When we launch a container in a new user namespace, we cannot create
devices, so we bind mount the host's devices into place instead.
If we are running in a user namespace (i.e. nested in a container),
then we need to do the same thing. Add a function to detect that
and check for it before doing mknod.
Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>
---
Changelog - add a comment clarifying what's going on with the
uidmap file.
It may be desirable to receive memory pressure levels notifications
before the container depletes all memory. This may be useful for
handling cases where the system thrashes when reaching the container's
memory limits.
Signed-off-by: Ido Yariv <ido@wizery.com>
Due to the fact that the init is implemented in Go (which seemingly
randomly spawns new processes and loves eating memory), most cgroup
configurations are required to have an arbitrary minimum dictated by the
init. This confuses users and makes configuration more annoying than it
should. An example of this is pids.max, where Go spawns multiple
processes that then cause init to violate the pids cgroup constraint
before the container can even start.
Solve this problem by setting the cgroup configurations as late as
possible, to avoid hitting as many of the resources hogged by the Go
init as possible. This has to be done before seccomp rules are applied,
as the parent and child must synchronise in order for the parent to
correctly set the configurations (and writes might be blocked by seccomp).
Signed-off-by: Aleksa Sarai <asarai@suse.com>
It is vital to loudly fail when a user attempts to set a cgroup limit
(rather than using the system default). Otherwise the user will assume
they have security they do not actually have. This mirrors the original
Apply() (that would set cgroup configs) semantics.
Signed-off-by: Aleksa Sarai <asarai@suse.com>
Apply and Set are two separate operations, and it doesn't make sense to
group the two together (especially considering that the bootstrap
process is added to the cgroup as well). The only exception to this is
the memory cgroup, which requires the configuration to be set before
processes can join.
Signed-off-by: Aleksa Sarai <asarai@suse.com>
Add support for the pids cgroup controller to libcontainer, a recent
feature that is available in Linux 4.3+.
Unfortunately, due to the init process being written in Go, it can spawn
an an unknown number of threads due to blocked syscalls. This results in
the init process being unable to run properly, and thus small pids.max
configs won't work properly.
Signed-off-by: Aleksa Sarai <asarai@suse.com>
syscall.NLA_HDRLEN is not in gccgo (as of 5.3), so in the meantime
use the #defines taken from linux/netlink.h.
See https://github.com/golang/go/issues/13629
Signed-off-by: Christy Perez <christy@linux.vnet.ibm.com>
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
Add state status() method
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
Allow multiple checkpoint on restore
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
Handle leave-running state
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
Fix state transitions for inprocess
Because the tests use libcontainer in process between the various states
we need to ensure that that usecase works as well as the out of process
one.
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
Remove isDestroyed method
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
Handling Pausing from freezer state
Signed-off-by: Rajasekaran <rajasec79@gmail.com>
freezer status
Signed-off-by: Rajasekaran <rajasec79@gmail.com>
Fixing review comments
Signed-off-by: Rajasekaran <rajasec79@gmail.com>
Added comment when freezer not available
Signed-off-by: Rajasekaran <rajasec79@gmail.com>
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
Conflicts:
libcontainer/container_linux.go
Change checkFreezer logic to isPaused()
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
Remove state base and factor out destroy func
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
Add unit test for state transitions
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
This allows us to distinguish cases where a container
needs to just join the paths or also additionally
set cgroups settings. This will help in implementing
cgroupsPath support in the spec.
Signed-off-by: Mrunal Patel <mrunalp@gmail.com>
replace passing of pid and console path via environment variable with passing
them with netlink message via an established pipe.
this change requires us to set _LIBCONTAINER_INITTYPE and
_LIBCONTAINER_INITPIPE as the env environment of the bootstrap process as we
only send the bootstrap data for setns process right now. When init and setns
bootstrap process are unified (i.e., init use nsexec instead of Go to clone new
process), we can remove _LIBCONTAINER_INITTYPE.
Note:
- we read nlmsghdr first before reading the content so we can get the total
length of the payload and allocate buffer properly instead of allocating
one large buffer.
- check read bytes vs the wanted number. It's an error if we failed to read
the desired number of bytes from the pipe into the buffer.
