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
23 ACK 02, 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. */
117 SWIFT_SIGNED_HASH = 7,
119 SWIFT_MSGTYPE_RCVD = 9,
120 SWIFT_MESSAGE_COUNT = 10
124 class CongestionController;
126 typedef void (*TransferProgressCallback) (int transfer, bin64_t bin);
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. */
190 #define SWFT_MAX_TRANSFER_CB 8
191 TransferProgressCallback callbacks[SWFT_MAX_TRANSFER_CB];
195 void OnDataIn (bin64_t pos);
196 void OnPexIn (const Address& addr);
198 friend class Channel;
199 friend uint64_t Size (int fdes);
200 friend bool IsComplete (int fdes);
201 friend uint64_t Complete (int fdes);
202 friend uint64_t SeqComplete (int fdes);
203 friend int Open (const char* filename, const Sha1Hash& hash) ;
204 friend void Close (int fd) ;
205 friend void AddProgressCallback (int transfer,TransferProgressCallback cb);
206 friend void RemoveProgressCallback (int transfer,TransferProgressCallback cb);
207 friend void ExternallyRetrieved (int transfer,bin64_t piece);
211 /** PiecePicker implements some strategy of choosing (picking) what
212 to request next, given the possible range of choices:
213 data acknowledged by the peer minus data already retrieved.
214 May pick sequentially, do rarest first or in some other way. */
217 virtual void Randomize (uint64_t twist) = 0;
218 /** The piece picking method itself.
219 * @param offered the daata acknowledged by the peer
220 * @param max_width maximum number of packets to ask for
221 * @param expires (not used currently) when to consider request expired
222 * @return the bin number to request */
223 virtual bin64_t Pick (binmap_t& offered, uint64_t max_width, tint expires) = 0;
224 virtual void LimitRange (bin64_t range) = 0;
225 virtual ~PiecePicker() {}
231 virtual void AddPeer (const Address& addr, const Sha1Hash& root) = 0;
232 virtual Address GetPeer (const Sha1Hash& for_root) = 0;
238 DataStorer (const Sha1Hash& id, size_t size);
239 virtual size_t ReadData (bin64_t pos,uint8_t** buf) = 0;
240 virtual size_t WriteData (bin64_t pos, uint8_t* buf, size_t len) = 0;
244 /** swift channel's "control block"; channels loosely correspond to TCP
245 connections or FTP sessions; one channel is created for one file
246 being transferred between two peers. As we don't need buffers and
247 lots of other TCP stuff, sizeof(Channel+members) must be below 1K.
248 Normally, API users do not deal with this class. */
251 Channel (FileTransfer* file, int socket=INVALID_SOCKET, Address peer=Address());
263 static const char* SEND_CONTROL_MODES[];
265 static void RecvDatagram (SOCKET socket);
266 static void Loop (tint till);
268 void Recv (Datagram& dgram);
272 void OnAck (Datagram& dgram);
273 void OnHave (Datagram& dgram);
274 bin64_t OnData (Datagram& dgram);
275 void OnHint (Datagram& dgram);
276 void OnHash (Datagram& dgram);
277 void OnPex (Datagram& dgram);
278 void OnHandshake (Datagram& dgram);
279 void AddHandshake (Datagram& dgram);
280 bin64_t AddData (Datagram& dgram);
281 void AddAck (Datagram& dgram);
282 void AddHave (Datagram& dgram);
283 void AddHint (Datagram& dgram);
284 void AddUncleHashes (Datagram& dgram, bin64_t pos);
285 void AddPeakHashes (Datagram& dgram);
286 void AddPex (Datagram& dgram);
288 void BackOffOnLosses (float ratio=0.5);
289 tint SwitchSendControl (int control_mode);
290 tint NextSendTime ();
291 tint KeepAliveNextSendTime ();
292 tint PingPongNextSendTime ();
293 tint CwndRateNextSendTime ();
294 tint SlowStartNextSendTime ();
295 tint AimdNextSendTime ();
296 tint LedbatNextSendTime ();
298 static int MAX_REORDERING;
301 static tint MAX_SEND_INTERVAL;
302 static tint LEDBAT_TARGET;
303 static float LEDBAT_GAIN;
304 static tint LEDBAT_DELAY_BIN;
305 static bool SELF_CONN_OK;
306 static tint MAX_POSSIBLE_RTT;
307 static FILE* debug_file;
309 const std::string id_string () const;
310 /** A channel is "established" if had already sent and received packets. */
311 bool is_established () { return peer_channel_id_ && own_id_mentioned_; }
312 FileTransfer& transfer() { return *transfer_; }
313 HashTree& file () { return transfer_->file(); }
314 const Address& peer() const { return peer_; }
315 tint ack_timeout () {
316 tint dev = dev_avg_ < MIN_DEV ? MIN_DEV : dev_avg_;
317 tint tmo = rtt_avg_ + dev * 4;
318 return tmo < 30*TINT_SEC ? tmo : 30*TINT_SEC;
320 uint32_t id () const { return id_; }
322 static int DecodeID(int scrambled);
323 static int EncodeID(int unscrambled);
324 static Channel* channel(int i) {
325 return i<channels.size()?channels[i]:NULL;
327 static void CloseTransfer (FileTransfer* trans);
