3 * the main header file for libswift, normally you should only read this one
5 * Created by Victor Grishchenko on 3/6/09.
6 * Copyright 2009 Delft University of Technology. All rights reserved.
15 HANDSHAKE 00, channelid
16 Communicates the channel id of the sender. The
17 initial handshake packet also has the root hash
20 DATA 01, bin_32, buffer
24 ACKTS 08, bin_32, timestamp_32
25 Confirms successfull delivery of data. Used for
26 congestion control, as well.
29 Practical value of "hints" is to avoid overlap, mostly.
30 Hints might be lost in the network or ignored.
31 Peer might send out data without a hint.
32 Hint which was not responded (by DATA) in some RTTs
33 is considered to be ignored.
34 As peers cant pick randomly kilobyte here and there,
35 they send out "long hints" for non-base bins.
37 HASH 04, bin_32, sha1hash
38 SHA1 hash tree hashes for data verification. The
39 connection to a fresh peer starts with bootstrapping
40 him with peak hashes. Later, before sending out
41 any data, a peer sends the necessary uncle hashes.
43 PEX+/PEX- 05/06, ipv4 addr, port
44 Peer exchange messages; reports all connected and
45 disconected peers. Might has special meaning (as
46 in the case with swarm supervisors).
63 #define NOW Datagram::now
65 /** tintbin is basically a pair<tint,bin64_t> plus some nice operators.
66 Most frequently used in different queues (acknowledgements, requests,
71 tintbin(const tintbin& b) : time(b.time), bin(b.bin) {}
72 tintbin() : time(TINT_NEVER), bin(bin64_t::NONE) {}
73 tintbin(tint time_, bin64_t bin_) : time(time_), bin(bin_) {}
74 tintbin(bin64_t bin_) : time(NOW), bin(bin_) {}
75 bool operator < (const tintbin& b) const
76 { return time > b.time; }
77 bool operator == (const tintbin& b) const
78 { return time==b.time && bin==b.bin; }
79 bool operator != (const tintbin& b) const
80 { return !(*this==b); }
83 typedef std::deque<tintbin> tbqueue;
84 typedef std::deque<bin64_t> binqueue;
85 typedef Address Address;
87 /** A heap (priority queue) for timestamped bin numbers (tintbins). */
91 int size () const { return data_.size(); }
92 bool is_empty () const { return data_.empty(); }
94 tintbin ret = data_.front();
95 std::pop_heap(data_.begin(),data_.end());
99 void push(const tintbin& tb) {
101 push_heap(data_.begin(),data_.end());
103 const tintbin& peek() const {
104 return data_.front();
108 /** swift protocol message types; these are used on the wire. */
118 SWIFT_SIGNED_HASH = 7,
119 SWIFT_MSGTYPE_SENT = 8,
120 SWIFT_MSGTYPE_RCVD = 9,
121 SWIFT_MESSAGE_COUNT = 10
125 class CongestionController;
129 /** A class representing single file transfer. */
134 /** A constructor. Open/submit/retrieve a file.
135 * @param file_name the name of the file
136 * @param root_hash the root hash of the file; zero hash if the file
137 is newly submitted */
138 FileTransfer(const char *file_name, const Sha1Hash& root_hash=Sha1Hash::ZERO);
140 /** Close everything. */
144 /** While we need to feed ACKs to every peer, we try (1) avoid
145 unnecessary duplication and (2) keep minimum state. Thus,
146 we use a rotating queue of bin completion events. */
147 //bin64_t RevealAck (uint64_t& offset);
148 /** Rotating queue read for channels of this transmission. */
149 int RevealChannel (int& i);
151 /** Find transfer by the root hash. */
152 static FileTransfer* Find (const Sha1Hash& hash);
153 /** Find transfer by the file descriptor. */
154 static FileTransfer* file (int fd) {
155 return fd<files.size() ? files[fd] : NULL;
158 /** The binmap for data already retrieved and checked. */
159 binmap_t& ack_out () { return file_.ack_out(); }
160 /** Piece picking strategy used by this transfer. */
161 PiecePicker& picker () { return *picker_; }
162 /** The number of channels working for this transfer. */
163 int channel_count () const { return hs_in_.size(); }
164 /** Hash tree checked file; all the hashes and data are kept here. */
165 HashTree& file() { return file_; }
166 /** File descriptor for the data file. */
167 int fd () const { return file_.file_descriptor(); }
168 /** Root SHA1 hash of the transfer (and the data file). */
169 const Sha1Hash& root_hash () const { return file_.root_hash(); }
173 static std::vector<FileTransfer*> files;
177 /** Piece picker strategy. */
178 PiecePicker* picker_;
180 /** Channels working for this transfer. */
183 std::deque<Address> pex_in_;
185 /** Messages we are accepting. */
191 void OnDataIn (bin64_t pos);
192 void OnPexIn (const Address& addr);
194 friend class Channel;
195 friend uint64_t Size (int fdes);
196 friend bool IsComplete (int fdes);
197 friend uint64_t Complete (int fdes);
198 friend uint64_t SeqComplete (int fdes);
199 friend int Open (const char* filename, const Sha1Hash& hash) ;
200 friend void Close (int fd) ;
204 /** PiecePicker implements some strategy of choosing (picking) what
205 to request next, given the possible range of choices:
206 data acknowledged by the peer minus data already retrieved.
