runtime/race: instrument Sendmsg/Recvmsg and Flock #5363
Labels
Milestone
Comments
|
Owner changed to @dvyukov. Status changed to Accepted. |
FTR, here is the change: https://golang.org/cl/8819050/ You can fix the false positive for now, by explicitly synchronizing the state inside of the process. I.e. on send: 1. touch the call context 2. put the call context into a map under the mutex 3. send call id on receive: 1. recv call id 2. find the call context in the map under the mutex 3. touch the call context this way the call context is synchronized with the map mutex. Labels changed: added racedetector. |
Sometimes I find that I was too quick making conclusions, and the race detector actually says me completely different thing. Read the report carefully, understand what exactly variable is involved in the race, what is the order or memory access (the first access in the report happened last) and what are involved goroutines. |
dvyukov
changed the title
|
Another case that come up is synchronization via syscall.Flock. It's effectively a mutex, so treating it as synchronization in race detector looks reasonable. |
|
@randall77 would the work that you did in CL 205461 aka commit golang/net@2180aed#diff-d1efad72dc5b17dc66a46767c32fff40 be similar to what would solve this issue? Thanks and happy new year! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
What steps will reproduce the problem? write a program that synchronizes due to calls to Sendmsg and Recvmsg build and test with -race What do you see instead? a race is detected Which compiler are you using (5g, 6g, 8g, gccgo)? 6g Which operating system are you using? linux Which version are you using? (run 'go version') tip Please provide any additional information below. Dmitry helped me debug this. Recvmsg needs: if rsa.Addr.Family != AF_UNSPEC { from, err = anyToSockaddr(&rsa) } + if raceenabled && err == nil { + raceAcquire(unsafe.Pointer(&ioSync)) + } Sendmsg needs: msg.Iov = &iov msg.Iovlen = 1 + if raceenabled { + raceReleaseMerge(unsafe.Pointer(&ioSync)) + } Dmitry suggested that this is for after Go 1.1. Since it's a small fix, it would be nice to do it before, but we can wait.The text was updated successfully, but these errors were encountered: