]> git.proxmox.com Git - ceph.git/blame - ceph/src/msg/async/dpdk/TCP.cc
update sources to ceph Nautilus 14.2.1
[ceph.git] / ceph / src / msg / async / dpdk / TCP.cc
CommitLineData
7c673cae
FG
1// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
2/*
3 * This file is open source software, licensed to you under the terms
4 * of the Apache License, Version 2.0 (the "License"). See the NOTICE file
5 * distributed with this work for additional information regarding copyright
6 * ownership. You may not use this file except in compliance with the License.
7 *
8 * You may obtain a copy of the License at
9 *
10 * http://www.apache.org/licenses/LICENSE-2.0
11 *
12 * Unless required by applicable law or agreed to in writing,
13 * software distributed under the License is distributed on an
14 * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
15 * KIND, either express or implied. See the License for the
16 * specific language governing permissions and limitations
17 * under the License.
18 */
19/*
20 * Copyright (C) 2014 Cloudius Systems, Ltd.
21 */
22
23#include "align.h"
24#include "TCP.h"
25#include "IP.h"
26#include "DPDKStack.h"
27
28#include "common/dout.h"
11fdf7f2 29#include "include/ceph_assert.h"
7c673cae
FG
30
31#define dout_subsys ceph_subsys_dpdk
32#undef dout_prefix
33#define dout_prefix *_dout << "tcp "
34
35void tcp_option::parse(uint8_t* beg, uint8_t* end)
36{
37 while (beg < end) {
38 auto kind = option_kind(*beg);
39 if (kind != option_kind::nop && kind != option_kind::eol) {
40 // Make sure there is enough room for this option
41 auto len = *(beg + 1);
42 if (beg + len > end) {
43 return;
44 }
45 }
46 switch (kind) {
47 case option_kind::mss:
48 _mss_received = true;
49 _remote_mss = ntoh(reinterpret_cast<mss*>(beg)->mss);
50 beg += option_len::mss;
51 break;
52 case option_kind::win_scale:
53 _win_scale_received = true;
54 _remote_win_scale = reinterpret_cast<win_scale*>(beg)->shift;
55 // We can turn on win_scale option, 7 is Linux's default win scale size
56 _local_win_scale = 7;
57 beg += option_len::win_scale;
58 break;
59 case option_kind::sack:
60 _sack_received = true;
61 beg += option_len::sack;
62 break;
63 case option_kind::nop:
64 beg += option_len::nop;
65 break;
66 case option_kind::eol:
67 return;
68 default:
69 // Ignore options we do not understand
70 auto len = *(beg + 1);
71 beg += len;
72 // Prevent infinite loop
73 if (len == 0) {
74 return;
75 }
76 break;
77 }
78 }
79}
80
81uint8_t tcp_option::fill(tcp_hdr* th, uint8_t options_size)
82{
83 auto hdr = reinterpret_cast<uint8_t*>(th);
84 auto off = hdr + sizeof(tcp_hdr);
85 uint8_t size = 0;
86 bool syn_on = th->f_syn;
87 bool ack_on = th->f_ack;
88
89 if (syn_on) {
90 if (_mss_received || !ack_on) {
91 auto mss = new (off) tcp_option::mss;
92 mss->mss = _local_mss;
93 off += mss->len;
94 size += mss->len;
95 *mss = mss->hton();
96 }
97 if (_win_scale_received || !ack_on) {
98 auto win_scale = new (off) tcp_option::win_scale;
99 win_scale->shift = _local_win_scale;
100 off += win_scale->len;
101 size += win_scale->len;
102 }
103 }
104 if (size > 0) {
105 // Insert NOP option
106 auto size_max = align_up(uint8_t(size + 1), tcp_option::align);
107 while (size < size_max - uint8_t(option_len::eol)) {
108 new (off) tcp_option::nop;
109 off += option_len::nop;
110 size += option_len::nop;
111 }
112 new (off) tcp_option::eol;
113 size += option_len::eol;
114 }
11fdf7f2 115 ceph_assert(size == options_size);
7c673cae
FG
116
117 return size;
118}
119
120uint8_t tcp_option::get_size(bool syn_on, bool ack_on)
121{
122 uint8_t size = 0;
123 if (syn_on) {
124 if (_mss_received || !ack_on) {
125 size += option_len::mss;
126 }
127 if (_win_scale_received || !ack_on) {
128 size += option_len::win_scale;
129 }
130 }
131 if (size > 0) {
132 size += option_len::eol;
133 // Insert NOP option to align on 32-bit