Signed-off-by: Daniel, Dao Quang Minh <dqminh89@gmail.com>
This patch fixes the following go vet warnings:
```
libcontainer/network_linux.go:96: github.com/vishvananda/netlink.Device
composite literal uses unkeyed fields
libcontainer/network_linux.go:114: github.com/vishvananda/netlink.Device
composite literal uses unkeyed fields
```
Signed-off-by: Antonio Murdaca <runcom@redhat.com>
add bootstrap data to setns process. If we have any bootstrap data then copy it
to the bootstrap process (i.e. nsexec) using the sync pipe. This will allow us
to eventually replace environment variable usage with more structured data
to setup namespaces, write pid/gid map, setgroup etc.
Signed-off-by: Daniel, Dao Quang Minh <dqminh89@gmail.com>
Enables launching userns containers by catching EPERM errors for writing
to devices cgroups, and for mknod invocations.
Signed-off-by: Abin Shahab <ashahab@altiscale.com>
When starting and quering for pids a container can start and exit before
this is set. So set the opts after the process is started and while
libcontainer still has the container's process blocking on the pipe.
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
The former cgroup entry is confusing, separate it to parent
and name.
Rename entry `c` to `config`.
Signed-off-by: Qiang Huang <h.huangqiang@huawei.com>
'parent' function is confusing with parent cgroup, it's actually
parent path, so rename it to parentPath.
The name 'data' is too common to be identified, rename it to cgroupData
which is exactly what it is.
Signed-off-by: Qiang Huang <h.huangqiang@huawei.com>
The spec uses symlinks to "/proc/1/..." but the implementation uses
"/proc/self/...": see setupDevSymlinks (libcontainer/rootfs_linux.go).
The implementation is more correct, so I'm changing the spec to match
the implementation.
Signed-off-by: Alban Crequy <alban.crequy@coreos.com>
Minor fix, the former setupDev=true means not setup dev,
which is contrary to intuition, just correct it.
Signed-off-by: Qiang Huang <h.huangqiang@huawei.com>
We have a rule that for optional cgroups, don't fail if some
of them are not mounted, but we want it fail hard when a
user specifies an option and we are unable to fulfill the
request.
Memory cgroup should also follow this rule.
Signed-off-by: Qiang Huang <h.huangqiang@huawei.com>
Also add cpuset as the first in the list to address issues setting the
pid in any cgroup before the cpuset is populated.
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
It can avoid unnecessary task migrataion, see this scenario:
- container init task is on cpu 1, and we assigned it to cpu 1,
but parent cgroup's cpuset.cpus=2
- we created the cgroup dir and inherited cpuset.cpus from parent as 2
- write container init task's pid to cgroup.procs
- [it's possibile the container init task migrated to cpu 2 here]
- set cpuset.cpus as assigned to cpu 1
- [the container init task has to be migrated back to cpu 1]
So we should set cpuset.cpus and cpuset.mems before writing pids
to cgroup.procs to aviod such problem.
Signed-off-by: Qiang Huang <h.huangqiang@huawei.com>
While testing different versions of criu it helps to know which
criu binary with which options is currently used. Therefore additional
debug output to display these information is added.
v2: increase readability of printed out criu options
Signed-off-by: Adrian Reber <adrian@lisas.de>
Only valid options to --security-opt for label should be
disable, user, role, type, level.
Return error on invalid entry
Signed-off-by: Dan Walsh <dwalsh@redhat.com>
This rather naively fixes an error observed where a processes stdio
streams are not written to when there is an error upon starting up the
process, such as when the executable doesn't exist within the
container's rootfs.
Before the "fix", when an error occurred on start, `terminate` is called
immediately, which calls `cmd.Process.Kill()`, then calling `Wait()` on
the process. In some cases when this `Kill` is called the stdio stream
have not yet been written to, causing non-deterministic output. The
error itself is properly preserved but users attached to the process
will not see this error.
With the fix it is just calling `Wait()` when an error occurs rather
than trying to `Kill()` the process first. This seems to preserve stdio.
Signed-off-by: Brian Goff <cpuguy83@gmail.com>
Docker pkgs were updated while golinting the whole docker code base.
Now when trying to bump libcontainer/runc in docker, it fails compiling
with the following error:
``
vendor/src/github.com/opencontainers/runc/libcontainer/rootfs_linux.go:424:
undefined: mount.MountInfo
``
This is because, for instance, the mount pkg was updated here
0f5c9d301b (diff-49294d05afa48e2f7c0d2f02c6f7614c)
and now that type is only `mount.Info`.
This patch bump docker pkgs commit and adapt code to it.