207 May pick sequentially, do rarest first or in some other way. */
210 virtual void Randomize (uint64_t twist) = 0;
211 /** The piece picking method itself.
212 * @param offered the daata acknowledged by the peer
213 * @param max_width maximum number of packets to ask for
214 * @param expires (not used currently) when to consider request expired
215 * @return the bin number to request */
216 virtual bin64_t Pick (binmap_t& offered, uint64_t max_width, tint expires) = 0;
217 virtual ~PiecePicker() {}
223 virtual void AddPeer (const Address& addr, const Sha1Hash& root) = 0;
224 virtual Address GetPeer (const Sha1Hash& for_root) = 0;
230 DataStorer (const Sha1Hash& id, size_t size);
231 virtual size_t ReadData (bin64_t pos,uint8_t** buf) = 0;
232 virtual size_t WriteData (bin64_t pos, uint8_t* buf, size_t len) = 0;
236 /** swift channel's "control block"; channels loosely correspond to TCP
237 connections or FTP sessions; one channel is created for one file
238 being transferred between two peers. As we don't need buffers and
239 lots of other TCP stuff, sizeof(Channel+members) must be below 1K.
240 Normally, API users do not deal with this class. */
243 Channel (FileTransfer* file, int socket=-1, Address peer=Address());
255 static const char* SEND_CONTROL_MODES[];
258 RecvDatagram (int socket);
259 static void Loop (tint till);
261 void Recv (Datagram& dgram);
265 void OnAck (Datagram& dgram);
266 void OnTs (Datagram& dgram);
267 bin64_t OnData (Datagram& dgram);
268 void OnHint (Datagram& dgram);
269 void OnHash (Datagram& dgram);
270 void OnPex (Datagram& dgram);
271 void OnHandshake (Datagram& dgram);
272 void AddHandshake (Datagram& dgram);
273 bin64_t AddData (Datagram& dgram);
274 void AddAck (Datagram& dgram);
275 void AddTs (Datagram& dgram);
276 void AddHint (Datagram& dgram);
277 void AddUncleHashes (Datagram& dgram, bin64_t pos);
278 void AddPeakHashes (Datagram& dgram);
279 void AddPex (Datagram& dgram);
281 void BackOffOnLosses (float ratio=0.5);
282 tint SwitchSendControl (int control_mode);
283 tint NextSendTime ();
284 tint KeepAliveNextSendTime ();
285 tint PingPongNextSendTime ();
286 tint CwndRateNextSendTime ();
287 tint SlowStartNextSendTime ();
288 tint AimdNextSendTime ();
289 tint LedbatNextSendTime ();
291 static int MAX_REORDERING;
294 static tint MAX_SEND_INTERVAL;
295 static tint LEDBAT_TARGET;
296 static float LEDBAT_GAIN;
297 static tint LEDBAT_DELAY_BIN;
298 static bool SELF_CONN_OK;
299 static tint MAX_POSSIBLE_RTT;
300 static FILE* debug_file;
302 const std::string id_string () const;
303 /** A channel is "established" if had already sent and received packets. */
304 bool is_established () { return peer_channel_id_ && own_id_mentioned_; }
305 FileTransfer& transfer() { return *transfer_; }
306 HashTree& file () { return transfer_->file(); }
307 const Address& peer() const { return peer_; }
308 tint ack_timeout () {
309 tint dev = dev_avg_ < MIN_DEV ? MIN_DEV : dev_avg_;
310 tint tmo = rtt_avg_ + dev * 4;
311 return tmo < 30*TINT_SEC ? tmo : 30*TINT_SEC;
313 uint32_t id () const { return id_; }
315 static int DecodeID(int scrambled);
316 static int EncodeID(int unscrambled);
317 static Channel* channel(int i) {
318 return i<channels.size()?channels[i]:NULL;
320 static void CloseTransfer (FileTransfer* trans);
321 static SOCKET default_socket() { return sockets[0]; }
324 /** Channel id: index in the channel array. */
326 /** Socket address of the peer. */
328 /** The UDP socket fd. */
330 /** Descriptor of the file in question. */
331 FileTransfer* transfer_;
332 /** Peer channel id; zero if we are trying to open a channel. */
333 uint32_t peer_channel_id_;
334 bool own_id_mentioned_;
335 /** Peer's progress, based on acknowledgements. */
337 /** Last data received; needs to be acked immediately. */
339 bin64_t data_in_dbl_;
340 /** The history of data sent and still unacknowledged. */
