The rootlessport forwarder requires a child IP to be set. This must be a
valid ip in the container network namespace. The problem is that after a
network disconnect and connect the eth0 ip changed. Therefore the
packages are dropped since the source ip does no longer exists in the
netns.
One solution is to set the child IP to 127.0.0.1, however this is a
security problem. [1]
To fix this we have to recreate the ports after network connect and
disconnect. To make this work the rootlessport process exposes a socket
where podman network connect/disconnect connect to and send to new child
IP to rootlessport. The rootlessport process will remove all ports and
recreate them with the new correct child IP.
Also bump rootlesskit to v0.14.3 to fix a race with RemovePort().
Fixes#10052
[1] https://nvd.nist.gov/vuln/detail/CVE-2021-20199
Signed-off-by: Paul Holzinger <pholzing@redhat.com>
There was an race condition when calling `GetRootlessCNINetNs()`. It
created the rootless cni directory before it got locked. Therefore
another process could have called cleanup and removed this directory
before it was used resulting in errors. The lockfile got moved into the
XDG_RUNTIME_DIR directory to prevent a panic when the parent dir was
removed by cleanup.
Fixes#10930Fixes#10922
To make this even more robust `GetRootlessCNINetNs()` will now return
locked. This guarantees that we can run `Do()` after `GetRootlessCNINetNs()`
before another process could have called `Cleanup()` in between.
[NO TESTS NEEDED] CI is flaking, hopefully this will fix it.
Signed-off-by: Paul Holzinger <pholzing@redhat.com>
Fix issue 10929 : `[Regression in 3.2.0] CNI-in-slirp4netns DNS gets broken when running a rootful container after running a rootless container`
When /etc/resolv.conf on the host is a symlink to /run/systemd/resolve/stub-resolv.conf,
we have to mount an empty filesystem on /run/systemd/resolve in the child namespace,
so as to isolate the directory from the host mount namespace.
Otherwise our bind-mount for /run/systemd/resolve/stub-resolv.conf is unmounted
when systemd-resolved unlinks and recreates /run/systemd/resolve/stub-resolv.conf on the host.
[NO TESTS NEEDED]
Signed-off-by: Akihiro Suda <akihiro.suda.cz@hco.ntt.co.jp>
The rootless cni namespace needs a valid /etc/resolv.conf file. On some
distros is a symlink to somewhere under /run. Because the kernel will
follow the symlink before mounting, it is not possible to mount a file
at exactly /etc/resolv.conf. We have to ensure that the link target will
be available in the rootless cni mount ns.
Fixes#10855
Also fixed a bug in the /var/lib/cni directory lookup logic. It used
`filepath.Base` instead of `filepath.Dir` and thus looping infinitely.
Fixes#10857
[NO TESTS NEEDED]
Signed-off-by: Paul Holzinger <pholzing@redhat.com>
`syncContainer()` requires the container to be locked, otherwise we can
end up with undefined behavior.
[NO TESTS NEEDED]
Signed-off-by: Paul Holzinger <pholzing@redhat.com>
Podman does not need to watch the cni config directory. If a network is
not found in the cache, OCICNI will reload the networks anyway and thus
even podman system service should work as expected.
Also include a change to not mount a "new" /var by default in the
rootless cni ns, instead try to use /var/lib/cni first and then the
parent dir. This allows users to store cni configs under /var/... which
is the case for the CI compose test.
[NO TESTS NEEDED]
Fixes#10686
Signed-off-by: Paul Holzinger <pholzing@redhat.com>
Network connect/disconnect has to call the cni plugins when the network
namespace is already configured. This is the case for `ContainerStateRunning`
and `ContainerStateCreated`. This is important otherwise the network is
not attached to this network namespace and libpod will throw errors like
`network inspection mismatch...` This problem happened when using
`docker-compose up` in attached mode.
Signed-off-by: Paul Holzinger <pholzing@redhat.com>
Using the gvproxy application on the host, we can now port forward from
the machine vm on the host. It requires that 'gvproxy' be installed in
an executable location. gvproxy can be found in the
containers/gvisor-tap-vsock github repo.
[NO TESTS NEEDED]
Signed-off-by: Brent Baude <bbaude@redhat.com>
This change adds the entry `host.containers.internal` to the `/etc/hosts`
file within a new containers filesystem. The ip address is determined by
the containers networking configuration and points to the gateway address
for the containers networking namespace.
