http gateway in progress
[swift-upb.git] / sendrecv.cpp
1 /*
2  *  sendrecv.cpp
3  *  most of the swift's state machine
4  *
5  *  Created by Victor Grishchenko on 3/6/09.
6  *  Copyright 2009 Delft University of Technology. All rights reserved.
7  *
8  */
9 #include "swift.h"
10 #include <algorithm>  // kill it
11
12 using namespace swift;
13 using namespace std;
14
15 /*
16  TODO  25 Oct 18:55
17  - range: ALL
18  - randomized testing of advanced ops (new testcase)
19  */
20
21 void    Channel::AddPeakHashes (Datagram& dgram) {
22     for(int i=0; i<file().peak_count(); i++) {
23         bin64_t peak = file().peak(i);
24         dgram.Push8(SWIFT_HASH);
25         dgram.Push32((uint32_t)peak);
26         dgram.PushHash(file().peak_hash(i));
27         //DLOG(INFO)<<"#"<<id<<" +pHASH"<<file().peak(i);
28         dprintf("%s #%u +phash %s\n",tintstr(),id_,peak.str());
29     }
30 }
31
32
33 void    Channel::AddUncleHashes (Datagram& dgram, bin64_t pos) {
34     bin64_t peak = file().peak_for(pos);
35     while (pos!=peak && ((NOW&3)==3 || !data_out_cap_.within(pos.parent())) &&
36             ack_in_.get(pos.parent())==binmap_t::EMPTY  ) {
37         bin64_t uncle = pos.sibling();
38         dgram.Push8(SWIFT_HASH);
39         dgram.Push32((uint32_t)uncle);
40         dgram.PushHash( file().hash(uncle) );
41         //DLOG(INFO)<<"#"<<id<<" +uHASH"<<uncle;
42         dprintf("%s #%u +hash %s\n",tintstr(),id_,uncle.str());
43         pos = pos.parent();
44     }
45 }
46
47
48 bin64_t           Channel::ImposeHint () {
49     uint64_t twist = peer_channel_id_;  // got no hints, send something randomly
50     twist &= file().peak(0); // FIXME may make it semi-seq here
51     file().ack_out().twist(twist);
52     ack_in_.twist(twist);
53     bin64_t my_pick =
54         file().ack_out().find_filtered(ack_in_,bin64_t::ALL,binmap_t::FILLED);
55     while (my_pick.width()>max(1,(int)cwnd_))
56         my_pick = my_pick.left();
57     file().ack_out().twist(0);
58     ack_in_.twist(0);
59     return my_pick.twisted(twist);
60 }
61
62
63 bin64_t        Channel::DequeueHint () {
64     if (hint_in_.empty() && last_recv_time_>NOW-rtt_avg_-TINT_SEC) {
65         bin64_t my_pick = ImposeHint(); // FIXME move to the loop
66         if (my_pick!=bin64_t::NONE) {
67             hint_in_.push_back(my_pick);
68             dprintf("%s #%u *hint %s\n",tintstr(),id_,my_pick.str());
69         }
70     }
71     bin64_t send = bin64_t::NONE;
72     while (!hint_in_.empty() && send==bin64_t::NONE) {
73         bin64_t hint = hint_in_.front().bin;
74         tint time = hint_in_.front().time;
75         hint_in_.pop_front();
76         while (!hint.is_base()) { // FIXME optimize; possible attack
77             hint_in_.push_front(tintbin(time,hint.right()));
78             hint = hint.left();
79         }
80         //if (time < NOW-TINT_SEC*3/2 )
81         //    continue;  bad idea
82         if (ack_in_.get(hint)!=binmap_t::FILLED)
83             send = hint;
84     }
85     uint64_t mass = 0;
86     for(int i=0; i<hint_in_.size(); i++)
87         mass += hint_in_[i].bin.width();
88     dprintf("%s #%u dequeued %s [%lli]\n",tintstr(),id_,send.str(),mass);
89     return send;
90 }
91
92
93 void    Channel::AddHandshake (Datagram& dgram) {
94     if (!peer_channel_id_) { // initiating
95         dgram.Push8(SWIFT_HASH);
96         dgram.Push32(bin64_t::ALL32);
97         dgram.PushHash(file().root_hash());
98         dprintf("%s #%u +hash ALL %s\n",
99                 tintstr(),id_,file().root_hash().hex().c_str());
100     }
101     dgram.Push8(SWIFT_HANDSHAKE);