342 /** Timeouted data (potentially to be retransmitted). */
343 tbqueue data_out_tmo_;
344 bin64_t data_out_cap_;
345 /** Index in the history array. */
347 /** Transmit schedule: in most cases filled with the peer's hints */
349 /** Hints sent (to detect and reschedule ignored hints). */
351 uint64_t hint_out_size_;
352 /** Types of messages the peer accepts. */
355 //tint last_send_time, last_recv_time;
358 /** Smoothed averages for RTT, RTT deviation and data interarrival periods. */
359 tint rtt_avg_, dev_avg_, dip_avg_;
360 tint last_send_time_;
361 tint last_recv_time_;
362 tint last_data_out_time_;
363 tint last_data_in_time_;
364 tint last_loss_time_;
365 tint next_send_time_;
366 tint peer_send_time_;
367 /** Congestion window; TODO: int, bytes. */
369 /** Data sending interval. */
371 /** The congestion control strategy. */
373 /** Datagrams (not data) sent since last recv. */
374 int sent_since_recv_;
375 /** Recent acknowlegements for data previously sent. */
376 int ack_rcvd_recent_;
377 /** Recent non-acknowlegements (losses) of data previously sent. */
378 int ack_not_rcvd_recent_;
379 /** LEDBAT one-way delay machinery */
380 tint owd_min_bins_[4];
382 tint owd_min_bin_start_;
383 tint owd_current_[4];
389 int PeerBPS() const {
390 return TINT_SEC / dip_avg_ * 1024;
392 /** Get a request for one packet from the queue of peer's requests. */
393 bin64_t DequeueHint();
394 void CleanDataOut (bin64_t acks_pos=bin64_t::NONE);
395 void CleanStaleHintOut();
396 void CleanHintOut(bin64_t pos);
399 static PeerSelector* peer_selector;
401 static SOCKET sockets[8];
402 static int socket_count;
403 static tint last_tick;
404 static tbheap send_queue;
406 static Address tracker;
407 static std::vector<Channel*> channels;
409 friend int Listen (Address addr);
410 friend void Shutdown (int sock_des);
411 friend void AddPeer (Address address, const Sha1Hash& root);
412 friend void SetTracker(const Address& tracker);
413 friend int Open (const char*, const Sha1Hash&) ; // FIXME
419 /*************** The top-level API ****************/
420 /** Start listening a port. Returns socket descriptor. */
421 int Listen (Address addr);
422 /** Run send/receive loop for the specified amount of time. */
423 void Loop (tint till);
424 /** Stop listening to a port. */
425 void Shutdown (int sock_des=-1);
427 /** Open a file, start a transmission; fill it with content for a given root hash;
428 in case the hash is omitted, the file is a fresh submit. */
429 int Open (const char* filename, const Sha1Hash& hash=Sha1Hash::ZERO) ;
430 /** Get the root hash for the transmission. */
431 const Sha1Hash& RootMerkleHash (int file) ;
432 /** Close a file and a transmission. */
433 void Close (int fd) ;
434 /** Add a possible peer which participares in a given transmission. In the case
435 root hash is zero, the peer might be talked to regarding any transmission
436 (likely, a tracker, cache or an archive). */
437 void AddPeer (Address address, const Sha1Hash& root=Sha1Hash::ZERO);
439 void SetTracker(const Address& tracker);
441 /** Returns size of the file in bytes, 0 if unknown. Might be rounded up to a kilobyte
442 before the transmission is complete. */
443 uint64_t Size (int fdes);
444 /** Returns the amount of retrieved and verified data, in bytes.
445 A 100% complete transmission has Size()==Complete(). */
446 uint64_t Complete (int fdes);
447 bool IsComplete (int fdes);
448 /** Returns the number of bytes that are complete sequentially, starting from the
449 beginning, till the first not-yet-retrieved packet. */
450 uint64_t SeqComplete (int fdes);
453 //uint32_t Width (const tbinvec& v);
456 /** Must be called by any client using the library */
457 void LibraryInit(void);
463 #define dprintf(...) { if (Channel::debug_file) fprintf(Channel::debug_file,__VA_ARGS__); }
465 #define dprintf(...) {}
467 #define eprintf(...) fprintf(stderr,__VA_ARGS__)