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
|
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
|
9d7e7288c6
|
start migrating the router to an actor
|
2019-08-23 18:47:15 -05:00 |
|
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 |
|
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 |
|
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
|
5d5486049b
|
add Conn.ReadNoCopy and Conn.WriteNoCopy that transfer ownership of a slice instead of copying, have Read and Write use the NoCopy versions under the hood and just manage copying as needed
|
2019-08-04 15:53:34 -05:00 |
|
Arceliar
|
6da5802ae5
|
don't block forever in Write if the session is cancelled, cleanup Conn.Read slightly
|
2019-08-04 02:08:47 -05:00 |
|
Arceliar
|
72ed541bf3
|
a little cleanup to Conn functions
|
2019-08-03 22:07:38 -05:00 |
|
Arceliar
|
df0090e32a
|
Add per-session read/write workers, work in progress, they still unfortunately need to take a mutex for safety
|
2019-08-03 21:46:18 -05:00 |
|
Neil Alexander
|
853054eb62
|
Merge pull request #475 from Arceliar/misc
Misc
|
2019-07-29 20:24:49 +01:00 |
|
Arceliar
|
b66bea813b
|
rename a couple of things and move a PutBytes so it happens sooner
|
2019-07-27 18:23:55 -05:00 |
|
Arceliar
|
9e118884d4
|
remove some commented code
|
2019-07-27 18:12:06 -05:00 |
|
Arceliar
|
e0a3055c2f
|
get rid of session workers, new util.PutBytes/GetBytes logic
|
2019-07-27 18:10:32 -05:00 |
|
Neil Alexander
|
de1005e4fa
|
Various API changes and simplifications to fix mobile builds
|
2019-07-27 15:00:09 +01:00 |
|
Arceliar
|
cf3ebe04a7
|
have Conn use Cancellation instead of manually setting up timers
|
2019-07-17 21:37:45 -05:00 |
|
Arceliar
|
5301207480
|
fix possible unsafe memory use in Conn.Read
|
2019-07-17 18:25:38 -05:00 |
|
Neil Alexander
|
307b24d8cb
|
Fix Conn.Read/Conn.Write behavior after Conn.Close, get rid of second TUN/TAP conn reader goroutine, no longer use deadlines
|
2019-07-17 21:42:17 +01:00 |
|
Neil Alexander
|
747b50bb7c
|
Try to improve handling of timeouts
|
2019-07-17 11:13:53 +01:00 |
|
Arceliar
|
86c30a1fc4
|
fix another panic from a send on a closed session worker channel, from races between Conn.Read/Write/Close
|
2019-07-01 18:55:07 -05:00 |
|
Arceliar
|
d39428735d
|
recover if we try to send to a closed session worker due to a race between a Conn.Write call and a Conn.Close call
|
2019-06-29 18:50:21 -05:00 |
|
Arceliar
|
7d58a7ef3e
|
fix channel multiple close bug and concurrency bug in the way sessionInfo.close was being called
|
2019-06-29 17:44:28 -05:00 |
|
Arceliar
|
818eca90db
|
fix nil pointer deref if searches fail, block dial until a search exceeds or a timeout passes (todo: replace timer with context)
|
2019-06-29 16:10:02 -05:00 |
|
Arceliar
|
784acba823
|
I think this fixes the concurrent map read/write panic
|
2019-06-29 12:14:44 -05:00 |
|
Arceliar
|
e88bef35c0
|
get rid of old buffered session packets
|
2019-06-28 20:02:58 -05:00 |
|
Arceliar
|
5df110ac79
|
make Dial block until the search finishes, and use it as such
|
2019-06-28 18:42:31 -05:00 |
|
Arceliar
|
29a0f8b572
|
some minor refactoring to dht callbacks and searches, work in progress
|
2019-06-25 19:31:29 -05:00 |
|
Arceliar
|
4b56849b08
|
fix issue with sessions dying and never being fixed
|
2019-06-10 22:09:12 -05:00 |
|
Arceliar
|
1addf08ccd
|
don't have Conn.Read return an error for temorary crypto failures from e.g. out of order packets, just drop the packet and keep blocking until there's usable traffic
|
2019-05-31 17:51:01 -05:00 |
|
Neil Alexander
|
9e086e70f0
|
Don't indefinitely block TUN/TAP reader goroutine when a conn error happens
|
2019-05-30 12:44:47 +01:00 |
|
Neil Alexander
|
0096d1ae3e
|
Re-add ICMPv6 packet too big handling
|
2019-05-29 20:16:17 +01:00 |
|
Arceliar
|
5ea864869a
|
don't spam searches for unused connections. todo: timeout old connections somehow
|
2019-05-23 20:27:52 -05:00 |
|
Arceliar
|
6469e39ff1
|
workaround to random timeouts
|
2019-04-26 22:42:05 -05:00 |
|
Arceliar
|
5d323861f0
|
properly fix the memory errors, it was caused by a function returning and PutBytes-ing a buffer before a worker had a chance to decrypt the buffer, so it would GetBytes the same buffer by dumb luck and then get an illegal overlap
|
2019-04-26 22:21:31 -05:00 |
|
Arceliar
|
01ea6d3d80
|
somehow this doesn't seem to deadlock or crash from buffer reuse (util.PutBytes), but I have no idea why it was doing that before and not now
|
2019-04-26 21:49:11 -05:00 |
|
Arceliar
|
15051b0a3c
|
Add deadline timers, keep searches alive until they complete (or the conn is closed) to keep Write from blocking forever
|
2019-04-26 19:31:47 -05:00 |
|
Arceliar
|
0059baf36c
|
add a newConn function that returns a pointer to a Conn with atomics properly initialized
|
2019-04-26 18:07:57 -05:00 |
|
Neil Alexander
|
d7a1c04748
|
It works, sort of, amazingly
|
2019-04-22 23:58:59 +01:00 |
|
Neil Alexander
|
e1a2d666bf
|
Clean up router, tweaks
|
2019-04-22 23:12:13 +01:00 |
|
Neil Alexander
|
6e528799e9
|
Conn Read/Write operations will block while search completes
|
2019-04-22 22:38:37 +01:00 |
|
Neil Alexander
|
ea8948f378
|
TUN/TAP addr/subnet to Conn mappings, other fixes
|
2019-04-22 20:06:39 +01:00 |
|
Neil Alexander
|
9778f5d2b8
|
Fix search behaviour on closed Conns, various other fixes
|
2019-04-22 15:00:19 +01:00 |
|
Neil Alexander
|
47eb2fc47f
|
Break deadlock by creating session recv queue when session is created instead of repointing at search completion, also make expired atomic
|
2019-04-22 11:20:35 +01:00 |
|
Arceliar
|
5dada3952c
|
use a session worker to try to avoid mutex hell. compiles, but incomplete and doesn't work yet
|
2019-04-21 20:38:14 -05:00 |
|
Neil Alexander
|
0b8f5b5dda
|
Tweaks
|
2019-04-21 12:28:46 +01:00 |
|
Neil Alexander
|
781cd7571f
|
Fix race on tun conns, but still deadlocks if more than one connection is opened
|
2019-04-21 12:00:31 +01:00 |
|
Neil Alexander
|
79bcfbf175
|
Change some mutexes to atomics, change conns map to pointers, sort of works but seems to deadlock very easily
|
2019-04-21 11:50:41 +01:00 |
|