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).
53 #include "compat/stdint.h"
68 #define NOW Datagram::now
70 /** tintbin is basically a pair<tint,bin64_t> plus some nice operators.
71 Most frequently used in different queues (acknowledgements, requests,
76 tintbin(const tintbin& b) : time(b.time), bin(b.bin) {}
77 tintbin() : time(TINT_NEVER), bin(bin64_t::NONE) {}
78 tintbin(tint time_, bin64_t bin_) : time(time_), bin(bin_) {}
79 tintbin(bin64_t bin_) : time(NOW), bin(bin_) {}
80 bool operator < (const tintbin& b) const
81 { return time > b.time; }
82 bool operator == (const tintbin& b) const
83 { return time==b.time && bin==b.bin; }
84 bool operator != (const tintbin& b) const
85 { return !(*this==b); }
88 typedef std::deque<tintbin> tbqueue;
89 typedef std::deque<bin64_t> binqueue;
90 typedef Address Address;
92 /** A heap (priority queue) for timestamped bin numbers (tintbins). */
96 int size () const { return data_.size(); }
97 bool is_empty () const { return data_.empty(); }
99 tintbin ret = data_.front();
100 std::pop_heap(data_.begin(),data_.end());
104 void push(const tintbin& tb) {
106 push_heap(data_.begin(),data_.end());
108 const tintbin& peek() const {
109 return data_.front();
113 /** swift protocol message types; these are used on the wire. */
123 SWIFT_SIGNED_HASH = 7,
124 SWIFT_MSGTYPE_SENT = 8,
125 SWIFT_MSGTYPE_RCVD = 9,
126 SWIFT_MESSAGE_COUNT = 10
130 class CongestionController;
134 /** A class representing single file transfer. */
139 /** A constructor. Open/submit/retrieve a file.
140 * @param file_name the name of the file
141 * @param root_hash the root hash of the file; zero hash if the file
142 is newly submitted */
143 FileTransfer(const char *file_name, const Sha1Hash& root_hash=Sha1Hash::ZERO);
145 /** Close everything. */
149 /** While we need to feed ACKs to every peer, we try (1) avoid
150 unnecessary duplication and (2) keep minimum state. Thus,
151 we use a rotating queue of bin completion events. */
152 //bin64_t RevealAck (uint64_t& offset);
153 /** Rotating queue read for channels of this transmission. */
154 int RevealChannel (int& i);
156 /** Find transfer by the root hash. */
157 static FileTransfer* Find (const Sha1Hash& hash);
158 /** Find transfer by the file descriptor. */
159 static FileTransfer* file (int fd) {
160 return fd<files.size() ? files[fd] : NULL;
163 /** The binmap for data already retrieved and checked. */
164 binmap_t& ack_out () { return file_.ack_out(); }
165 /** Piece picking strategy used by this transfer. */
166 PiecePicker& picker () { return *picker_; }
167 /** The number of channels working for this transfer. */
168 int channel_count () const { return hs_in_.size(); }
169 /** Hash tree checked file; all the hashes and data are kept here. */
170 HashTree& file() { return file_; }
171 /** File descriptor for the data file. */
172 int fd () const { return file_.file_descriptor(); }
173 /** Root SHA1 hash of the transfer (and the data file). */
174 const Sha1Hash& root_hash () const { return file_.root_hash(); }
178 static std::vector<FileTransfer*> files;
182 /** Piece picker strategy. */
183 PiecePicker* picker_;
185 /** Channels working for this transfer. */
188 std::deque<Address> pex_in_;
190 /** Messages we are accepting. */
196 void OnDataIn (bin64_t pos);
197 void OnPexIn (const Address& addr);
199 friend class Channel;
200 friend uint64_t Size (int fdes);
201 friend bool IsComplete (int fdes);
202 friend uint64_t Complete (int fdes);
203 friend uint64_t SeqComplete (int fdes);
204 friend int Open (const char* filename, const Sha1Hash& hash) ;
205 friend void Close (int fd) ;
209 /** PiecePicker implements some strategy of choosing (picking) what
210 to request next, given the possible range of choices:
211 data acknowledged by the peer minus data already retrieved.
212 May pick sequentially, do rarest first or in some other way. */
215 virtual void Randomize (uint64_t twist) = 0;
216 /** The piece picking method itself.
