Matt Joiner
ebbd555e7b
Move a bunch of Client methods onto more appropriate types
2016-11-22 21:12:53 +11:00
Matt Joiner
90ca45dd2d
Don't accept pieces with bad indexes
...
Fixes #132
2016-11-22 14:20:48 +11:00
Matt Joiner
84d243afca
Rename connection.downloadedChunk -> receiveChunk
2016-11-22 14:18:09 +11:00
Matt Joiner
6f9bcff8b2
Make downloadedChunk a method of connection
2016-11-22 14:17:30 +11:00
Matt Joiner
0d9348c5d1
Move Client.downloadedChunk to connection.go
2016-11-22 14:16:18 +11:00
Matt Joiner
a49e133d8a
Comment on connection.updatePiecePriority
...
There's also Torrent.updatePiecePriority, and it's always confusing what the difference is.
2016-10-31 16:22:07 +11:00
Matt Joiner
1725133111
Move the implementation of connection.requestPiecePendingChunks out of Torrent
2016-10-25 14:58:17 +11:00
Matt Joiner
ed0dbba384
Make newConnection a method on Client
2016-10-10 16:55:56 +11:00
Matt Joiner
c34234bf34
Split connection.rw into separate Reader and Writer
...
This will make it easier to write hooks for Read and Write separately.
2016-10-10 16:30:51 +11:00
Matt Joiner
de761fb506
Do chunk pooling at Torrent instead of connection level
2016-10-05 15:57:00 +11:00
Ye Yin
d3963eedfd
Use sync.pool for decode buffer
2016-09-12 17:32:04 +10:00
Matt Joiner
2fc814b5eb
Move Client.connectionLoop to connection.mainReadLoop
2016-09-11 14:32:56 +10:00
Matt Joiner
e6d7b52638
Speed up connection.lastHelpful
...
Moving to reflection and interfaces has made it a bottleneck.
2016-08-14 22:39:23 +10:00
Matt Joiner
fbe0ded844
Add connection read stats
2016-07-12 16:42:04 +10:00
Matt Joiner
a4e140b939
New slices package
2016-07-12 16:40:14 +10:00
Matt Joiner
240abaf512
Add Torrent.SetMaxEstablishedConns
2016-07-06 00:42:16 +10:00
Matt Joiner
326b36545b
Prepare to allow max conns per torrent to be configured
2016-07-05 16:23:17 +10:00
Matt Joiner
12191dbfa3
Add counters for uploads
...
Fixes #71 .
2016-07-05 15:52:33 +10:00
Matt Joiner
874bba43f8
Fix crash causing deadlock in connection writer when no messages have been posted yet
2016-06-27 15:45:15 +10:00
Matt Joiner
6cac882030
Reduce the impact of preferring earlier pieces
...
I think urgent pieces at the end of a torrent were getting fairly starved.
2016-05-28 22:04:33 +10:00
Matt Joiner
b55e67dc22
Ban IPs that contribute to failed piece checks
2016-05-24 02:09:47 +10:00
Matt Joiner
d2951c6c4b
Move requestPendingMetadata onto connection from Client
2016-05-16 18:46:38 +10:00
Matt Joiner
dcfee93f96
Fix race marshalling a bitfield after a Have has also been posted
2016-05-11 23:50:21 +10:00
Matt Joiner
69643860ea
Move some behaviour into Client.newConnection
2016-05-07 18:57:38 +10:00
Matt Joiner
0fd73396fd
missinggo.Event changed, connection.writeOptimizer changes
2016-05-07 18:56:44 +10:00
Matt Joiner
63c5ca1f01
Fix crash when peers have sent undersized bitfields and we get the torrent info
2016-04-30 04:08:02 +10:00
Seth Hoenig
9f4bf70e17
refactor struct identifiers to follow conventional go names
2016-04-18 23:11:11 -05:00
Matt Joiner
d3a1c79c79
Merge Torrent and torrent types
2016-04-03 18:40:43 +10:00
Matt Joiner
bfd0b86441
Tidy up the peer have{,all,none},bitfield handling
2016-03-22 12:07:03 +11:00
Matt Joiner
99fa0c26aa
Make peerGotPiece a method on connection
2016-03-21 14:19:51 +11:00
Matt Joiner
1508e02935
Move peerGotPiece
2016-03-21 14:17:15 +11:00
Matt Joiner
409d3695a6
Favour earlier pieces slightly, helps with in-order downloads
2016-03-19 17:40:18 +11:00
Matt Joiner
a6759b321a
Only assert we should be requesting a piece in debug mode
2016-02-15 17:47:04 +11:00
Matt Joiner
5b5ecb8f3b
Still update a connections piece priority even if the peer doesn't have the piece
2016-02-10 00:47:53 +11:00
Matt Joiner
26fa2c4af8
Track posted message type counts
2016-02-10 00:45:47 +11:00
Matt Joiner
3df7d61836
Reintroduce 64 outbound request cap
2016-02-08 16:09:12 +11:00
Matt Joiner
77d6e9e5cb
Use new missinggo iterator style, and speed up torrent.connHasWantedPieces()
2016-02-07 01:22:31 +11:00
Matt Joiner
19a304b8ea
connection.Close must be called under the Client lock
2016-02-02 00:45:26 +11:00
Matt Joiner
afa264e6c5
Reintroduce connection piece inclinations, and begin caching piece priorities
2016-02-01 21:11:41 +11:00
Matt Joiner
3ed628356b
Use missinggo.Event for connection closing event
2016-02-01 21:08:52 +11:00
Matt Joiner
b90dacd324
Abstract the pendingPieces as a bitmap
2016-02-01 01:46:28 +11:00
Matt Joiner
ae9bba2611
Fix downloading of unwanted chunks, and write out downloaded chunks using the connection loop goroutine
...
Doing the writes with the connection loop goroutine prevents memory use blowing out when chunks aren't written out quickly enough.
2016-01-28 05:54:48 +11:00
Matt Joiner
0ad4dda9fc
Move fillRequests and replenishConnRequests into connection.go
2016-01-24 15:21:17 +11:00
Matt Joiner
06445f2a1e
It's working and the tests are usually passing
...
I still need to handle "prefetch"-style downloading, and some functions haven't been committed to force this issue.
2016-01-18 18:35:14 +11:00
Matt Joiner
09050ff2c5
Some missing hunks that fix build
2016-01-04 22:56:36 +11:00
Matt Joiner
3c446086af
Add Have and Bitfield connection methods
2016-01-04 22:37:49 +11:00
Matt Joiner
58c997210e
Track peer sources with a dict, and don't exceed high water mark
2015-09-28 15:30:13 +10:00
Matt Joiner
53259397fc
Initialize connection piece priorities lazily, and using a pool
2015-09-26 17:27:35 +10:00
Matt Joiner
54339a30e3
Drop connections that contributed to pieces that failed to hash
...
There are peers out there that seem to intentionally spam bad data, and the same pieces will often fail continuously. This really puts a bee in their bonnet.
2015-08-04 01:12:09 +10:00
Matt Joiner
1cc50f85a6
connection.writer wasn't working efficiently
...
Because of the pseudo-random selection of a communication in a select statement, flushing was occuring prematurely. Also the buffer was needlessly large. For large messages, like outgoing pieces, it's probably better they just go straight to the wire anyway.
This change will make it easier to implement sending of pieces that doesn't block control messages in the future. Any time that the buffer is empty, we'll be able to signal that it's an optimal time to send a piece.
2015-08-02 20:39:31 +10:00
