Mirage3 #262

samoht · 2017-07-28T14:50:26Z

This is built on top of the major cleanup that I've pushed in #259

Everything compiles but it seems that there are 3 missing functions in the tcpip stack:

Stack_ipv4.set_ip_netmask
Stack_ipv4.set_ip_gateways
Stack_tcp_wire.src_of_id (e.g. there is no way to get the source IP)

Also the tests are compiling but not passing, either because of my very flaky connection or because I broke something with the dns ref that I modified to be None by default. I will investigate more next week.

djs55 · 2017-07-31T18:09:06Z

Error: External library "tcpip.unix" not found.
-> required by "src/hostnet/jbuild (context default)"
Hint: try: jbuilder external-lib-deps --missing --dev src/bin/main.exe
make: *** [vpnkit.exe] Error 1

samoht · 2017-07-31T18:27:39Z

It needs the dev version of mirage-tcpip and charrua-core. I'll fix this.

samoht · 2017-08-01T08:50:20Z

Hum, it seems that some of the tests in mirage-tcpip do not work on windows:

--------------------------------------------------------------------------------
- ASSERT sender address
- --------------------------------------------------------------------------------
- Fatal error: exception (Failure "Error sender address: expecting fc00::23, got ::.")

And the vpnkit tests are failing on macOS (I can reproduce this locally, and @djs55 identified the error message to come from here):

main_uwt.exe: [WARNING] IP module couldn't send UDP packet to 8.8.8.8: no route to destination: no response for IP on local network
Fatal error: exception Dns__Protocol.Dns_resolve_error(_)
Raised at file "src/core/lwt.ml", line 2719, characters 24-25
Called from file "uwt.ml", line 1848, characters 10-23
Called from file "list.ml", line 77, characters 12-15
Called from file "list.ml", line 77, characters 12-15
Called from file "src/hostnet_test/main_uwt.ml", line 19, characters 2-282
make: *** [test] Error 2

djs55 · 2017-08-01T10:32:17Z

I'll try to fix the routing problem by switching to DHCP in the test case. This requires: mirage/charrua#65

samoht · 2017-08-01T14:58:37Z

I can produce a different failure for ipv6 on Windows. Trying to understand what happens ...

samoht · 2017-08-01T15:57:24Z

There are 2 failing tests on windows:

ipv6 0:

test.exe: [INFO] Connected Ethernet interface 02:50:00:00:00:01
test.exe: [INFO] IP6: Starting
test.exe: [DEBUG] ND6: Sending RS
test.exe: [DEBUG] ND6: Sending NS src=:: dst=ff02::1:ff00:1 tgt=fe80::50:ff:fe00:1
test.exe: [DEBUG] no size limit, sending
test.exe: [DEBUG] no size limit, sending
test.exe: [DEBUG] ND6: Sending NS src=:: dst=ff02::1:ff00:23 tgt=fc00::23
test.exe: [DEBUG] no size limit, sending
test.exe: [INFO] UDP interface connected on
test.exe: [INFO] Connected Ethernet interface 02:50:00:00:00:02
test.exe: [INFO] IP6: Starting
test.exe: [DEBUG] ND6: Sending RS
test.exe: [DEBUG] ND6: Sending NS src=:: dst=ff02::1:ff00:2 tgt=fe80::50:ff:fe00:2
test.exe: [DEBUG] no size limit, sending
test.exe: [DEBUG] no size limit, sending
test.exe: [DEBUG] ND6: Sending NS src=:: dst=ff02::1:ff00:45 tgt=fc00::45
test.exe: [DEBUG] no size limit, sending
test.exe: [INFO] UDP interface connected on
test.exe: [DEBUG] IP6: Queueing packet: dst=fc00::45
test.exe: [DEBUG] ND6: Sending NS src=:: dst=ff02::1:ff00:45 tgt=fc00::45
test.exe: [DEBUG] no size limit, sending
test.exe: [DEBUG] ND: Received NS: src=:: dst=ff02::1:ff00:45 tgt=fc00::45
test.exe: [DEBUG] SLAAC: fe80::50:ff:fe00:2 --> PREFERRED
test.exe: [DEBUG] SLAAC: fc00::45 --> PREFERRED
test.exe: [DEBUG] SLAAC: fe80::50:ff:fe00:1 --> PREFERRED
test.exe: [DEBUG] SLAAC: fc00::23 --> PREFERRED
--------------------------------------------------------------------------------------------------------------------
ASSERT UDP packet should have been received
--------------------------------------------------------------------------------------------------------------------
Test error: Error UDP packet should have been received.

socket 1:

_build/_tests\socket.001.output:
test.exe: [INFO] Manager: connect
test.exe: [INFO] Manager: configuring
test.exe: [WARNING] Manager: sockets currently bind to all available IPs. IPs 127.0.0.1 were specified, but this will be ignored
test.exe: [INFO] Manager: connect
test.exe: [INFO] Manager: configuring
test.exe: [WARNING] Manager: sockets currently bind to all available IPs. IPs 127.0.0.1 were specified, but this will be ignored
[exception] Unix.Unix_error(Unix.EPROTONOSUPPORT, "socket", "")

I will follow up on mirage/mirage-tcpip

djs55 · 2017-08-01T15:59:19Z

One of the tcpip tests just failed on CircleCI too:

[ERROR]             connect              1   connect two stacks, uniform packet loss of packets with no payload x 100.