Closes#5651
Signed-off-by: Baron Lenardson <lenardson.baron@gmail.com>
Allow podman network reload to be run as rootless user. While it is
unlikely that the iptable rules are flushed inside the rootless cni
namespace, it could still happen. Also fix podman network reload --all
to ignore errors when a container does not have the bridge network mode,
e.g. slirp4netns.
Signed-off-by: Paul Holzinger <paul.holzinger@web.de>
The source ip for the rootlesskit port forwarder was hardcoded to the
standard slirp4netns ip. This is incorrect since users can change the
subnet used by slirp4netns with `--network slirp4netns:cidr=10.5.0.0/24`.
The container interface ip is always the .100 in the subnet. Only when
the rootlesskit port forwarder child ip matches the container interface
ip the port forwarding will work.
Fixes#9828
Signed-off-by: Paul Holzinger <paul.holzinger@web.de>
Add a new --rootless-cni option to podman unshare to also join the
rootless-cni network namespace. This is useful if you want to connect
to a rootless container via IP address. This is only possible from the
rootless-cni namespace and not from the host namespace. This option also
helps to debug problems in the rootless-cni namespace.
Signed-off-by: Paul Holzinger <paul.holzinger@web.de>
The CNI plugins need access to iptables in $PATH. On debian /usr/sbin
is not added to $PATH for rootless users. This will break rootless
cni completely. To prevent breaking existing users add /usr/sbin to
$PATH in podman if needed.
Signed-off-by: Paul Holzinger <paul.holzinger@web.de>
If a user only has a local dns server in the resolv.conf file the dns
resolution will fail. Instead we create a new resolv.conf which will use
the slirp4netns dns.
Signed-off-by: Paul Holzinger <paul.holzinger@web.de>
With the new rootless cni supporting network connect/disconnect is easy.
Combine common setps into extra functions to prevent code duplication.
Signed-off-by: Paul Holzinger <paul.holzinger@web.de>
Instead of creating an extra container create a network and mount
namespace inside the podman user namespace. This ns is used to
for rootless cni operations.
This helps to align the rootless and rootful network code path.
If we run as rootless we just have to set up a extra net ns and
initialize slirp4netns in it. The ocicni lib will be called in
that net ns.
This design allows allows easier maintenance, no extra container
with pause processes, support for rootless cni with --uidmap
and possibly more.
The biggest problem is backwards compatibility. I don't think
live migration can be possible. If the user reboots or restart
all cni containers everything should work as expected again.
The user is left with the rootless-cni-infa container and image
but this can safely be removed.
To make the existing cni configs work we need execute the cni plugins
in a extra mount namespace. This ensures that we can safely mount over
/run and /var which have to be writeable for the cni plugins without
removing access to these files by the main podman process. One caveat
is that we need to keep the netns files at `XDG_RUNTIME_DIR/netns`
accessible.
`XDG_RUNTIME_DIR/rootless-cni/{run,var}` will be mounted to `/{run,var}`.
To ensure that we keep the netns directory we bind mount this relative
to the new root location, e.g. XDG_RUNTIME_DIR/rootless-cni/run/user/1000/netns
before we mount the run directory. The run directory is mounted recursive,
this makes the netns directory at the same path accessible as before.
This also allows iptables-legacy to work because /run/xtables.lock is
now writeable.
Signed-off-by: Paul Holzinger <paul.holzinger@web.de>
Make sure we do not display the expected error when using podman network
reload. This is already done for iptables-legacy however iptables-nft
creates a slightly different error message so check for this as well.
The error is logged at info level.
[NO TESTS NEEDED] The test VMs do not use iptables-nft so there is no
way to test this. It is already tested for iptables-legacy.
Signed-off-by: Paul Holzinger <paul.holzinger@web.de>
Make sure to pass the cni interface descriptions to cni teardowns.
Otherwise cni cannot find the correct cache files because the
interface name might not match the networks. This can only happen
when network disconnect was used.
Fixes#9602
Signed-off-by: Paul Holzinger <paul.holzinger@web.de>
Only the the network mode bridge supports cni networks.
Other network modes cannot use network connect/disconnect
so we should throw a error.
Fixes#9496
Signed-off-by: Paul Holzinger <paul.holzinger@web.de>
The libpod network logic knows about networks IDs but OCICNI
does not. We cannot pass the network ID to OCICNI. Instead we
need to make sure we only use network names internally. This
is also important for libpod since we also only store the
network names in the state. If we would add a ID there the
same networks could accidentally be added twice.
