when loaded from router.info, and discard if invalid or when family changed.
This fixes some routers out there with bad sigs.
Don't loop forever through LoadRouterInfoJob.
See previous checkin for SSU 1.
This change dramatically reduces the measured RTT to get it much closer to the real RTT by reducing the ack delay.
With the ack delay dependent on the measured RTT, the measured RTT will increase.
Simulation results: For a denominator D in the ack delay calculation (currently 2) and a true RTT = TRTT:
For D <= about 2, measured RTT -> TRTT + 2 * TRTT/D For D >= about 4, measured RTT -> TRTT + TRTT/D
In other words, for D == 2, measured RTT -> 2 * TRTT. For a TRTT of 50, the measured RTT is 100.
D == 2 is too small. It's been like that since 2005.
Testnet results: measured RTT 80-90
This changes D to 6. measured RTT -> 6/5 * TRTT.
Simulation results: For a TRTT of 50, the measured RTT is 58.
Testnet results: measured RTT 60
Synch fields in the data messages both directions,
after one report of zero session ID in MessagePayloadMessage
to an external client on fast hardware with Java 18.
Add new constructors for efficiency.
Deprecated setters, unit tests not changed, TODO.
Should fix all the other messages also, TODO.
Store OB tokens by IP/port
Centralize token generation and expiration in Establishment Manager
Add methods to flush OB tokens on IP/port change
Log tweaks
Javadoc fix
Don't overwrite an existing family keystore file
Don't allow starting a new family with an existing family keystore file
Don't add family to RI until we have restarted and have a keystore file
Don't fail a netdb store for no family sig
Don't ever fail our own netdb store for family errors, to avoid rekey/restart
Return a result code from verify()
Load all known certs at startup rather than continually reloading them
Only give full verified status to known keys
Enforce signatures in netdb store when key is available
Show family verification status on /netdb
Export our cert to disk if missing
Add stormycloud family cert
Bypass /24 Sybil penalty
Add min packet size constants and notes
Don't attempt to decrypt a packet as a new inbound session if too small
Remove unused payload[] in IES2, now decrypted in-place
Log level tweaks in Establisher
Save data messages received before or immediately after session confirmed
by queueing them for processing after the PeerState2 is created.
The fragments for the first I2NP message from Alice to Bob are frequently lost,
this will hopefully fix it.
Not fully tested, needs wider network testing.
Avoid relaying until it's implemented:
Don't use SSU2 for an SSU address with introducers (fixes NPE)
Don't connect to SSU2 address without host/port
Don't ask for relay tag
Log tweaks
Fix length calculation to see if new token block will fit in session confirmed
Extend timeout after sending retry
Fix retx timer for sess req after sending token req
Remove dup call to confirmedPacketsSent()
Cancel ack timer when sending acks
Include intro key in firewalled addresses too
Use SSU2 version of ping packet for SSU2 peers
Reduce max padding
Log tweaks
MTU and related fixes:
Fix max fragment size
Fix max space for acks
Fix max packet size
Pull MTU from best address in Session Confirmed
Pass MTU from establish state to peer state
Use SSU2 min/default/max MTU in PeerState
Stop looping when out of space in OMF (SSU 1 too)
Correctly calculate skew after retry in establish state
Pass calculated skew from establish state to peer state
Round time in DateTime block
Periodically send DateTime block
Handle DateTime block in peer state
after initial testnet testing
Use correct intro key for Session/Token request
Fix state management in EstablishmentManager, OES2, IES2
Fix next send time during handshake
Fix header decryption in PacketHandler
Add additional packet checks in IES2 handling
Remove expired IES immediately (SSU1 also)
Failsafe sleep in EstablishmentManager on exception
Remove dup requestSent() calls
Don't release packet in PS2
Log tweaks and javadocs
Only affects NTCP for now.
This bug had little impact when both transports were enabled,
but significantly harmed connectivity and tunnel building for NTCP2-only.
Decrease unreachable cache time in testnet
Make cache cleaner a little more efficient
zzz
zzz