p2p/net/conn: reuse listen tcp port to dial

[jbenet]

This PR finally makes it possible to dial out from our TCP listen addr. I had to sink low into net to make this possible. The result is the go-reuseport package. Sadly golang/net doesnt let us add tcp socket options, so we have to do it this way (setting up the sockets manually, which is truly painful). can see the same approach used in various modules:

• https://github.com/kavu/go_reuseport
• https://github.com/benburkert/net.tune/ [1]

Anyway, this PR:

• adds go-reuseport (and its dep, go-sockaddr/net)
• changes p2p/net/conn to optimistically use go-reuseport (it can't always work because we may be dialing the same 5-tuple), or fallback to reg dial.
• changed the identify receiver to register the observed address, if and only if it is for a connection whose LocalAddress() is one of our ListenAddrs().

[1] wow i just discovered this one... after rolling my own. fffff! it looks good-- we might want to switch to this one if my go-reuseport gives us problems

[jbenet]

https://build.protocol-dev.com/job/ipfs_test_all_commits_on_branch/185/console <-- tells me tests_all_commits fails, but i cant tell which commit. make test_all_commits passes locally

[jbenet]

RFCR

[whyrusleeping]

this PR shouldnt be against ipfs, you should file it against Go.

[whyrusleeping]

whats this?

[whyrusleeping]

Is much of this code directly copied from go's net lib? If so, it might be good to mark any of that as such.

[jbenet]

this PR shouldnt be against ipfs, you should file it against Go.

Right!?

Is much of this code directly copied from go's net lib? If so, it might be good to mark any of that as such.

Yeah, the stuff copied straight out is marked with the copyright notice. The connect() function is a stripped down version of theirs.

[jbenet]

This amount of copy pasting is horrible. This PR is an exercise in why internal packages / forcibly unaccessible symbols are a bad idea. There's a significant difference between "not exporting something to my interface" and "making it absolutely impossible for you to use this code". It should be possible to import things if you really mean it. Maybe as a different package altogether with a dependency like ibaz "foo/bar/baz/internal".

[jbenet]

https://build.protocol-dev.com/job/build%20docker%20image/923/console

Sending build context to Docker daemon 
Step 0 : FROM golang:1.3
 ---> ebd3fd90ae2e
Step 1 : MAINTAINER Brian Tiger Chow <btc@perfmode.com>
 ---> Using cache
 ---> 6c893fd2f62d
Step 2 : ADD . /go/src/github.com/jbenet/go-ipfs
 ---> a18697f6bce6
Removing intermediate container 764cfa2d8ed1
Step 3 : RUN cd /go/src/github.com/jbenet/go-ipfs/cmd/ipfs && go install
 ---> Running in 3452c2eb3913
 ---> 89b1045e2b93
Removing intermediate container 3452c2eb3913
Step 4 : EXPOSE 4001 5001 4002/udp
 ---> Running in ee9c69101f50
 ---> 2c6dc0ce7374
Removing intermediate container ee9c69101f50
Step 5 : ENTRYPOINT ipfs
 ---> Running in b5429cafd23b
2015/01/20 00:14:49 An error occured while creating the container
Build step 'Execute shell' marked build as failure
Warning: you have no plugins providing access control for builds, so falling back to legacy behavior of permitting any downstream builds to be triggered
Triggering a new build of chore_reclaim-docker-space
Finished: FAILURE

Maybe the chore should happen before the docker image tries to build and fails? or this may be unrelated.

cc @briantigerchow

[whyrusleeping]

Looks pretty good to me, let's hope it solves our issues!

[jbenet]

@whyrusleeping is that a LGTM on the PR?

[whyrusleeping]

Yeah, merge it! (before i change my mind! :P )