134 size = align_up(size, tcp_option::align);
135 }
136 return size;
137}
138
139ipv4_tcp::ipv4_tcp(ipv4& inet, EventCenter *c)
140 : _inet_l4(inet), _tcp(std::unique_ptr<tcp<ipv4_traits>>(new tcp<ipv4_traits>(inet.cct, _inet_l4, c)))
141{ }
142
143ipv4_tcp::~ipv4_tcp() { }
144
145void ipv4_tcp::received(Packet p, ipv4_address from, ipv4_address to)
146{
147 _tcp->received(std::move(p), from, to);
148}
149
150bool ipv4_tcp::forward(forward_hash& out_hash_data, Packet& p, size_t off)
151{
152 return _tcp->forward(out_hash_data, p, off);
153}
154
155int tcpv4_listen(tcp<ipv4_traits>& tcpv4, uint16_t port, const SocketOptions &opts,
11fdf7f2 156 int type, ServerSocket *sock)
7c673cae 157{
11fdf7f2 158 auto p = new DPDKServerSocketImpl<tcp<ipv4_traits>>(tcpv4, port, opts, type);
7c673cae
FG
159 int r = p->listen();
160 if (r < 0) {
161 delete p;
162 return r;
163 }
164 *sock = ServerSocket(std::unique_ptr<ServerSocketImpl>(p));
165 return 0;
166}
167
168int tcpv4_connect(tcp<ipv4_traits>& tcpv4, const entity_addr_t &addr,
169 ConnectedSocket *sock)
170{
171 auto conn = tcpv4.connect(addr);
172 *sock = ConnectedSocket(std::unique_ptr<ConnectedSocketImpl>(
173 new NativeConnectedSocketImpl<tcp<ipv4_traits>>(std::move(conn))));
174 return 0;
175}
176
177template <typename InetTraits>
178void tcp<InetTraits>::respond_with_reset(tcp_hdr* rth, ipaddr local_ip, ipaddr foreign_ip)
179{
180 ldout(cct, 20) << __func__ << " tcp header rst=" << bool(rth->f_rst) << " fin=" << bool(rth->f_fin)
181 << " syn=" << bool(rth->f_syn) << dendl;
182 if (rth->f_rst) {
183 return;
184 }
185 Packet p;
186 auto th = p.prepend_header<tcp_hdr>();
187 th->src_port = rth->dst_port;
188 th->dst_port = rth->src_port;
189 if (rth->f_ack) {
190 th->seq = rth->ack;
191 }
192 // If this RST packet is in response to a SYN packet. We ACK the ISN.
193 if (rth->f_syn) {
194 th->ack = rth->seq + 1;
195 th->f_ack = true;
196 }
197 th->f_rst = true;
198 th->data_offset = sizeof(*th) / 4;
199 th->checksum = 0;
200 *th = th->hton();
201
202 checksummer csum;
203 offload_info oi;
204 InetTraits::tcp_pseudo_header_checksum(csum, local_ip, foreign_ip, sizeof(*th));
205 if (get_hw_features().tx_csum_l4_offload) {
206 th->checksum = ~csum.get();
207 oi.needs_csum = true;
208 } else {
209 csum.sum(p);
210 th->checksum = csum.get();
211 oi.needs_csum = false;
212 }
213
214 oi.protocol = ip_protocol_num::tcp;
215 oi.tcp_hdr_len = sizeof(tcp_hdr);
216 p.set_offload_info(oi);
217
218 send_packet_without_tcb(local_ip, foreign_ip, std::move(p));
219}
220
221#undef dout_prefix
222#define dout_prefix _prefix(_dout)
223template<typename InetTraits>
224ostream& tcp<InetTraits>::tcb::_prefix(std::ostream *_dout) {
225 return *_dout << "tcp " << _local_ip << ":" << _local_port << " -> " << _foreign_ip << ":" << _foreign_port
226 << " tcb(" << this << " fd=" << fd << " s=" << _state << ").";
227}
228
229template<typename InetTraits>
230void tcp<InetTraits>::tcb::input_handle_listen_state(tcp_hdr* th, Packet p)
231{
232 auto opt_len = th->data_offset * 4 - sizeof(tcp_hdr);
233 auto opt_start = reinterpret_cast<uint8_t*>(p.get_header(0, th->data_offset * 4)) + sizeof(tcp_hdr);
234 auto opt_end = opt_start + opt_len;
235 p.trim_front(th->data_offset * 4);
236 tcp_sequence seg_seq = th->seq;
237
238 // Set RCV.NXT to SEG.SEQ+1, IRS is set to SEG.SEQ
239 _rcv.next = seg_seq + 1;
240 _rcv.initial = seg_seq;
241
242 // ISS should be selected and a SYN segment sent of the form:
243 // <SEQ=ISS><ACK=RCV.NXT><CTL=SYN,ACK>
244 // SND.NXT is set to ISS+1 and SND.UNA to ISS
245 // NOTE: In previous code, _snd.next is set to ISS + 1 only when SYN is
246 // ACKed. Now, we set _snd.next to ISS + 1 here, so in output_one(): we
247 // have
248 // th->seq = syn_on ? _snd.initial : _snd.next
249 // to make sure retransmitted SYN has correct SEQ number.