330 /** Channel id: index in the channel array. */
332 /** Socket address of the peer. */
334 /** The UDP socket fd. */
336 /** Descriptor of the file in question. */
337 FileTransfer* transfer_;
338 /** Peer channel id; zero if we are trying to open a channel. */
339 uint32_t peer_channel_id_;
340 bool own_id_mentioned_;
341 /** Peer's progress, based on acknowledgements. */
343 /** Last data received; needs to be acked immediately. */
345 bin64_t data_in_dbl_;
346 /** The history of data sent and still unacknowledged. */
348 /** Timeouted data (potentially to be retransmitted). */
349 tbqueue data_out_tmo_;
350 bin64_t data_out_cap_;
351 /** Index in the history array. */
353 /** Transmit schedule: in most cases filled with the peer's hints */
355 /** Hints sent (to detect and reschedule ignored hints). */
357 uint64_t hint_out_size_;
358 /** Types of messages the peer accepts. */
361 //tint last_send_time, last_recv_time;
364 /** Smoothed averages for RTT, RTT deviation and data interarrival periods. */
365 tint rtt_avg_, dev_avg_, dip_avg_;
366 tint last_send_time_;
367 tint last_recv_time_;
368 tint last_data_out_time_;
369 tint last_data_in_time_;
370 tint last_loss_time_;
371 tint next_send_time_;
372 /** Congestion window; TODO: int, bytes. */
374 /** Data sending interval. */
376 /** The congestion control strategy. */
378 /** Datagrams (not data) sent since last recv. */
379 int sent_since_recv_;
380 /** Recent acknowlegements for data previously sent. */
381 int ack_rcvd_recent_;
382 /** Recent non-acknowlegements (losses) of data previously sent. */
383 int ack_not_rcvd_recent_;
384 /** LEDBAT one-way delay machinery */
385 tint owd_min_bins_[4];
387 tint owd_min_bin_start_;
388 tint owd_current_[4];
394 int PeerBPS() const {
395 return TINT_SEC / dip_avg_ * 1024;
397 /** Get a request for one packet from the queue of peer's requests. */
398 bin64_t DequeueHint();
399 bin64_t ImposeHint();
400 void TimeoutDataOut ();
401 void CleanStaleHintOut();
402 void CleanHintOut(bin64_t pos);
405 static PeerSelector* peer_selector;
407 static tint last_tick;
408 static tbheap send_queue;
410 static Address tracker;
411 static std::vector<Channel*> channels;
413 friend int Listen (Address addr);
414 friend void Shutdown (int sock_des);
415 friend void AddPeer (Address address, const Sha1Hash& root);
416 friend void SetTracker(const Address& tracker);
417 friend int Open (const char*, const Sha1Hash&) ; // FIXME
423 /*************** The top-level API ****************/
424 /** Start listening a port. Returns socket descriptor. */
425 int Listen (Address addr);
426 /** Run send/receive loop for the specified amount of time. */
427 void Loop (tint till);
428 bool Listen3rdPartySocket (sckrwecb_t);
429 /** Stop listening to a port. */
430 void Shutdown (int sock_des=-1);
432 /** Open a file, start a transmission; fill it with content for a given root hash;
433 in case the hash is omitted, the file is a fresh submit. */
434 int Open (const char* filename, const Sha1Hash& hash=Sha1Hash::ZERO) ;
435 /** Get the root hash for the transmission. */
436 const Sha1Hash& RootMerkleHash (int file) ;
437 /** Close a file and a transmission. */
438 void Close (int fd) ;
439 /** Add a possible peer which participares in a given transmission. In the case
440 root hash is zero, the peer might be talked to regarding any transmission
441 (likely, a tracker, cache or an archive). */
442 void AddPeer (Address address, const Sha1Hash& root=Sha1Hash::ZERO);
444 void SetTracker(const Address& tracker);
446 /** Returns size of the file in bytes, 0 if unknown. Might be rounded up to a kilobyte
447 before the transmission is complete. */
448 uint64_t Size (int fdes);
449 /** Returns the amount of retrieved and verified data, in bytes.
450 A 100% complete transmission has Size()==Complete(). */
451 uint64_t Complete (int fdes);
452 bool IsComplete (int fdes);
453 /** Returns the number of bytes that are complete sequentially, starting from the
454 beginning, till the first not-yet-retrieved packet. */
455 uint64_t SeqComplete (int fdes);
457 int Find (Sha1Hash hash);
459 void AddProgressCallback (int transfer,TransferProgressCallback cb);
460 void RemoveProgressCallback (int transfer,TransferProgressCallback cb);
461 void ExternallyRetrieved (int transfer,bin64_t piece);
463 //uint32_t Width (const tbinvec& v);
466 /** Must be called by any client using the library */
467 void LibraryInit(void);
473 #define dprintf(...) { if (Channel::debug_file) fprintf(Channel::debug_file,__VA_ARGS__); }
475 #define dprintf(...) {}
477 #define eprintf(...) fprintf(stderr,__VA_ARGS__)