Integration tests were failing due to proc filter behavior
changes with new apparmor policies.
Also include the missing docker-unconfined policy resolving
potential startup errors. This policy is complain-only so
it should behave identically to the standard unconfined policy,
but will not apply system path-based policies within containers.
Signed-off-by: Eric Windisch <eric@windisch.us>
Will attempt to load profiles automatically. If loading fails
but the profiles are already loaded, execution will continue.
A hard failure will only occur if Docker cannot load
the profiles *and* they have not already been loaded via
some other means.
Also introduces documentation for AppArmor.
Signed-off-by: Eric Windisch <eric@windisch.us>
By using the 'unconfined' policy for privileged
containers, we have inherited the host's apparmor
policies, which really make no sense in the
context of the container's filesystem.
For instance, policies written against
the paths of binaries such as '/usr/sbin/tcpdump'
can be easily circumvented by moving the binary
within the container filesystem.
Fixes GH#5490
Signed-off-by: Eric Windisch <eric@windisch.us>
It's introduced in
68ba5f0b69 (Execdriver implementation on new libcontainer API)
But I don't see reson why we need it.
Signed-off-by: Qiang Huang <h.huangqiang@huawei.com>
The automatic installation of AppArmor policies prevents the
management of custom, site-specific apparmor policies for the
default container profile. Furthermore, this change will allow
a future policy for the engine itself to be written without demanding
the engine be able to arbitrarily create and manage AppArmor policies.
- Add deb package suggests for apparmor.
- Ubuntu postinst use aa-status & fix policy path
- Add the policies to the debian packages.
- Add apparmor tests for writing proc files
Additional restrictions against modifying files in proc
are enforced by AppArmor. Ensure that AppArmor is preventing
access to these files, not simply Docker's configuration of proc.
- Remove /proc/k?mem from AA policy
The path to mem and kmem are in /dev, not /proc
and cannot be restricted successfully through AppArmor.
The device cgroup will need to be sufficient here.
- Load contrib/apparmor during integration tests
Note that this is somewhat dirty because we
cannot restore the host to its original configuration.
However, it should be noted that prior to this patch
series, the Docker daemon itself was loading apparmor
policy from within the tests, so this is no dirtier or
uglier than the status-quo.
Signed-off-by: Eric Windisch <eric@windisch.us>
Replaced github.com/docker/libcontainer with
github.com/opencontainers/runc/libcontaier.
Also I moved AppArmor profile generation to docker.
Main idea of this update is to fix mounting cgroups inside containers.
After updating docker on CI we can even remove dind.
Signed-off-by: Alexander Morozov <lk4d4@docker.com>
Libcontainer already supported mount container's own cgroup into
container, with this patch, we can see container's own cgroup info
in container.
Signed-off-by: Qiang Huang <h.huangqiang@huawei.com>
If a container is read-only, also set /proc, /sys,
& /dev to read-only. This should apply to both privileged and
unprivileged containers.
Note that when /dev is read-only, device files may still be
written to. This change will simply prevent the device paths
from being modified, or performing mknod of new devices within
the /dev path.
Tests are included for all cases. Also adds a test to ensure
that /dev/pts is always mounted read/write, even in the case of a
read-write rootfs. The kernel restricts writes here naturally and
bad things will happen if we mount it ro.
Signed-off-by: Eric Windisch <eric@windisch.us>
By default, the cgroup setting in libcontainer's configs.Cgroup for
memory swappiness will default to 0, which is a valid choice for memory
swappiness, but that means by default every container's memory
swappiness will be set to zero instead of the default 60, which is
probably not what users are expecting.
When the swappiness UI PR comes into Docker, there will be docker run
controls to set this per container, but for now we want to make sure
*not* to change the default, as well as work around an older kernel
issue that refuses to allow it to be set when cgroup hiearchies are in
use.
Docker-DCO-1.1-Signed-off-by: Phil Estes <estesp@linux.vnet.ibm.com> (github: estesp)
* Don't AllocateNetwork when network is disabled
* Don't createNetwork in execdriver when network is disabled
Signed-off-by: Alexander Morozov <lk4d4@docker.com>
- Updated Dockerfile to satisfy libnetwork GOPATH requirements.