(probably intermittent)

samoht · 2017-08-01T16:03:13Z

yes sorry I have changed the default test flags in my tcpip branch, I will create a more stable dev branch ...

samoht · 2017-08-01T16:13:18Z

I've pushed a new Git url for tcpip dev package, where only quick tests are running. Not sure why the slow test is failing, but this is probably due to a timeout/slow machine.

samoht · 2017-08-01T16:14:33Z

src/hostnet/mux.ml

-    Lwt.return r >|= lift_error
+    Lwt.async
+      (fun () ->
+        Netif.listen netif @@ callback t >>= function


you can write this as:

Netif.listen netif @@ callback t >|= function | Ok () -> () | Error _ -> Log.err (fun f -> ... )

samoht · 2017-08-01T16:16:39Z

src/hostnet/slirp.ml

        ~gateway:None
-        ~network:Ipaddr.V4.(Prefix.of_addr unspecified)
+        ~network:Ipaddr.V4.Prefix.global


good catch! I wasn't really sure what to use there :-)

samoht · 2017-08-01T18:21:04Z

@djs55 it now seems that we have genuine test errors on both windows and macOS:

**** Starting test Host resolver: lookup 
Starting test case 
main_uwt.exe: [INFO] Made a loopback connection
main_uwt.exe: [INFO] PPP.negotiate: received { magic = VMN3T; version = 1; commit = 0123456789012345678901234567890123456789 }
main_uwt.exe: [INFO] Client.negotiate: received { magic = VMN3T; version = 1; commit = 0123456789012345678901234567890123456789 }
main_uwt.exe: [INFO] PPP.negotiate: received Ethernet d1d9cd61-d0dc-4715-9bb3-4c11da7ad7a5
main_uwt.exe: [INFO] PPP.negotiate: sending { mtu = 1500; max_packet_size = 1550; client_macaddr = c0:ff:ee:c0:ff:ee }
main_uwt.exe: [INFO] PPP.listen: rebinding the primary listen callback
main_uwt.exe: [INFO] Client mac: c0:ff:ee:c0:ff:ee server mac: f6:16:36:bc:f9:c6
main_uwt.exe: [INFO] TCP/IP ready
main_uwt.exe: [INFO] stack connected
main_uwt.exe: [INFO] Initialising client TCP/IP stack
main_uwt.exe: [INFO] PPP.listen: rebinding the primary listen callback
main_uwt.exe: [ERROR] Socket.TCPv4.read tcp:127.0.0.1:49510: caught Lwt.Completion_loop.Canceled returning Eof
Fatal error: exception (Failure timeout)
Raised at file "src/core/lwt.ml", line 2719, characters 24-25
Called from file "uwt.ml", line 1848, characters 10-23
Called from file "list.ml", line 77, characters 12-15
Called from file "list.ml", line 77, characters 12-15
Called from file "src/hostnet_test/main_uwt.ml", line 19, characters 2-282

djs55 · 2017-08-02T13:29:52Z

Tests are still failing in CI but I've not managed to reproduce it locally yet. I've pushed some patches to improve the debug logging to see if that helps. I'll see if I can make myself a more exact replica of the CI environment.

djs55 · 2017-08-02T13:54:56Z

OK it reproduced for me when I ran scripts/common.sh. A bit of environment-bisecting later and the culpit appears to be: mirage/charrua#66 -- the tests work for me when this is merged.

samoht · 2017-08-02T14:10:33Z

The new errors are related to missing license files, which means the tests passed, yay!!

I'll fix the licenses :-)

samoht · 2017-08-02T16:18:15Z

ALL GREEN!

The last worrying (or not) bit are the failing TCPIP tests on windows. Probably not really severe, but would be great to check that we understand what happens. I'll have a look tomorrow if I find some time.

djs55 · 2017-08-02T16:22:42Z

Was it only the IPv6 tests? If so that's not too bad, as we don't use IPv6 (yet).