250 do_setup_isn();
251
252 _rcv.urgent = _rcv.next;
253
254 ldout(_tcp.cct, 10) << __func__ << " listen: LISTEN -> SYN_RECEIVED" << dendl;
255 init_from_options(th, opt_start, opt_end);
256 do_syn_received();
257}
258
259template <typename InetTraits>
260void tcp<InetTraits>::tcb::input_handle_syn_sent_state(tcp_hdr* th, Packet p)
261{
262 auto opt_len = th->data_offset * 4 - sizeof(tcp_hdr);
263 auto opt_start = reinterpret_cast<uint8_t*>(p.get_header(0, th->data_offset * 4)) + sizeof(tcp_hdr);
264 auto opt_end = opt_start + opt_len;
265 p.trim_front(th->data_offset * 4);
266 tcp_sequence seg_seq = th->seq;
267 auto seg_ack = th->ack;
268
269 ldout(_tcp.cct, 20) << __func__ << " tcp header seq " << seg_seq.raw << " ack " << seg_ack.raw
270 << " fin=" << bool(th->f_fin) << " syn=" << bool(th->f_syn) << dendl;
271
272 bool acceptable = false;
273 // 3.1 first check the ACK bit
274 if (th->f_ack) {
275 // If SEG.ACK =< ISS, or SEG.ACK > SND.NXT, send a reset (unless the
276 // RST bit is set, if so drop the segment and return)
277 if (seg_ack <= _snd.initial || seg_ack > _snd.next) {
278 return respond_with_reset(th);
279 }
280
281 // If SND.UNA =< SEG.ACK =< SND.NXT then the ACK is acceptable.
282 acceptable = _snd.unacknowledged <= seg_ack && seg_ack <= _snd.next;
283 }
284
285 // 3.2 second check the RST bit
286 if (th->f_rst) {
287 // If the ACK was acceptable then signal the user "error: connection
288 // reset", drop the segment, enter CLOSED state, delete TCB, and
289 // return. Otherwise (no ACK) drop the segment and return.
290 if (acceptable) {
291 return do_reset();
292 } else {
293 return;
294 }
295 }
296
297 // 3.3 third check the security and precedence
298 // NOTE: Ignored for now
299
300 // 3.4 fourth check the SYN bit
301 if (th->f_syn) {
302 // RCV.NXT is set to SEG.SEQ+1, IRS is set to SEG.SEQ. SND.UNA should
303 // be advanced to equal SEG.ACK (if there is an ACK), and any segments
304 // on the retransmission queue which are thereby acknowledged should be
305 // removed.
306 _rcv.next = seg_seq + 1;
307 _rcv.initial = seg_seq;
308 if (th->f_ack) {
309 // TODO: clean retransmission queue
310 _snd.unacknowledged = seg_ack;
311 }
312 if (_snd.unacknowledged > _snd.initial) {
313 // If SND.UNA > ISS (our SYN has been ACKed), change the connection
314 // state to ESTABLISHED, form an ACK segment
315 // <SEQ=SND.NXT><ACK=RCV.NXT><CTL=ACK>
316 ldout(_tcp.cct, 20) << __func__ << " syn: SYN_SENT -> ESTABLISHED" << dendl;
317 init_from_options(th, opt_start, opt_end);
318 do_established();
319 output();
320 } else {
321 // Otherwise enter SYN_RECEIVED, form a SYN,ACK segment
322 // <SEQ=ISS><ACK=RCV.NXT><CTL=SYN,ACK>
323 ldout(_tcp.cct, 20) << __func__ << " syn: SYN_SENT -> SYN_RECEIVED" << dendl;
324 do_syn_received();
325 }
326 }
327
328 // 3.5 fifth, if neither of the SYN or RST bits is set then drop the
329 // segment and return.
