Aleksander Mistewicz
|
21b236771b
|
Add a simple transfer benchmark
|
2019-09-28 14:25:42 +02:00 |
|
Aleksander Mistewicz
|
8677a042cf
|
Wait for nodes to negotiate
|
2019-09-28 14:25:42 +02:00 |
|
Aleksander Mistewicz
|
fffbbbcbd3
|
Pass message between nodes
|
2019-09-28 14:25:42 +02:00 |
|
Aleksander Mistewicz
|
d96fb27ab8
|
Add simple connection test
|
2019-09-28 14:25:42 +02:00 |
|
Arceliar
|
0f99d590a1
|
typo, ipv6->ipv4
|
2019-09-26 18:15:26 -05:00 |
|
Arceliar
|
e16d3efb0a
|
check packet length before checking if it's an ipv6 packet, and add some trace level logging whenever a packet is rejected for being too short to parse
|
2019-09-26 18:11:58 -05:00 |
|
Arceliar
|
a87581b0fa
|
Merge pull request #556 from Arceliar/switch
Switch hack
|
2019-09-25 17:58:01 -05:00 |
|
Arceliar
|
ac58c3586e
|
cleanup/comments
|
2019-09-25 17:53:25 -05:00 |
|
Neil Alexander
|
d27891aaf6
|
Merge pull request #528 from yggdrasil-network/documentation
Documentation updates
|
2019-09-25 17:09:09 +01:00 |
|
Arceliar
|
b9e74f34ec
|
replace the send-to-self with a timer and an arbitrary timeout; i don't really like this but it seems to work better (1 ms is fast by human standards but an eternity for a syscall or the scheduler, so i think that's reasonable)
|
2019-09-24 18:28:13 -05:00 |
|
Arceliar
|
8c64e6fa09
|
explicitly notify the switch when a link appears to be blocked in a send instead of assuming this is the case for all idle links. how we decide when it's really blocked still needs testing/optimizing
|
2019-09-24 18:01:35 -05:00 |
|
Arceliar
|
691192ff5a
|
weird scheduler hack, seems to tend to make things more stable without actually locking streams to any particular link
|
2019-09-21 14:33:45 -05:00 |
|
Arceliar
|
87658f83e9
|
Revert "force things to buffer in the switch if the best link is currently busy. note that other links can end up sending if they become non-idle for other reasons. this is a temporary workaround to packet reordering, until we can figure out a better solution"
This reverts commit 80ba24d512.
|
2019-09-20 23:09:12 -05:00 |
|
Arceliar
|
8003ea0f3e
|
use a separate multicast beacon interval per multicast interface
|
2019-09-20 17:42:42 -05:00 |
|
Neil Alexander
|
1cd4b6e8dd
|
Increase multicast interval at startup from 1s to 15s
|
2019-09-20 10:08:41 +01:00 |
|
Arceliar
|
f9163a56b6
|
fix race between listener accepting and shutting down
|
2019-09-19 19:50:45 -05:00 |
|
Arceliar
|
eeb34ce4e4
|
modify TcpListener
|
2019-09-19 19:45:17 -05:00 |
|
Arceliar
|
93e81867fd
|
have link.stop signal active links to close, have tcp.stop wait for all listeners and active connections to close
|
2019-09-19 19:15:59 -05:00 |
|
Neil Alexander
|
681e9afc79
|
Merge develop into bugfixes
|
2019-09-19 09:05:56 +01:00 |
|
Neil Alexander
|
7b1678a11d
|
Goroutines in _addPeerLoop from bugfixes
|
2019-09-19 09:04:25 +01:00 |
|
Neil Alexander
|
5a382e7e0b
|
Cherrypick fixes for _addPeerLoop memory leak for now
|
2019-09-19 08:55:55 +01:00 |
|
Arceliar
|
995d67cca8
|
fix leak in _addPeerLoop
|
2019-09-18 18:46:03 -05:00 |
|
Arceliar
|
92d9274f3f
|
resolve conflicts
|
2019-09-18 18:40:01 -05:00 |
|
Arceliar
|
2d64a6380a
|
misc other fixes
|
2019-09-18 18:33:51 -05:00 |
|
Neil Alexander
|
909e4e29a8
|
Don't spawn goroutines for addPeerLoop, TCP connect timeout of 5 seconds for now
|
2019-09-18 23:44:28 +01:00 |
|
Neil Alexander
|
64570a8d3e
|
Merge pull request #542 from Arceliar/switch
Switch
|
2019-09-18 20:26:48 +01:00 |
|
Neil Alexander
|
0a12e4b1c1
|
Revert "Catch a nil pointer when sending a session packet to a conn, this shouldn't happen but it's caused multiple crashes in conn.recvMsg"
This reverts commit be35675d0f.