102     int encoded = EncodeID(id_);
103     dgram.Push32(encoded);
104     dprintf("%s #%u +hs %x\n",tintstr(),id_,encoded);
105     have_out_.clear();
106     AddHave(dgram);
107 }
108
109
110 void    Channel::Send () {
111     Datagram dgram(socket_,peer());
112     dgram.Push32(peer_channel_id_);
113     bin64_t data = bin64_t::NONE;
114     if ( is_established() ) {
115         // FIXME: seeder check
116         AddHave(dgram);
117         AddAck(dgram);
118         if (!file().is_complete())
119             AddHint(dgram);
120         AddPex(dgram);
121         TimeoutDataOut();
122         data = AddData(dgram);
123     } else {
124         AddHandshake(dgram);
125         AddHave(dgram);
126         AddAck(dgram);
127     }
128     dprintf("%s #%u sent %ib %s:%x\n",
129             tintstr(),id_,dgram.size(),peer().str(),peer_channel_id_);
130     if (dgram.size()==4) {// only the channel id; bare keep-alive
131         data = bin64_t::ALL;
132     }
133     if (dgram.Send()==-1)
134         print_error("can't send datagram");
135     last_send_time_ = NOW;
136     sent_since_recv_++;
137     dgrams_sent_++;
138     Reschedule();
139 }
140
141
142 void    Channel::AddHint (Datagram& dgram) {
143
144     tint plan_for = max(TINT_SEC,rtt_avg_*4);
145
146     tint timed_out = NOW - plan_for*2;
147     while ( !hint_out_.empty() && hint_out_.front().time < timed_out ) {
148         hint_out_size_ -= hint_out_.front().bin.width();
149         hint_out_.pop_front();
150     }
151
152     int plan_pck = max ( (tint)1, plan_for / dip_avg_ );
153
154     if ( hint_out_size_ < plan_pck ) {
155
156         int diff = plan_pck - hint_out_size_; // TODO: aggregate
157         bin64_t hint = transfer().picker().Pick(ack_in_,diff,NOW+plan_for*2);
158
159         if (hint!=bin64_t::NONE) {
160             dgram.Push8(SWIFT_HINT);
161             dgram.Push32(hint);
162             dprintf("%s #%u +hint %s [%lli]\n",tintstr(),id_,hint.str(),hint_out_size_);
163             hint_out_.push_back(hint);
164             hint_out_size_ += hint.width();
165         } else
166             dprintf("%s #%u Xhint\n",tintstr(),id_);
167
168     }
169 }
170
171
172 bin64_t        Channel::AddData (Datagram& dgram) {
173
174     if (!file().size()) // know nothing
175         return bin64_t::NONE;
176
177     bin64_t tosend = bin64_t::NONE;
178     tint luft = send_interval_>>4; // may wake up a bit earlier
179     if (data_out_.size()<cwnd_ &&
180             last_data_out_time_+send_interval_<=NOW+luft) {
181         tosend = DequeueHint();
182         if (tosend==bin64_t::NONE) {
183             dprintf("%s #%u sendctrl no idea what to send\n",tintstr(),id_);
184             if (send_control_!=KEEP_ALIVE_CONTROL)
185                 SwitchSendControl(KEEP_ALIVE_CONTROL);
186         }
187     } else
188         dprintf("%s #%u sendctrl wait cwnd %f data_out %i next %s\n",
189                 tintstr(),id_,cwnd_,data_out_.size(),tintstr(last_data_out_time_+NOW-send_interval_));
190
191     if (tosend==bin64_t::NONE)// && (last_data_out_time_>NOW-TINT_SEC || data_out_.empty()))
192         return bin64_t::NONE; // once in a while, empty data is sent just to check rtt FIXED
193
194     if (ack_in_.is_empty() && file().size())
195         AddPeakHashes(dgram);
196     AddUncleHashes(dgram,tosend);
197     if (!ack_in_.is_empty()) // TODO: cwnd_>1
198         data_out_cap_ = tosend;
199
200     if (dgram.size()>254) {
201         dgram.Send(); // kind of fragmentation
202         dgram.Push32(peer_channel_id_);
203     }
204
205     dgram.Push8(SWIFT_DATA);
