Building off insights from the great work Cory Snider has been doing,
this replaces a reexec with a much lower overhead implementation which
performs the `Chddir` in a new goroutine that is locked to a specific
thread with CLONE_FS unshared.
The thread is thrown away afterwards and the Chdir does effectively the
same thing as what the reexec was being used for.
Signed-off-by: Brian Goff <cpuguy83@gmail.com>
Commit 955c1f881a (Docker v17.12.0) replaced
detection of support for multiple lowerdirs (as required by overlay2) to not
depend on the kernel version. The `overlay2.override_kernel_check` was still
used to print a warning that older kernel versions may not have full support.
After this, commit e226aea280 (Docker v20.10.0,
backported to v19.03.7) removed uses of the `overlay2.override_kernel_check`
option altogether, but we were still parsing it.
This patch changes the `parseOptions()` function to not parse the option,
printing a deprecation warning instead. We should change this to be an error,
but the `overlay2.override_kernel_check` option was not deprecated in the
documentation, so keeping it around for one more release.
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
On Linux, when (os/exec.Cmd).SysProcAttr.Pdeathsig is set, the signal
will be sent to the process when the OS thread on which cmd.Start() was
executed dies. The runtime terminates an OS thread when a goroutine
exits after being wired to the thread with runtime.LockOSThread(). If
other goroutines are allowed to be scheduled onto a thread which called
cmd.Start(), an unrelated goroutine could cause the thread to be
terminated and prematurely signal the command. See
https://github.com/golang/go/issues/27505 for more information.
Prevent started subprocesses with Pdeathsig from getting signaled
prematurely by wiring the starting goroutine to the OS thread until the
subprocess has exited. No other goroutines can be scheduled onto a
locked thread so it will remain alive until unlocked or the daemon
process exits.
Signed-off-by: Cory Snider <csnider@mirantis.com>
The pkg/fsutils package was forked in containerd, and later moved to
containerd/continuity/fs. As we're moving more bits to containerd, let's also
use the same implementation to reduce code-duplication and to prevent them from
diverging.
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
Support for overlay on a backing filesystem without d_type was deprecated in
0abb8dec3f (Docker 17.12), with an exception
for existing installations (0a4e793a3d).
That deprecation was nearly 5 years ago, and running without d_type is known to
cause serious issues (so users will likely already have run into other problems).
This patch removes support for running overlay and overlay2 on these filesystems,
returning the error instead of logging it.
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
Finish the refactor which was partially completed with commit
34536c498d, passing around IdentityMapping structs instead of pairs of
[]IDMap slices.
Existing code which uses []IDMap relies on zero-valued fields to be
valid, empty mappings. So in order to successfully finish the
refactoring without introducing bugs, their replacement therefore also
needs to have a useful zero value which represents an empty mapping.
Change IdentityMapping to be a pass-by-value type so that there are no
nil pointers to worry about.
The functionality provided by the deprecated NewIDMappingsFromMaps
function is required by unit tests to to construct arbitrary
IdentityMapping values. And the daemon will always need to access the
mappings to pass them to the Linux kernel. Accommodate these use cases
by exporting the struct fields instead. BuildKit currently depends on
the UIDs and GIDs methods so we cannot get rid of them yet.
Signed-off-by: Cory Snider <csnider@mirantis.com>
pkg/system historically has been a bit of a kitchen-sink of things that were
somewhat "system" related, but didn't have a good place for. EnsureRemoveAll()
is one of those utilities. EnsureRemoveAll() is used to both unmount and remove
a path, for which it depends on both github.com/moby/sys/mount, which in turn
depends on github.com/moby/sys/mountinfo.
pkg/system is imported in the CLI, but neither EnsureRemoveAll(), nor any of its
moby/sys dependencies are used on the client side, so let's move this function
somewhere else, to remove those dependencies from the CLI.
I looked for plausible locations that were related; it's used in:
- daemon
- daemon/graphdriver/XXX/
- plugin
I considered moving it into a (e.g.) "utils" package within graphdriver (but not
a huge fan of "utils" packages), and given that it felt (mostly) related to
cleaning up container filesystems, I decided to move it there.
Some things to follow-up on after this:
- Verify if this function is still needed (it feels a bit like a big hammer in
a "YOLO, let's try some things just in case it fails")
- Perhaps it should be integrated in `containerfs.Remove()` (so that it's used
automatically)
- Look if there's other implementations (and if they should be consolidated),
although (e.g.) the one in containerd is a copy of ours:
https://github.com/containerd/containerd/blob/v1.5.9/pkg/cri/server/helpers_linux.go#L200
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
Do not use 0701 perms.