330 return;
331}
332
333template <typename InetTraits>
334void tcp<InetTraits>::tcb::input_handle_other_state(tcp_hdr* th, Packet p)
335{
336 p.trim_front(th->data_offset * 4);
337 bool do_output = false;
338 bool do_output_data = false;
339 tcp_sequence seg_seq = th->seq;
340 auto seg_ack = th->ack;
341 auto seg_len = p.len();
342 ldout(_tcp.cct, 20) << __func__ << " tcp header seq " << seg_seq.raw << " ack " << seg_ack.raw
343 << " snd next " << _snd.next.raw << " unack " << _snd.unacknowledged.raw
344 << " rcv next " << _rcv.next.raw << " len " << seg_len
345 << " fin=" << bool(th->f_fin) << " syn=" << bool(th->f_syn) << dendl;
346
347 // 4.1 first check sequence number
348 if (!segment_acceptable(seg_seq, seg_len)) {
349 //<SEQ=SND.NXT><ACK=RCV.NXT><CTL=ACK>
350 return output();
351 }
352
353 // In the following it is assumed that the segment is the idealized
354 // segment that begins at RCV.NXT and does not exceed the window.
355 if (seg_seq < _rcv.next) {
356 // ignore already acknowledged data
357 auto dup = std::min(uint32_t(_rcv.next - seg_seq), seg_len);
358 ldout(_tcp.cct, 10) << __func__ << " dup segment len " << dup << dendl;
359 p.trim_front(dup);
360 seg_len -= dup;
361 seg_seq += dup;
362 }
363 // FIXME: We should trim data outside the right edge of the receive window as well
364
365 if (seg_seq != _rcv.next) {
366 ldout(_tcp.cct, 10) << __func__ << " out of order, expect " << _rcv.next.raw
367 << " actual " << seg_seq.raw
368 << " out of order size " << _rcv.out_of_order.map.size()
369 << dendl;
370 insert_out_of_order(seg_seq, std::move(p));
371 // A TCP receiver SHOULD send an immediate duplicate ACK
372 // when an out-of-order segment arrives.
373 return output();
374 }
375
376 // 4.2 second check the RST bit
377 if (th->f_rst) {
378 if (in_state(SYN_RECEIVED)) {
379 // If this connection was initiated with a passive OPEN (i.e.,
380 // came from the LISTEN state), then return this connection to
381 // LISTEN state and return. The user need not be informed. If
382 // this connection was initiated with an active OPEN (i.e., came
383 // from SYN_SENT state) then the connection was refused, signal
384 // the user "connection refused". In either case, all segments
385 // on the retransmission queue should be removed. And in the
386 // active OPEN case, enter the CLOSED state and delete the TCB,
387 // and return.
388 errno = -ECONNREFUSED;
389 return do_reset();
390 }
391 if (in_state(ESTABLISHED | FIN_WAIT_1 | FIN_WAIT_2 | CLOSE_WAIT)) {
392 // If the RST bit is set then, any outstanding RECEIVEs and SEND
393 // should receive "reset" responses. All segment queues should be
394 // flushed. Users should also receive an unsolicited general
395 // "connection reset" signal. Enter the CLOSED state, delete the
396 // TCB, and return.
397 return do_reset();
398 }
399 if (in_state(CLOSING | LAST_ACK | TIME_WAIT)) {
400 // If the RST bit is set then, enter the CLOSED state, delete the
401 // TCB, and return.
402 return do_closed();
403 }
404 }
405
406 // 4.3 third check security and precedence
407 // NOTE: Ignored for now
408
409 // 4.4 fourth, check the SYN bit
410 if (th->f_syn) {
411 // SYN_RECEIVED, ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2
412 // CLOSE_WAIT, CLOSING, LAST_ACK, TIME_WAIT
413
414 // If the SYN is in the window it is an error, send a reset, any
415 // outstanding RECEIVEs and SEND should receive "reset" responses,
416 // all segment queues should be flushed, the user should also
417 // receive an unsolicited general "connection reset" signal, enter
418 // the CLOSED state, delete the TCB, and return.
419 respond_with_reset(th);
420 return do_reset();
421
422 // If the SYN is not in the window this step would not be reached
423 // and an ack would have been sent in the first step (sequence
424 // number check).
