ok &= !ireq->sack_ok;
669 670
	}
#endif
A
Adam Langley 已提交
671

672
	/* We always send an MSS option. */
A
Adam Langley 已提交
673
	opts->mss = mss;
674
	remaining -= TCPOLEN_MSS_ALIGNED;
A
Adam Langley 已提交
675 676 677

	if (likely(ireq->wscale_ok)) {
		opts->ws = ireq->rcv_wscale;
678
		opts->options |= OPTION_WSCALE;
679
		remaining -= TCPOLEN_WSCALE_ALIGNED;
A
Adam Langley 已提交
680
	}
E
Eric Dumazet 已提交
681
	if (likely(ireq->tstamp_ok)) {
A
Adam Langley 已提交
682
		opts->options |= OPTION_TS;
683
		opts->tsval = tcp_skb_timestamp(skb) + tcp_rsk(req)->ts_off;
A
Adam Langley 已提交
684
		opts->tsecr = req->ts_recent;
685
		remaining -= TCPOLEN_TSTAMP_ALIGNED;
A
Adam Langley 已提交
686 687 688
	}
	if (likely(ireq->sack_ok)) {
		opts->options |= OPTION_SACK_ADVERTISE;
E
Eric Dumazet 已提交
689
		if (unlikely(!ireq->tstamp_ok))
690
			remaining -= TCPOLEN_SACKPERM_ALIGNED;
A
Adam Langley 已提交
691
	}
692 693 694 695 696
	if (foc != NULL && foc->len >= 0) {
		u32 need = foc->len;

		need += foc->exp ? TCPOLEN_EXP_FASTOPEN_BASE :
				   TCPOLEN_FASTOPEN_BASE;
697 698 699 700 701 702 703
		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
704

705 706
	smc_set_option_cond(tcp_sk(sk), ireq, opts, &remaining);

707
	return MAX_TCP_OPTION_SPACE - remaining;
A
Adam Langley 已提交
708 709
}

710 711 712
/* Compute TCP options for ESTABLISHED sockets. This is not the
 * final wire format yet.
 */
713
static unsigned int tcp_established_options(struct sock *sk, struct sk_buff *skb,
A
Adam Langley 已提交
714
					struct tcp_out_options *opts,
715 716
					struct tcp_md5sig_key **md5)
{
A
Adam Langley 已提交
717
	struct tcp_sock *tp = tcp_sk(sk);
718
	unsigned int size = 0;
719
	unsigned int eff_sacks;
A
Adam Langley 已提交
720

721 722
	opts->options = 0;

A
Adam Langley 已提交
723 724 725 726 727 728 729 730 731 732 733 734
#ifdef CONFIG_TCP_MD5SIG
	*md5 = tp->af_specific->md5_lookup(sk, sk);
	if (unlikely(*md5)) {
		opts->options |= OPTION_MD5;
		size += TCPOLEN_MD5SIG_ALIGNED;
	}
#else
	*md5 = NULL;
#endif

	if (likely(tp->rx_opt.tstamp_ok)) {
		opts->options |= OPTION_TS;
E
Eric Dumazet 已提交
735
		opts->tsval = skb ? tcp_skb_timestamp(skb) + tp->tsoffset : 0;
A
Adam Langley 已提交
736 737 738 739
		opts->tsecr = tp->rx_opt.ts_recent;
		size += TCPOLEN_TSTAMP_ALIGNED;
	}

740 741
	eff_sacks = tp->rx_opt.num_sacks + tp->rx_opt.dsack;
	if (unlikely(eff_sacks)) {
742
		const unsigned int remaining = MAX_TCP_OPTION_SPACE - size;
A
Adam Langley 已提交
743
		opts->num_sack_blocks =
744
			min_t(unsigned int, eff_sacks,
A
Adam Langley 已提交
745 746 747 748 749 750 751
			      (remaining - TCPOLEN_SACK_BASE_ALIGNED) /
			      TCPOLEN_SACK_PERBLOCK);
		size += TCPOLEN_SACK_BASE_ALIGNED +
			opts->num_sack_blocks * TCPOLEN_SACK_PERBLOCK;
	}

	return size;
S
Stephen Hemminger 已提交
752
}
L
Linus Torvalds 已提交
753

E
Eric Dumazet 已提交
754 755 756 757 758 759 760 761 762

/* 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
763
 * The invariant being skb->truesize subtracted from sk->sk_wmem_alloc
E
Eric Dumazet 已提交
764 765 766 767 768 769 770 771 772 773 774
 *
 * 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);

775 776 777 778
static void tcp_tsq_handler(struct sock *sk)
{
	if ((1 << sk->sk_state) &
	    (TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_CLOSING |
779 780 781 782
	     TCPF_CLOSE_WAIT  | TCPF_LAST_ACK)) {
		struct tcp_sock *tp = tcp_sk(sk);

		if (tp->lost_out > tp->retrans_out &&
783 784
		    tp->snd_cwnd > tcp_packets_in_flight(tp)) {
			tcp_mstamp_refresh(tp);
785
			tcp_xmit_retransmit_queue(sk);
786
		}
787 788

		tcp_write_xmit(sk, tcp_current_mss(sk), tp->nonagle,
J
John Ogness 已提交
789
			       0, GFP_ATOMIC);
790
	}
791
}
E
Eric Dumazet 已提交
792
/*
S
stephen hemminger 已提交
793
 * One tasklet per cpu tries to send more skbs.
E
Eric Dumazet 已提交
794
 * We run in tasklet context but need to disable irqs when
S
stephen hemminger 已提交
795
 * transferring tsq->head because tcp_wfree() might
E
Eric Dumazet 已提交
796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815
 * interrupt us (non NAPI drivers)
 */
static void tcp_tasklet_func(unsigned long data)
{
	struct tsq_tasklet *tsq = (struct tsq_tasklet *)data;
	LIST_HEAD(list);
	unsigned long flags;
	struct list_head *q, *n;
	struct tcp_sock *tp;
	struct sock *sk;

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

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

		sk = (struct sock *)tp;
816
		smp_mb__before_atomic();
817 818
		clear_bit(TSQ_QUEUED, &sk->sk_tsq_flags);

819
		if (!sk->sk_lock.owned &&
820
		    test_bit(TCP_TSQ_DEFERRED, &sk->sk_tsq_flags)) {
821 822
			bh_lock_sock(sk);
			if (!sock_owned_by_user(sk)) {
823
				clear_bit(TCP_TSQ_DEFERRED, &sk->sk_tsq_flags);
824 825 826
				tcp_tsq_handler(sk);
			}
			bh_unlock_sock(sk);
E
Eric Dumazet 已提交
827 828 829 830 831 832
		}

		sk_free(sk);
	}
}

E
Eric Dumazet 已提交
833 834 835 836
#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
837 838 839 840 841 842 843 844 845
/**
 * 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)
{
846
	unsigned long flags, nflags;
E
Eric Dumazet 已提交
847

848 849
	/* perform an atomic operation only if at least one flag is set */
	do {
850
		flags = sk->sk_tsq_flags;
851 852 853
		if (!(flags & TCP_DEFERRED_ALL))
			return;
		nflags = flags & ~TCP_DEFERRED_ALL;
854
	} while (cmpxchg(&sk->sk_tsq_flags, flags, nflags) != flags);
855

E
Eric Dumazet 已提交
856
	if (flags & TCPF_TSQ_DEFERRED)
857 858
		tcp_tsq_handler(sk);

859 860 861 862 863 864 865 866 867 868 869
	/* 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
870
	if (flags & TCPF_WRITE_TIMER_DEFERRED) {
871
		tcp_write_timer_handler(sk);
872 873
		__sock_put(sk);
	}
E
Eric Dumazet 已提交
874
	if (flags & TCPF_DELACK_TIMER_DEFERRED) {
875
		tcp_delack_timer_handler(sk);
876 877
		__sock_put(sk);
	}
E
Eric Dumazet 已提交
878
	if (flags & TCPF_MTU_REDUCED_DEFERRED) {
879
		inet_csk(sk)->icsk_af_ops->mtu_reduced(sk);
880 881
		__sock_put(sk);
	}
E
Eric Dumazet 已提交
882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900
}
EXPORT_SYMBOL(tcp_release_cb);

void __init tcp_tasklet_init(void)
{
	int i;

	for_each_possible_cpu(i) {
		struct tsq_tasklet *tsq = &per_cpu(tsq_tasklet, i);

		INIT_LIST_HEAD(&tsq->head);
		tasklet_init(&tsq->tasklet,
			     tcp_tasklet_func,
			     (unsigned long)tsq);
	}
}

/*
 * Write buffer destructor automatically called from kfree_skb.
S
stephen hemminger 已提交
901
 * We can't xmit new skbs from this context, as we might already
E
Eric Dumazet 已提交
902 903
 * hold qdisc lock.
 */
E
Eric Dumazet 已提交
904
void tcp_wfree(struct sk_buff *skb)
E
Eric Dumazet 已提交
905 906 907
{
	struct sock *sk = skb->sk;
	struct tcp_sock *tp = tcp_sk(sk);
908
	unsigned long flags, nval, oval;
909 910 911 912

	/* Keep one reference on sk_wmem_alloc.
	 * Will be released by sk_free() from here or tcp_tasklet_func()
	 */
913
	WARN_ON(refcount_sub_and_test(skb->truesize - 1, &sk->sk_wmem_alloc));
914 915 916 917 918 919 920 921

	/* 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)
	 */
922
	if (refcount_read(&sk->sk_wmem_alloc) >= SKB_TRUESIZE(1) && this_cpu_ksoftirqd() == current)
923
		goto out;
E
Eric Dumazet 已提交
924

925
	for (oval = READ_ONCE(sk->sk_tsq_flags);; oval = nval) {
E
Eric Dumazet 已提交
926
		struct tsq_tasklet *tsq;
927
		bool empty;
E
Eric Dumazet 已提交
928

929 930 931
		if (!(oval & TSQF_THROTTLED) || (oval & TSQF_QUEUED))
			goto out;

932
		nval = (oval & ~TSQF_THROTTLED) | TSQF_QUEUED | TCPF_TSQ_DEFERRED;
933
		nval = cmpxchg(&sk->sk_tsq_flags, oval, nval);
934 935 936
		if (nval != oval)
			continue;

E
Eric Dumazet 已提交
937 938
		/* queue this socket to tasklet queue */
		local_irq_save(flags);
939
		tsq = this_cpu_ptr(&tsq_tasklet);
940
		empty = list_empty(&tsq->head);
E
Eric Dumazet 已提交
941
		list_add(&tp->tsq_node, &tsq->head);
942 943
		if (empty)
			tasklet_schedule(&tsq->tasklet);
E
Eric Dumazet 已提交
944
		local_irq_restore(flags);
945
		return;
E
Eric Dumazet 已提交
946
	}
947 948
out:
	sk_free(sk);
E
Eric Dumazet 已提交
949 950
}

951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971
/* Note: Called under hard irq.
 * We can not call TCP stack right away.
 */
enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer)
{
	struct tcp_sock *tp = container_of(timer, struct tcp_sock, pacing_timer);
	struct sock *sk = (struct sock *)tp;
	unsigned long nval, oval;

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

		if (oval & TSQF_QUEUED)
			break;

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

972
		if (!refcount_inc_not_zero(&sk->sk_wmem_alloc))
973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016
			break;
		/* queue this socket to tasklet queue */
		tsq = this_cpu_ptr(&tsq_tasklet);
		empty = list_empty(&tsq->head);
		list_add(&tp->tsq_node, &tsq->head);
		if (empty)
			tasklet_schedule(&tsq->tasklet);
		break;
	}
	return HRTIMER_NORESTART;
}

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

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

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

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

1017 1018 1019 1020 1021 1022
static void tcp_update_skb_after_send(struct tcp_sock *tp, struct sk_buff *skb)
{
	skb->skb_mstamp = tp->tcp_mstamp;
	list_move_tail(&skb->tcp_tsorted_anchor, &tp->tsorted_sent_queue);
}

L
Linus Torvalds 已提交
1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033
/* 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.
 */
1034 1035
static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it,
			    gfp_t gfp_mask)
L
Linus Torvalds 已提交
1036
{
1037 1038 1039 1040
	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
1041
	struct tcp_out_options opts;
1042
	unsigned int tcp_options_size, tcp_header_size;
1043
	struct sk_buff *oskb = NULL;
1044
	struct tcp_md5sig_key *md5;
1045 1046 1047 1048
	struct tcphdr *th;
	int err;

	BUG_ON(!skb || !tcp_skb_pcount(skb));
1049
	tp = tcp_sk(sk);
1050

1051
	if (clone_it) {
1052 1053
		TCP_SKB_CB(skb)->tx.in_flight = TCP_SKB_CB(skb)->end_seq
			- tp->snd_una;
1054
		oskb = skb;
1055 1056 1057 1058 1059 1060 1061 1062

		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);

1063 1064 1065
		if (unlikely(!skb))
			return -ENOBUFS;
	}
1066
	skb->skb_mstamp = tp->tcp_mstamp;
L
Linus Torvalds 已提交
1067

1068 1069
	inet = inet_sk(sk);
	tcb = TCP_SKB_CB(skb);
A
Adam Langley 已提交
1070
	memset(&opts, 0, sizeof(opts));
L
Linus Torvalds 已提交
1071

E
Eric Dumazet 已提交
1072
	if (unlikely(tcb->tcp_flags & TCPHDR_SYN))
A
Adam Langley 已提交
1073 1074 1075 1076 1077
		tcp_options_size = tcp_syn_options(sk, skb, &opts, &md5);
	else
		tcp_options_size = tcp_established_options(sk, skb, &opts,
							   &md5);
	tcp_header_size = tcp_options_size + sizeof(struct tcphdr);
1078

E
Eric Dumazet 已提交
1079
	/* if no packet is in qdisc/device queue, then allow XPS to select
1080 1081 1082 1083 1084
	 * another queue. We can be called from tcp_tsq_handler()
	 * which holds one reference to sk_wmem_alloc.
	 *
	 * TODO: Ideally, in-flight pure ACK packets should not matter here.
	 * One way to get this would be to set skb->truesize = 2 on them.
E
Eric Dumazet 已提交
1085
	 */
1086
	skb->ooo_okay = sk_wmem_alloc_get(sk) < SKB_TRUESIZE(1);
1087