206     dgram.Push32(tosend.to32());
207
208     uint8_t buf[1024];
209     size_t r = pread(file().file_descriptor(),buf,1024,tosend.base_offset()<<10);
210     // TODO: corrupted data, retries, caching
211     if (r<0) {
212         print_error("error on reading");
213         return bin64_t::NONE;
214     }
215     assert(dgram.space()>=r+4+1);
216     dgram.Push(buf,r);
217
218     last_data_out_time_ = NOW;
219     data_out_.push_back(tosend);
220     dprintf("%s #%u +data %s\n",tintstr(),id_,tosend.str());
221
222     return tosend;
223 }
224
225
226 void    Channel::AddAck (Datagram& dgram) {
227     if (data_in_==tintbin())
228         return;
229     dgram.Push8(SWIFT_ACK);
230     dgram.Push32(data_in_.bin.to32()); // FIXME not cover
231     dgram.Push64(data_in_.time); // FIXME 32
232     have_out_.set(data_in_.bin);
233     dprintf("%s #%u +ack %s %s\n",
234         tintstr(),id_,data_in_.bin.str(),tintstr(data_in_.time));
235     if (data_in_.bin.layer()>2)
236         data_in_dbl_ = data_in_.bin;
237     data_in_ = tintbin();
238 }
239
240
241 void    Channel::AddHave (Datagram& dgram) {
242     if (data_in_dbl_!=bin64_t::NONE) { // TODO: do redundancy better
243         dgram.Push8(SWIFT_HAVE);
244         dgram.Push32(data_in_dbl_.to32());
245         data_in_dbl_=bin64_t::NONE;
246     }
247     for(int count=0; count<4; count++) {
248         bin64_t ack = file().ack_out().find_filtered // FIXME: do rotating queue
249             (have_out_, bin64_t::ALL, binmap_t::FILLED);
250         if (ack==bin64_t::NONE)
251             break;
252         ack = file().ack_out().cover(ack);
253         have_out_.set(ack);
254         dgram.Push8(SWIFT_HAVE);
255         dgram.Push32(ack.to32());
256         dprintf("%s #%u +have %s\n",tintstr(),id_,ack.str());
257     }
258 }
259
260
261 void    Channel::Recv (Datagram& dgram) {
262     dprintf("%s #%u recvd %ib\n",tintstr(),id_,dgram.size()+4);
263     dgrams_rcvd_++;
264     if (last_send_time_ && rtt_avg_==TINT_SEC && dev_avg_==0) {
265         rtt_avg_ = NOW - last_send_time_;
266         dev_avg_ = rtt_avg_;
267         dip_avg_ = rtt_avg_;
268         dprintf("%s #%u sendctrl rtt init %lli\n",tintstr(),id_,rtt_avg_);
269     }
270     bin64_t data = dgram.size() ? bin64_t::NONE : bin64_t::ALL;
271     while (dgram.size()) {
272         uint8_t type = dgram.Pull8();
273         switch (type) {
274             case SWIFT_HANDSHAKE: OnHandshake(dgram); break;
275             case SWIFT_DATA:      data=OnData(dgram); break;
276             case SWIFT_HAVE:      OnHave(dgram); break;
277             case SWIFT_ACK:       OnAck(dgram); break;
278             case SWIFT_HASH:      OnHash(dgram); break;
279             case SWIFT_HINT:      OnHint(dgram); break;
280             case SWIFT_PEX_ADD:   OnPex(dgram); break;
281             default:
282                 eprintf("%s #%u ?msg id unknown %i\n",tintstr(),id_,(int)type);
283                 return;
284         }
285     }
286     last_recv_time_ = NOW;
287     sent_since_recv_ = 0;
288     Reschedule();
289 }
290
291
292 void    Channel::OnHash (Datagram& dgram) {
293     bin64_t pos = dgram.Pull32();
294     Sha1Hash hash = dgram.PullHash();
295     file().OfferHash(pos,hash);
296     dprintf("%s #%u -hash %s\n",tintstr(),id_,pos.str());
297 }
298
299
300 void    Channel::CleanHintOut (bin64_t pos) {
301     int hi = 0;
302     while (hi<hint_out_.size() && !pos.within(hint_out_[hi].bin))
303         hi++;
304     if (hi==hint_out_.size())
305         return; // something not hinted or hinted in far past