425 }
426
427 // 4.5 fifth check the ACK field
428 if (!th->f_ack) {
429 // if the ACK bit is off drop the segment and return
430 return;
431 } else {
432 // SYN_RECEIVED STATE
433 if (in_state(SYN_RECEIVED)) {
434 // If SND.UNA =< SEG.ACK =< SND.NXT then enter ESTABLISHED state
435 // and continue processing.
436 if (_snd.unacknowledged <= seg_ack && seg_ack <= _snd.next) {
437 ldout(_tcp.cct, 20) << __func__ << " SYN_RECEIVED -> ESTABLISHED" << dendl;
438 do_established();
439 if (_tcp.push_listen_queue(_local_port, this)) {
440 ldout(_tcp.cct, 20) << __func__ << " successfully accepting socket" << dendl;
441 } else {
442 ldout(_tcp.cct, 5) << __func__ << " not exist listener or full queue, reset" << dendl;
443 return respond_with_reset(th);
444 }
445 } else {
446 // <SEQ=SEG.ACK><CTL=RST>
447 return respond_with_reset(th);
448 }
449 }
450 auto update_window = [this, th, seg_seq, seg_ack] {
451 ldout(_tcp.cct, 20) << __func__ << " window update seg_seq=" << seg_seq
452 << " seg_ack=" << seg_ack << " old window=" << th->window
453 << " new window=" << int(_snd.window_scale) << dendl;
454 _snd.window = th->window << _snd.window_scale;
455 _snd.wl1 = seg_seq;
456 _snd.wl2 = seg_ack;
457 if (_snd.window == 0) {
458 _persist_time_out = _rto;
459 start_persist_timer();
460 } else {
461 stop_persist_timer();
462 }
463 };
464 // ESTABLISHED STATE or
465 // CLOSE_WAIT STATE: Do the same processing as for the ESTABLISHED state.
466 if (in_state(ESTABLISHED | CLOSE_WAIT)) {
467 // If SND.UNA < SEG.ACK =< SND.NXT then, set SND.UNA <- SEG.ACK.
468 if (_snd.unacknowledged < seg_ack && seg_ack <= _snd.next) {
469 // Remote ACKed data we sent
470 auto acked_bytes = data_segment_acked(seg_ack);
471
472 // If SND.UNA < SEG.ACK =< SND.NXT, the send window should be updated.
473 if (_snd.wl1 < seg_seq || (_snd.wl1 == seg_seq && _snd.wl2 <= seg_ack)) {
474 update_window();
475 }
476
477 // some data is acked, try send more data
478 do_output_data = true;
479
480 auto set_retransmit_timer = [this] {
481 if (_snd.data.empty()) {
482 // All outstanding segments are acked, turn off the timer.
483 stop_retransmit_timer();
484 // Signal the waiter of this event
485 signal_all_data_acked();
486 } else {
487 // Restart the timer becasue new data is acked.
488 start_retransmit_timer();
489 }
490 };
491
492 if (_snd.dupacks >= 3) {
493 // We are in fast retransmit / fast recovery phase
494 uint32_t smss = _snd.mss;
495 if (seg_ack > _snd.recover) {
496 ldout(_tcp.cct, 20) << __func__ << " ack: full_ack" << dendl;
497 // Set cwnd to min (ssthresh, max(FlightSize, SMSS) + SMSS)
498 _snd.cwnd = std::min(_snd.ssthresh, std::max(flight_size(), smss) + smss);
499 // Exit the fast recovery procedure
500 exit_fast_recovery();
501 set_retransmit_timer();
502 } else {
503 ldout(_tcp.cct, 20) << __func__ << " ack: partial_ack" << dendl;
504 // Retransmit the first unacknowledged segment
505 fast_retransmit();
506 // Deflate the congestion window by the amount of new data
507 // acknowledged by the Cumulative Acknowledgment field
508 _snd.cwnd -= acked_bytes;
509 // If the partial ACK acknowledges at least one SMSS of new
510 // data, then add back SMSS bytes to the congestion window
511 if (acked_bytes >= smss) {
512 _snd.cwnd += smss;
513 }
514 // Send a new segment if permitted by the new value of
515 // cwnd. Do not exit the fast recovery procedure For
516 // the first partial ACK that arrives during fast
517 // recovery, also reset the retransmit timer.
