p2p: race condition in dialer scheduler

[fxfactorial]

WARNING: DATA RACE                                                                                                    
Read at 0x00c03c44d2a0 by goroutine 152:                                                                              
  github.com/ethereum/go-ethereum/p2p.(*dialScheduler).updateStaticPool()                                             
      github.com/ethereum/go-ethereum/p2p/dial.go:433 +0x2e4                                                          
  github.com/ethereum/go-ethereum/p2p.(*dialScheduler).expireHistory.func1()                                          
      github.com/ethereum/go-ethereum/p2p/dial.go:378 +0xcf                                                           
  github.com/ethereum/go-ethereum/p2p.(*expHeap).expire()                                                             
      github.com/ethereum/go-ethereum/p2p/util.go:59 +0x10a 
  github.com/ethereum/go-ethereum/p2p.(*dialScheduler).expireHistory()
      github.com/ethereum/go-ethereum/p2p/dial.go:375 +0x124
  github.com/ethereum/go-ethereum/p2p.(*dialScheduler).loop()
      github.com/ethereum/go-ethereum/p2p/dial.go:308 +0x239a
  github.com/ethereum/go-ethereum/p2p.newDialScheduler.func2()
      github.com/ethereum/go-ethereum/p2p/dial.go:186 +0x53 
  github.com/panjf2000/ants/v2.(*goWorker).run.func1()
      github.com/panjf2000/ants/v2@v2.4.5/worker.go:70 +0xd9

Previous write at 0x00c03c44d2a0 by goroutine 1038:
  github.com/ethereum/go-ethereum/p2p.(*dialTask).resolve() 
      github.com/ethereum/go-ethereum/p2p/dial.go:537 +0x204
  github.com/ethereum/go-ethereum/p2p.(*dialTask).run()
      github.com/ethereum/go-ethereum/p2p/dial.go:498 +0xf9 
  github.com/ethereum/go-ethereum/p2p.(*dialScheduler).startDial.func1()
      github.com/ethereum/go-ethereum/p2p/dial.go:465 +0x58 
  github.com/panjf2000/ants/v2.(*goWorker).run.func1()
      github.com/panjf2000/ants/v2@v2.4.5/worker.go:70 +0xd9

off of latest master -

[MariusVanDerWijden]

Hmm that is not latest master afaict. Could you link the commit you're using?

[MariusVanDerWijden]

I fixed some races here: #23434
Don't think it's the one you found though

[fjl]

The race in this issue is on t.dest. When resolve() runs, it may update t.dest. Meanwhile, on the dialer loop, we run updateStaticPool when a peer is removed, and this reads t.dest.

[fxfactorial]

My bad , was latest master off bsc and should have written the description @fjl wrote , yes that’s the problem -

[fjl]

I'm actually not sure how to fix this race correctly. The check on t.dest in updateStaticPool seems off to me, maybe that should just be removed.

[fxfactorial]

Agree - it wasn’t clear to me how to fix it immediately either , and don’t want locks for just this - if node being removed anyway then .dest being updated shouldn’t be needed then

[MariusVanDerWijden]

This issue has become quite stale now. I can't reproduce the race condition anymore. Feel free to reopen if this is still an issue with the current master.

[fjl]

I'm sure this race is not fixed. It's just hard to reproduce because it requires resolving the node, which takes variable time.