Age | Commit message (Collapse) | Author |
|
gonet.TCPConn implements CloseRead and CloseWrite, but it is not a
net.TCPConn. Use an interface to look for CloseRead and CloseWrite so
that they are called on gonet.TCPConn.
Updates tailscale/corp#25169
Signed-off-by: James Tucker <[email protected]>
|
|
Fixes #39
Signed-off-by: Brad Fitzpatrick <[email protected]>
|
|
Signed-off-by: James Tucker <[email protected]>
|
|
No point keeping it around. We can look at the git history to do
something similar later if we end up doing TLS-ALPN-01 in a similar
way.
|
|
|
|
|
|
Signed-off-by: GitHub <[email protected]>
|
|
protocol header format
original code sent the header in the format: PROXY <family> <srcIP>
<srcPort> <dstIP> <dstPort>
according to docs header format should be:
PROXY <family> <srcIP> <dstIP> <srcPort> <dstPort>
this is according
to:
https://www.haproxy.org/download/1.8/doc/proxy-protocol.txt
section
2.1. Human-readable header format (Version 1).
|
|
original code sent the header in the format: PROXY <family> <srcIP>
<srcPort> <dstIP> <dstPort>
according to docs header format should be:
PROXY <family> <srcIP> <dstIP> <srcPort> <dstPort>
this is according
to:
https://www.haproxy.org/download/1.8/doc/proxy-protocol.txt
section
2.1. Human-readable header format (Version 1).
|
|
- add go.mod/go.sum
- update .travis.yml to test against supported Go versions
- drop golint from CI checks
To satisfy modern Go crypto in the tlsrouter e2e_test.go:
- bump RSA keysize to minimum 2048 bits
- set NotBefore to valid recent timestamp
- include CommonName in SANs
Signed-off-by: Dominic Evans <[email protected]>
|
|
|
|
Fixes #19
Signed-off-by: Vincent Batts <[email protected]>
|
|
|
|
Not sure the root cause yet. See golang/go#25985.
|
|
|
|
Changing the internal-only match interface to return any parsed
hostnames.
It can be useful for implementers of Target to be able to inspect
the already-parsed SNI header (in the case of TLS) or host header
(in the case of http) to know what host was asked for by the client
in order to make additional routing decisions. This can be used
by transparent reverse proxies where the destination is not known
in advance.
|
|
Link to docs from the top level README.
Fixes https://github.com/google/tcpproxy/issues/13
|
|
Add{HTTPHost,SNI}Route remain so that the common case of exact matches
remains trivial to use. Add{HTTPHost,SNI}MatchRoute allow you to specify
your own matching function.
Fixes #9
|
|
By design, the tls-sni-01 challenge does not reveal information
about the domain being verified, so the proxy cannot "naively" route
such requests. Instead, it probes the Targets of all SNI routes, looking
for one that responds plausibly to the challenge hostname, and routes the
client connection to that.
ACME support can be turned off by inserting AddStopAcmeSearch in the route
chain. Subsequently registered SNI routes will not be probed by ACME challenges.
|
|
This allows routes to compute a target at match time, instead of being
statically mapped to a Target at register time.
|
|
|
|
|
|
|
|
Drop Go 1.7 from the build matrix, tcpproxy uses a 1.8 feature of crypto/tls.
Make `go get` fetch the right test packages for cmd/tlsrouter.
|
|
|
|
|
|
tlsrouter.
Brad's code will be the place for future development, and the base for the binary
formerly known as tlsrouter. This merge is the first step towards converging the
codebases.
|
|
|
|
This is prep work for merging with google/tcpproxy.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
This should only package and deploy if the build is clean on *all*
Go versions in the test matrix.
|
|
|
|
This allows backends that support it to receive the client's true
ip:port as out-of-band information, despite the connection being
proxied.
|
|
|
|
|
|
|
|
|
|
1.6 lacks the builtin context package, and 1.8 will be out any day
now, so 1.6 is on its last legs anyway.
|
|
TLS connections that look like ACME verification get fanned out to
all known backends, and the one that responds with the right cert
to continue ACME verification is the winner.
|
|
|
|
|
|
|
|
|