|
2019-09-18 20:26:06 +01:00 |
|
Neil Alexander
|
ddaaa865cb
|
Be more verbose when a peer or listener is badly formatted
|
2019-09-18 19:58:41 +01:00 |
|
Neil Alexander
|
94cf2854a9
|
Fix panic where slice goes out of bounds because iface.Read returns less than zero (which might happen when the TUN/TAP interface is closed)
|
2019-09-18 19:48:53 +01:00 |
|
Neil Alexander
|
ae0b2672ff
|
Fix #539
|
2019-09-18 19:48:16 +01:00 |
|
Neil Alexander
|
2dc136f94a
|
Multicast actor to prevent races
|
2019-09-18 16:51:46 +01:00 |
|
Neil Alexander
|
b959f53fee
|
Shut down listeners when stopping
|
2019-09-18 16:32:22 +01:00 |
|
Neil Alexander
|
b0df9e2f31
|
Fix race when adding peers
|
2019-09-18 16:15:33 +01:00 |
|
Neil Alexander
|
c78a4cb28f
|
Only stop timers if they are running
|
2019-09-18 15:34:26 +01:00 |
|
Neil Alexander
|
366fe7e772
|
Allow multicast to be shut down more sanely
|
2019-09-18 15:31:43 +01:00 |
|
Neil Alexander
|
00a972b74e
|
Disconnect peers when stopping, stop modules before core
|
2019-09-18 15:22:17 +01:00 |
|
Neil Alexander
|
846df4789a
|
Be more verbose when a peer or listener is badly formatted
|
2019-09-18 15:01:19 +01:00 |
|
Neil Alexander
|
200b3623b2
|
Fix #539
|
2019-09-18 14:32:28 +01:00 |
|
Neil Alexander
|
e9bacda0b3
|
Catch a nil pointer when sending a session packet to a conn, this shouldn't happen but it's caused multiple crashes in conn.recvMsg
|
2019-09-18 14:07:26 +01:00 |
|
Neil Alexander
|
c3016e680c
|
Fix panic where slice goes out of bounds because iface.Read returns less than zero (which might happen when the TUN/TAP interface is closed)
|
2019-09-18 14:05:18 +01:00 |
|
Neil Alexander
|
40204caab6
|
Try to fix race condition in sessions.reset
|
2019-09-18 14:03:31 +01:00 |
|
Neil Alexander
|
be35675d0f
|
Catch a nil pointer when sending a session packet to a conn, this shouldn't happen but it's caused multiple crashes in conn.recvMsg
|
2019-09-18 13:37:01 +01:00 |
|
Arceliar
|
80ba24d512
|
force things to buffer in the switch if the best link is currently busy. note that other links can end up sending if they become non-idle for other reasons. this is a temporary workaround to packet reordering, until we can figure out a better solution
|
2019-09-17 19:42:07 -05:00 |
|
Arceliar
|
0141180279
|
cleanup
|
2019-09-09 19:25:10 -05:00 |
|
Arceliar
|
10a828af2c
|
when forwarding traffic, break distance ties by favoring the link that sent the most recent switch update the fastest
|
2019-09-09 19:20:46 -05:00 |
|
Arceliar
|
2426a87ccc
|
really finish initializing the session before returning it / giving up control of the router, in the Conn.search function used by Dial
|
2019-09-03 19:03:12 -05:00 |
|
Neil Alexander
|
af3dcb44d8
|
Update config.go godoc
|
2019-09-02 09:45:11 +01:00 |
|
Arceliar
|
b3361d4bbc
|
package level documentation for address/crypto/util
|
2019-09-01 19:01:33 -05:00 |
|
Arceliar
|
cd99d04bd4
|
document address, crypto, and util
|
2019-09-01 18:53:45 -05:00 |
|
Neil Alexander
|
903a8921fc
|
Update api.go godoc
|
2019-09-01 23:47:47 +01:00 |
|
Neil Alexander
|
935324efe1
|
Update conn.go godoc
|
2019-09-01 23:33:51 +01:00 |
|
Neil Alexander
|
9e8e1c5a41
|
Documentation updates
|
2019-09-01 23:10:46 +01:00 |