1088 1089 1090 1091 1092 1093 1094
	/* 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;

1095 1096
	skb_push(skb, tcp_header_size);
	skb_reset_transport_header(skb);
E
Eric Dumazet 已提交
1097 1098 1099

	skb_orphan(skb);
	skb->sk = sk;
E
Eric Dumazet 已提交
1100
	skb->destructor = skb_is_tcp_pure_ack(skb) ? __sock_wfree : tcp_wfree;
1101
	skb_set_hash_from_sk(skb, sk);
1102
	refcount_add(skb->truesize, &sk->sk_wmem_alloc);
1103

1104 1105
	skb_set_dst_pending_confirm(skb, sk->sk_dst_pending_confirm);

1106
	/* Build TCP header and checksum it. */
1107
	th = (struct tcphdr *)skb->data;
E
Eric Dumazet 已提交
1108 1109
	th->source		= inet->inet_sport;
	th->dest		= inet->inet_dport;
1110 1111
	th->seq			= htonl(tcb->seq);
	th->ack_seq		= htonl(tp->rcv_nxt);
1112
	*(((__be16 *)th) + 6)	= htons(((tcp_header_size >> 2) << 12) |
E
Eric Dumazet 已提交
1113
					tcb->tcp_flags);
1114 1115 1116

	th->check		= 0;
	th->urg_ptr		= 0;
L
Linus Torvalds 已提交
1117

I
Ilpo Järvinen 已提交
1118
	/* The urg_mode check is necessary during a below snd_una win probe */
1119 1120 1121 1122 1123
	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)) {
1124
			th->urg_ptr = htons(0xFFFF);
1125 1126
			th->urg = 1;
		}
1127
	}
L
Linus Torvalds 已提交
1128

1129
	tcp_options_write((__be32 *)(th + 1), tp, &opts);
1130
	skb_shinfo(skb)->gso_type = sk->sk_gso_type;
1131 1132 1133 1134 1135 1136 1137 1138 1139
	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));
	}
1140 1141 1142
#ifdef CONFIG_TCP_MD5SIG
	/* Calculate the MD5 hash, as we have all we need now */
	if (md5) {
E
Eric Dumazet 已提交
1143
		sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1144
		tp->af_specific->calc_md5_hash(opts.hash_location,
1145
					       md5, sk, skb);
1146 1147 1148
	}
#endif

1149
	icsk->icsk_af_ops->send_check(sk, skb);
L
Linus Torvalds 已提交
1150

E
Eric Dumazet 已提交
1151
	if (likely(tcb->tcp_flags & TCPHDR_ACK))
1152
		tcp_event_ack_sent(sk, tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
1153

1154
	if (skb->len != tcp_header_size) {
1155
		tcp_event_data_sent(tp, sk);
1156
		tp->data_segs_out += tcp_skb_pcount(skb);
1157
		tcp_internal_pacing(sk, skb);
1158
	}
L
Linus Torvalds 已提交
1159

1160
	if (after(tcb->end_seq, tp->snd_nxt) || tcb->seq == tcb->end_seq)
1161 1162
		TCP_ADD_STATS(sock_net(sk), TCP_MIB_OUTSEGS,
			      tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
1163

1164
	tp->segs_out += tcp_skb_pcount(skb);
1165
	/* OK, its time to fill skb_shinfo(skb)->gso_{segs|size} */
1166
	skb_shinfo(skb)->gso_segs = tcp_skb_pcount(skb);
1167
	skb_shinfo(skb)->gso_size = tcp_skb_mss(skb);
1168

E
Eric Dumazet 已提交
1169
	/* Our usage of tstamp should remain private */
T
Thomas Gleixner 已提交
1170
	skb->tstamp = 0;
1171 1172 1173 1174 1175

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

1176
	err = icsk->icsk_af_ops->queue_xmit(sk, skb, &inet->cork.fl);
E
Eric Dumazet 已提交
1177

1178 1179 1180 1181
	if (unlikely(err > 0)) {
		tcp_enter_cwr(sk);
		err = net_xmit_eval(err);
	}
E
Eric Dumazet 已提交
1182
	if (!err && oskb) {
1183
		tcp_update_skb_after_send(tp, oskb);
E
Eric Dumazet 已提交
1184 1185
		tcp_rate_skb_sent(sk, oskb);
	}
1186
	return err;
L
Linus Torvalds 已提交
1187 1188
}

1189
/* This routine just queues the buffer for sending.
L
Linus Torvalds 已提交
1190 1191 1192 1193 1194 1195 1196 1197 1198 1199
 *
 * NOTE: probe0 timer is not checked, do not forget tcp_push_pending_frames,
 * otherwise socket can stall.
 */
static void tcp_queue_skb(struct sock *sk, struct sk_buff *skb)
{
	struct tcp_sock *tp = tcp_sk(sk);

	/* Advance write_seq and place onto the write_queue. */
	tp->write_seq = TCP_SKB_CB(skb)->end_seq;
1200
	__skb_header_release(skb);
1201
	tcp_add_write_queue_tail(sk, skb);
1202 1203
	sk->sk_wmem_queued += skb->truesize;
	sk_mem_charge(sk, skb->truesize);
L
Linus Torvalds 已提交
1204 1205
}

1206
/* Initialize TSO segments for a packet. */
1207
static void tcp_set_skb_tso_segs(struct sk_buff *skb, unsigned int mss_now)
1208
{
1209
	if (skb->len <= mss_now || skb->ip_summed == CHECKSUM_NONE) {
1210 1211 1212
		/* Avoid the costly divide in the normal
		 * non-TSO case.
		 */
1213
		tcp_skb_pcount_set(skb, 1);
1214
		TCP_SKB_CB(skb)->tcp_gso_size = 0;
1215
	} else {
1216
		tcp_skb_pcount_set(skb, DIV_ROUND_UP(skb->len, mss_now));
1217
		TCP_SKB_CB(skb)->tcp_gso_size = mss_now;
L
Linus Torvalds 已提交
1218 1219 1220
	}
}

I
Ilpo Järvinen 已提交
1221
/* When a modification to fackets out becomes necessary, we need to check
1222
 * skb is counted to fackets_out or not.
I
Ilpo Järvinen 已提交
1223
 */
1224
static void tcp_adjust_fackets_out(struct sock *sk, const struct sk_buff *skb,
I
Ilpo Järvinen 已提交
1225 1226
				   int decr)
{
1227 1228
	struct tcp_sock *tp = tcp_sk(sk);

1229
	if (!tp->sacked_out || tcp_is_reno(tp))
I
Ilpo Järvinen 已提交
1230 1231
		return;

1232
	if (after(tcp_highest_sack_seq(tp), TCP_SKB_CB(skb)->seq))
I
Ilpo Järvinen 已提交
1233 1234 1235
		tp->fackets_out -= decr;
}

1236 1237 1238
/* Pcount in the middle of the write queue got changed, we need to do various
 * tweaks to fix counters
 */
1239
static void tcp_adjust_pcount(struct sock *sk, const struct sk_buff *skb, int decr)
1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259
{
	struct tcp_sock *tp = tcp_sk(sk);

	tp->packets_out -= decr;

	if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)
		tp->sacked_out -= decr;
	if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS)
		tp->retrans_out -= decr;
	if (TCP_SKB_CB(skb)->sacked & TCPCB_LOST)
		tp->lost_out -= decr;

	/* Reno case is special. Sigh... */
	if (tcp_is_reno(tp) && decr > 0)
		tp->sacked_out -= min_t(u32, tp->sacked_out, decr);

	tcp_adjust_fackets_out(sk, skb, decr);

	if (tp->lost_skb_hint &&
	    before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(tp->lost_skb_hint)->seq) &&
1260
	    (tcp_is_fack(tp) || (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)))
1261 1262 1263 1264 1265
		tp->lost_cnt_hint -= decr;

	tcp_verify_left_out(tp);
}

1266 1267 1268 1269 1270 1271
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);
}

1272 1273 1274 1275
static void tcp_fragment_tstamp(struct sk_buff *skb, struct sk_buff *skb2)
{
	struct skb_shared_info *shinfo = skb_shinfo(skb);

1276
	if (unlikely(tcp_has_tx_tstamp(skb)) &&
1277 1278 1279 1280 1281 1282 1283
	    !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);
1284 1285
		TCP_SKB_CB(skb2)->txstamp_ack = TCP_SKB_CB(skb)->txstamp_ack;
		TCP_SKB_CB(skb)->txstamp_ack = 0;
1286 1287 1288
	}
}

1289 1290 1291 1292 1293 1294
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;
}

1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306
/* 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
1307 1308
/* 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
1309
 * packet to the list.  This won't be called frequently, I hope.
L
Linus Torvalds 已提交
1310 1311
 * Remember, these are still headerless SKBs at this point.
 */
1312 1313
int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
		 struct sk_buff *skb, u32 len,
1314
		 unsigned int mss_now, gfp_t gfp)
L
Linus Torvalds 已提交
1315 1316 1317
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *buff;
1318
	int nsize, old_factor;
1319
	int nlen;
1320
	u8 flags;
L
Linus Torvalds 已提交
1321

1322 1323
	if (WARN_ON(len > skb->len))
		return -EINVAL;
1324

L
Linus Torvalds 已提交
1325 1326 1327 1328
	nsize = skb_headlen(skb) - len;
	if (nsize < 0)
		nsize = 0;

1329
	if (skb_unclone(skb, gfp))
L
Linus Torvalds 已提交
1330 1331 1332
		return -ENOMEM;

	/* Get a new skb... force flag on. */
1333
	buff = sk_stream_alloc_skb(sk, nsize, gfp, true);
1334
	if (!buff)
L
Linus Torvalds 已提交
1335
		return -ENOMEM; /* We'll just try again later. */
H
Herbert Xu 已提交
1336

1337 1338
	sk->sk_wmem_queued += buff->truesize;
	sk_mem_charge(sk, buff->truesize);
1339 1340 1341
	nlen = skb->len - len - nsize;
	buff->truesize += nlen;
	skb->truesize -= nlen;
L
Linus Torvalds 已提交
1342 1343 1344 1345 1346 1347 1348

	/* 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
1349 1350 1351
	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;
1352
	TCP_SKB_CB(buff)->sacked = TCP_SKB_CB(skb)->sacked;
1353
	tcp_skb_fragment_eor(skb, buff);
L
Linus Torvalds 已提交
1354

1355
	if (!skb_shinfo(skb)->nr_frags && skb->ip_summed != CHECKSUM_PARTIAL) {
L
Linus Torvalds 已提交
1356
		/* Copy and checksum data tail into the new buffer. */
1357 1358
		buff->csum = csum_partial_copy_nocheck(skb->data + len,
						       skb_put(buff, nsize),
L
Linus Torvalds 已提交
1359 1360 1361 1362 1363 1364
						       nsize, 0);

		skb_trim(skb, len);

		skb->csum = csum_block_sub(skb->csum, buff->csum, len);
	} else {
1365
		skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
1366 1367 1368 1369 1370
		skb_split(skb, buff, len);
	}

	buff->ip_summed = skb->ip_summed;

1371
	buff->tstamp = skb->tstamp;
1372
	tcp_fragment_tstamp(skb, buff);
L
Linus Torvalds 已提交
1373

1374 1375
	old_factor = tcp_skb_pcount(skb);

L
Linus Torvalds 已提交
1376
	/* Fix up tso_factor for both original and new SKB.  */
1377 1378
	tcp_set_skb_tso_segs(skb, mss_now);
	tcp_set_skb_tso_segs(buff, mss_now);
L
Linus Torvalds 已提交
1379

1380 1381 1382
	/* Update delivered info for the new segment */
	TCP_SKB_CB(buff)->tx = TCP_SKB_CB(skb)->tx;

1383 1384 1385
	/* If this packet has been sent out already, we must
	 * adjust the various packet counters.
	 */
1386
	if (!before(tp->snd_nxt, TCP_SKB_CB(buff)->end_seq)) {
1387 1388
		int diff = old_factor - tcp_skb_pcount(skb) -
			tcp_skb_pcount(buff);
L
Linus Torvalds 已提交
1389

1390 1391
		if (diff)
			tcp_adjust_pcount(sk, skb, diff);
L
Linus Torvalds 已提交
1392 1393 1394
	}

	/* Link BUFF into the send queue. */
1395
	__skb_header_release(buff);
1396
	tcp_insert_write_queue_after(skb, buff, sk, tcp_queue);
1397 1398
	if (tcp_queue == TCP_FRAG_IN_RTX_QUEUE)
		list_add(&buff->tcp_tsorted_anchor, &skb->tcp_tsorted_anchor);
L
Linus Torvalds 已提交
1399 1400 1401 1402

	return 0;
}

1403 1404
/* 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
1405
 */
1406
static int __pskb_trim_head(struct sk_buff *skb, int len)
L
Linus Torvalds 已提交
1407
{
1408
	struct skb_shared_info *shinfo;
L
Linus Torvalds 已提交
1409 1410
	int i, k, eat;

E
Eric Dumazet 已提交
1411 1412 1413 1414 1415
	eat = min_t(int, len, skb_headlen(skb));
	if (eat) {
		__skb_pull(skb, eat);
		len -= eat;
		if (!len)
1416
			return 0;
E
Eric Dumazet 已提交
1417
	}
L
Linus Torvalds 已提交
1418 1419
	eat = len;
	k = 0;
1420 1421 1422
	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
1423 1424

		if (size <= eat) {
1425
			skb_frag_unref(skb, i);
E
Eric Dumazet 已提交
1426
			eat -= size;
L
Linus Torvalds 已提交
1427
		} else {
1428
			shinfo->frags[k] = shinfo->frags[i];
L
Linus Torvalds 已提交
1429
			if (eat) {
1430 1431
				shinfo->frags[k].page_offset += eat;
				skb_frag_size_sub(&shinfo->frags[k], eat);
L
Linus Torvalds 已提交
1432 1433 1434 1435 1436
				eat = 0;
			}
			k++;
		}
	}
1437
	shinfo->nr_frags = k;
L
Linus Torvalds 已提交
1438 1439 1440

	skb->data_len -= len;
	skb->len = skb->data_len;
1441
	return len;
L
Linus Torvalds 已提交
1442 1443
}