- Reworked daemon to allocate network resources using libnetwork.
- Reworked remove link code to also update network resources in libnetwork.
- Adjusted the exec driver command population to reflect libnetwork design.
- Adjusted the exec driver create command steps.
- Updated a few test cases to reflect the change in design.
- Removed the dns setup code from docker as resolv.conf is entirely managed
in libnetwork.
- Integrated with lxc exec driver.
Signed-off-by: Jana Radhakrishnan <mrjana@docker.com>
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
Add tests for mounting into /proc and /sys
These two locations should be prohibited from mounting volumes into
those destinations.
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
The `--userland-proxy` daemon flag makes it possible to rely on hairpin
NAT and additional iptables routes instead of userland proxy for port
publishing and inter-container communication.
Usage of the userland proxy remains the default as hairpin NAT is
unsupported by older kernels.
Signed-off-by: Arnaud Porterie <arnaud.porterie@docker.com>
To help avoid version mismatches between libcontainer and Docker, this updates libcontainer to be the source of truth for which version of logrus the project is using. This should help avoid potential incompatibilities in the future, too. 👍
Signed-off-by: Andrew "Tianon" Page <admwiggin@gmail.com>
This also moves `exec -i` test to _unix_test.go because it seems to need a
pty to reliably reproduce the behavior.
Signed-off-by: Daniel, Dao Quang Minh <dqminh89@gmail.com>
When working with Go channels you must not set it to nil or else the
channel will block forever. It will not panic reading from a nil chan
but it blocks. The correct way to do this is to create the channel then
close it as the correct results to the caller will be returned.
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
container.
docker run -v /dev:/dev should stop mounting other default mounts in i
libcontainer otherwise directories and devices like /dev/ptx get mishandled.
We want to be able to run libvirtd for launching vms and it needs
access to the hosts /dev. This is a key componant of OpenStack.
Docker-DCO-1.1-Signed-off-by: Dan Walsh <dwalsh@redhat.com> (github: rhatdan)
This ensures that the libcontainer state is fully removed for a
container after it is terminated.
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
The default for rlimit handling should be to inherit the rlimit of the
daemon unless explicitly set.
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
Add a --readonly flag to allow the container's root filesystem to be
mounted as readonly. This can be used in combination with volumes to
force a container's process to only write to locations that will be
persisted. This is useful in many cases where the admin controls where
they would like developers to write files and error on any other
locations.
Closes#7923Closes#8752
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
We want to be able to use container without the PID namespace. We basically
want containers that can manage the host os, which I call Super Privileged
Containers. We eventually would like to get to the point where the only
namespace we use is the MNT namespace to bring the Apps userspace with it.
By eliminating the PID namespace we can get better communication between the
host and the clients and potentially tools like strace and gdb become easier
to use. We also see tools like libvirtd running within a container telling
systemd to place a VM in a particular cgroup, we need to have communications of the PID.
I don't see us needing to share PID namespaces between containers, since this
is really what docker exec does.
So currently I see us just needing docker run --pid=host
Docker-DCO-1.1-Signed-off-by: Dan Walsh <dwalsh@redhat.com> (github: rhatdan)
This commit contains changes for docker:
* user.GetGroupFile to user.GetGroupPath docker/libcontainer#301
* Add systemd support for OOM docker/libcontainer#307
* Support for custom namespaces docker/libcontainer#279, docker/libcontainer#312
* Fixes#9699docker/libcontainer#308
Signed-off-by: Alexander Morozov <lk4d4@docker.com>
Some workloads rely on IPC for communications with other processes. We
would like to split workloads between two container but still allow them
to communicate though shared IPC.
This patch mimics the --net code to allow --ipc=host to not split off
the IPC Namespace. ipc=container:CONTAINERID to share ipc between containers
If you share IPC between containers, then you need to make sure SELinux labels
match.
Docker-DCO-1.1-Signed-off-by: Dan Walsh <dwalsh@redhat.com> (github: rhatdan)
Since the containers can handle the out of memory kernel kills gracefully, docker
will only provide out of memory information as an additional metadata as part of
container status.