518 if (++_snd.partial_ack == 1) {
519 start_retransmit_timer();
520 }
521 }
522 } else {
523 // RFC5681: The fast retransmit algorithm uses the arrival
524 // of 3 duplicate ACKs (as defined in section 2, without
525 // any intervening ACKs which move SND.UNA) as an
526 // indication that a segment has been lost.
527 //
528 // So, here we reset dupacks to zero becasue this ACK moves
529 // SND.UNA.
530 exit_fast_recovery();
531 set_retransmit_timer();
532 }
533 } else if (!_snd.data.empty() && seg_len == 0 &&
534 th->f_fin == 0 && th->f_syn == 0 &&
535 th->ack == _snd.unacknowledged &&
536 uint32_t(th->window << _snd.window_scale) == _snd.window) {
537 // Note:
538 // RFC793 states:
539 // If the ACK is a duplicate (SEG.ACK < SND.UNA), it can be ignored
540 // RFC5681 states:
541 // The TCP sender SHOULD use the "fast retransmit" algorithm to detect
542 // and repair loss, based on incoming duplicate ACKs.
543 // Here, We follow RFC5681.
544 _snd.dupacks++;
545 uint32_t smss = _snd.mss;
546 // 3 duplicated ACKs trigger a fast retransmit
547 if (_snd.dupacks == 1 || _snd.dupacks == 2) {
548 // RFC5681 Step 3.1
549 // Send cwnd + 2 * smss per RFC3042
550 do_output_data = true;
551 } else if (_snd.dupacks == 3) {
552 // RFC6582 Step 3.2
553 if (seg_ack - 1 > _snd.recover) {
554 _snd.recover = _snd.next - 1;
555 // RFC5681 Step 3.2
556 _snd.ssthresh = std::max((flight_size() - _snd.limited_transfer) / 2, 2 * smss);
557 fast_retransmit();
558 } else {
559 // Do not enter fast retransmit and do not reset ssthresh
560 }
561 // RFC5681 Step 3.3
562 _snd.cwnd = _snd.ssthresh + 3 * smss;
563 } else if (_snd.dupacks > 3) {
564 // RFC5681 Step 3.4
565 _snd.cwnd += smss;
566 // RFC5681 Step 3.5
567 do_output_data = true;
568 }
569 } else if (seg_ack > _snd.next) {
570 // If the ACK acks something not yet sent (SEG.ACK > SND.NXT)
571 // then send an ACK, drop the segment, and return
572 return output();
573 } else if (_snd.window == 0 && th->window > 0) {
574 update_window();
575 do_output_data = true;
576 }
577 }
578 // FIN_WAIT_1 STATE
579 if (in_state(FIN_WAIT_1)) {
580 // In addition to the processing for the ESTABLISHED state, if
581 // our FIN is now acknowledged then enter FIN-WAIT-2 and continue
582 // processing in that state.
583 if (seg_ack == _snd.next + 1) {
584 ldout(_tcp.cct, 20) << __func__ << " ack: FIN_WAIT_1 -> FIN_WAIT_2" << dendl;
585 _state = FIN_WAIT_2;
586 do_local_fin_acked();
587 }
588 }
589 // FIN_WAIT_2 STATE
590 if (in_state(FIN_WAIT_2)) {
591 // In addition to the processing for the ESTABLISHED state, if
592 // the retransmission queue is empty, the user’s CLOSE can be
593 // acknowledged ("ok") but do not delete the TCB.
594 // TODO
595 }
596 // CLOSING STATE
597 if (in_state(CLOSING)) {
598 if (seg_ack == _snd.next + 1) {
599 ldout(_tcp.cct, 20) << __func__ << " ack: CLOSING -> TIME_WAIT" << dendl;
600 do_local_fin_acked();
601 return do_time_wait();
602 } else {
603 return;
604 }
605 }
606 // LAST_ACK STATE
607 if (in_state(LAST_ACK)) {
608 if (seg_ack == _snd.next + 1) {
609 ldout(_tcp.cct, 20) << __func__ << " ack: LAST_ACK -> CLOSED" << dendl;
610 do_local_fin_acked();
611 return do_closed();
612 }
613 }
614 // TIME_WAIT STATE
615 if (in_state(TIME_WAIT)) {
616 // The only thing that can arrive in this state is a
617 // retransmission of the remote FIN. Acknowledge it, and restart
618 // the 2 MSL timeout.