1444
/* Remove acked data from a packet in the transmit queue. */
L
Linus Torvalds 已提交
1445 1446
int tcp_trim_head(struct sock *sk, struct sk_buff *skb, u32 len)
{
1447 1448
	u32 delta_truesize;

1449
	if (skb_unclone(skb, GFP_ATOMIC))
L
Linus Torvalds 已提交
1450 1451
		return -ENOMEM;

1452
	delta_truesize = __pskb_trim_head(skb, len);
L
Linus Torvalds 已提交
1453 1454

	TCP_SKB_CB(skb)->seq += len;
1455
	skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
1456

1457 1458 1459 1460 1461 1462
	if (delta_truesize) {
		skb->truesize	   -= delta_truesize;
		sk->sk_wmem_queued -= delta_truesize;
		sk_mem_uncharge(sk, delta_truesize);
		sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
	}
L
Linus Torvalds 已提交
1463

1464
	/* Any change of skb->len requires recalculation of tso factor. */
L
Linus Torvalds 已提交
1465
	if (tcp_skb_pcount(skb) > 1)
1466
		tcp_set_skb_tso_segs(skb, tcp_skb_mss(skb));
L
Linus Torvalds 已提交
1467 1468 1469 1470

	return 0;
}

1471 1472
/* 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
1473
{
1474 1475
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1476 1477 1478 1479 1480 1481 1482
	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);

1483 1484 1485 1486 1487 1488 1489 1490
	/* 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
1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503
	/* Clamp it (mss_clamp does not include tcp options) */
	if (mss_now > tp->rx_opt.mss_clamp)
		mss_now = tp->rx_opt.mss_clamp;

	/* Now subtract optional transport overhead */
	mss_now -= icsk->icsk_ext_hdr_len;

	/* Then reserve room for full set of TCP options and 8 bytes of data */
	if (mss_now < 48)
		mss_now = 48;
	return mss_now;
}

1504 1505 1506 1507 1508 1509 1510 1511
/* Calculate MSS. Not accounting for SACKs here.  */
int tcp_mtu_to_mss(struct sock *sk, int pmtu)
{
	/* Subtract TCP options size, not including SACKs */
	return __tcp_mtu_to_mss(sk, pmtu) -
	       (tcp_sk(sk)->tcp_header_len - sizeof(struct tcphdr));
}

J
John Heffner 已提交
1512
/* Inverse of above */
1513
int tcp_mss_to_mtu(struct sock *sk, int mss)
J
John Heffner 已提交
1514
{
1515 1516
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1517 1518 1519 1520 1521 1522 1523
	int mtu;

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

1524 1525 1526 1527 1528 1529 1530
	/* 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
1531 1532
	return mtu;
}
1533
EXPORT_SYMBOL(tcp_mss_to_mtu);
J
John Heffner 已提交
1534

1535
/* MTU probing init per socket */
J
John Heffner 已提交
1536 1537 1538 1539
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
1540
	struct net *net = sock_net(sk);
J
John Heffner 已提交
1541

F
Fan Du 已提交
1542
	icsk->icsk_mtup.enabled = net->ipv4.sysctl_tcp_mtu_probing > 1;
J
John Heffner 已提交
1543
	icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp + sizeof(struct tcphdr) +
1544
			       icsk->icsk_af_ops->net_header_len;
F
Fan Du 已提交
1545
	icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, net->ipv4.sysctl_tcp_base_mss);
J
John Heffner 已提交
1546
	icsk->icsk_mtup.probe_size = 0;
1547
	if (icsk->icsk_mtup.enabled)
1548
		icsk->icsk_mtup.probe_timestamp = tcp_jiffies32;
J
John Heffner 已提交
1549
}
E
Eric Dumazet 已提交
1550
EXPORT_SYMBOL(tcp_mtup_init);
J
John Heffner 已提交
1551

L
Linus Torvalds 已提交
1552 1553 1554 1555 1556 1557
/* 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
1558
   It is minimum of user_mss and mss received with SYN.
L
Linus Torvalds 已提交
1559 1560
   It also does not include TCP options.

1561
   inet_csk(sk)->icsk_pmtu_cookie is last pmtu, seen by this function.
L
Linus Torvalds 已提交
1562 1563 1564 1565 1566 1567 1568 1569 1570

   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.

1571 1572
   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
1573 1574 1575 1576
 */
unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu)
{
	struct tcp_sock *tp = tcp_sk(sk);
1577
	struct inet_connection_sock *icsk = inet_csk(sk);
J
John Heffner 已提交
1578
	int mss_now;
L
Linus Torvalds 已提交
1579

J
John Heffner 已提交
1580 1581
	if (icsk->icsk_mtup.search_high > pmtu)
		icsk->icsk_mtup.search_high = pmtu;
L
Linus Torvalds 已提交
1582

J
John Heffner 已提交
1583
	mss_now = tcp_mtu_to_mss(sk, pmtu);
1584
	mss_now = tcp_bound_to_half_wnd(tp, mss_now);
L
Linus Torvalds 已提交
1585 1586

	/* And store cached results */
1587
	icsk->icsk_pmtu_cookie = pmtu;
J
John Heffner 已提交
1588 1589
	if (icsk->icsk_mtup.enabled)
		mss_now = min(mss_now, tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low));
1590
	tp->mss_cache = mss_now;
L
Linus Torvalds 已提交
1591 1592 1593

	return mss_now;
}
E
Eric Dumazet 已提交
1594
EXPORT_SYMBOL(tcp_sync_mss);
L
Linus Torvalds 已提交
1595 1596 1597 1598

/* Compute the current effective MSS, taking SACKs and IP options,
 * and even PMTU discovery events into account.
 */
I
Ilpo Järvinen 已提交
1599
unsigned int tcp_current_mss(struct sock *sk)
L
Linus Torvalds 已提交
1600
{
1601 1602
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct dst_entry *dst = __sk_dst_get(sk);
1603
	u32 mss_now;
1604
	unsigned int header_len;
A
Adam Langley 已提交
1605 1606
	struct tcp_out_options opts;
	struct tcp_md5sig_key *md5;
1607 1608 1609

	mss_now = tp->mss_cache;

L
Linus Torvalds 已提交
1610 1611
	if (dst) {
		u32 mtu = dst_mtu(dst);
1612
		if (mtu != inet_csk(sk)->icsk_pmtu_cookie)
L
Linus Torvalds 已提交
1613 1614 1615
			mss_now = tcp_sync_mss(sk, mtu);
	}

A
Adam Langley 已提交
1616 1617 1618 1619 1620 1621 1622 1623 1624 1625
	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;
	}
1626

L
Linus Torvalds 已提交
1627 1628 1629
	return mss_now;
}

1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648
/* 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;
	}
1649
	tp->snd_cwnd_stamp = tcp_jiffies32;
1650 1651
}

1652
static void tcp_cwnd_validate(struct sock *sk, bool is_cwnd_limited)
1653
{
1654
	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1655
	struct tcp_sock *tp = tcp_sk(sk);
1656

1657 1658 1659 1660 1661 1662 1663 1664 1665
	/* Track the maximum number of outstanding packets in each
	 * window, and remember whether we were cwnd-limited then.
	 */
	if (!before(tp->snd_una, tp->max_packets_seq) ||
	    tp->packets_out > tp->max_packets_out) {
		tp->max_packets_out = tp->packets_out;
		tp->max_packets_seq = tp->snd_nxt;
		tp->is_cwnd_limited = is_cwnd_limited;
	}
1666

1667
	if (tcp_is_cwnd_limited(sk)) {
1668 1669
		/* Network is feed fully. */
		tp->snd_cwnd_used = 0;
1670
		tp->snd_cwnd_stamp = tcp_jiffies32;
1671 1672 1673 1674 1675
	} else {
		/* Network starves. */
		if (tp->packets_out > tp->snd_cwnd_used)
			tp->snd_cwnd_used = tp->packets_out;

1676
		if (sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle &&
1677
		    (s32)(tcp_jiffies32 - tp->snd_cwnd_stamp) >= inet_csk(sk)->icsk_rto &&
1678
		    !ca_ops->cong_control)
1679
			tcp_cwnd_application_limited(sk);
1680 1681 1682 1683 1684

		/* 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)
1685
		 * 3) no more data to send (tcp_write_queue_empty())
1686 1687
		 * 4) application is hitting buffer limit (SOCK_NOSPACE)
		 */
1688
		if (tcp_write_queue_empty(sk) && sk->sk_socket &&
1689 1690 1691
		    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);
1692 1693 1694
	}
}

E
Eric Dumazet 已提交
1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708
/* 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
1709
 */
E
Eric Dumazet 已提交
1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724
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,
1725
			    int nonagle)
E
Eric Dumazet 已提交
1726 1727 1728 1729 1730
{
	return partial &&
		((nonagle & TCP_NAGLE_CORK) ||
		 (!nonagle && tp->packets_out && tcp_minshall_check(tp)));
}
E
Eric Dumazet 已提交
1731 1732 1733 1734

/* Return how many segs we'd like on a TSO packet,
 * to send one TSO packet per ms
 */
1735 1736
u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
		     int min_tso_segs)
E
Eric Dumazet 已提交
1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747
{
	u32 bytes, segs;

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

	/* Goal is to send at least one packet per ms,
	 * not one big TSO packet every 100 ms.
	 * This preserves ACK clocking and is consistent
	 * with tcp_tso_should_defer() heuristic.
	 */
1748
	segs = max_t(u32, bytes / mss_now, min_tso_segs);
E
Eric Dumazet 已提交
1749 1750 1751

	return min_t(u32, segs, sk->sk_gso_max_segs);
}
1752
EXPORT_SYMBOL(tcp_tso_autosize);
E
Eric Dumazet 已提交
1753

1754 1755 1756 1757 1758 1759 1760 1761
/* Return the number of segments we want in the skb we are transmitting.
 * See if congestion control module wants to decide; otherwise, autosize.
 */
static u32 tcp_tso_segs(struct sock *sk, unsigned int mss_now)
{
	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
	u32 tso_segs = ca_ops->tso_segs_goal ? ca_ops->tso_segs_goal(sk) : 0;

1762
	return tso_segs ? :
1763 1764
		tcp_tso_autosize(sk, mss_now,
				 sock_net(sk)->ipv4.sysctl_tcp_min_tso_segs);
1765 1766
}

E
Eric Dumazet 已提交
1767 1768 1769 1770 1771 1772
/* 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)
1773
{
1774
	const struct tcp_sock *tp = tcp_sk(sk);
E
Eric Dumazet 已提交
1775
	u32 partial, needed, window, max_len;
1776

1777
	window = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
1778
	max_len = mss_now * max_segs;
1779

1780 1781
	if (likely(max_len <= window && skb != tcp_write_queue_tail(sk)))
		return max_len;
1782

1783 1784
	needed = min(skb->len, window);

1785 1786
	if (max_len <= needed)
		return max_len;
1787

E
Eric Dumazet 已提交
1788 1789 1790 1791 1792
	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
	 */
1793
	if (tcp_nagle_check(partial != 0, tp, nonagle))
E
Eric Dumazet 已提交
1794 1795 1796
		return needed - partial;

	return needed;
1797 1798 1799 1800 1801
}

/* 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.
 */
1802 1803
static inline unsigned int tcp_cwnd_test(const struct tcp_sock *tp,
					 const struct sk_buff *skb)
1804
{
E
Eric Dumazet 已提交
1805
	u32 in_flight, cwnd, halfcwnd;
1806 1807

	/* Don't be strict about the congestion window for the final FIN.  */
E
Eric Dumazet 已提交
1808 1809
	if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) &&
	    tcp_skb_pcount(skb) == 1)
1810 1811 1812 1813
		return 1;

	in_flight = tcp_packets_in_flight(tp);
	cwnd = tp->snd_cwnd;
E
Eric Dumazet 已提交
1814 1815
	if (in_flight >= cwnd)
		return 0;
1816

E
Eric Dumazet 已提交
1817 1818 1819 1820 1821
	/* For better scheduling, ensure we have at least
	 * 2 GSO packets in flight.
	 */
	halfcwnd = max(cwnd >> 1, 1U);
	return min(halfcwnd, cwnd - in_flight);
1822 1823
}

1824
/* Initialize TSO state of a skb.
1825
 * This must be invoked the first time we consider transmitting
1826 1827
 * SKB onto the wire.
 */
1828
static int tcp_init_tso_segs(struct sk_buff *skb, unsigned int mss_now)
1829 1830 1831
{
	int tso_segs = tcp_skb_pcount(skb);

1832
	if (!tso_segs || (tso_segs > 1 && tcp_skb_mss(skb) != mss_now)) {
1833
		tcp_set_skb_tso_segs(skb, mss_now);
1834 1835 1836 1837 1838 1839
		tso_segs = tcp_skb_pcount(skb);
	}
	return tso_segs;
}


E
Eric Dumazet 已提交
1840
/* Return true if the Nagle test allows this packet to be
1841 1842
 * sent now.
 */
E
Eric Dumazet 已提交
1843 1844
static inline bool tcp_nagle_test(const struct tcp_sock *tp, const struct sk_buff *skb,
				  unsigned int cur_mss, int nonagle)
1845 1846 1847 1848 1849 1850 1851 1852
{
	/* 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
1853
		return true;
1854

Y
Yuchung Cheng 已提交
1855 1856
	/* 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
1857
		return true;
1858

1859
	if (!tcp_nagle_check(skb->len < cur_mss, tp, nonagle))
E
Eric Dumazet 已提交
1860
		return true;
1861

E
Eric Dumazet 已提交
1862
	return false;
1863 1864 1865
}

/* Does at least the first segment of SKB fit into the send window? */
E
Eric Dumazet 已提交
1866 1867 1868
static bool tcp_snd_wnd_test(const struct tcp_sock *tp,
			     const struct sk_buff *skb,
			     unsigned int cur_mss)
1869 1870 1871 1872 1873 1874
{
	u32 end_seq = TCP_SKB_CB(skb)->end_seq;

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

1875
	return !after(end_seq, tcp_wnd_end(tp));
1876 1877 1878 1879 1880 1881 1882 1883 1884
}