|
Neil Alexander
|
01517e5dc3
|
Create doc.go for godoc preamble
|
2019-09-01 22:43:27 +01:00 |
|
Arceliar
|
8c52ccadf9
|
make dial fail if a session to the same node already exists, fixes race between simultaneous connections to a node's 200 address and one of its 300 addresses, should also fix races between a search and an accepted listen
|
2019-09-01 14:07:00 -05:00 |
|
Arceliar
|
8d2c31d39c
|
add some artifical delay to windows netsh commands, since it seems like maybe they don't take effect immediately, and this was leading to races when setting MTU
|
2019-09-01 13:20:48 -05:00 |
|
Arceliar
|
c53831696b
|
make tun stop check that iface is not nil, in case it wasn't set for some reason (windows bugs)
|
2019-09-01 13:06:25 -05:00 |
|
Arceliar
|
d08c2eb237
|
stop exporting ReadNoCopy and WriteNoCopy, since we use the actor functions / callbacks and everything else should use Read and Write instead...
|
2019-09-01 13:04:10 -05:00 |
|
Arceliar
|
3a493fe894
|
gc more often on mobile
|
2019-09-01 11:08:25 -05:00 |
|
Arceliar
|
cabdc27a54
|
change how nonce is tracked, so we allow packets if we've recently received a highest nonce ever, but don't bother tracking all received nonce values, this means duplicate packets are possible but only for a small window of time (and significantly reduces memory usage per session)
|
2019-08-31 17:39:05 -05:00 |
|
Arceliar
|
a64f7320d8
|
update phony, add mobile versions of util bytes functions that don't try to store anything
|
2019-08-31 16:27:36 -05:00 |
|
Arceliar
|
7649ea0f9f
|
remove sessionInfo.doFunc, have the api just use phony.Block instead
|
2019-08-29 21:59:28 -05:00 |
|
Neil Alexander
|
1f658cce76
|
Add Core actor
|
2019-08-28 19:53:52 +01:00 |
|
Neil Alexander
|
aa0770546e
|
Move responsibility for configuring max queue size into switch
|
2019-08-28 19:39:23 +01:00 |
|
Neil Alexander
|
fc9a1c6c31
|
Simplify reconfiguration
|
2019-08-28 19:31:04 +01:00 |
|
Neil Alexander
|
764f9c8e11
|
Remove legacy debug functions
|
2019-08-28 17:24:41 +01:00 |
|
Neil Alexander
|
881d0a1ada
|
Fix DEBUG_getDHTSize
|
2019-08-28 12:46:49 +01:00 |
|
Neil Alexander
|
e553f3e013
|
Reconfigure functions now ran by actors
|
2019-08-28 12:46:12 +01:00 |
|
Neil Alexander
|
607c906820
|
Pointer receivers for phony.Block
|
2019-08-28 12:26:44 +01:00 |
|
Neil Alexander
|
5d7d84f827
|
Remove router.doAdmin and switchTable.doAdmin
|
2019-08-28 12:17:19 +01:00 |
|
Arceliar
|
a8b323acdd
|
have an actor manage the crypto worker pool instead of each session trying to use it directly, this should result in a fairer round-robin behavior in cases where crypto congestion is the bottleneck
|
2019-08-27 20:01:37 -05:00 |
|
Arceliar
|
3845f81357
|
update to latest phony, adjust interface use accordingly
|
2019-08-27 19:43:54 -05:00 |
|
Arceliar
|
4d9c6342a7
|
more link updates
|
2019-08-26 18:37:38 -05:00 |
|
Arceliar
|
c97dd4ad28
|
fix dial bug
|
2019-08-26 00:38:14 -05:00 |
|
Arceliar
|
ab59129557
|
have the writer clean things up. note that their still seem to be bugs in the main linkInterface actor's state machine--links sometimes just die, presumably because they're dropped from the switch and never replaced
|
2019-08-25 23:24:18 -05:00 |
|