Docker-DCO-1.1-Signed-off-by: Vishnu Kannan <vishnuk@google.com> (github: vishh)
We removed the syncpipe package and replaced it with specific calls to
create a new *os.File from a specified fd passed to the process. This
reduced code and an extra object to manage the container's init
lifecycle.
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
Right now, MAC addresses are randomly generated by the kernel when
creating the veth interfaces.
This causes different issues related to ARP, such as #4581, #5737 and #8269.
This change adds support for consistent MAC addresses, guaranteeing that
an IP address will always end up with the same MAC address, no matter
what.
Since IP addresses are already guaranteed to be unique by the
IPAllocator, MAC addresses will inherit this property as well for free.
Consistent mac addresses is also a requirement for stable networking (#8297)
since re-using the same IP address on a different MAC address triggers the ARP
issue.
Finally, this change makes the MAC address accessible through docker
inspect, which fixes#4033.
Signed-off-by: Andrea Luzzardi <aluzzardi@gmail.com>
This also removes dead code in the native driver for a past feature that
was never fully implemented.
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
This changes the way the exec drivers work by not specifing a -driver
flag on reexec. For each of the exec drivers they register their own
functions that will be matched aginst the argv 0 on exec and called if
they match.
This also allows any functionality to be added to docker so that the
binary can be reexec'd and any type of function can be called. I moved
the flag parsing on docker exec to the specific initializers so that the
implementations do not bleed into one another. This also allows for
more flexability within reexec initializers to specify their own flags
and options.
Signed-off-by: Michael Crosby <michael@docker.com>
lxc is special in that we cannot create the master outside of the
container without opening the slave because we have nothing to provide to the
cmd. We have to open both then do the crazy setup on command right now instead of
passing the console path to lxc and telling it to open up that console. we save a couple of
openfiles in the native driver because we can do this.
Docker-DCO-1.1-Signed-off-by: Michael Crosby <michael@docker.com> (github: crosbymichael)
This patch adds pause/unpause to the command line, api, and drivers
for use on containers. This is implemented using the cgroups/freeze
utility in libcontainer and lxc freeze/unfreeze.
Co-Authored-By: Eric Windisch <ewindisch@docker.com>
Co-Authored-By: Chris Alfonso <calfonso@redhat.com>
Docker-DCO-1.1-Signed-off-by: Ian Main <imain@redhat.com> (github: imain)
This also makes sure that devices are pointers to avoid copies
Docker-DCO-1.1-Signed-off-by: Michael Crosby <michael@crosbymichael.com> (github: crosbymichael)
We now have one place that keeps track of (most) devices that are allowed and created within the container. That place is pkg/libcontainer/devices/devices.go
This fixes several inconsistencies between which devices were created in the lxc backend and the native backend. It also fixes inconsistencies between wich devices were created and which were allowed. For example, /dev/full was being created but it was not allowed within the cgroup. It also declares the file modes and permissions of the default devices, rather than copying them from the host. This is in line with docker's philosphy of not being host dependent.
Docker-DCO-1.1-Signed-off-by: Timothy Hobbs <timothyhobbs@seznam.cz> (github: https://github.com/timthelion)
Add specific types for Required and Optional DeviceNodes
Docker-DCO-1.1-Signed-off-by: Michael Crosby <michael@crosbymichael.com> (github: crosbymichael)
Michael Crosby
Qiang Huang
Eric Windisch
Jessie Frazelle
David Calavera
Alexander Morozov
Mrunal Patel
unclejack
Phil Estes
Arnaud Porterie
Alexander Morozov
Jessica Frazelle
Antonio Murdaca
Jana Radhakrishnan
Darren Shepherd
John Howard
Tianon Gravi
Jessica Frazelle
Daniel, Dao Quang Minh
Ma Shimiao
Michael Crosby
Lei Jitang
Yan Feng
guoxiuyan
Dan Walsh
Brian Goff
Abin Shahab
Malte Janduda
Vishnu Kannan
Andrea Luzzardi
Victor Vieux
Erik Hollensbe
Michael Crosby
Tibor Vass
Michael Crosby
Chris Alfonso
Ian Main
unclejack
Timothy Hobbs
Victor Vieux