/* 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).
 */
1885 1886
static int tso_fragment(struct sock *sk, enum tcp_queue tcp_queue,
			struct sk_buff *skb, unsigned int len,
1887
			unsigned int mss_now, gfp_t gfp)
1888 1889 1890
{
	struct sk_buff *buff;
	int nlen = skb->len - len;
1891
	u8 flags;
1892 1893

	/* All of a TSO frame must be composed of paged data.  */
1894
	if (skb->len != skb->data_len)
1895
		return tcp_fragment(sk, tcp_queue, skb, len, mss_now, gfp);
1896

1897
	buff = sk_stream_alloc_skb(sk, 0, gfp, true);
1898
	if (unlikely(!buff))
1899 1900
		return -ENOMEM;

1901 1902
	sk->sk_wmem_queued += buff->truesize;
	sk_mem_charge(sk, buff->truesize);
1903
	buff->truesize += nlen;
1904 1905 1906 1907 1908 1909 1910 1911
	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
1912 1913 1914
	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;
1915 1916 1917 1918

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

1919 1920
	tcp_skb_fragment_eor(skb, buff);

1921
	buff->ip_summed = skb->ip_summed = CHECKSUM_PARTIAL;
1922
	skb_split(skb, buff, len);
1923
	tcp_fragment_tstamp(skb, buff);
1924 1925

	/* Fix up tso_factor for both original and new SKB.  */
1926 1927
	tcp_set_skb_tso_segs(skb, mss_now);
	tcp_set_skb_tso_segs(buff, mss_now);
1928 1929

	/* Link BUFF into the send queue. */
1930
	__skb_header_release(buff);
1931
	tcp_insert_write_queue_after(skb, buff, sk, tcp_queue);
1932 1933 1934 1935 1936 1937 1938 1939 1940

	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.
 */
1941
static bool tcp_tso_should_defer(struct sock *sk, struct sk_buff *skb,
E
Eric Dumazet 已提交
1942
				 bool *is_cwnd_limited, u32 max_segs)
1943
{
1944
	const struct inet_connection_sock *icsk = inet_csk(sk);
1945 1946 1947
	u32 age, send_win, cong_win, limit, in_flight;
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *head;
1948
	int win_divisor;
1949

E
Eric Dumazet 已提交
1950
	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
J
John Heffner 已提交
1951
		goto send_now;
1952

1953
	if (icsk->icsk_ca_state >= TCP_CA_Recovery)
J
John Heffner 已提交
1954 1955
		goto send_now;

E
Eric Dumazet 已提交
1956 1957 1958
	/* Avoid bursty behavior by allowing defer
	 * only if the last write was recent.
	 */
1959
	if ((s32)(tcp_jiffies32 - tp->lsndtime) > 0)
J
John Heffner 已提交
1960
		goto send_now;
1961

1962 1963
	in_flight = tcp_packets_in_flight(tp);

1964
	BUG_ON(tcp_skb_pcount(skb) <= 1 || (tp->snd_cwnd <= in_flight));
1965

1966
	send_win = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
1967 1968 1969 1970 1971 1972

	/* 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);

1973
	/* If a full-sized TSO skb can be sent, do it. */
E
Eric Dumazet 已提交
1974
	if (limit >= max_segs * tp->mss_cache)
J
John Heffner 已提交
1975
		goto send_now;
1976

1977 1978 1979 1980
	/* 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;

1981
	win_divisor = ACCESS_ONCE(sock_net(sk)->ipv4.sysctl_tcp_tso_win_divisor);
1982
	if (win_divisor) {
1983 1984 1985 1986 1987
		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.
		 */
1988
		chunk /= win_divisor;
1989
		if (limit >= chunk)
J
John Heffner 已提交
1990
			goto send_now;
1991 1992 1993 1994 1995 1996
	} 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.
		 */
1997
		if (limit > tcp_max_tso_deferred_mss(tp) * tp->mss_cache)
J
John Heffner 已提交
1998
			goto send_now;
1999 2000
	}

2001 2002 2003 2004
	/* TODO : use tsorted_sent_queue ? */
	head = tcp_rtx_queue_head(sk);
	if (!head)
		goto send_now;
2005
	age = tcp_stamp_us_delta(tp->tcp_mstamp, head->skb_mstamp);
2006 2007 2008 2009
	/* If next ACK is likely to come too late (half srtt), do not defer */
	if (age < (tp->srtt_us >> 4))
		goto send_now;

E
Eric Dumazet 已提交
2010
	/* Ok, it looks like it is advisable to defer. */
J
John Heffner 已提交
2011

2012
	if (cong_win < send_win && cong_win <= skb->len)
2013 2014
		*is_cwnd_limited = true;

E
Eric Dumazet 已提交
2015
	return true;
J
John Heffner 已提交
2016 2017

send_now:
E
Eric Dumazet 已提交
2018
	return false;
2019 2020
}

2021 2022 2023 2024 2025 2026 2027 2028 2029
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;
2030
	delta = tcp_jiffies32 - icsk->icsk_mtup.probe_timestamp;
2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041
	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 */
2042
		icsk->icsk_mtup.probe_timestamp = tcp_jiffies32;
2043 2044 2045
	}
}

J
John Heffner 已提交
2046
/* Create a new MTU probe if we are ready.
2047 2048 2049 2050
 * 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
2051 2052
 * Returns 0 if we should wait to probe (no cwnd available),
 *         1 if a probe was sent,
2053 2054
 *         -1 otherwise
 */
J
John Heffner 已提交
2055 2056 2057
static int tcp_mtu_probe(struct sock *sk)
{
	struct inet_connection_sock *icsk = inet_csk(sk);
2058
	struct tcp_sock *tp = tcp_sk(sk);
J
John Heffner 已提交
2059
	struct sk_buff *skb, *nskb, *next;
2060
	struct net *net = sock_net(sk);
J
John Heffner 已提交
2061
	int probe_size;
2062
	int size_needed;
2063
	int copy, len;
J
John Heffner 已提交
2064
	int mss_now;
2065
	int interval;
J
John Heffner 已提交
2066 2067 2068 2069

	/* Not currently probing/verifying,
	 * not in recovery,
	 * have enough cwnd, and
2070 2071 2072 2073 2074 2075 2076
	 * 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
2077 2078
		return -1;

2079 2080 2081 2082
	/* 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
2083
	mss_now = tcp_current_mss(sk);
2084 2085
	probe_size = tcp_mtu_to_mss(sk, (icsk->icsk_mtup.search_high +
				    icsk->icsk_mtup.search_low) >> 1);
2086
	size_needed = probe_size + (tp->reordering + 1) * tp->mss_cache;
2087
	interval = icsk->icsk_mtup.search_high - icsk->icsk_mtup.search_low;
2088 2089 2090 2091
	/* 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.
	 */
2092
	if (probe_size > tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_high) ||
2093 2094 2095 2096 2097
		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
2098 2099 2100 2101
		return -1;
	}

	/* Have enough data in the send queue to probe? */
2102
	if (tp->write_seq - tp->snd_nxt < size_needed)
J
John Heffner 已提交
2103 2104
		return -1;

2105 2106
	if (tp->snd_wnd < size_needed)
		return -1;
2107
	if (after(tp->snd_nxt + size_needed, tcp_wnd_end(tp)))
2108
		return 0;
J
John Heffner 已提交
2109

2110 2111 2112
	/* 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
2113 2114 2115 2116 2117 2118
			return -1;
		else
			return 0;
	}

	/* We're allowed to probe.  Build it now. */
2119
	nskb = sk_stream_alloc_skb(sk, probe_size, GFP_ATOMIC, false);
2120
	if (!nskb)
J
John Heffner 已提交
2121
		return -1;
2122 2123
	sk->sk_wmem_queued += nskb->truesize;
	sk_mem_charge(sk, nskb->truesize);
J
John Heffner 已提交
2124

2125
	skb = tcp_send_head(sk);
J
John Heffner 已提交
2126 2127 2128

	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
2129
	TCP_SKB_CB(nskb)->tcp_flags = TCPHDR_ACK;
J
John Heffner 已提交
2130 2131
	TCP_SKB_CB(nskb)->sacked = 0;
	nskb->csum = 0;
2132
	nskb->ip_summed = skb->ip_summed;
J
John Heffner 已提交
2133

2134
	tcp_insert_write_queue_before(nskb, skb, sk);
2135
	tcp_highest_sack_replace(sk, skb, nskb);
2136

J
John Heffner 已提交
2137
	len = 0;
2138
	tcp_for_write_queue_from_safe(skb, next, sk) {
J
John Heffner 已提交
2139
		copy = min_t(int, skb->len, probe_size - len);
2140
		if (nskb->ip_summed) {
J
John Heffner 已提交
2141
			skb_copy_bits(skb, 0, skb_put(nskb, copy), copy);
2142 2143 2144 2145 2146 2147
		} else {
			__wsum csum = skb_copy_and_csum_bits(skb, 0,
							     skb_put(nskb, copy),
							     copy, 0);
			nskb->csum = csum_block_add(nskb->csum, csum, len);
		}
J
John Heffner 已提交
2148 2149 2150 2151

		if (skb->len <= copy) {
			/* We've eaten all the data from this skb.
			 * Throw it away. */
E
Eric Dumazet 已提交
2152
			TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;
2153
			tcp_unlink_write_queue(skb, sk);
2154
			sk_wmem_free_skb(sk, skb);
J
John Heffner 已提交
2155
		} else {
E
Eric Dumazet 已提交
2156
			TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags &
C
Changli Gao 已提交
2157
						   ~(TCPHDR_FIN|TCPHDR_PSH);
J
John Heffner 已提交
2158 2159
			if (!skb_shinfo(skb)->nr_frags) {
				skb_pull(skb, copy);
2160
				if (skb->ip_summed != CHECKSUM_PARTIAL)
2161 2162
					skb->csum = csum_partial(skb->data,
								 skb->len, 0);
J
John Heffner 已提交
2163 2164
			} else {
				__pskb_trim_head(skb, copy);
2165
				tcp_set_skb_tso_segs(skb, mss_now);
J
John Heffner 已提交
2166 2167 2168 2169 2170
			}
			TCP_SKB_CB(skb)->seq += copy;
		}

		len += copy;
2171 2172 2173

		if (len >= probe_size)
			break;
J
John Heffner 已提交
2174
	}
2175
	tcp_init_tso_segs(nskb, nskb->len);
J
John Heffner 已提交
2176 2177

	/* We're ready to send.  If this fails, the probe will
E
Eric Dumazet 已提交
2178 2179
	 * be resegmented into mss-sized pieces by tcp_write_xmit().
	 */
J
John Heffner 已提交
2180 2181
	if (!tcp_transmit_skb(sk, nskb, 1, GFP_ATOMIC)) {
		/* Decrement cwnd here because we are sending
2182
		 * effectively two packets. */
J
John Heffner 已提交
2183
		tp->snd_cwnd--;
2184
		tcp_event_new_data_sent(sk, nskb);
J
John Heffner 已提交
2185 2186

		icsk->icsk_mtup.probe_size = tcp_mss_to_mtu(sk, nskb->len);
2187 2188
		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
2189 2190 2191 2192 2193 2194 2195

		return 1;
	}

	return -1;
}

2196 2197 2198 2199 2200 2201
static bool tcp_pacing_check(const struct sock *sk)
{
	return tcp_needs_internal_pacing(sk) &&
	       hrtimer_active(&tcp_sk(sk)->pacing_timer);
}

2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218
/* TCP Small Queues :
 * Control number of packets in qdisc/devices to two packets / or ~1 ms.
 * (These limits are doubled for retransmits)
 * This allows for :
 *  - better RTT estimation and ACK scheduling
 *  - faster recovery
 *  - high rates
 * Alas, some drivers / subsystems require a fair amount
 * of queued bytes to ensure line rate.
 * One example is wifi aggregation (802.11 AMPDU)
 */
static bool tcp_small_queue_check(struct sock *sk, const struct sk_buff *skb,
				  unsigned int factor)
{
	unsigned int limit;

	limit = max(2 * skb->truesize, sk->sk_pacing_rate >> 10);
2219 2220
	limit = min_t(u32, limit,
		      sock_net(sk)->ipv4.sysctl_tcp_limit_output_bytes);
2221 2222
	limit <<= factor;

2223
	if (refcount_read(&sk->sk_wmem_alloc) > limit) {
2224
		/* Always send skb if rtx queue is empty.
2225 2226 2227 2228
		 * No need to wait for TX completion to call us back,
		 * after softirq/tasklet schedule.
		 * This helps when TX completions are delayed too much.
		 */
2229
		if (tcp_rtx_queue_empty(sk))
2230 2231
			return false;

2232
		set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags);
2233 2234 2235 2236 2237
		/* It is possible TX completion already happened
		 * before we set TSQ_THROTTLED, so we must
		 * test again the condition.
		 */
		smp_mb__after_atomic();
2238
		if (refcount_read(&sk->sk_wmem_alloc) > limit)
2239 2240 2241 2242 2243
			return true;
	}
	return false;
}

2244 2245
static void tcp_chrono_set(struct tcp_sock *tp, const enum tcp_chrono new)
{
2246
	const u32 now = tcp_jiffies32;
2247
	enum tcp_chrono old = tp->chrono_type;
2248

2249 2250
	if (old > TCP_CHRONO_UNSPEC)
		tp->chrono_stat[old - 1] += now - tp->chrono_start;
2251 2252 2253 2254 2255 2256 2257 2258 2259
	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
2260 2261
	 * chronograph then the highest priority enum takes precedence
	 * over the other conditions. So that if something "more interesting"
2262 2263 2264 2265 2266 2267 2268 2269 2270 2271
	 * 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);

2272 2273 2274 2275 2276 2277 2278 2279

	/* 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.
	 */
2280
	if (tcp_rtx_and_write_queues_empty(sk))
2281 2282 2283
		tcp_chrono_set(tp, TCP_CHRONO_UNSPEC);
	else if (type == tp->chrono_type)
		tcp_chrono_set(tp, TCP_CHRONO_BUSY);
2284 2285
}