Since I'm paranoid, I'll put this vpnkit build through the Docker for Desktop end-to-end tests too and report back.

samoht · 2017-08-02T16:24:15Z

yea, the ipv6 test is failing but also using an unsuported protocol over a socket: mirage/mirage-tcpip#331

djs55 · 2017-08-02T17:47:14Z

Docker Windows tests all passed but Mac tests all failed -- I'll investigate tomorrow. It's probably a startup issue somehow related to the Vmnet Unix domain socket.

Signed-off-by: Thomas Gazagnaire <thomas@gazagnaire.org>

Remove the ones which have been released and add new ones Signed-off-by: Thomas Gazagnaire <thomas@gazagnaire.org>

Signed-off-by: Thomas Gazagnaire <thomas@gazagnaire.org>

Signed-off-by: David Scott <dave@recoil.org>

- use the `connect ~ip` argument rather than `set_ip` to set the stack's IP - set the network to `global` (i.e. "all addresses are on my local network) instead of `unspecified` (i.e. "no addresses are on my local network). Signed-off-by: David Scott <dave@recoil.org>

Previously the test cases were using an (unspecified?) static IP when talking to the loopback vpnkit server. This patch makes the client use DHCP which is - more realistic: this is what we expect real clients to do - more likely to work: previously the client's IP address appeared to be undefined Signed-off-by: David Scott <dave@recoil.org>

Signed-off-by: Thomas Gazagnaire <thomas@gazagnaire.org>

because of mirage/mirage-tcpip#330 and mirage/mirage-tcpip#331 Signed-off-by: Thomas Gazagnaire <thomas@gazagnaire.org>

Signed-off-by: David Scott <dave.scott@docker.com>

Signed-off-by: Thomas Gazagnaire <thomas@gazagnaire.org>

Recall that `listen` can be called multiple times with a "last-callback-wins" semantic. Previously we would start additional copies of the packet receive loop() which can lead to I/O problems on the channel. This patch starts at most one receive loop; other `listen` calls block until disconnect. Signed-off-by: David Scott <dave.scott@docker.com>

Although this is a "Point to Point Protocol" it is not *the* "Point to Point Protocol" so call it "Vmnet" instead (which at least matches the module name, but it's still not ideal) Signed-off-by: David Scott <dave.scott@docker.com>

Previously we would fail the thread with `Too_many_connections` but this meant that the SYN was never responed to. With this change, we send explicit RST packets to the caller. This makes the `test_max_connections` test pass -- previously the TCP `connect` call would resend SYNs for a long time. Signed-off-by: David Scott <dave.scott@docker.com>

It's useful to know what the connection limit is supposed to be. Signed-off-by: David Scott <dave.scott@docker.com>

When debugging traces from the test cases it's useful to know whether a `Vmnet.listen` call is from the client end or the server. This patch adds a logging prefix and sets this to either `Vmnet.Client` or `Vmnet.Server`. Signed-off-by: David Scott <dave.scott@docker.com>

When tests time out in CI it's useful to see where the time was spent. Signed-off-by: David Scott <dave.scott@docker.com>

Signed-off-by: Thomas Gazagnaire <thomas@gazagnaire.org>

The mirage-http package depends on mirage-conduit which depends on tls which depends on libgmp etc. We only need the definitions of `read_line` `read` `write` and `flush` so include them directly. Signed-off-by: David Scott <dave.scott@docker.com>

Signed-off-by: Thomas Gazagnaire <thomas@gazagnaire.org>

Signed-off-by: David Scott <dave.scott@docker.com>

…pnkit Signed-off-by: Thomas Gazagnaire <thomas@gazagnaire.org>

djs55 · 2017-08-03T12:28:55Z

@samoht sorry I force pushed this and lost 2 of your patches. I've re-applied them, hopefully it's ok now.

Signed-off-by: David Scott <dave@recoil.org>

djs55 · 2017-08-03T15:03:29Z

OK, the Docker Mac tests are passing now. I think we are safe to merge this. Any objections? (@samoht?)

samoht force-pushed the mirage3 branch 3 times, most recently from 188527e to 5487366 Compare July 31, 2017 17:25

samoht force-pushed the mirage3 branch 2 times, most recently from b002878 to 854b6d4 Compare August 1, 2017 08:22

samoht commented Aug 1, 2017

View reviewed changes

samoht mentioned this pull request Aug 2, 2017

disconnect sockets when they're removed #260

Open

samoht force-pushed the mirage3 branch from 7ce8451 to efcdd35 Compare August 2, 2017 08:10

samoht mentioned this pull request Aug 2, 2017

Reduce the number of dependencies on forked packages. #27

Closed

samoht added 3 commits August 3, 2017 11:55

Update opam dependencies

9023077

Signed-off-by: Thomas Gazagnaire <thomas@gazagnaire.org>

Update dev dependencies

2705064

Remove the ones which have been released and add new ones Signed-off-by: Thomas Gazagnaire <thomas@gazagnaire.org>