Arceliar
|
bd3eaefb72
|
more link migration
|
2019-08-25 22:55:17 -05:00 |
|
Arceliar
|
b5b179904b
|
ugly work-in-progress to migrate link to the actor model
|
2019-08-25 22:19:20 -05:00 |
|
Arceliar
|
dffd70119d
|
remove session shutdown goroutine, just send a message instead
|
2019-08-25 19:13:47 -05:00 |
|
Arceliar
|
b2a2e251ad
|
more TunAdapter migration
|
2019-08-25 18:53:11 -05:00 |
|
Arceliar
|
aaf34c6304
|
start migrating the TunAdapter to the actor model
|
2019-08-25 18:08:43 -05:00 |
|
Arceliar
|
502f2937a9
|
a couple race fixes and use timer.AfterFunc instead of sleeping goroutines or ticker in a few places
|
2019-08-25 17:00:02 -05:00 |
|
Arceliar
|
a3d4d8125b
|
make the main library reconfiguration more actor-friendly
|
2019-08-25 12:10:59 -05:00 |
|
Arceliar
|
aa30c6cc98
|
upgrade phony dependency and switch to its new interface
|
2019-08-25 10:36:09 -05:00 |
|
Arceliar
|
5312b21665
|
Merge branch 'develop' of https://github.com/yggdrasil-network/yggdrasil-go into actors
|
2019-08-24 18:30:15 -05:00 |
|
Arceliar
|
f62bc842ae
|
fix memory leak in session nonce map
|
2019-08-24 18:23:54 -05:00 |
|
Arceliar
|
48bbdac9b3
|
add a helper actor to the link reader to make it play nicer with backpressure
|
2019-08-24 16:27:12 -05:00 |
|
Arceliar
|
99be6b037d
|
stop synchronizing message reads for now, not 100% safe but I don't have any better ideas
|
2019-08-24 16:13:34 -05:00 |
|
Arceliar
|
209d2ffea5
|
correctly call peer.sendPacketsFrom in the switch
|
2019-08-24 16:04:05 -05:00 |
|
Arceliar
|
8c7e9ec7c0
|
fix debug builds
|
2019-08-24 15:32:19 -05:00 |
|
Arceliar
|
c573170886
|
remove switch doworker loop, start a dummy loop to respond to (unused) reconfiguration instead
|
2019-08-24 15:27:56 -05:00 |
|
Arceliar
|
998c76fd8c
|
more switch migration
|
2019-08-24 15:22:46 -05:00 |
|
Arceliar
|
555b4c18d4
|
a little switch cleanup
|
2019-08-24 15:05:18 -05:00 |
|
Arceliar
|
498bc395e2
|
start migrating switch to the actor model
|
2019-08-24 14:56:33 -05:00 |
|
Arceliar
|
b337228aa4
|
minor fixes to peer stuff
|
2019-08-24 14:24:42 -05:00 |
|
Arceliar
|
0539dee900
|
warning about possible deadlock in legacy channel send, need to migrate the link code to fix it
|
2019-08-24 13:25:38 -05:00 |
|
Arceliar
|
034fece33f
|
more peer migration
|
2019-08-24 13:15:29 -05:00 |
|
Arceliar
|
ecd23ce9fc
|
safer linkloop
|
2019-08-24 12:59:20 -05:00 |
|
Arceliar
|
88161009e9
|
more peer migration
|
2019-08-24 12:55:49 -05:00 |
|
Arceliar
|
775fb535dc
|
start converting the peer struct into an actor
|
2019-08-24 12:46:24 -05:00 |
|
Arceliar
|
ef15a6bd79
|
tunConn cleanup
|
2019-08-24 11:44:21 -05:00 |
|
Arceliar
|
4893a07696
|
start migrating tunConn to the actor model
|
2019-08-24 11:38:47 -05:00 |
|
Arceliar
|
b582c444f8
|
minor cleanup
|
2019-08-24 01:57:08 -05:00 |
|
Arceliar
|
1e346aaad0
|
have the conn actor receive messages from the session actor and either pass them to a callback or buffer them in a channel for Read to use if no callback was set
|
2019-08-24 01:52:21 -05:00 |
|
Arceliar
|
9948e3d659
|
add Conn.WriteFrom to allow actor-based sending
|
2019-08-24 00:44:02 -05:00 |
|
Arceliar
|
da9f7151e3
|
more conn migration
|
2019-08-24 00:17:37 -05:00 |