L
Linus Torvalds 已提交
2286 2287 2288 2289
/* This routine writes packets to the network.  It advances the
 * send_head.  This happens as incoming acks open up the remote
 * window for us.
 *
2290 2291 2292 2293
 * 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
2294 2295 2296
 * 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
2297 2298
 * 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
2299
 */
E
Eric Dumazet 已提交
2300 2301
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
2302 2303
{
	struct tcp_sock *tp = tcp_sk(sk);
2304
	struct sk_buff *skb;
2305 2306
	unsigned int tso_segs, sent_pkts;
	int cwnd_quota;
J
John Heffner 已提交
2307
	int result;
2308
	bool is_cwnd_limited = false, is_rwnd_limited = false;
E
Eric Dumazet 已提交
2309
	u32 max_segs;
L
Linus Torvalds 已提交
2310

2311
	sent_pkts = 0;
J
John Heffner 已提交
2312

2313
	tcp_mstamp_refresh(tp);
2314 2315 2316 2317
	if (!push_one) {
		/* Do MTU probing. */
		result = tcp_mtu_probe(sk);
		if (!result) {
E
Eric Dumazet 已提交
2318
			return false;
2319 2320 2321
		} else if (result > 0) {
			sent_pkts = 1;
		}
J
John Heffner 已提交
2322 2323
	}

2324
	max_segs = tcp_tso_segs(sk, mss_now);
2325
	while ((skb = tcp_send_head(sk))) {
2326 2327
		unsigned int limit;

2328 2329 2330
		if (tcp_pacing_check(sk))
			break;

2331
		tso_segs = tcp_init_tso_segs(skb, mss_now);
2332
		BUG_ON(!tso_segs);
2333

2334
		if (unlikely(tp->repair) && tp->repair_queue == TCP_SEND_QUEUE) {
E
Eric Dumazet 已提交
2335
			/* "skb_mstamp" is used as a start point for the retransmit timer */
2336
			tcp_update_skb_after_send(tp, skb);
2337
			goto repair; /* Skip network transmission */
2338
		}
2339

2340
		cwnd_quota = tcp_cwnd_test(tp, skb);
N
Nandita Dukkipati 已提交
2341 2342 2343 2344 2345 2346 2347
		if (!cwnd_quota) {
			if (push_one == 2)
				/* Force out a loss probe pkt. */
				cwnd_quota = 1;
			else
				break;
		}
2348

2349 2350
		if (unlikely(!tcp_snd_wnd_test(tp, skb, mss_now))) {
			is_rwnd_limited = true;
2351
			break;
2352
		}
2353

2354
		if (tso_segs == 1) {
2355 2356 2357 2358 2359
			if (unlikely(!tcp_nagle_test(tp, skb, mss_now,
						     (tcp_skb_is_last(sk, skb) ?
						      nonagle : TCP_NAGLE_PUSH))))
				break;
		} else {
2360
			if (!push_one &&
E
Eric Dumazet 已提交
2361 2362
			    tcp_tso_should_defer(sk, skb, &is_cwnd_limited,
						 max_segs))
2363 2364
				break;
		}
2365

E
Eric Dumazet 已提交
2366
		limit = mss_now;
2367
		if (tso_segs > 1 && !tcp_urg_mode(tp))
E
Eric Dumazet 已提交
2368 2369 2370 2371 2372 2373 2374
			limit = tcp_mss_split_point(sk, skb, mss_now,
						    min_t(unsigned int,
							  cwnd_quota,
							  max_segs),
						    nonagle);

		if (skb->len > limit &&
2375 2376
		    unlikely(tso_fragment(sk, TCP_FRAG_IN_WRITE_QUEUE,
					  skb, limit, mss_now, gfp)))
E
Eric Dumazet 已提交
2377 2378
			break;

2379 2380
		if (test_bit(TCP_TSQ_DEFERRED, &sk->sk_tsq_flags))
			clear_bit(TCP_TSQ_DEFERRED, &sk->sk_tsq_flags);
2381 2382
		if (tcp_small_queue_check(sk, skb, 0))
			break;
E
Eric Dumazet 已提交
2383

2384
		if (unlikely(tcp_transmit_skb(sk, skb, 1, gfp)))
2385
			break;
L
Linus Torvalds 已提交
2386

2387
repair:
2388 2389 2390
		/* Advance the send_head.  This one is sent out.
		 * This call will increment packets_out.
		 */
2391
		tcp_event_new_data_sent(sk, skb);
L
Linus Torvalds 已提交
2392

2393
		tcp_minshall_update(tp, mss_now, skb);
2394
		sent_pkts += tcp_skb_pcount(skb);
2395 2396 2397

		if (push_one)
			break;
2398
	}
L
Linus Torvalds 已提交
2399

2400 2401 2402 2403 2404
	if (is_rwnd_limited)
		tcp_chrono_start(sk, TCP_CHRONO_RWND_LIMITED);
	else
		tcp_chrono_stop(sk, TCP_CHRONO_RWND_LIMITED);

2405
	if (likely(sent_pkts)) {
2406 2407
		if (tcp_in_cwnd_reduction(sk))
			tp->prr_out += sent_pkts;
N
Nandita Dukkipati 已提交
2408 2409 2410 2411

		/* Send one loss probe per tail loss episode. */
		if (push_one != 2)
			tcp_schedule_loss_probe(sk);
2412
		is_cwnd_limited |= (tcp_packets_in_flight(tp) >= tp->snd_cwnd);
2413
		tcp_cwnd_validate(sk, is_cwnd_limited);
E
Eric Dumazet 已提交
2414
		return false;
L
Linus Torvalds 已提交
2415
	}
2416
	return !tp->packets_out && !tcp_write_queue_empty(sk);
N
Nandita Dukkipati 已提交
2417 2418 2419 2420 2421 2422
}

bool tcp_schedule_loss_probe(struct sock *sk)
{
	struct inet_connection_sock *icsk = inet_csk(sk);
	struct tcp_sock *tp = tcp_sk(sk);
2423
	u32 timeout, rto_delta_us;
2424
	int early_retrans;
N
Nandita Dukkipati 已提交
2425 2426 2427 2428

	/* Don't do any loss probe on a Fast Open connection before 3WHS
	 * finishes.
	 */
2429
	if (tp->fastopen_rsk)
N
Nandita Dukkipati 已提交
2430 2431
		return false;

2432
	early_retrans = sock_net(sk)->ipv4.sysctl_tcp_early_retrans;
N
Nandita Dukkipati 已提交
2433 2434 2435
	/* Schedule a loss probe in 2*RTT for SACK capable connections
	 * in Open state, that are either limited by cwnd or application.
	 */
2436
	if ((early_retrans != 3 && early_retrans != 4) ||
Y
Yuchung Cheng 已提交
2437 2438
	    !tp->packets_out || !tcp_is_sack(tp) ||
	    icsk->icsk_ca_state != TCP_CA_Open)
N
Nandita Dukkipati 已提交
2439 2440 2441
		return false;

	if ((tp->snd_cwnd > tcp_packets_in_flight(tp)) &&
2442
	     !tcp_write_queue_empty(sk))
N
Nandita Dukkipati 已提交
2443 2444
		return false;

2445
	/* Probe timeout is 2*rtt. Add minimum RTO to account
2446 2447
	 * 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
2448
	 */
2449 2450 2451 2452 2453 2454 2455 2456 2457
	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
2458

2459 2460 2461 2462
	/* If the RTO formula yields an earlier time, then use that time. */
	rto_delta_us = tcp_rto_delta_us(sk);  /* How far in future is RTO? */
	if (rto_delta_us > 0)
		timeout = min_t(u32, timeout, usecs_to_jiffies(rto_delta_us));
N
Nandita Dukkipati 已提交
2463 2464 2465 2466 2467 2468

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

2469 2470 2471 2472 2473 2474 2475
/* Thanks to skb fast clones, we can detect if a prior transmit of
 * a packet is still in a qdisc or driver queue.
 * In this case, there is very little point doing a retransmit !
 */
static bool skb_still_in_host_queue(const struct sock *sk,
				    const struct sk_buff *skb)
{
2476
	if (unlikely(skb_fclone_busy(sk, skb))) {
2477 2478
		NET_INC_STATS(sock_net(sk),
			      LINUX_MIB_TCPSPURIOUS_RTX_HOSTQUEUES);
2479 2480 2481 2482 2483
		return true;
	}
	return false;
}

2484
/* When probe timeout (PTO) fires, try send a new segment if possible, else
N
Nandita Dukkipati 已提交
2485 2486 2487 2488
 * retransmit the last segment.
 */
void tcp_send_loss_probe(struct sock *sk)
{
N
Nandita Dukkipati 已提交
2489
	struct tcp_sock *tp = tcp_sk(sk);
N
Nandita Dukkipati 已提交
2490 2491 2492 2493
	struct sk_buff *skb;
	int pcount;
	int mss = tcp_current_mss(sk);

2494
	skb = tcp_send_head(sk);
2495 2496 2497 2498 2499 2500
	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
2501
	}
2502
	skb = skb_rb_last(&sk->tcp_rtx_queue);
N
Nandita Dukkipati 已提交
2503

N
Nandita Dukkipati 已提交
2504 2505 2506 2507
	/* At most one outstanding TLP retransmission. */
	if (tp->tlp_high_seq)
		goto rearm_timer;

N
Nandita Dukkipati 已提交
2508 2509 2510 2511
	/* Retransmit last segment. */
	if (WARN_ON(!skb))
		goto rearm_timer;

2512 2513 2514
	if (skb_still_in_host_queue(sk, skb))
		goto rearm_timer;

N
Nandita Dukkipati 已提交
2515 2516 2517 2518 2519
	pcount = tcp_skb_pcount(skb);
	if (WARN_ON(!pcount))
		goto rearm_timer;

	if ((pcount > 1) && (skb->len > (pcount - 1) * mss)) {
2520 2521
		if (unlikely(tcp_fragment(sk, TCP_FRAG_IN_RTX_QUEUE, skb,
					  (pcount - 1) * mss, mss,
2522
					  GFP_ATOMIC)))
N
Nandita Dukkipati 已提交
2523
			goto rearm_timer;
2524
		skb = skb_rb_next(skb);
N
Nandita Dukkipati 已提交
2525 2526 2527 2528 2529
	}

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

2530
	if (__tcp_retransmit_skb(sk, skb, 1))
2531
		goto rearm_timer;
N
Nandita Dukkipati 已提交
2532

N
Nandita Dukkipati 已提交
2533
	/* Record snd_nxt for loss detection. */
2534
	tp->tlp_high_seq = tp->snd_nxt;
N
Nandita Dukkipati 已提交
2535

2536
probe_sent:
2537
	NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPLOSSPROBES);
2538 2539
	/* Reset s.t. tcp_rearm_rto will restart timer from now */
	inet_csk(sk)->icsk_pending = 0;
N
Nandita Dukkipati 已提交
2540
rearm_timer:
2541
	tcp_rearm_rto(sk);
L
Linus Torvalds 已提交
2542 2543
}

2544 2545 2546 2547
/* 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.
 */
2548 2549
void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
			       int nonagle)
2550
{
2551 2552 2553 2554 2555 2556 2557
	/* 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;

2558
	if (tcp_write_xmit(sk, cur_mss, nonagle, 0,
2559
			   sk_gfp_mask(sk, GFP_ATOMIC)))
2560
		tcp_check_probe_timer(sk);
2561 2562
}

2563 2564 2565 2566 2567
/* 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)
{
2568
	struct sk_buff *skb = tcp_send_head(sk);
2569 2570 2571

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

2572
	tcp_write_xmit(sk, mss_now, TCP_NAGLE_PUSH, 1, sk->sk_allocation);
2573 2574
}

L
Linus Torvalds 已提交
2575 2576
/* This function returns the amount that we can raise the
 * usable window based on the following constraints
2577
 *
L
Linus Torvalds 已提交
2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595
 * 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.
2596
 *
L
Linus Torvalds 已提交
2597 2598
 * Of course, if the sender implements sender side SWS prevention
 * then this will not be a problem.
2599
 *
L
Linus Torvalds 已提交
2600
 * BSD seems to make the following compromise:
2601
 *
L
Linus Torvalds 已提交
2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628
 *	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)
{
2629
	struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
2630
	struct tcp_sock *tp = tcp_sk(sk);
S
Stephen Hemminger 已提交
2631
	/* MSS for the peer's data.  Previous versions used mss_clamp
L
Linus Torvalds 已提交
2632 2633 2634 2635 2636
	 * 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
	 */
2637
	int mss = icsk->icsk_ack.rcv_mss;
L
Linus Torvalds 已提交
2638
	int free_space = tcp_space(sk);
2639 2640
	int allowed_space = tcp_full_space(sk);
	int full_space = min_t(int, tp->window_clamp, allowed_space);
L
Linus Torvalds 已提交
2641 2642
	int window;

2643
	if (unlikely(mss > full_space)) {
2644
		mss = full_space;
2645 2646 2647
		if (mss <= 0)
			return 0;
	}
2648
	if (free_space < (full_space >> 1)) {
2649
		icsk->icsk_ack.quick = 0;
L
Linus Torvalds 已提交
2650

2651
		if (tcp_under_memory_pressure(sk))
2652 2653
			tp->rcv_ssthresh = min(tp->rcv_ssthresh,
					       4U * tp->advmss);
L
Linus Torvalds 已提交
2654

2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667
		/* 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
2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683
			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.
		 */
2684
		window = ALIGN(window, (1 << tp->rx_opt.rcv_wscale));
L
Linus Torvalds 已提交
2685
	} else {
2686
		window = tp->rcv_wnd;
L
Linus Torvalds 已提交
2687 2688 2689 2690 2691 2692 2693 2694 2695
		/* 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)
2696
			window = rounddown(free_space, mss);
2697
		else if (mss == full_space &&
2698
			 free_space > window + (full_space >> 1))
2699
			window = free_space;
L
Linus Torvalds 已提交
2700 2701 2702 2703 2704
	}

	return window;
}

2705 2706
void tcp_skb_collapse_tstamp(struct sk_buff *skb,
			     const struct sk_buff *next_skb)
2707
{
2708 2709 2710
	if (unlikely(tcp_has_tx_tstamp(next_skb))) {
		const struct skb_shared_info *next_shinfo =
			skb_shinfo(next_skb);
2711 2712
		struct skb_shared_info *shinfo = skb_shinfo(skb);

2713
		shinfo->tx_flags |= next_shinfo->tx_flags & SKBTX_ANY_TSTAMP;
2714
		shinfo->tskey = next_shinfo->tskey;
2715 2716
		TCP_SKB_CB(skb)->txstamp_ack |=
			TCP_SKB_CB(next_skb)->txstamp_ack;
2717 2718 2719
	}
}