306     while (hi--) { // removing likely snubbed hints
307         hint_out_size_ -= hint_out_.front().bin.width();
308         hint_out_.pop_front();
309     }
310     while (hint_out_.front().bin!=pos) {
311         tintbin f = hint_out_.front();
312         f.bin = f.bin.towards(pos);
313         hint_out_.front().bin = f.bin.sibling();
314         hint_out_.push_front(f);
315     }
316     hint_out_.pop_front();
317     hint_out_size_--;
318 }
319
320
321 bin64_t Channel::OnData (Datagram& dgram) {  // TODO: HAVE NONE for corrupted data
322     bin64_t pos = dgram.Pull32();
323     uint8_t *data;
324     int length = dgram.Pull(&data,1024);
325     bool ok = (pos==bin64_t::NONE) || file().OfferData(pos, (char*)data, length) ;
326     dprintf("%s #%u %cdata %s\n",tintstr(),id_,ok?'-':'!',pos.str());
327     data_in_ = tintbin(NOW,bin64_t::NONE);
328     if (!ok)
329         return bin64_t::NONE;
330     data_in_.bin = pos;
331     if (pos!=bin64_t::NONE) {
332         if (last_data_in_time_) {
333             tint dip = NOW - last_data_in_time_;
334             dip_avg_ = ( dip_avg_*3 + dip ) >> 2;
335         }
336         last_data_in_time_ = NOW;
337     }
338     CleanHintOut(pos);
339     return pos;
340 }
341
342
343 void    Channel::OnAck (Datagram& dgram) {
344     bin64_t ackd_pos = dgram.Pull32();
345     tint peer_time = dgram.Pull64(); // FIXME 32
346     // FIXME FIXME: wrap around here
347     if (ackd_pos==bin64_t::NONE)
348         return; // likely, brocken packet / insufficient hashes
349     if (file().size() && ackd_pos.base_offset()>=file().packet_size()) {
350         eprintf("invalid ack: %s\n",ackd_pos.str());
351         return;
352     }
353     ack_in_.set(ackd_pos);
354     int di = 0, ri = 0;
355     // find an entry for the send (data out) event
356     while (  di<data_out_.size() && ( data_out_[di]==tintbin() ||
357            !data_out_[di].bin.within(ackd_pos) )  )
358         di++;
359     // FUTURE: delayed acks
360     // rule out retransmits
361     while (  ri<data_out_tmo_.size() && !data_out_tmo_[ri].bin.within(ackd_pos) )
362         ri++;
363     dprintf("%s #%u %cack %s %lli\n",tintstr(),id_,
364             di==data_out_.size()?'?':'-',ackd_pos.str(),peer_time);
365     if (di!=data_out_.size() && ri==data_out_tmo_.size()) { // not a retransmit
366             // round trip time calculations
367         tint rtt = NOW-data_out_[di].time;
368         rtt_avg_ = (rtt_avg_*7 + rtt) >> 3;
369         dev_avg_ = ( dev_avg_*3 + abs(rtt-rtt_avg_) ) >> 2;
370         assert(data_out_[di].time!=TINT_NEVER);
371             // one-way delay calculations
372         tint owd = peer_time - data_out_[di].time;
373         owd_cur_bin_ = 0;//(owd_cur_bin_+1) & 3;
374         owd_current_[owd_cur_bin_] = owd;
375         if ( owd_min_bin_start_+TINT_SEC*30 < NOW ) {
376             owd_min_bin_start_ = NOW;
377             owd_min_bin_ = (owd_min_bin_+1) & 3;
378             owd_min_bins_[owd_min_bin_] = TINT_NEVER;
379         }
380         if (owd_min_bins_[owd_min_bin_]>owd)
381             owd_min_bins_[owd_min_bin_] = owd;
382         dprintf("%s #%u sendctrl rtt %lli dev %lli based on %s\n",
383                 tintstr(),id_,rtt_avg_,dev_avg_,data_out_[di].bin.str());
384         ack_rcvd_recent_++;
385         // early loss detection by packet reordering
386         for (int re=0; re<di-MAX_REORDERING; re++) {
387             if (data_out_[re]==tintbin())
388                 continue;
389             ack_not_rcvd_recent_++;
390             data_out_tmo_.push_back(data_out_[re].bin);