|
Arceliar
|
6ecbc439f0
|
start migrating Conn to be an actor
|
2019-08-23 23:36:00 -05:00 |
|
Arceliar
|
cac3444d9a
|
fix debug builds
|
2019-08-23 22:40:13 -05:00 |
|
Arceliar
|
cf9880464b
|
explicitly consider the session finished case, and make a note that we could fix the packet drop situation by making the Conn into an actor too
|
2019-08-23 22:36:59 -05:00 |
|
Arceliar
|
e3603c0462
|
clean up unused session code
|
2019-08-23 22:25:40 -05:00 |
|
Arceliar
|
533da351f9
|
fix actor EnqueueFrom stack overflow (use nil now to send from self) and replace session send/recv workers with actor functions
|
2019-08-23 22:23:01 -05:00 |
|
Arceliar
|
436c84ca33
|
refactor sessions to store a pointer to router instead of core
|
2019-08-23 20:53:00 -05:00 |
|
Arceliar
|
5bb85cf07b
|
refactor searches to store a pointer to router instead of core
|
2019-08-23 20:42:38 -05:00 |
|
Arceliar
|
e7024a00e7
|
have dht store a pointer to router instead of core
|
2019-08-23 20:35:54 -05:00 |
|
Arceliar
|
ebd806f27a
|
move router member initialization into router.init
|
2019-08-23 20:29:16 -05:00 |
|
Arceliar
|
9835c63818
|
refactor things the router owns (dht, sessions, searches) into that struct, to make the ownership more explicit
|
2019-08-23 20:26:15 -05:00 |
|
Arceliar
|
bbcbbaf3b1
|
start migrating sessionInfo to be an actor
|
2019-08-23 20:05:18 -05:00 |
|
Arceliar
|
8e89816099
|
more router migration: rename functions that should only be called internally by the actor
|
2019-08-23 18:59:34 -05:00 |
|
Arceliar
|
232e6d3cb3
|
more router migration
|
2019-08-23 18:55:41 -05:00 |
|
Arceliar
|
9d7e7288c6
|
start migrating the router to an actor
|
2019-08-23 18:47:15 -05:00 |
|
Arceliar
|
12ce8c6a0a
|
Merge pull request #512 from neilalexander/cryptokey
Cryptokey routing changes
|
2019-08-20 20:23:00 -05:00 |
|
Arceliar
|
226dd6170d
|
hopefully prevent a deadlock
|
2019-08-20 18:49:53 -05:00 |
|
Arceliar
|
4156aa3003
|
move ckr checks into the tunConn code
|
2019-08-20 18:10:08 -05:00 |
|
Neil Alexander
|
b79829c43b
|
Merge branch 'develop' into cryptokey
|
2019-08-20 09:43:17 +01:00 |
|
Neil Alexander
|
b6e67bc0ba
|
Check CKR remotes when receiving traffic
|
2019-08-20 09:38:46 +01:00 |
|
Neil Alexander
|
2b6462c8a9
|
Strict checking of Yggdrasil source/destination addresses
|
2019-08-20 09:38:27 +01:00 |
|
Arceliar
|
834a6a6f1a
|
don't allocate a new child cancellation in Conn read/write calls if no deadline is set
|
2019-08-19 18:06:05 -05:00 |
|
Neil Alexander
|
2a629880fd
|
Rename crypto-key config options, improve control flow
|
2019-08-19 10:28:30 +01:00 |
|
Arceliar
|
8af1a7086c
|
when a link becomes idle and packet are buffered that the link could send, send at least 65535 bytes worth instead of 1 packet, this reduces syscall overhead when small packets are sent through the network
|
2019-08-18 12:29:07 -05:00 |
|
Arceliar
|
62337bcd64
|
allow links to send multiple packets at once, currently we still only bother to send 1 at a time from the switch level
|
2019-08-18 12:17:54 -05:00 |
|
Arceliar
|
fd5f3ca764
|
fix heap pop order
|
2019-08-16 23:07:40 -05:00 |
|
Arceliar
|
03b8af9f1a
|
keep track of recent nonces with a heap and a map instead of a fixed-size bitmask