2720
/* Collapses two adjacent SKB's during retransmission. */
2721
static bool tcp_collapse_retrans(struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
2722 2723
{
	struct tcp_sock *tp = tcp_sk(sk);
2724
	struct sk_buff *next_skb = skb_rb_next(skb);
2725
	int skb_size, next_skb_size;
L
Linus Torvalds 已提交
2726

2727 2728
	skb_size = skb->len;
	next_skb_size = next_skb->len;
L
Linus Torvalds 已提交
2729

2730
	BUG_ON(tcp_skb_pcount(skb) != 1 || tcp_skb_pcount(next_skb) != 1);
2731

2732 2733 2734 2735 2736 2737 2738
	if (next_skb_size) {
		if (next_skb_size <= skb_availroom(skb))
			skb_copy_bits(next_skb, 0, skb_put(skb, next_skb_size),
				      next_skb_size);
		else if (!skb_shift(skb, next_skb, next_skb_size))
			return false;
	}
2739
	tcp_highest_sack_replace(sk, next_skb, skb);
L
Linus Torvalds 已提交
2740

2741 2742
	if (next_skb->ip_summed == CHECKSUM_PARTIAL)
		skb->ip_summed = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
2743

2744 2745
	if (skb->ip_summed != CHECKSUM_PARTIAL)
		skb->csum = csum_block_add(skb->csum, next_skb->csum, skb_size);
L
Linus Torvalds 已提交
2746

2747 2748
	/* 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
2749

2750
	/* Merge over control information. This moves PSH/FIN etc. over */
E
Eric Dumazet 已提交
2751
	TCP_SKB_CB(skb)->tcp_flags |= TCP_SKB_CB(next_skb)->tcp_flags;
2752 2753 2754 2755 2756

	/* 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;
2757
	TCP_SKB_CB(skb)->eor = TCP_SKB_CB(next_skb)->eor;
2758 2759

	/* changed transmit queue under us so clear hints */
2760 2761 2762
	tcp_clear_retrans_hints_partial(tp);
	if (next_skb == tp->retransmit_skb_hint)
		tp->retransmit_skb_hint = skb;
2763

2764 2765
	tcp_adjust_pcount(sk, next_skb, tcp_skb_pcount(next_skb));

2766 2767
	tcp_skb_collapse_tstamp(skb, next_skb);

2768
	tcp_rtx_queue_unlink_and_free(next_skb, sk);
2769
	return true;
L
Linus Torvalds 已提交
2770 2771
}

2772
/* Check if coalescing SKBs is legal. */
E
Eric Dumazet 已提交
2773
static bool tcp_can_collapse(const struct sock *sk, const struct sk_buff *skb)
2774 2775
{
	if (tcp_skb_pcount(skb) > 1)
E
Eric Dumazet 已提交
2776
		return false;
2777
	if (skb_cloned(skb))
E
Eric Dumazet 已提交
2778
		return false;
E
Eric Dumazet 已提交
2779
	/* Some heuristics for collapsing over SACK'd could be invented */
2780
	if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)
E
Eric Dumazet 已提交
2781
		return false;
2782

E
Eric Dumazet 已提交
2783
	return true;
2784 2785
}

2786 2787 2788
/* Collapse packets in the retransmit queue to make to create
 * less packets on the wire. This is only done on retransmission.
 */
2789 2790 2791 2792 2793
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
2794
	bool first = true;
2795

2796
	if (!sock_net(sk)->ipv4.sysctl_tcp_retrans_collapse)
2797
		return;
E
Eric Dumazet 已提交
2798
	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)
2799 2800
		return;

2801
	skb_rbtree_walk_from_safe(skb, tmp) {
2802 2803 2804
		if (!tcp_can_collapse(sk, skb))
			break;

2805 2806 2807
		if (!tcp_skb_can_collapse_to(to))
			break;

2808 2809 2810
		space -= skb->len;

		if (first) {
E
Eric Dumazet 已提交
2811
			first = false;
2812 2813 2814 2815 2816 2817 2818 2819 2820
			continue;
		}

		if (space < 0)
			break;

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

2821 2822
		if (!tcp_collapse_retrans(sk, to))
			break;
2823 2824 2825
	}
}

L
Linus Torvalds 已提交
2826 2827 2828 2829
/* 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.
 */
2830
int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs)
L
Linus Torvalds 已提交
2831
{
J
John Heffner 已提交
2832
	struct inet_connection_sock *icsk = inet_csk(sk);
2833
	struct tcp_sock *tp = tcp_sk(sk);
2834
	unsigned int cur_mss;
2835 2836
	int diff, len, err;

L
Linus Torvalds 已提交
2837

2838 2839
	/* Inconclusive MTU probe */
	if (icsk->icsk_mtup.probe_size)
J
John Heffner 已提交
2840 2841
		icsk->icsk_mtup.probe_size = 0;

L
Linus Torvalds 已提交
2842
	/* Do not sent more than we queued. 1/4 is reserved for possible
S
Stephen Hemminger 已提交
2843
	 * copying overhead: fragmentation, tunneling, mangling etc.
L
Linus Torvalds 已提交
2844
	 */
2845
	if (refcount_read(&sk->sk_wmem_alloc) >
2846 2847
	    min_t(u32, sk->sk_wmem_queued + (sk->sk_wmem_queued >> 2),
		  sk->sk_sndbuf))
L
Linus Torvalds 已提交
2848 2849
		return -EAGAIN;

2850 2851 2852
	if (skb_still_in_host_queue(sk, skb))
		return -EBUSY;

L
Linus Torvalds 已提交
2853 2854 2855 2856 2857 2858 2859
	if (before(TCP_SKB_CB(skb)->seq, tp->snd_una)) {
		if (before(TCP_SKB_CB(skb)->end_seq, tp->snd_una))
			BUG();
		if (tcp_trim_head(sk, skb, tp->snd_una - TCP_SKB_CB(skb)->seq))
			return -ENOMEM;
	}

2860 2861 2862
	if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk))
		return -EHOSTUNREACH; /* Routing failure or similar. */

I
Ilpo Järvinen 已提交
2863
	cur_mss = tcp_current_mss(sk);
2864

L
Linus Torvalds 已提交
2865 2866 2867 2868 2869
	/* 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.
	 */
2870 2871
	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
2872 2873
		return -EAGAIN;

2874 2875
	len = cur_mss * segs;
	if (skb->len > len) {
2876 2877
		if (tcp_fragment(sk, TCP_FRAG_IN_RTX_QUEUE, skb, len,
				 cur_mss, GFP_ATOMIC))
L
Linus Torvalds 已提交
2878
			return -ENOMEM; /* We'll try again later. */
2879
	} else {
2880 2881
		if (skb_unclone(skb, GFP_ATOMIC))
			return -ENOMEM;
2882

2883 2884 2885 2886 2887 2888 2889
		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
2890 2891
	}

2892 2893 2894 2895
	/* 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);

2896 2897 2898 2899 2900 2901 2902
	/* 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;

2903 2904 2905 2906 2907 2908
	/* 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)) {
2909 2910
		struct sk_buff *nskb;

2911 2912 2913 2914 2915 2916
		tcp_skb_tsorted_save(skb) {
			nskb = __pskb_copy(skb, MAX_TCP_HEADER, GFP_ATOMIC);
			err = nskb ? tcp_transmit_skb(sk, nskb, 0, GFP_ATOMIC) :
				     -ENOBUFS;
		} tcp_skb_tsorted_restore(skb);

2917
		if (!err) {
2918
			tcp_update_skb_after_send(tp, skb);
2919 2920
			tcp_rate_skb_sent(sk, skb);
		}
E
Eric Dumazet 已提交
2921
	} else {
2922
		err = tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
E
Eric Dumazet 已提交
2923
	}
2924

2925
	if (likely(!err)) {
2926
		TCP_SKB_CB(skb)->sacked |= TCPCB_EVER_RETRANS;
2927
		trace_tcp_retransmit_skb(sk, skb);
2928 2929
	} else if (err != -EBUSY) {
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPRETRANSFAIL);
2930
	}
2931
	return err;
2932 2933
}

2934
int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs)
2935 2936
{
	struct tcp_sock *tp = tcp_sk(sk);
2937
	int err = __tcp_retransmit_skb(sk, skb, segs);
L
Linus Torvalds 已提交
2938 2939 2940

	if (err == 0) {
#if FASTRETRANS_DEBUG > 0
2941
		if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) {
2942
			net_dbg_ratelimited("retrans_out leaked\n");
L
Linus Torvalds 已提交
2943 2944 2945 2946 2947 2948 2949
		}
#endif
		TCP_SKB_CB(skb)->sacked |= TCPCB_RETRANS;
		tp->retrans_out += tcp_skb_pcount(skb);

		/* Save stamp of the first retransmit. */
		if (!tp->retrans_stamp)
E
Eric Dumazet 已提交
2950
			tp->retrans_stamp = tcp_skb_timestamp(skb);
L
Linus Torvalds 已提交
2951 2952

	}
Y
Yuchung Cheng 已提交
2953 2954 2955 2956

	if (tp->undo_retrans < 0)
		tp->undo_retrans = 0;
	tp->undo_retrans += tcp_skb_pcount(skb);
L
Linus Torvalds 已提交
2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969
	return err;
}

/* This gets called after a retransmit timeout, and the initially
 * retransmitted data is acknowledged.  It tries to continue
 * resending the rest of the retransmit queue, until either
 * we've sent it all or the congestion window limit is reached.
 * If doing SACK, the first ACK which comes back for a timeout
 * based retransmit packet might feed us FACK information again.
 * If so, we use it to avoid unnecessarily retransmissions.
 */
void tcp_xmit_retransmit_queue(struct sock *sk)
{
2970
	const struct inet_connection_sock *icsk = inet_csk(sk);
2971
	struct sk_buff *skb, *rtx_head, *hole = NULL;
L
Linus Torvalds 已提交
2972
	struct tcp_sock *tp = tcp_sk(sk);
2973
	u32 max_segs;
2974
	int mib_idx;
2975

2976 2977 2978
	if (!tp->packets_out)
		return;

2979 2980
	rtx_head = tcp_rtx_queue_head(sk);
	skb = tp->retransmit_skb_hint ?: rtx_head;
2981
	max_segs = tcp_tso_segs(sk, tcp_current_mss(sk));
2982
	skb_rbtree_walk_from(skb) {
E
Eric Dumazet 已提交
2983
		__u8 sacked;
2984
		int segs;
L
Linus Torvalds 已提交
2985

2986 2987 2988
		if (tcp_pacing_check(sk))
			break;

2989
		/* we could do better than to assign each time */
2990
		if (!hole)
2991
			tp->retransmit_skb_hint = skb;
2992

2993 2994
		segs = tp->snd_cwnd - tcp_packets_in_flight(tp);
		if (segs <= 0)
2995
			return;
E
Eric Dumazet 已提交
2996
		sacked = TCP_SKB_CB(skb)->sacked;
2997 2998 2999 3000
		/* 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
3001

3002 3003
		if (tp->retrans_out >= tp->lost_out) {
			break;
3004
		} else if (!(sacked & TCPCB_LOST)) {
3005
			if (!hole && !(sacked & (TCPCB_SACKED_RETRANS|TCPCB_SACKED_ACKED)))
3006 3007
				hole = skb;
			continue;
L
Linus Torvalds 已提交
3008

3009 3010 3011 3012 3013 3014
		} 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
3015

3016
		if (sacked & (TCPCB_SACKED_ACKED|TCPCB_SACKED_RETRANS))
L
Linus Torvalds 已提交
3017 3018
			continue;

3019 3020 3021
		if (tcp_small_queue_check(sk, skb, 1))
			return;

3022
		if (tcp_retransmit_skb(sk, skb, segs))
3023
			return;
3024

3025
		NET_ADD_STATS(sock_net(sk), mib_idx, tcp_skb_pcount(skb));
L
Linus Torvalds 已提交
3026

3027
		if (tcp_in_cwnd_reduction(sk))
3028 3029
			tp->prr_out += tcp_skb_pcount(skb);

3030
		if (skb == rtx_head &&
3031
		    icsk->icsk_pending != ICSK_TIME_REO_TIMEOUT)
3032 3033 3034
			inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
						  inet_csk(sk)->icsk_rto,
						  TCP_RTO_MAX);
L
Linus Torvalds 已提交
3035 3036 3037
	}
}

3038 3039
/* We allow to exceed memory limits for FIN packets to expedite
 * connection tear down and (memory) recovery.
3040 3041
 * Otherwise tcp_send_fin() could be tempted to either delay FIN
 * or even be forced to close flow without any FIN.
3042 3043
 * In general, we want to allow one skb per socket to avoid hangs
 * with edge trigger epoll()
3044
 */
3045
void sk_forced_mem_schedule(struct sock *sk, int size)
3046
{
3047
	int amt;
3048 3049 3050 3051 3052

	if (size <= sk->sk_forward_alloc)
		return;
	amt = sk_mem_pages(size);
	sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
3053 3054
	sk_memory_allocated_add(sk, amt);

3055 3056
	if (mem_cgroup_sockets_enabled && sk->sk_memcg)
		mem_cgroup_charge_skmem(sk->sk_memcg, amt);
3057 3058
}

3059 3060
/* 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
3061 3062 3063
 */
void tcp_send_fin(struct sock *sk)
{
3064
	struct sk_buff *skb, *tskb = tcp_write_queue_tail(sk);
3065 3066
	struct tcp_sock *tp = tcp_sk(sk);