619 // TODO
620 }
621 }
622
623 // 4.6 sixth, check the URG bit
624 if (th->f_urg) {
625 // TODO
626 }
627
628 // 4.7 seventh, process the segment text
629 if (in_state(ESTABLISHED | FIN_WAIT_1 | FIN_WAIT_2)) {
630 if (p.len()) {
631 // Once the TCP takes responsibility for the data it advances
632 // RCV.NXT over the data accepted, and adjusts RCV.WND as
633 // apporopriate to the current buffer availability. The total of
634 // RCV.NXT and RCV.WND should not be reduced.
635 _rcv.data.push_back(std::move(p));
636 _rcv.next += seg_len;
637 auto merged = merge_out_of_order();
638 signal_data_received();
639 // Send an acknowledgment of the form:
640 // <SEQ=SND.NXT><ACK=RCV.NXT><CTL=ACK>
641 // This acknowledgment should be piggybacked on a segment being
642 // transmitted if possible without incurring undue delay.
643 if (merged) {
644 // TCP receiver SHOULD send an immediate ACK when the
645 // incoming segment fills in all or part of a gap in the
646 // sequence space.
647 do_output = true;
648 } else {
649 do_output = should_send_ack(seg_len);
650 }
651 ldout(_tcp.cct, 20) << __func__ << " merged=" << merged << " do_output=" << do_output << dendl;
652 }
653 } else if (in_state(CLOSE_WAIT | CLOSING | LAST_ACK | TIME_WAIT)) {
654 // This should not occur, since a FIN has been received from the
655 // remote side. Ignore the segment text.
656 return;
657 }
658
659 // 4.8 eighth, check the FIN bit
660 if (th->f_fin) {
661 if (in_state(CLOSED | LISTEN | SYN_SENT)) {
662 // Do not process the FIN if the state is CLOSED, LISTEN or SYN-SENT
663 // since the SEG.SEQ cannot be validated; drop the segment and return.
664 return;
665 }
666 auto fin_seq = seg_seq + seg_len;
667 if (fin_seq == _rcv.next) {
668 _rcv.next = fin_seq + 1;
669
670 // If this <FIN> packet contains data as well, we can ACK both data
671 // and <FIN> in a single packet, so canncel the previous ACK.
672 clear_delayed_ack();
673 do_output = false;
674 // Send ACK for the FIN!
675 output();
676 signal_data_received();
677 _errno = 0;
678
679 if (in_state(SYN_RECEIVED | ESTABLISHED)) {
680 ldout(_tcp.cct, 20) << __func__ << " fin: SYN_RECEIVED or ESTABLISHED -> CLOSE_WAIT" << dendl;
681 _state = CLOSE_WAIT;
682 // EOF
683 }
684 if (in_state(FIN_WAIT_1)) {
685 // If our FIN has been ACKed (perhaps in this segment), then
686 // enter TIME-WAIT, start the time-wait timer, turn off the other
687 // timers; otherwise enter the CLOSING state.
688 // Note: If our FIN has been ACKed, we should be in FIN_WAIT_2
689 // not FIN_WAIT_1 if we reach here.
690 ldout(_tcp.cct, 20) << __func__ << " fin: FIN_WAIT_1 -> CLOSING" << dendl;
691 _state = CLOSING;
692 }
693 if (in_state(FIN_WAIT_2)) {
694 ldout(_tcp.cct, 20) << __func__ << " fin: FIN_WAIT_2 -> TIME_WAIT" << dendl;
695 return do_time_wait();
696 }
697 }
698 }
699 if (do_output || (do_output_data && can_send())) {
700 // Since we will do output, we can canncel scheduled delayed ACK.
701 clear_delayed_ack();
702 output();
703 }
704}
705
706template <typename InetTraits>
707void tcp<InetTraits>::tcb::connect()
708{
709 ldout(_tcp.cct, 20) << __func__ << dendl;
710 // An initial send sequence number (ISS) is selected. A SYN segment of the
711 // form <SEQ=ISS><CTL=SYN> is sent. Set SND.UNA to ISS, SND.NXT to ISS+1,
712 // enter SYN-SENT state, and return.