Signed-off-by: Antonio Murdaca <amurdaca@redhat.com>
This allows getting the path to the subsystem and so is subsequently
used in EnterPid by an exec process.
Signed-off-by: Mrunal Patel <mrunalp@gmail.com>
This is meant to be used in retrieving the paths so an exec
process enters all the cgroup paths correctly.
Signed-off-by: Mrunal Patel <mrunalp@gmail.com>
/etc/groups is not needed when specifying numeric group ids. This
change allows containers without /etc/groups to specify numeric
supplemental groups.
Signed-off-by: Sami Wagiaalla <swagiaal@redhat.com>
Godeps: Vendor opencontainers/specs 96bcd043aa
Fix a bug where it's impossible to pass multiple devices to blkio
cgroup controller files. See https://github.com/opencontainers/runc/issues/274
Signed-off-by: Antonio Murdaca <runcom@linux.com>
pivotDir is the one where pivot_root() call puts the old root. We will
unmount pivotDir() and delete it.
Previously we were making / always rslave or rprivate. That will mean
that pivotDir() could never have mounts which would be shared with
parent mount namespace. That also means that unmounting pivotDir() was
safe and none of the unmount will propagate to parent namespace and
unmount things which we did not want to.
But now user can specify that apply private, shared, slave on /. That
means some of the mounts we inherited from parent could be shared and that
also means if we umount pivotDir/, those mounts will get unmounted in
parent too. That's not what we want.
Instead make pivotDir rprivate so that unmounts don't propagate back to
parent.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
pivot_root() introduces bunch of restrictions otherwise it fails. parent
mount of container root can not be shared otherwise pivot_root() will
fail.
So far parent could not be shared as we marked everything either private
or slave. But now we have introduced new propagation modes where parent
mount of container rootfs could be shared and pivot_root() will fail.
So check if parent mount is shared and if yes, make it private. This will
make sure pivot_root() works.
Also it will make sure that when we bind mount container rootfs, it does
not propagate to parent mount namespace. Otherwise cleanup becomes a
problem.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Right now config.Privatefs is a boolean which determines if / is applied
with propagation flag syscall.MS_PRIVATE | syscall.MS_REC or not.
Soon we want to represent other propagation states like private, [r]slave,
and [r]shared. So either we can introduce more boolean variable or keep
track of propagation flags in an integer variable. Keeping an integer
variable is more versatile and can allow various kind of propagation flags
to be specified. So replace Privatefs with RootPropagation which is an
integer.
Note, this will require changes in docker. Instead of setting Privatefs
to true, they will need to set.
config.RootPropagation = syscall.MS_PRIVATE | syscall.MS_REC
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Do not remount a bind mount to enable flags unless non-default flags are
provided for the requested mount. This solves a problem with user
namespaces and remount of bind mount permissions.
Docker-DCO-1.1-Signed-off-by: Phil Estes <estesp@linux.vnet.ibm.com> (github: estesp)
Do not have methods and actions that require syscalls in the configs
package because it breaks cross compile.
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
This commit allows additional architectures to be added to Seccomp filters
created by containers. This allows containers to make syscalls using these
architectures. For example, in a container on an AMD64 system, only AMD64
syscalls would be usable unless x86 was added to the filter using this patch,
which would allow both 32-bit and 64-bit syscalls to be used.
Signed-off-by: Matthew Heon <mheon@redhat.com>
We need to update the mount's destination after we resolve symlinks so
that it properly creates and mounts the correct location.
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
Whenever dev/null is used as one of the main processes STDIO, do not try
to change the permissions on it via fchown because we should not do it
in the first place and also this will fail if the container is supposed
to be readonly.
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
When executing an additional process in a container, all namespaces are
entered but the user namespace. As a result, the process may be
executed as the host's root user. This has both functionality and
security implications.
Fix this by adding the missing user namespace to the array of
namespaces. Since joining a user namespace in which the caller is
already a member yields an error, skip namespaces we're already in.
Last, remove a needless and buggy AT_SYMLINK_NOFOLLOW in the code.
Signed-off-by: Ido Yariv <ido@wizery.com>
Fix the permissions of the container's main processes STDIO when the
process is not run as the root user. This changes the permissions right
before switching to the specified user so that it's STDIO matches it's
UID and GID.
Add a test for checking that the STDIO of the process is owned by the
specified user.
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
Right now if one passes a mount propagation flag in spec file, it
does not take effect. For example, try following in spec json file.