217 * @param offered the daata acknowledged by the peer
218 * @param max_width maximum number of packets to ask for
219 * @param expires (not used currently) when to consider request expired
220 * @return the bin number to request */
221 virtual bin64_t Pick (binmap_t& offered, uint64_t max_width, tint expires) = 0;
222 virtual ~PiecePicker() {}
228 virtual void AddPeer (const Address& addr, const Sha1Hash& root) = 0;
229 virtual Address GetPeer (const Sha1Hash& for_root) = 0;
235 DataStorer (const Sha1Hash& id, size_t size);
236 virtual size_t ReadData (bin64_t pos,uint8_t** buf) = 0;
237 virtual size_t WriteData (bin64_t pos, uint8_t* buf, size_t len) = 0;
241 /** swift channel's "control block"; channels loosely correspond to TCP
242 connections or FTP sessions; one channel is created for one file
243 being transferred between two peers. As we don't need buffers and
244 lots of other TCP stuff, sizeof(Channel+members) must be below 1K.
245 Normally, API users do not deal with this class. */
248 Channel (FileTransfer* file, int socket=-1, Address peer=Address());
260 static const char* SEND_CONTROL_MODES[];
263 RecvDatagram (int socket);
264 static void Loop (tint till);
266 void Recv (Datagram& dgram);
270 void OnAck (Datagram& dgram);
271 void OnTs (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 AddTs (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 return std::min(30*TINT_SEC,rtt_avg_ + std::max(dev_avg_,MIN_DEV)*4);
316 uint32_t id () const { return id_; }
318 static int DecodeID(int scrambled);
319 static int EncodeID(int unscrambled);
320 static Channel* channel(int i) {
321 return i<channels.size()?channels[i]:NULL;
323 static void CloseTransfer (FileTransfer* trans);
324 static SOCKET default_socket() { return sockets[0]; }
327 /** Channel id: index in the channel array. */
329 /** Socket address of the peer. */
331 /** The UDP socket fd. */
333 /** Descriptor of the file in question. */
334 FileTransfer* transfer_;
335 /** Peer channel id; zero if we are trying to open a channel. */
336 uint32_t peer_channel_id_;
337 bool own_id_mentioned_;
338 /** Peer's progress, based on acknowledgements. */
340 /** Last data received; needs to be acked immediately. */
342 bin64_t data_in_dbl_;
343 /** The history of data sent and still unacknowledged. */
345 /** Timeouted data (potentially to be retransmitted). */
346 tbqueue data_out_tmo_;
347 bin64_t data_out_cap_;
348 /** Index in the history array. */
350 /** Transmit schedule: in most cases filled with the peer's hints */
352 /** Hints sent (to detect and reschedule ignored hints). */
354 uint64_t hint_out_size_;
355 /** Types of messages the peer accepts. */
358 //tint last_send_time, last_recv_time;
361 /** Smoothed averages for RTT, RTT deviation and data interarrival periods. */
362 tint rtt_avg_, dev_avg_, dip_avg_;
363 tint last_send_time_;
364 tint last_recv_time_;
365 tint last_data_out_time_;
366 tint last_data_in_time_;
367 tint last_loss_time_;
368 tint next_send_time_;
369 tint peer_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 void CleanDataOut (bin64_t acks_pos=bin64_t::NONE);
398 void CleanStaleHintOut();
399 void CleanHintOut(bin64_t pos);
402 static PeerSelector* peer_selector;
404 static SOCKET sockets[8];
405 static int socket_count;
406 static tint last_tick;
407 static tbheap send_queue;
409 static Address tracker;
410 static std::vector<Channel*> channels;
412 friend int Listen (Address addr);
413 friend void Shutdown (int sock_des);
414 friend void AddPeer (Address address, const Sha1Hash& root);
415 friend void SetTracker(const Address& tracker);
416 friend int Open (const char*, const Sha1Hash&) ; // FIXME
422 /*************** The top-level API ****************/
423 /** Start listening a port. Returns socket descriptor. */
424 int Listen (Address addr);
425 /** Run send/receive loop for the specified amount of time. */
426 void Loop (tint till);
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);
456 //uint32_t Width (const tbinvec& v);
459 /** Must be called by any client using the library */
460 void LibraryInit(void);
466 #define dprintf(...) { if (Channel::debug_file) fprintf(Channel::debug_file,__VA_ARGS__); }
468 #define dprintf(...) {}
470 #define eprintf(...) fprintf(stderr,__VA_ARGS__)