AUTO: Update upstream packages

dcf7d50

Signed-off-by: Thomas Gazagnaire <thomas@gazagnaire.org>

djs55 and others added 16 commits August 3, 2017 11:55

Reindent Vmnet.listen

51824bd

Signed-off-by: David Scott <dave@recoil.org>

Use a more stable dev branch for mirage-tcpip

d94f49a

Signed-off-by: Thomas Gazagnaire <thomas@gazagnaire.org>

disable tests on windows

929616c

because of mirage/mirage-tcpip#330 and mirage/mirage-tcpip#331 Signed-off-by: Thomas Gazagnaire <thomas@gazagnaire.org>

ci: install test dependencies

e3ca5a2

Signed-off-by: David Scott <dave.scott@docker.com>

Add charrua-client dev packages on windows too

06bbf44

Signed-off-by: Thomas Gazagnaire <thomas@gazagnaire.org>

test: improve the logging in the test_max_connections_test

83cfa1d

It's useful to know what the connection limit is supposed to be. Signed-off-by: David Scott <dave.scott@docker.com>

test: include timestamps on the log lines

803fff9

When tests time out in CI it's useful to see where the time was spent. Signed-off-by: David Scott <dave.scott@docker.com>

Improve the license script

e881091

Update license metadata

788c60a

Fix lwt license

8eeb842

Signed-off-by: Thomas Gazagnaire <thomas@gazagnaire.org>

samoht force-pushed the mirage3 branch from a081ca6 to 8eeb842 Compare August 3, 2017 09:56

samoht mentioned this pull request Aug 3, 2017

Fix dns dev version #267

Closed

djs55 force-pushed the mirage3 branch from c770ffd to afaed4c Compare August 3, 2017 12:21

djs55 added 4 commits August 3, 2017 13:27

Fix cohttp-lwt constraint

330a96e

Signed-off-by: Thomas Gazagnaire <thomas@gazagnaire.org>

AUTO: Update upstream packages

4698868

Signed-off-by: David Scott <dave.scott@docker.com>

Remove watermarking for hostnet, as it is the exact same version as v…

76bdeeb

…pnkit Signed-off-by: Thomas Gazagnaire <thomas@gazagnaire.org>

djs55 force-pushed the mirage3 branch from afaed4c to 76bdeeb Compare August 3, 2017 12:28

djs55 added 2 commits August 3, 2017 13:52

opam: depend on mirage-dns

c4d7e96

Signed-off-by: David Scott <dave@recoil.org>

AUTO: Update upstream packages

333d9a0

Signed-off-by: David Scott <dave@recoil.org>

djs55 merged commit a06c975 into moby:master Aug 3, 2017

djs55 mentioned this pull request Sep 15, 2017

Failure to start on some Windows machines #286

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Mirage3 #262

Mirage3 #262

samoht commented Jul 28, 2017 •

edited

Loading

djs55 commented Jul 31, 2017

samoht commented Jul 31, 2017

samoht commented Aug 1, 2017

djs55 commented Aug 1, 2017

samoht commented Aug 1, 2017

samoht commented Aug 1, 2017

djs55 commented Aug 1, 2017

samoht commented Aug 1, 2017

samoht commented Aug 1, 2017

samoht Aug 1, 2017

samoht Aug 1, 2017

samoht commented Aug 1, 2017

djs55 commented Aug 2, 2017

djs55 commented Aug 2, 2017

samoht commented Aug 2, 2017

samoht commented Aug 2, 2017 •

edited

Loading

djs55 commented Aug 2, 2017

samoht commented Aug 2, 2017

djs55 commented Aug 2, 2017

djs55 commented Aug 3, 2017

djs55 commented Aug 3, 2017

Mirage3 #262

Mirage3 #262

Conversation

samoht commented Jul 28, 2017 • edited Loading

djs55 commented Jul 31, 2017

samoht commented Jul 31, 2017

samoht commented Aug 1, 2017

djs55 commented Aug 1, 2017

samoht commented Aug 1, 2017

samoht commented Aug 1, 2017

djs55 commented Aug 1, 2017

samoht commented Aug 1, 2017

samoht commented Aug 1, 2017

samoht Aug 1, 2017

Choose a reason for hiding this comment

samoht Aug 1, 2017

Choose a reason for hiding this comment

samoht commented Aug 1, 2017

djs55 commented Aug 2, 2017

djs55 commented Aug 2, 2017

samoht commented Aug 2, 2017

samoht commented Aug 2, 2017 • edited Loading

djs55 commented Aug 2, 2017

samoht commented Aug 2, 2017

djs55 commented Aug 2, 2017

djs55 commented Aug 3, 2017

djs55 commented Aug 3, 2017

samoht commented Jul 28, 2017 •

edited

Loading

samoht commented Aug 2, 2017 •

edited

Loading