There are many use cases where you want to just mount an image
without creating a container on it. For example you might want
to just examine the content in an image after you pull it for
security analysys. Or you might want to just use the executables
on the image without running it in a container.
The image is mounted readonly since we do not want people changing
images.
Signed-off-by: Daniel J Walsh <dwalsh@redhat.com>
We should default to the user name unmount rather then the internal
name of umount.
Also User namespace was not being handled correctly. We want to inform
the user that if they do a mount when in rootless mode that they have
to be first in the podman unshare state.
Signed-off-by: Daniel J Walsh <dwalsh@redhat.com>
commit 788fdc685b introduced a race
where the target process dies before the child process opens the
namespace files. Move the open before the fork so if it fails the
parent process can attempt to join a different container instead of
failing.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
when /proc is mounted with hidepid=1 a process doesn't see processes
from the outer user namespace. This causes an issue reading the
cmdline from the parent process.
To address it, always read the command line from /proc/self instead of
using /proc/PARENT_PID.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
if there are more than FD_SETSIZE open fds passed down to the Podman
process, the initialization code could crash as it attempts to store
them into a fd_set. Use an array of fd_set structs, each of them
holding only FD_SETSIZE file descriptors.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
make sure the rootless env variables are set also when we are joining
directly the user+mount namespace without creating a new process.
It is required by pkg/unshare in containers/common.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
the renameat2 syscall might be defined in the C library but lacking
support in the kernel.
In such case, let it fallback to open(O_CREAT)+rename as it does on
systems lacking the definition for renameat2.
Closes: https://github.com/containers/libpod/issues/4570
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
do not attempt to join the rootless namespace if it is running already
with euid == 0.
Closes: https://github.com/containers/libpod/issues/3463
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
Currently pause process blocks all signals which may cause its
termination, including SIGTERM. This behavior hangs init(1) during
system shutdown, until pause process gets SIGKILLed after some grace
period. To avoid this hanging, SIGTERM is excluded from list of blocked
signals.
Fixes#3440
Signed-off-by: Danila Kiver <danila.kiver@mail.ru>
The second argument of `execlp` should be of type `char *`, so we need
to add an additional argument there.
Signed-off-by: Sascha Grunert <sgrunert@suse.com>
we are allowed to use only signal safe functions between a fork of a
multithreaded application and the next execve. Since setenv(3) is not
signal safe, block signals. We are already doing it for creating a
new namespace.
This is mostly a cleanup since reexec_in_user_namespace_wait is used
only only to join existing namespaces when we have not a pause.pid
file.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
after we read from the pause PID file, NUL terminate the buffer to
avoid reading garbage from the stack.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
move the logic for joining existing namespaces down to the rootless
package. In main_local we still retrieve the list of conmon pid files
and use it from the rootless package.
In addition, create a temporary user namespace for reading these
files, as the unprivileged user might not have enough privileges for
reading the conmon pid file, for example when running with a different
uidmap and root in the container is different than the rootless user.
Closes: https://github.com/containers/libpod/issues/3187
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
add a shortcut for joining immediately the namespace so we don't need
to re-exec Podman.
With the pause process simplificaton, we can now attempt to join the
namespaces as soon as Podman starts (and before the Go runtime kicks
in), so that we don't need to re-exec and use just one process.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
use a pause process to keep the user and mount namespace alive.
The pause process is created immediately on reload, and all successive
Podman processes will refer to it for joining the user&mount
namespace.
This solves all the race conditions we had on joining the correct
namespaces using the conmon processes.
As a fallback if the join fails for any reason (e.g. the pause process
was killed), then we try to join the running containers as we were
doing before.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
we were previously closing as many FDs as they were open when we first
started Podman in the range (3-MAX-FD). This would cause issues if
there were empty intervals, as these FDs are later on used by the
Golang runtime. Store exactly what FDs were first open in a fd_set,
so that we can close exactly the FDs that were open at startup.
Closes: https://github.com/containers/libpod/issues/2964
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
simplify the rootless implementation to use a single user namespace
for all the running containers.
This makes the rootless implementation behave more like root Podman,
where each container is created in the host environment.
There are multiple advantages to it: 1) much simpler implementation as
there is only one namespace to join. 2) we can join namespaces owned
by different containers. 3) commands like ps won't be limited to what
container they can access as previously we either had access to the
storage from a new namespace or access to /proc when running from the
host. 4) rootless varlink works. 5) there are only two ways to enter
in a namespace, either by creating a new one if no containers are
running or joining the existing one from any container.
Containers created by older Podman versions must be restarted.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
from _LIBPOD to _CONTAINERS. The same change was done in buildah
unshare.
This is necessary for podman to detect we are running in a rootless
environment and work properly from a "buildah unshare" session.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
we use "podman info" to reconfigure the runtime after a reboot, but we
don't propagate the error message back if something goes wrong.
Closes: https://github.com/containers/libpod/issues/2584
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
from the clone man page:
On the cris and s390 architectures, the order of the first two
arguments is reversed:
long clone(void *child_stack, unsigned long flags,
int *ptid, int *ctid,
unsigned long newtls);
Closes: https://bugzilla.redhat.com/show_bug.cgi?id=1672714
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
when joining an existing namespace, we were not maintaining the
current working directory, causing commands like export -o to fail
when they weren't referring to absolute paths.
Closes: https://github.com/containers/libpod/issues/2381
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
Add the possibility to join directly the user and mount namespace
without looking up the parent of the user namespace.
We need this in order to be able the conmon process, as the mount
namespace is kept alive only there.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
the issue is caused by the Go Runtime that messes up with the process
signals, overriding SIGSETXID and SIGCANCEL which are used internally
by glibc. They are used to inform all the threads to update their
stored uid/gid information. This causes a hang on the set*id glibc
wrappers since the handler installed by glibc is never invoked.
Since we are running with only one thread, we don't really need to
update other threads or even the current thread as we are not using
getuid/getgid before the execvp.
Closes: https://github.com/containers/libpod/issues/1625
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
change the tests to use chroot to set a numeric UID/GID.
Go syscall.Credential doesn't change the effective UID/GID of the
process.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
Closes: #1372
Approved by: mheon
Qi Wang
Giuseppe Scrivano
Daniel J Walsh
OpenShift Merge Robot
Erik Sjölund
Danila Kiver
Sascha Grunert
Harald Hoyer