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