391             dprintf("%s #%u Rdata %s\n",tintstr(),id_,data_out_.front().bin.str());
392             data_out_cap_ = bin64_t::ALL;
393             data_out_[re] = tintbin();
394         }
395     }
396     if (di!=data_out_.size())
397         data_out_[di]=tintbin();
398     // clear zeroed items
399     while (!data_out_.empty() && ( data_out_.front()==tintbin() ||
400             ack_in_.is_filled(data_out_.front().bin) ) )
401         data_out_.pop_front();
402     assert(data_out_.empty() || data_out_.front().time!=TINT_NEVER);
403 }
404
405
406 void Channel::TimeoutDataOut ( ) {
407     // losses: timeouted packets
408     tint timeout = NOW - ack_timeout();
409     while (!data_out_.empty() && 
410         ( data_out_.front().time<timeout || data_out_.front()==tintbin() ) ) {
411         if (data_out_.front()!=tintbin() && ack_in_.is_empty(data_out_.front().bin)) {
412             ack_not_rcvd_recent_++;
413             data_out_cap_ = bin64_t::ALL;
414             data_out_tmo_.push_back(data_out_.front().bin);
415             dprintf("%s #%u Tdata %s\n",tintstr(),id_,data_out_.front().bin.str());
416         }
417         data_out_.pop_front();
418     }
419     // clear retransmit queue of older items
420     while (!data_out_tmo_.empty() && data_out_tmo_.front().time<NOW-MAX_POSSIBLE_RTT)
421         data_out_tmo_.pop_front();
422 }
423
424
425 void Channel::OnHave (Datagram& dgram) {
426     bin64_t ackd_pos = dgram.Pull32();
427     if (ackd_pos==bin64_t::NONE)
428         return; // wow, peer has hashes
429     ack_in_.set(ackd_pos);
430     dprintf("%s #%u -have %s\n",tintstr(),id_,ackd_pos.str());
431 }
432
433
434 void    Channel::OnHint (Datagram& dgram) {
435     bin64_t hint = dgram.Pull32();
436     // FIXME: wake up here
437     hint_in_.push_back(hint);
438     dprintf("%s #%u -hint %s\n",tintstr(),id_,hint.str());
439 }
440
441
442 void Channel::OnHandshake (Datagram& dgram) {
443     peer_channel_id_ = dgram.Pull32();
444     dprintf("%s #%u -hs %x\n",tintstr(),id_,peer_channel_id_);
445     // self-connection check
446     if (!SELF_CONN_OK) {
447         uint32_t try_id = DecodeID(peer_channel_id_);
448         if (channel(try_id) && !channel(try_id)->peer_channel_id_) {
449             peer_channel_id_ = 0;
450             Close();
451             return; // this is a self-connection
452         }
453     }
454     // FUTURE: channel forking
455 }
456
457
458 void Channel::OnPex (Datagram& dgram) {
459     uint32_t ipv4 = dgram.Pull32();
460     uint16_t port = dgram.Pull16();
461     Address addr(ipv4,port);
462     dprintf("%s #%u -pex %s\n",tintstr(),id_,addr.str());
463     transfer().OnPexIn(addr);
464 }
465
466
467 void    Channel::AddPex (Datagram& dgram) {
468     int chid = transfer().RevealChannel(pex_out_);
469     if (chid==-1 || chid==id_)
470         return;
471     Address a = channels[chid]->peer();
472     dgram.Push8(SWIFT_PEX_ADD);
473     dgram.Push32(a.ipv4());
474     dgram.Push16(a.port());
475     dprintf("%s #%u +pex %s\n",tintstr(),id_,a.str());
476 }
477
478
479 void    Channel::RecvDatagram (SOCKET socket) {
480     Datagram data(socket);
481     data.Recv();
482     const Address& addr = data.address();
483 #define return_log(...) { printf(__VA_ARGS__); }
484     if (data.size()<4)
485         return_log("datagram shorter than 4 bytes %s\n",addr.str());
486     uint32_t mych = data.Pull32();
487     Sha1Hash hash;
488     Channel* channel = NULL;
489     if (!mych) { // handshake initiated
490         if (data.size()<1+4+1+4+Sha1Hash::SIZE)
491             return_log ("%s #0 incorrect size %i initial handshake packet %s\n",