|
2019-08-16 18:37:16 -05:00 |
|
Neil Alexander
|
5b054766a2
|
Update comments in handleIn, add switch_getFlowLabelFromCoords helper (in case it is useful if we try to consider flowlabels in multi-link scenarios)
|
2019-08-15 10:54:04 +01:00 |
|
Arceliar
|
1a2b7a8b60
|
test a change to how switch hops are selected when multiple links are idle
|
2019-08-14 17:57:36 -05:00 |
|
Neil Alexander
|
f26f071901
|
Merge pull request #497 from Slex/issues/488
Implement feature from https://github.com/yggdrasil-network/yggdrasil…
|
2019-08-14 20:11:15 +01:00 |
|
Neil Alexander
|
33cd10c463
|
Merge branch 'issues/488' of github.com:slex/yggdrasil-go into issues/488
|
2019-08-14 19:58:45 +01:00 |
|
Neil Alexander
|
4702da2bcb
|
Use new netlink library (fixes #493)
|
2019-08-14 19:32:40 +01:00 |
|
Neil Alexander
|
d9fabad8bc
|
Merge pull request #502 from Arceliar/linkleak
Try to fix leaks in #501
|
2019-08-14 07:17:39 +01:00 |
|
Arceliar
|
46c5df1c23
|
when we abandon a link because we already have a connection to that peer, only wait for the connection to close if it's an *outgoing* link, otherwise incomming connection attempts can cause us to leak links
|
2019-08-13 18:49:49 -05:00 |
|
Arceliar
|
b2cb1d965c
|
avoid leaking sessions when no listener exists, or blocking if it's busy
|
2019-08-12 18:22:30 -05:00 |
|
Arceliar
|
277da1fe60
|
make sure searches don't end if try to continue (in parallel) with nowhere left to send, but we just sent a search and are still waiting for a response
|
2019-08-11 13:11:14 -05:00 |
|
Arceliar
|
7a28eb787e
|
try to fix a few edge cases with searches that could lead them to ending without the callback being run or without cleaning up the old search info
|
2019-08-11 13:00:19 -05:00 |
|
Slex
|
589ad638ea
|
Implement feature from https://github.com/yggdrasil-network/yggdrasil-go/issues/488
|
2019-08-11 00:31:22 +03:00 |
|
Arceliar
|
5e81a0c421
|
Use a separate buffer per session for incoming packets, so 1 session that floods won't block other sessions
|
2019-08-07 18:08:31 -05:00 |
|
Arceliar
|
9ab08446ff
|
make sure the sessionInfo.recvWorker doesn't block if sinfo.recv somehow fills
|
2019-08-07 17:40:50 -05:00 |
|
Neil Alexander
|
c99ed9fb60
|
Merge pull request #491 from Arceliar/flowkey
Fix the old flowkey stuff so congestion control actually works...
|
2019-08-07 10:33:17 +01:00 |
|
Arceliar
|
d795ab1b65
|
minor allocation fix
|
2019-08-06 20:51:38 -05:00 |
|
Arceliar
|
790524bd1c
|
copy/paste old flowkey logic into a util function, add a struct of key and packet, make WriteNoCopy accept this instead of a slice
|
2019-08-06 19:25:55 -05:00 |
|
Arceliar
|
679866d5ff
|
have createSession fill the sessionInfo.cancel field, have Conn use Conn.session.cancel instead of storing its own cancellation, this should prevent any of these things from being both nil and reachable at the same time
|
2019-08-05 19:11:28 -05:00 |
|
Arceliar
|
8a85149817
|
remove src/.DS_Store
|
2019-08-05 18:50:08 -05:00 |
|
Arceliar
|
84a4f54217
|
temporary fix to nil pointer, better to make sure it's never nil
|
2019-08-05 18:49:15 -05:00 |
|
Neil Alexander
|
bd3b42022b
|
Merge pull request #480 from Arceliar/speedup
Speedup
|
2019-08-05 10:24:54 +01:00 |
|