{
"type": "bind",
"source": "/root/mnt-source",
"destination": "/root/mnt-dest",
"options": "rbind,shared"
}
One would expect that /root/mnt-dest will be shared inside the container
but that's not the case.
#findmnt -o TARGET,PROPAGATION
`-/root/mnt-dest private
Reason being that propagation flags can't be passed in along with other
regular flags. They need to be passed in a separate call to mount syscall.
That too, one propagation flag at a time. (from mount man page).
Hence, store propagation flags separately in a slice and apply these
in that order after the mount call wherever appropriate. This allows
user to control the propagation property of mount point inside
the container.
Storing them separately also solves another problem where recursive flag
(syscall.MS_REC) can get mixed up. For example, options "rbind,private"
and "bind,rprivate" will be same and there will be no way to differentiate
between these if all the flags are stored in a single integer.
This patch would allow one to pass propagation flags "[r]shared,[r]slave,
[r]private,[r]unbindable" in spec file as per mount property.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Here are two reasons:
* If we use systemd, we need to ask it to create cgroups
* If a container is restored with another ID, we need to
change paths to cgroups.
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Bug was introduced in #250
According to: http://man7.org/linux/man-pages/man5/proc.5.html
36 35 98:0 /mnt1 /mnt2 rw,noatime master:1 - ext3 /dev/root rw,errors=continue
(1)(2)(3) (4) (5) (6) (7) (8) (9) (10) (11)
...
(7) optional fields: zero or more fields of the form
"tag[:value]".
The 7th field is optional. We should skip it when parsing mount info.
Signed-off-by: Qiang Huang <h.huangqiang@huawei.com>
I got:
```
exec_test.go:823: Mode expected to contain 'ro,nosuid,nodev,noexec': tmpfs on /sys/fs/cgroup type tmpfs (ro,seclabel,nosuid,nodev,noexec,relatime,mode=755
```wq
Signed-off-by: Qiang Huang <h.huangqiang@huawei.com>
Again. It looks like a build tag was somehow dropped between
the PR here: https://github.com/docker/libcontainer/pull/625
and the move to runc.
Signed-off-by: Christy Perez <clnperez@linux.vnet.ibm.com>
As v2.1.0 is no longer required for successful testing, do not build it in the
Dockerfile - instead just use the version Ubuntu ships.
Signed-off-by: Matthew Heon <mheon@redhat.com>
This removes the existing, native Go seccomp filter generation and replaces it
with Libseccomp. Libseccomp is a C library which provides architecture
independent generation of Seccomp filters for the Linux kernel.
This adds a dependency on v2.2.1 or above of Libseccomp.
Signed-off-by: Matthew Heon <mheon@redhat.com>
Simplify the code introduced by the commit d1f0d5705deb:
Return actual ProcessState on Wait error
Cc: Alexander Morozov <lk4d4@docker.com>
Signed-off-by: Lai Jiangshan <jiangshanlai@gmail.com>
This adds a `Signal()` method to the container interface so that the
initial process can be signaled after a Load or operation. It also
implements signaling the init process from a nonChildProcess.
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
A boolean field named GidMappingsEnableSetgroups was added to
SysProcAttr in Go1.5. This field determines the value of the process's
setgroups proc entry.
Since the default is to set the entry to 'deny', calling setgroups will
fail on systems running kernels 3.19+.
Set GidMappingsEnableSetgroups to true so setgroups wont be set to
'deny'.
Signed-off-by: Ido Yariv <ido@wizery.com>
Alexander Morozov
Serge Hallyn
Jimmi Dyson
Ahmet Alp Balkan
xlgao-zju
Mrunal Patel
Qiang Huang
Kai Qiang WU(Kennan)
Ido Yariv
Aleksa Sarai
Michael Crosby
Christy Perez
Mrunal Patel
David Calavera
Daniel, Dao Quang Minh
Antonio Murdaca
Rajasekaran
Abin Shahab
John Howard
Doug Davis
Dan Walsh
yangshukui
Alban Crequy
Alexander Morozov
Adrian Reber
Hui Kang
Brian Goff
Bogdan Purcareata
Antonio Murdaca
Sami Wagiaalla
Antonio Murdaca
Vivek Goyal
Chun Chen
Phil Estes
Matthew Heon
Andrey Vagin
Vishnu Kannan
Matthew Heon
boucher
Lai Jiangshan
Shijiang Wei
Michael Chase-Salerno