3067 3068 3069 3070
	/* 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
3071
	 */
3072 3073 3074 3075
	if (!tskb && tcp_under_memory_pressure(sk))
		tskb = skb_rb_last(&sk->tcp_rtx_queue);

	if (tskb) {
3076 3077 3078
coalesce:
		TCP_SKB_CB(tskb)->tcp_flags |= TCPHDR_FIN;
		TCP_SKB_CB(tskb)->end_seq++;
L
Linus Torvalds 已提交
3079
		tp->write_seq++;
3080
		if (tcp_write_queue_empty(sk)) {
3081 3082 3083 3084 3085 3086 3087 3088 3089
			/* This means tskb was already sent.
			 * Pretend we included the FIN on previous transmit.
			 * We need to set tp->snd_nxt to the value it would have
			 * if FIN had been sent. This is because retransmit path
			 * does not change tp->snd_nxt.
			 */
			tp->snd_nxt++;
			return;
		}
L
Linus Torvalds 已提交
3090
	} else {
3091 3092 3093 3094 3095
		skb = alloc_skb_fclone(MAX_TCP_HEADER, sk->sk_allocation);
		if (unlikely(!skb)) {
			if (tskb)
				goto coalesce;
			return;
L
Linus Torvalds 已提交
3096
		}
3097
		INIT_LIST_HEAD(&skb->tcp_tsorted_anchor);
3098
		skb_reserve(skb, MAX_TCP_HEADER);
3099
		sk_forced_mem_schedule(sk, skb->truesize);
L
Linus Torvalds 已提交
3100
		/* FIN eats a sequence byte, write_seq advanced by tcp_queue_skb(). */
3101
		tcp_init_nondata_skb(skb, tp->write_seq,
C
Changli Gao 已提交
3102
				     TCPHDR_ACK | TCPHDR_FIN);
L
Linus Torvalds 已提交
3103 3104
		tcp_queue_skb(sk, skb);
	}
3105
	__tcp_push_pending_frames(sk, tcp_current_mss(sk), TCP_NAGLE_OFF);
L
Linus Torvalds 已提交
3106 3107 3108 3109 3110
}

/* 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
3111
 * by RFC 2525, section 2.17.  -DaveM
L
Linus Torvalds 已提交
3112
 */
A
Al Viro 已提交
3113
void tcp_send_active_reset(struct sock *sk, gfp_t priority)
L
Linus Torvalds 已提交
3114 3115 3116
{
	struct sk_buff *skb;

3117 3118
	TCP_INC_STATS(sock_net(sk), TCP_MIB_OUTRSTS);

L
Linus Torvalds 已提交
3119 3120 3121
	/* 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
3122
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTFAILED);
L
Linus Torvalds 已提交
3123 3124 3125 3126 3127
		return;
	}

	/* Reserve space for headers and prepare control bits. */
	skb_reserve(skb, MAX_TCP_HEADER);
3128
	tcp_init_nondata_skb(skb, tcp_acceptable_seq(sk),
C
Changli Gao 已提交
3129
			     TCPHDR_ACK | TCPHDR_RST);
3130
	tcp_mstamp_refresh(tcp_sk(sk));
L
Linus Torvalds 已提交
3131
	/* Send it off. */
3132
	if (tcp_transmit_skb(sk, skb, 0, priority))
P
Pavel Emelyanov 已提交
3133
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTFAILED);
3134 3135 3136 3137 3138

	/* 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
3139 3140
}

3141 3142
/* 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
3143 3144 3145 3146 3147 3148
 * 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)
{
3149
	struct sk_buff *skb;
L
Linus Torvalds 已提交
3150

3151
	skb = tcp_rtx_queue_head(sk);
3152
	if (!skb || !(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
3153
		pr_err("%s: wrong queue state\n", __func__);
L
Linus Torvalds 已提交
3154 3155
		return -EFAULT;
	}
E
Eric Dumazet 已提交
3156
	if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)) {
L
Linus Torvalds 已提交
3157
		if (skb_cloned(skb)) {
3158 3159 3160 3161 3162
			struct sk_buff *nskb;

			tcp_skb_tsorted_save(skb) {
				nskb = skb_copy(skb, GFP_ATOMIC);
			} tcp_skb_tsorted_restore(skb);
3163
			if (!nskb)
L
Linus Torvalds 已提交
3164
				return -ENOMEM;
3165
			INIT_LIST_HEAD(&nskb->tcp_tsorted_anchor);
3166
			tcp_rtx_queue_unlink_and_free(skb, sk);
3167
			__skb_header_release(nskb);
3168
			tcp_rbtree_insert(&sk->tcp_rtx_queue, nskb);
3169 3170
			sk->sk_wmem_queued += nskb->truesize;
			sk_mem_charge(sk, nskb->truesize);
L
Linus Torvalds 已提交
3171 3172 3173
			skb = nskb;
		}

E
Eric Dumazet 已提交
3174
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ACK;
3175
		tcp_ecn_send_synack(sk, skb);
L
Linus Torvalds 已提交
3176
	}
3177
	return tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
L
Linus Torvalds 已提交
3178 3179
}

3180 3181 3182 3183 3184 3185 3186 3187 3188
/**
 * tcp_make_synack - Prepare a SYN-ACK.
 * sk: listener socket
 * dst: dst entry attached to the SYNACK
 * req: request_sock pointer
 *
 * Allocate one skb and build a SYNACK packet.
 * @dst is consumed : Caller should not use it again.
 */
3189
struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
3190
				struct request_sock *req,
3191
				struct tcp_fastopen_cookie *foc,
3192
				enum tcp_synack_type synack_type)
L
Linus Torvalds 已提交
3193
{
3194
	struct inet_request_sock *ireq = inet_rsk(req);
3195
	const struct tcp_sock *tp = tcp_sk(sk);
E
Eric Dumazet 已提交
3196
	struct tcp_md5sig_key *md5 = NULL;
3197 3198
	struct tcp_out_options opts;
	struct sk_buff *skb;
3199
	int tcp_header_size;
3200
	struct tcphdr *th;
3201
	int mss;
L
Linus Torvalds 已提交
3202

3203
	skb = alloc_skb(MAX_TCP_HEADER, GFP_ATOMIC);
3204 3205
	if (unlikely(!skb)) {
		dst_release(dst);
L
Linus Torvalds 已提交
3206
		return NULL;
3207
	}
L
Linus Torvalds 已提交
3208 3209 3210
	/* Reserve space for headers. */
	skb_reserve(skb, MAX_TCP_HEADER);

3211 3212
	switch (synack_type) {
	case TCP_SYNACK_NORMAL:
3213
		skb_set_owner_w(skb, req_to_sk(req));
3214 3215 3216 3217 3218 3219 3220
		break;
	case TCP_SYNACK_COOKIE:
		/* Under synflood, we do not attach skb to a socket,
		 * to avoid false sharing.
		 */
		break;
	case TCP_SYNACK_FASTOPEN:
3221 3222 3223 3224 3225
		/* 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);
3226
		break;
3227
	}
3228
	skb_dst_set(skb, dst);
L
Linus Torvalds 已提交
3229

E
Eric Dumazet 已提交
3230
	mss = tcp_mss_clamp(tp, dst_metric_advmss(dst));
3231

A
Adam Langley 已提交
3232
	memset(&opts, 0, sizeof(opts));
3233 3234
#ifdef CONFIG_SYN_COOKIES
	if (unlikely(req->cookie_ts))
3235
		skb->skb_mstamp = cookie_init_timestamp(req);
3236 3237
	else
#endif
3238
		skb->skb_mstamp = tcp_clock_us();
E
Eric Dumazet 已提交
3239 3240 3241

#ifdef CONFIG_TCP_MD5SIG
	rcu_read_lock();
3242
	md5 = tcp_rsk(req)->af_specific->req_md5_lookup(sk, req_to_sk(req));
E
Eric Dumazet 已提交
3243
#endif
3244
	skb_set_hash(skb, tcp_rsk(req)->txhash, PKT_HASH_TYPE_L4);
3245 3246
	tcp_header_size = tcp_synack_options(sk, req, mss, skb, &opts, md5,
					     foc) + sizeof(*th);
3247

3248 3249
	skb_push(skb, tcp_header_size);
	skb_reset_transport_header(skb);
L
Linus Torvalds 已提交
3250

3251
	th = (struct tcphdr *)skb->data;
L
Linus Torvalds 已提交
3252 3253 3254
	memset(th, 0, sizeof(struct tcphdr));
	th->syn = 1;
	th->ack = 1;
3255
	tcp_ecn_make_synack(req, th);
E
Eric Dumazet 已提交
3256
	th->source = htons(ireq->ir_num);
3257
	th->dest = ireq->ir_rmt_port;
3258
	skb->mark = ireq->ir_mark;
3259 3260
	skb->ip_summed = CHECKSUM_PARTIAL;
	th->seq = htonl(tcp_rsk(req)->snt_isn);
3261 3262
	/* 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
3263 3264

	/* RFC1323: The window in SYN & SYN/ACK segments is never scaled. */
3265
	th->window = htons(min(req->rsk_rcv_wnd, 65535U));
3266
	tcp_options_write((__be32 *)(th + 1), NULL, &opts);
L
Linus Torvalds 已提交
3267
	th->doff = (tcp_header_size >> 2);
E
Eric Dumazet 已提交
3268
	__TCP_INC_STATS(sock_net(sk), TCP_MIB_OUTSEGS);
3269 3270 3271

#ifdef CONFIG_TCP_MD5SIG
	/* Okay, we have all we need - do the md5 hash if needed */
E
Eric Dumazet 已提交
3272
	if (md5)
3273
		tcp_rsk(req)->af_specific->calc_md5_hash(opts.hash_location,
3274
					       md5, req_to_sk(req), skb);
E
Eric Dumazet 已提交
3275
	rcu_read_unlock();
3276 3277
#endif

3278
	/* Do not fool tcpdump (if any), clean our debris */
T
Thomas Gleixner 已提交
3279
	skb->tstamp = 0;
L
Linus Torvalds 已提交
3280 3281
	return skb;
}
E
Eric Dumazet 已提交
3282
EXPORT_SYMBOL(tcp_make_synack);
L
Linus Torvalds 已提交
3283

3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302
static void tcp_ca_dst_init(struct sock *sk, const struct dst_entry *dst)
{
	struct inet_connection_sock *icsk = inet_csk(sk);
	const struct tcp_congestion_ops *ca;
	u32 ca_key = dst_metric(dst, RTAX_CC_ALGO);

	if (ca_key == TCP_CA_UNSPEC)
		return;

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

3303
/* Do all connect socket setups that can be done AF independent. */
3304
static void tcp_connect_init(struct sock *sk)
L
Linus Torvalds 已提交
3305
{
3306
	const struct dst_entry *dst = __sk_dst_get(sk);
L
Linus Torvalds 已提交
3307 3308
	struct tcp_sock *tp = tcp_sk(sk);
	__u8 rcv_wscale;
3309
	u32 rcv_wnd;
L
Linus Torvalds 已提交
3310 3311 3312 3313

	/* 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.
	 */
3314 3315 3316
	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
3317

3318
#ifdef CONFIG_TCP_MD5SIG
3319
	if (tp->af_specific->md5_lookup(sk, sk))
3320 3321 3322
		tp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
#endif

L
Linus Torvalds 已提交
3323 3324 3325 3326
	/* 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
3327
	tcp_mtup_init(sk);
L
Linus Torvalds 已提交
3328 3329
	tcp_sync_mss(sk, dst_mtu(dst));

3330 3331
	tcp_ca_dst_init(sk, dst);

L
Linus Torvalds 已提交
3332 3333
	if (!tp->window_clamp)
		tp->window_clamp = dst_metric(dst, RTAX_WINDOW);
E
Eric Dumazet 已提交
3334
	tp->advmss = tcp_mss_clamp(tp, dst_metric_advmss(dst));
3335

L
Linus Torvalds 已提交
3336 3337
	tcp_initialize_rcv_mss(sk);

3338 3339 3340 3341 3342
	/* 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);

3343 3344 3345 3346
	rcv_wnd = tcp_rwnd_init_bpf(sk);
	if (rcv_wnd == 0)
		rcv_wnd = dst_metric(dst, RTAX_INITRWND);

3347
	tcp_select_initial_window(sk, tcp_full_space(sk),
L
Linus Torvalds 已提交
3348 3349 3350
				  tp->advmss - (tp->rx_opt.ts_recent_stamp ? tp->tcp_header_len - sizeof(struct tcphdr) : 0),
				  &tp->rcv_wnd,
				  &tp->window_clamp,
3351
				  sock_net(sk)->ipv4.sysctl_tcp_window_scaling,
3352
				  &rcv_wscale,
3353
				  rcv_wnd);
L
Linus Torvalds 已提交
3354 3355 3356 3357 3358 3359 3360

	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;
3361
	tcp_init_wl(tp, 0);
L
Linus Torvalds 已提交
3362 3363
	tp->snd_una = tp->write_seq;
	tp->snd_sml = tp->write_seq;
I
Ilpo Järvinen 已提交
3364
	tp->snd_up = tp->write_seq;
P
Pavel Emelyanov 已提交
3365
	tp->snd_nxt = tp->write_seq;
P
Pavel Emelyanov 已提交
3366 3367 3368

	if (likely(!tp->repair))
		tp->rcv_nxt = 0;
3369
	else
3370
		tp->rcv_tstamp = tcp_jiffies32;
P
Pavel Emelyanov 已提交
3371 3372
	tp->rcv_wup = tp->rcv_nxt;
	tp->copied_seq = tp->rcv_nxt;
L
Linus Torvalds 已提交
3373

3374
	inet_csk(sk)->icsk_rto = tcp_timeout_init(sk);
3375
	inet_csk(sk)->icsk_retransmits = 0;
L
Linus Torvalds 已提交
3376 3377 3378
	tcp_clear_retrans(tp);
}

3379 3380 3381 3382 3383 3384
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;
3385
	__skb_header_release(skb);
3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402
	sk->sk_wmem_queued += skb->truesize;
	sk_mem_charge(sk, skb->truesize);
	tp->write_seq = tcb->end_seq;
	tp->packets_out += tcp_skb_pcount(skb);
}

/* Build and send a SYN with data and (cached) Fast Open cookie. However,
 * queue a data-only packet after the regular SYN, such that regular SYNs
 * are retransmitted on timeouts. Also if the remote SYN-ACK acknowledges
 * only the SYN sequence, the data are retransmitted in the first ACK.
 * If cookie is not cached or other error occurs, falls back to send a
 * regular SYN with Fast Open cookie request option.
 */
static int tcp_send_syn_data(struct sock *sk, struct sk_buff *syn)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct tcp_fastopen_request *fo = tp->fastopen_req;
3403
	int space, err = 0;
3404
	struct sk_buff *syn_data;
3405