492                         tintstr(),data.size(),addr.str());
493         uint8_t hashid = data.Pull8();
494         if (hashid!=SWIFT_HASH)
495             return_log ("%s #0 no hash in the initial handshake %s\n",
496                         tintstr(),addr.str());
497         bin64_t pos = data.Pull32();
498         if (pos!=bin64_t::ALL)
499             return_log ("%s #0 that is not the root hash %s\n",tintstr(),addr.str());
500         hash = data.PullHash();
501         FileTransfer* file = FileTransfer::Find(hash);
502         if (!file)
503             return_log ("%s #0 hash %s unknown, no such file %s\n",tintstr(),hash.hex().c_str(),addr.str());
504         dprintf("%s #0 -hash ALL %s\n",tintstr(),hash.hex().c_str());
505         for(binqueue::iterator i=file->hs_in_.begin(); i!=file->hs_in_.end(); i++)
506             if (channels[*i] && channels[*i]->peer_==data.address() &&
507                 channels[*i]->last_recv_time_>NOW-TINT_SEC*2)
508                 return_log("%s #0 have a channel already to %s\n",tintstr(),addr.str());
509         channel = new Channel(file, socket, data.address());
510     } else {
511         mych = DecodeID(mych);
512         if (mych>=channels.size())
513             return_log("%s invalid channel #%u, %s\n",tintstr(),mych,addr.str());
514         channel = channels[mych];
515         if (!channel)
516             return_log ("%s #%u is already closed\n",tintstr(),mych,addr.str());
517         if (channel->peer() != addr)
518             return_log ("%s #%u invalid peer address %s!=%s\n",
519                         tintstr(),mych,channel->peer().str(),addr.str());
520         channel->own_id_mentioned_ = true;
521     }
522     //dprintf("recvd %i bytes for %i\n",data.size(),channel->id);
523     channel->Recv(data);
524 }
525
526
527 void    Channel::Loop (tint howlong) {
528
529     tint limit = Datagram::Time() + howlong;
530
531     do {
532
533         tint send_time(TINT_NEVER);
534         Channel* sender(NULL);
535         while (!sender && !send_queue.is_empty()) { // dequeue
536             tintbin next = send_queue.pop();
537             sender = channel((int)next.bin);
538             send_time = next.time;
539             if (sender && sender->next_send_time_!=send_time &&
540                      sender->next_send_time_!=TINT_NEVER )
541                 sender = NULL; // it was a stale entry
542         }
543
544         if ( sender!=NULL && send_time<=NOW ) { // it's time
545
546             dprintf("%s #%u sch_send %s\n",tintstr(),sender->id(),
547                     tintstr(send_time));
548             sender->Send();
549
550         } else {  // it's too early, wait
551
552             tint towait = min(limit,send_time) - NOW;
553             dprintf("%s #0 waiting %lliusec\n",tintstr(),towait);
554             Datagram::Wait(socket_count,sockets,towait);
555             if (sender)  // get back to that later
556                 send_queue.push(tintbin(send_time,sender->id()));
557
558         }
559
560     } while (NOW<limit);
561
562 }
563
564
565 void Channel::Close () {
566     this->SwitchSendControl(CLOSE_CONTROL);
567 }
568
569
570 void Channel::Reschedule () {
571     next_send_time_ = NextSendTime();
572     if (next_send_time_!=TINT_NEVER) {
573         assert(next_send_time_<NOW+TINT_MIN);
574         send_queue.push(tintbin(next_send_time_,id_));
575         dprintf("%s #%u requeue for %s\n",tintstr(),id_,tintstr(next_send_time_));
576     } else {
577         dprintf("%s #%u closed\n",tintstr(),id_);
578         delete this;
579     }
580 }