713 do_setup_isn();
714
715 // Local receive window scale factor
716 _rcv.window_scale = _option._local_win_scale = 7;
717 // Maximum segment size local can receive
718 _rcv.mss = _option._local_mss = local_mss();
719 // Linux's default window size
720 _rcv.window = 29200 << _rcv.window_scale;
721
722 do_syn_sent();
723}
724
725template <typename InetTraits>
726void tcp<InetTraits>::tcb::close_final_cleanup()
727{
728 if (_snd._all_data_acked_fd >= 0) {
729 center->delete_file_event(_snd._all_data_acked_fd, EVENT_READABLE);
730 _tcp.manager.close(_snd._all_data_acked_fd);
731 _snd._all_data_acked_fd = -1;
732 }
733
734 _snd.closed = true;
735 signal_data_received();
736 ldout(_tcp.cct, 20) << __func__ << " unsent_len=" << _snd.unsent_len << dendl;
737 if (in_state(CLOSE_WAIT)) {
738 ldout(_tcp.cct, 20) << __func__ << " CLOSE_WAIT -> LAST_ACK" << dendl;
739 _state = LAST_ACK;
740 } else if (in_state(ESTABLISHED)) {
741 ldout(_tcp.cct, 20) << __func__ << " ESTABLISHED -> FIN_WAIT_1" << dendl;
742 _state = FIN_WAIT_1;
743 }
744 // Send <FIN> to remote
745 // Note: we call output_one to make sure a packet with FIN actually
746 // sent out. If we only call output() and _packetq is not empty,
747 // tcp::tcb::get_packet(), packet with FIN will not be generated.
748 output_one();
749 output();
750 center->delete_file_event(fd, EVENT_READABLE|EVENT_WRITABLE);
751}
752
753template <typename InetTraits>
754void tcp<InetTraits>::tcb::retransmit()
755{
756 auto output_update_rto = [this] {
757 output();
758 // According to RFC6298, Update RTO <- RTO * 2 to perform binary exponential back-off
759 this->_rto = std::min(this->_rto * 2, this->_rto_max);
760 start_retransmit_timer();
761 };
762
763 // Retransmit SYN
764 if (syn_needs_on()) {
765 if (_snd.syn_retransmit++ < _max_nr_retransmit) {
766 output_update_rto();
767 } else {
768 _errno = -ECONNABORTED;
769 ldout(_tcp.cct, 5) << __func__ << " syn retransmit exceed max "
770 << _max_nr_retransmit << dendl;
771 _errno = -ETIMEDOUT;
772 cleanup();
773 return;
774 }
775 }
776
777 // Retransmit FIN
778 if (fin_needs_on()) {
779 if (_snd.fin_retransmit++ < _max_nr_retransmit) {
780 output_update_rto();
781 } else {
782 ldout(_tcp.cct, 5) << __func__ << " fin retransmit exceed max "
783 << _max_nr_retransmit << dendl;
784 _errno = -ETIMEDOUT;
785 cleanup();
786 return;
787 }
788 }
789
790 // Retransmit Data
791 if (_snd.data.empty()) {
792 return;
793 }
794
795 // If there are unacked data, retransmit the earliest segment
796 auto& unacked_seg = _snd.data.front();
797
798 // According to RFC5681
799 // Update ssthresh only for the first retransmit
800 uint32_t smss = _snd.mss;
801 if (unacked_seg.nr_transmits == 0) {
802 _snd.ssthresh = std::max(flight_size() / 2, 2 * smss);
803 }
804 // RFC6582 Step 4
805 _snd.recover = _snd.next - 1;
806 // Start the slow start process
807 _snd.cwnd = smss;
808 // End fast recovery
809 exit_fast_recovery();
810
811 ldout(_tcp.cct, 20) << __func__ << " unack data size " << _snd.data.size()
812 << " nr=" << unacked_seg.nr_transmits << dendl;
813 if (unacked_seg.nr_transmits < _max_nr_retransmit) {
814 unacked_seg.nr_transmits++;
815 } else {
816 // Delete connection when max num of retransmission is reached
817 ldout(_tcp.cct, 5) << __func__ << " seg retransmit exceed max "
818 << _max_nr_retransmit << dendl;
819 _errno = -ETIMEDOUT;
820 cleanup();
821 return;
822 }
823 retransmit_one();
824
825 output_update_rto();
826}
827
828template <typename InetTraits>
829void tcp<InetTraits>::tcb::persist() {
830 ldout(_tcp.cct, 20) << __func__ << " persist timer fired" << dendl;
831 // Send 1 byte packet to probe peer's window size
832 _snd.window_probe = true;
833 output_one();
834 _snd.window_probe = false;
835
836 output();
837 // Perform binary exponential back-off per RFC1122
838 _persist_time_out = std::min(_persist_time_out * 2, _rto_max);
839 start_persist_timer();
840}