Fixes#9451
Signed-off-by: Paul Holzinger <paul.holzinger@web.de>
We missed bumping the go module, so let's do it now :)
* Automated go code with github.com/sirkon/go-imports-rename
* Manually via `vgrep podman/v2` the rest
Signed-off-by: Valentin Rothberg <rothberg@redhat.com>
The allocated `tmpNetworkStatus` must be allocated with the length 0.
Otherwise append would add new elements to the end of the slice and
not at the beginning of the allocated memory.
This caused inspect to fail since the number of networks did not
matched the number of network statuses.
Fixes#9234
Signed-off-by: Paul Holzinger <paul.holzinger@web.de>
The mtu default value is currently forced to 65520.
This let the user control it using the config key network_cmd_options,
i.e.: network_cmd_options=["mtu=9000"]
Signed-off-by: bitstrings <pino.silvaggio@gmail.com>
when inspecting a container that is only connected to the default
network, we should populate the default network in the container inspect
information.
Fixes: #6618
Signed-off-by: baude <bbaude@redhat.com>
MH: Small fixes, added another test
Signed-off-by: Matthew Heon <matthew.heon@pm.me>
if a CNI network is added to the container, use the IP address in that
network instead of hard-coding the slirp4netns default.
commit 5e65f0ba30f3fca73f8c207825632afef08378c1 introduced this
regression.
Closes: https://github.com/containers/podman/issues/9065
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
set the source IP to the slirp4netns address instead of 127.0.0.1 when
using rootlesskit.
Closes: https://github.com/containers/podman/issues/5138
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
Systemd is now complaining or mentioning /var/run as a legacy directory.
It has been many years where /var/run is a symlink to /run on all
most distributions, make the change to the default.
Partial fix for https://github.com/containers/podman/issues/8369
Signed-off-by: Daniel J Walsh <dwalsh@redhat.com>
Just checking for `rootless.IsRootless()` does not catch all the
cases where slirp4netns is in use - we actually allow it to be
used as root as well. Fortify the conditional here so we don't
fail in the root + slirp case.
Fixes#7883
Signed-off-by: Matthew Heon <matthew.heon@pm.me>
This adds a new command, 'podman network reload', to reload the
networks of existing containers, forcing recreation of firewall
rules after e.g. `firewall-cmd --reload` wipes them out.
Under the hood, this works by calling CNI to tear down the
existing network, then recreate it using identical settings. We
request that CNI preserve the old IP and MAC address in most
cases (where the container only had 1 IP/MAC), but there will be
some downtime inherent to the teardown/bring-up approach. The
architecture of CNI doesn't really make doing this without
downtime easy (or maybe even possible...).
At present, this only works for root Podman, and only locally.
I don't think there is much of a point to adding remote support
(this is very much a local debugging command), but I think adding
rootless support (to kill/recreate slirp4netns) could be
valuable.
Signed-off-by: Matthew Heon <matthew.heon@pm.me>
Signed-off-by: Paul Holzinger <paul.holzinger@web.de>
a bug was being caused by the fact that the container network results
were not being updated properly.
given that jhon is on PTO, this PR will replace #8362
Signed-off-by: baude <bbaude@redhat.com>
* Make endpoint compatibile with docker-py network expectations
* Update specgen helper when called from compat endpoint
* Update godoc on types
* Add test for network/container create using docker-py method
* Add syslog logging when DEBUG=1 for tests
Fixes#8361
Signed-off-by: Jhon Honce <jhonce@redhat.com>
This makes things a lot more clear - if we are actually joining a
CNI network, we are guaranteed to get a non-zero length list of
networks.
We do, however, need to know if the network we are joining is the
default network for inspecting containers as it determines how we
populate the response struct. To handle this, add a bool to
indicate that the network listed was the default network, and
only the default network.
Signed-off-by: Matthew Heon <mheon@redhat.com>
this enables the ability to connect and disconnect a container from a
given network. it is only for the compatibility layer. some code had to
be refactored to avoid circular imports.
additionally, tests are being deferred temporarily due to some
incompatibility/bug in either docker-py or our stack.
Signed-off-by: baude <bbaude@redhat.com>
this enables the ability to connect and disconnect a container from a
given network. it is only for the compatibility layer. some code had to
be refactored to avoid circular imports.
additionally, tests are being deferred temporarily due to some
incompatibility/bug in either docker-py or our stack.
Signed-off-by: baude <bbaude@redhat.com>