0701 dir perms allows anyone to traverse the docker dir.
It happens to allow any user to execute, as an example, suid binaries
from image rootfs dirs because it allows traversal AND critically
container users need to be able to do execute things.
0701 on lower directories also happens to allow any user to modify
things in, for instance, the overlay upper dir which neccessarily
has 0755 permissions.
This changes to use 0710 which allows users in the group to traverse.
In userns mode the UID owner is (real) root and the GID is the remapped
root's GID.
This prevents anyone but the remapped root to traverse our directories
(which is required for userns with runc).
Signed-off-by: Brian Goff <cpuguy83@gmail.com>
(cherry picked from commit ef7237442147441a7cadcda0600be1186d81ac73)
Signed-off-by: Brian Goff <cpuguy83@gmail.com>
(cherry picked from commit 93ac040bf0)
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
The io/ioutil package has been deprecated in Go 1.16. This commit
replaces the existing io/ioutil functions with their new definitions in
io and os packages.
Signed-off-by: Eng Zer Jun <engzerjun@gmail.com>
overlay2 no longer sets `archive.OverlayWhiteoutFormat` when
running in UserNS, so we can remove the complicated logic in the
archive package.
Signed-off-by: Akihiro Suda <akihiro.suda.cz@hco.ntt.co.jp>
When running in userns, returns error (i.e. "use naive, not native")
immediately.
No substantial change to the logic.
Signed-off-by: Akihiro Suda <akihiro.suda.cz@hco.ntt.co.jp>
The "userxattr" option is needed for mounting overlayfs inside a user namespace with kernel >= 5.11.
The "userxattr" option is NOT needed for the initial user namespace (aka "the host").
Also, Ubuntu (since circa 2015) and Debian (since 10) with kernel < 5.11 can mount the overlayfs in a user namespace without the "userxattr" option.
The corresponding kernel commit: 2d2f2d7322ff43e0fe92bf8cccdc0b09449bf2e1
> **ovl: user xattr**
>
> Optionally allow using "user.overlay." namespace instead of "trusted.overlay."
> ...
> Disable redirect_dir and metacopy options, because these would allow privilege escalation through direct manipulation of the
> "user.overlay.redirect" or "user.overlay.metacopy" xattrs.
Fix issue 42055
Related to containerd/containerd PR 5076
Signed-off-by: Akihiro Suda <akihiro.suda.cz@hco.ntt.co.jp>
Various dirs in /var/lib/docker contain data that needs to be mounted
into a container. For this reason, these dirs are set to be owned by the
remapped root user, otherwise there can be permissions issues.
However, this uneccessarily exposes these dirs to an unprivileged user
on the host.
Instead, set the ownership of these dirs to the real root (or rather the
UID/GID of dockerd) with 0701 permissions, which allows the remapped
root to enter the directories but not read/write to them.
The remapped root needs to enter these dirs so the container's rootfs
can be configured... e.g. to mount /etc/resolve.conf.
This prevents an unprivileged user from having read/write access to
these dirs on the host.
The flip side of this is now any user can enter these directories.
Signed-off-by: Brian Goff <cpuguy83@gmail.com>
(cherry picked from commit e908cc3901)
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
This is a fix for https://github.com/docker/for-linux/issues/1012.
The code was not considering that C strings are NULL-terminated so
we need to leave one extra byte.
Without this fix, the testcase in https://github.com/docker/for-linux/issues/1012
fails with
```
Step 61/1001 : RUN echo 60 > 60
---> Running in dde85ac3b1e3
Removing intermediate container dde85ac3b1e3
---> 80a12a18a241
Step 62/1001 : RUN echo 61 > 61
error creating overlay mount to /23456789112345678921234/overlay2/d368abcc97d6c6ebcf23fa71225e2011d095295d5d8c9b31d6810bea748bdf07-init/merged: no such file or directory
```
with the output of `dmesg -T` as:
```
[Sat Dec 19 02:35:40 2020] overlayfs: failed to resolve '/23456789112345678921234/overlay2/89e435a1b24583c463abb73e8abfad8bf8a88312ef8253455390c5fa0a765517-init/wor': -2
```
with this fix, you get the expected:
```
Step 126/1001 : RUN echo 125 > 125
---> Running in 2f2e56da89e0
max depth exceeded
```
Signed-off-by: Oscar Bonilla <6f6231@gmail.com>
The implementation in libcontainer/system is quite complicated,
and we only use it to detect if user-namespaces are enabled.
In addition, the implementation in containerd uses a sync.Once,
so that detection (and reading/parsing `/proc/self/uid_map`) is
only performed once.