3406
	tp->rx_opt.mss_clamp = tp->advmss;  /* If MSS is not cached */
3407
	if (!tcp_fastopen_cookie_check(sk, &tp->rx_opt.mss_clamp, &fo->cookie))
3408 3409 3410 3411 3412 3413
		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
3414 3415
	tp->rx_opt.mss_clamp = tcp_mss_clamp(tp, tp->rx_opt.mss_clamp);

3416
	space = __tcp_mtu_to_mss(sk, inet_csk(sk)->icsk_pmtu_cookie) -
3417 3418
		MAX_TCP_OPTION_SPACE;

3419 3420 3421 3422 3423
	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));

3424
	syn_data = sk_stream_alloc_skb(sk, space, sk->sk_allocation, false);
3425
	if (!syn_data)
3426
		goto fallback;
3427 3428
	syn_data->ip_summed = CHECKSUM_PARTIAL;
	memcpy(syn_data->cb, syn->cb, sizeof(syn->cb));
3429 3430 3431 3432
	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
3433
			tcp_skb_tsorted_anchor_cleanup(syn_data);
3434 3435 3436 3437 3438 3439 3440
			kfree_skb(syn_data);
			goto fallback;
		}
		if (copied != space) {
			skb_trim(syn_data, copied);
			space = copied;
		}
3441
	}
3442 3443 3444 3445
	/* No more data pending in inet_wait_for_connect() */
	if (space == fo->size)
		fo->data = NULL;
	fo->copied = space;
3446

3447
	tcp_connect_queue_skb(sk, syn_data);
3448 3449
	if (syn_data->len)
		tcp_chrono_start(sk, TCP_CHRONO_BUSY);
3450

3451
	err = tcp_transmit_skb(sk, syn_data, 1, sk->sk_allocation);
3452

3453
	syn->skb_mstamp = syn_data->skb_mstamp;
E
Eric Dumazet 已提交
3454

3455 3456 3457 3458 3459 3460 3461 3462
	/* 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) {
3463
		tp->syn_data = (fo->copied > 0);
3464
		tcp_rbtree_insert(&sk->tcp_rtx_queue, syn_data);
3465
		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPORIGDATASENT);
3466 3467 3468
		goto done;
	}

3469 3470
	/* data was not sent, put it in write_queue */
	__skb_queue_tail(&sk->sk_write_queue, syn_data);
3471 3472
	tp->packets_out -= tcp_skb_pcount(syn_data);

3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484
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;
}

3485
/* Build a SYN and send it off. */
L
Linus Torvalds 已提交
3486 3487 3488 3489
int tcp_connect(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *buff;
3490
	int err;
L
Linus Torvalds 已提交
3491

3492
	tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_CONNECT_CB);
3493 3494 3495 3496

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

L
Linus Torvalds 已提交
3497 3498
	tcp_connect_init(sk);

A
Andrey Vagin 已提交
3499 3500 3501 3502 3503
	if (unlikely(tp->repair)) {
		tcp_finish_connect(sk, NULL);
		return 0;
	}

3504
	buff = sk_stream_alloc_skb(sk, 0, sk->sk_allocation, true);
3505
	if (unlikely(!buff))
L
Linus Torvalds 已提交
3506 3507
		return -ENOBUFS;

C
Changli Gao 已提交
3508
	tcp_init_nondata_skb(buff, tp->write_seq++, TCPHDR_SYN);
3509 3510
	tcp_mstamp_refresh(tp);
	tp->retrans_stamp = tcp_time_stamp(tp);
3511
	tcp_connect_queue_skb(sk, buff);
3512
	tcp_ecn_send_syn(sk, buff);
3513
	tcp_rbtree_insert(&sk->tcp_rtx_queue, buff);
L
Linus Torvalds 已提交
3514

3515 3516 3517
	/* 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);
3518 3519
	if (err == -ECONNREFUSED)
		return err;
3520 3521 3522 3523 3524 3525

	/* We change tp->snd_nxt after the tcp_transmit_skb() call
	 * in order to make this packet get counted in tcpOutSegs.
	 */
	tp->snd_nxt = tp->write_seq;
	tp->pushed_seq = tp->write_seq;
3526 3527 3528 3529 3530
	buff = tcp_send_head(sk);
	if (unlikely(buff)) {
		tp->snd_nxt	= TCP_SKB_CB(buff)->seq;
		tp->pushed_seq	= TCP_SKB_CB(buff)->seq;
	}
P
Pavel Emelyanov 已提交
3531
	TCP_INC_STATS(sock_net(sk), TCP_MIB_ACTIVEOPENS);
L
Linus Torvalds 已提交
3532 3533

	/* Timer for repeating the SYN until an answer. */
3534 3535
	inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
				  inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
L
Linus Torvalds 已提交
3536 3537
	return 0;
}
E
Eric Dumazet 已提交
3538
EXPORT_SYMBOL(tcp_connect);
L
Linus Torvalds 已提交
3539 3540 3541 3542 3543 3544 3545

/* 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)
{
3546 3547
	struct inet_connection_sock *icsk = inet_csk(sk);
	int ato = icsk->icsk_ack.ato;
L
Linus Torvalds 已提交
3548 3549
	unsigned long timeout;

3550 3551
	tcp_ca_event(sk, CA_EVENT_DELAYED_ACK);

L
Linus Torvalds 已提交
3552
	if (ato > TCP_DELACK_MIN) {
3553
		const struct tcp_sock *tp = tcp_sk(sk);
3554
		int max_ato = HZ / 2;
L
Linus Torvalds 已提交
3555

3556 3557
		if (icsk->icsk_ack.pingpong ||
		    (icsk->icsk_ack.pending & ICSK_ACK_PUSHED))
L
Linus Torvalds 已提交
3558 3559 3560 3561 3562
			max_ato = TCP_DELACK_MAX;

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

		/* If some rtt estimate is known, use it to bound delayed ack.
3563
		 * Do not use inet_csk(sk)->icsk_rto here, use results of rtt measurements
L
Linus Torvalds 已提交
3564 3565
		 * directly.
		 */
3566 3567 3568
		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
3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580

			if (rtt < max_ato)
				max_ato = rtt;
		}

		ato = min(ato, max_ato);
	}

	/* Stay within the limit we were given */
	timeout = jiffies + ato;

	/* Use new timeout only if there wasn't a older one earlier. */
3581
	if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
L
Linus Torvalds 已提交
3582 3583 3584
		/* If delack timer was blocked or is about to expire,
		 * send ACK now.
		 */
3585 3586
		if (icsk->icsk_ack.blocked ||
		    time_before_eq(icsk->icsk_ack.timeout, jiffies + (ato >> 2))) {
L
Linus Torvalds 已提交
3587 3588 3589 3590
			tcp_send_ack(sk);
			return;
		}

3591 3592
		if (!time_before(timeout, icsk->icsk_ack.timeout))
			timeout = icsk->icsk_ack.timeout;
L
Linus Torvalds 已提交
3593
	}
3594 3595 3596
	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
3597 3598 3599 3600 3601
}

/* This routine sends an ack and also updates the window. */
void tcp_send_ack(struct sock *sk)
{
3602
	struct sk_buff *buff;
L
Linus Torvalds 已提交
3603

3604 3605 3606
	/* If we have been reset, we may not send again. */
	if (sk->sk_state == TCP_CLOSE)
		return;
L
Linus Torvalds 已提交
3607

3608 3609
	tcp_ca_event(sk, CA_EVENT_NON_DELAYED_ACK);

3610 3611 3612 3613
	/* We are not putting this on the write queue, so
	 * tcp_transmit_skb() will set the ownership to this
	 * sock.
	 */
3614 3615 3616
	buff = alloc_skb(MAX_TCP_HEADER,
			 sk_gfp_mask(sk, GFP_ATOMIC | __GFP_NOWARN));
	if (unlikely(!buff)) {
3617 3618 3619 3620 3621
		inet_csk_schedule_ack(sk);
		inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
		inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
					  TCP_DELACK_MAX, TCP_RTO_MAX);
		return;
L
Linus Torvalds 已提交
3622
	}
3623 3624 3625

	/* Reserve space for headers and prepare control bits. */
	skb_reserve(buff, MAX_TCP_HEADER);
C
Changli Gao 已提交
3626
	tcp_init_nondata_skb(buff, tcp_acceptable_seq(sk), TCPHDR_ACK);
3627

E
Eric Dumazet 已提交
3628 3629 3630 3631 3632 3633
	/* 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);

3634
	/* Send it off, this clears delayed acks for us. */
3635
	tcp_transmit_skb(sk, buff, 0, (__force gfp_t)0);
L
Linus Torvalds 已提交
3636
}
3637
EXPORT_SYMBOL_GPL(tcp_send_ack);
L
Linus Torvalds 已提交
3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649

/* 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.
 */
3650
static int tcp_xmit_probe_skb(struct sock *sk, int urgent, int mib)
L
Linus Torvalds 已提交
3651 3652 3653 3654 3655
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb;

	/* We don't queue it, tcp_transmit_skb() sets ownership. */
3656 3657
	skb = alloc_skb(MAX_TCP_HEADER,
			sk_gfp_mask(sk, GFP_ATOMIC | __GFP_NOWARN));
3658
	if (!skb)
L
Linus Torvalds 已提交
3659 3660 3661 3662 3663 3664 3665 3666
		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
3667
	tcp_init_nondata_skb(skb, tp->snd_una - !urgent, TCPHDR_ACK);
3668
	NET_INC_STATS(sock_net(sk), mib);
3669
	return tcp_transmit_skb(sk, skb, 0, (__force gfp_t)0);
L
Linus Torvalds 已提交
3670 3671
}

3672
/* Called from setsockopt( ... TCP_REPAIR ) */
P
Pavel Emelyanov 已提交
3673 3674 3675 3676
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;
3677
		tcp_mstamp_refresh(tcp_sk(sk));
3678
		tcp_xmit_probe_skb(sk, 0, LINUX_MIB_TCPWINPROBE);
P
Pavel Emelyanov 已提交
3679 3680 3681
	}
}

3682
/* Initiate keepalive or window probe from timer. */
3683
int tcp_write_wakeup(struct sock *sk, int mib)
L
Linus Torvalds 已提交
3684
{
3685 3686
	struct tcp_sock *tp = tcp_sk(sk);
	struct sk_buff *skb;
L
Linus Torvalds 已提交
3687

3688 3689 3690
	if (sk->sk_state == TCP_CLOSE)
		return -1;

3691 3692
	skb = tcp_send_head(sk);
	if (skb && before(TCP_SKB_CB(skb)->seq, tcp_wnd_end(tp))) {
3693
		int err;
I
Ilpo Järvinen 已提交
3694
		unsigned int mss = tcp_current_mss(sk);
3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706
		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
3707
			TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
3708 3709
			if (tcp_fragment(sk, TCP_FRAG_IN_WRITE_QUEUE,
					 skb, seg_size, mss, GFP_ATOMIC))
3710 3711
				return -1;
		} else if (!tcp_skb_pcount(skb))
3712
			tcp_set_skb_tso_segs(skb, mss);
3713

E
Eric Dumazet 已提交
3714
		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
3715 3716 3717 3718 3719
		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
3720
		if (between(tp->snd_up, tp->snd_una + 1, tp->snd_una + 0xFFFF))
3721 3722
			tcp_xmit_probe_skb(sk, 1, mib);
		return tcp_xmit_probe_skb(sk, 0, mib);
L
Linus Torvalds 已提交
3723 3724 3725 3726 3727 3728 3729 3730
	}
}

/* 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)
{
3731
	struct inet_connection_sock *icsk = inet_csk(sk);
L
Linus Torvalds 已提交
3732
	struct tcp_sock *tp = tcp_sk(sk);
3733
	struct net *net = sock_net(sk);
3734
	unsigned long probe_max;
L
Linus Torvalds 已提交
3735 3736
	int err;

3737
	err = tcp_write_wakeup(sk, LINUX_MIB_TCPWINPROBE);
L
Linus Torvalds 已提交
3738

3739
	if (tp->packets_out || tcp_write_queue_empty(sk)) {
L
Linus Torvalds 已提交
3740
		/* Cancel probe timer, if it is not required. */
3741
		icsk->icsk_probes_out = 0;
3742
		icsk->icsk_backoff = 0;
L
Linus Torvalds 已提交
3743 3744 3745 3746
		return;
	}

	if (err <= 0) {
3747
		if (icsk->icsk_backoff < net->ipv4.sysctl_tcp_retries2)
3748
			icsk->icsk_backoff++;
3749
		icsk->icsk_probes_out++;
3750
		probe_max = TCP_RTO_MAX;
L
Linus Torvalds 已提交
3751 3752
	} else {
		/* If packet was not sent due to local congestion,
3753
		 * do not backoff and do not remember icsk_probes_out.
L
Linus Torvalds 已提交
3754 3755 3756 3757
		 * Let local senders to fight for local resources.
		 *
		 * Use accumulated backoff yet.
		 */
3758 3759
		if (!icsk->icsk_probes_out)
			icsk->icsk_probes_out = 1;
3760
		probe_max = TCP_RESOURCE_PROBE_INTERVAL;
L
Linus Torvalds 已提交
3761
	}
3762
	inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
3763
				  tcp_probe0_when(sk, probe_max),
3764
				  TCP_RTO_MAX);
L
Linus Torvalds 已提交
3765
}
3766

3767
int tcp_rtx_synack(const struct sock *sk, struct request_sock *req)
3768 3769 3770 3771 3772
{
	const struct tcp_request_sock_ops *af_ops = tcp_rsk(req)->af_specific;
	struct flowi fl;
	int res;

3773
	tcp_rsk(req)->txhash = net_tx_rndhash();
3774
	res = af_ops->send_synack(sk, NULL, &fl, req, NULL, TCP_SYNACK_NORMAL);
3775
	if (!res) {
E
Eric Dumazet 已提交
3776
		__TCP_INC_STATS(sock_net(sk), TCP_MIB_RETRANSSEGS);
3777
		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPSYNRETRANS);
3778 3779
		if (unlikely(tcp_passive_fastopen(sk)))
			tcp_sk(sk)->total_retrans++;
3780
		trace_tcp_retransmit_synack(sk, req);
3781 3782 3783 3784
	}
	return res;
}
EXPORT_SYMBOL(tcp_rtx_synack);