as discussed in #ls2 meeting
- Bob replies with Session Created even if skewed,
so that Alice finds out what the skew is
- Alice handles Session Created timestamp and drops if skewed,
bans Bob, and updates clock if NTP failed
- If Alice does reply with SessionConfirmed, Bob will send a
destroy with a skew error code
- Don't change skew error code if netdb store failed
- Fix skew adjustment for RTT by Bob
- Call setLastBadSkew() in the right places
- Fix ntcp.invalidInboundSkew and ntcp.invalidOutboundSkew stats
- Always use a lease as the reply tunnel when publishing LS through a client tunnel
This ensures we're not using about-to-expire tunnels for the reply,
and ensures the ff is able pick an alternate
- Don't count skipped peers as attempted in FloodfillVerify
- Pass failed and skipped peers to FloodfillVerify job to be skipped there also
- Pass failed and skipped peers from FloodfillVerify job to the next StoreJob on failure
- Consolidate common reply token generation code in StoreJob
- Ensure tunnel diversity in StoreJob retries by only
using tunnels closest to the target for the first request.
This increases reliability by not reusing the same tunnels for all retries.
- Refactor StoreState to rework inefficient methods and unused data
- Clean up commented-out code in StoreState
- Log tweaks
Actually update the stat for stores in dbStoreSent();
we are generally storing to different ffs than lookups, so we need the
stat for stores as well, since we use it as the timeout in StoreJob.
Change from 1-day to 1-hour stat.
Switch to avgOrLifetimeAvg() so the rate is always valid.
Reduce max time used for timeout.
This allows more peers to be tried before total timeout
Previously, the per-peer timeout was almost always the max.
Make sendStore() package private.
Javadocs and cleanups.
- Write directly from writer threads, except for during establishment and when write doesn't complete; throw those to the pumper as usual
- New NTCPCon writelock, readlock, and statlock (formerly all on NTCPCon.this) to prevent deadlocks
- NTCPCon chan and key now volatile, remove synch to prevent deadlocks
- All interestOps changes now lock on the key via setInterest() and clearInterest() since changes may now happen in multiple threads
- Set _paddingConfig at initialization to avoid NPE
Greatly reduces pumper loops and CPU
As proposed by jogger
Reviewed by zlatinb
Ref: http://zzz.i2p/topics/3192
Removed in May 2011 when we added fast tier slices
Also add support in exploratory tunnels
Create MaskedIPSet in peer selectors, pass to ProfileOrganizer.selectXXX() for each call.
Not required for one-hop tunnels.
Disable for test networks (i2np.allowLocal)
Reported by 'vulnerability_reports' http://zzz.i2p/topics/3215