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
Switch to moby/sys/mount and mountinfo. Keep the pkg/mount for potential
outside users.
This commit was generated by the following bash script:
```
set -e -u -o pipefail
for file in $(git grep -l 'docker/docker/pkg/mount"' | grep -v ^pkg/mount); do
sed -i -e 's#/docker/docker/pkg/mount"#/moby/sys/mount"#' \
-e 's#mount\.\(GetMounts\|Mounted\|Info\|[A-Za-z]*Filter\)#mountinfo.\1#g' \
$file
goimports -w $file
done
```
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
`fuse-overlayfs` provides rootless overlayfs functionality without depending
on any kernel patch.
Aside from rootless, `fuse-overlayfs` could be potentially used for eliminating
`chown()` calls that happen in userns-remap mode, because `fuse-overlayfs` also
provides shiftfs functionality.
System requirements:
* fuse-overlayfs needs to be installed. Tested with 0.7.6.
* kernel >= 4.18
Unit test: `go test -exec sudo -v ./daemon/graphdriver/fuse-overlayfs`
The implementation is based on Podman's `overlay` driver which supports
both kernel-mode overlayfs and fuse-overlayfs in the single driver instance:
https://github.com/containers/storage/blob/39a8d5ed/drivers/overlay/overlay.go
However, Moby's implementation aims to decouple `fuse-overlayfs` driver from the
kernel-mode driver (`overlay2`) for simplicity.
Fix#40218
Signed-off-by: Akihiro Suda <akihiro.suda.cz@hco.ntt.co.jp>
Now that we do check if overlay is working by performing an actual
overlayfs mount, there's no need in extra checks for the kernel version
or the filesystem type. Actual mount check is sufficient.
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
Before this commit, overlay check was performed by looking for
`overlay` in /proc/filesystem. This obviously might not work
for rootless Docker (fs is there, but one can't use it as non-root).
This commit changes the check to perform the actual mount, by reusing
the code previously written to check for multiple lower dirs support.
The old check is removed from both drivers, as well as the additional
check for the multiple lower dirs support in overlay2 since it's now
a part of the main check.
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
This moves supportsMultipleLowerDir() to overlayutils
so it can be used from both overlay and overlay2.
The only changes made were:
* replace logger with logrus
* don't use workDirName mergedDirName constants
* add mnt var to improve readability a bit
This is a preparation for the next commit.
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
Format the source according to latest goimports.
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
When mounting overlays which have children, enforce that
the mount is always performed as read only. Newer versions
of the kernel return a device busy error when a lower directory
is in use as an upper directory in another overlay mount.
Adds committed file to indicate when an overlay is being used
as a parent, ensuring it will no longer be mounted with an
upper directory.
Signed-off-by: Derek McGowan <derek@mcgstyle.net>
As pointed out in https://github.com/moby/moby/issues/37970,
Docker overlay driver can't work with index=on feature of
the Linux kernel "overlay" filesystem. In case the global
default is set to "yes", Docker will fail with EBUSY when
trying to mount, like this:
> error creating overlay mount to ...../merged: device or resource busy
and the kernel log should contain something like:
> overlayfs: upperdir is in-use by another mount, mount with
> '-o index=off' to override exclusive upperdir protection.
A workaround is to set index=off in overlay kernel module
parameters, or even recompile the kernel with
CONFIG_OVERLAY_FS_INDEX=n in .config. Surely this is not
always practical or even possible.
The solution, as pointed out my Amir Goldstein (as well as
the above kernel message:) is to use 'index=off' option
when mounting.
NOTE since older (< 4.13rc1) kernels do not support "index="
overlayfs parameter, try to figure out whether the option
is supported. In case it's not possible to figure out,
assume it is not.
NOTE the default can be changed anytime (by writing to
/sys/module/overlay/parameters/index) so we need to always
use index=off.
[v2: move the detection code to Init()]
[v3: don't set index=off if stat() failed]
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
This implements chown support on Windows. Built-in accounts as well
as accounts included in the SAM database of the container are supported.
NOTE: IDPair is now named Identity and IDMappings is now named
IdentityMapping.
The following are valid examples:
ADD --chown=Guest . <some directory>
COPY --chown=Administrator . <some directory>
COPY --chown=Guests . <some directory>
COPY --chown=ContainerUser . <some directory>
On Windows an owner is only granted the permission to read the security
descriptor and read/write the discretionary access control list. This
fix also grants read/write and execute permissions to the owner.
Signed-off-by: Salahuddin Khan <salah@docker.com>
