Before this change, volume management was relying on the fact that
everything the plugin mounts is visible on the host within the plugin's
rootfs. In practice this caused some issues with mount leaks, so we
changed the behavior such that mounts are not visible on the plugin's
rootfs, but available outside of it, which breaks volume management.
To fix the issue, allow the plugin to scope the path correctly rather
than assuming that everything is visible in `p.Rootfs`.
In practice this is just scoping the `PropagatedMount` paths to the
correct host path.
Signed-off-by: Brian Goff <cpuguy83@gmail.com>
Upstream-commit: 0e5eaf8ee32662182147f5f62c1bfebef66f5c47
Component: engine
Setting up the mounts on the host increases chances of mount leakage and
makes for more cleanup after the plugin has stopped.
With this change all mounts for the plugin are performed by the
container runtime and automatically cleaned up when the container exits.
Signed-off-by: Brian Goff <cpuguy83@gmail.com>
Upstream-commit: a53930a04fa81b082aa78e66b342ff19cc63cc5f
Component: engine
This was added in #36047 just as a way to make sure the tree is fully
unmounted on shutdown.
For ZFS this could be a breaking change since there was no unmount before.
Someone could have setup the zfs tree themselves. It would be better, if
we really do want the cleanup to actually the unpacked layers checking
for mounts rather than a blind recursive unmount of the root.
BTRFS does not use mounts and does not need to unmount anyway.
These was only an unmount to begin with because for some reason the
btrfs tree was being moutned with `private` propagation.
For the other graphdrivers that still have a recursive unmount here...
these were already being unmounted and performing the recursive unmount
shouldn't break anything. If anyone had anything mounted at the
graphdriver location it would have been unmounted on shutdown anyway.
Signed-off-by: Brian Goff <cpuguy83@gmail.com>
Upstream-commit: 2fe4f888bee52b1f256d6fa5e20f9b061d30221c
Component: engine
The idea behind making the graphdrivers private is to prevent leaking
mounts into other namespaces.
Unfortunately this is not really what happens.
There is one case where this does work, and that is when the namespace
was created before the daemon's namespace.
However with systemd each system servie winds up with it's own mount
namespace. This causes a race betwen daemon startup and other system
services as to if the mount is actually private.
This also means there is a negative impact when other system services
are started while the daemon is running.
Basically there are too many things that the daemon does not have
control over (nor should it) to be able to protect against these kinds
of leakages. One thing is certain, setting the graphdriver roots to
private disconnects the mount ns heirarchy preventing propagation of
unmounts... new mounts are of course not propagated either, but the
behavior is racey (or just bad in the case of restarting services)... so
it's better to just be able to keep mount propagation in tact.
It also does not protect situations like `-v
/var/lib/docker:/var/lib/docker` where all mounts are recursively bound
into the container anyway.
Signed-off-by: Brian Goff <cpuguy83@gmail.com>
Upstream-commit: 9803272f2db84df7955b16c0d847ad72cdc494d1
Component: engine
This is a fix to regression in vfs graph driver introduced by
commit 7a1618ced359a3ac92 ("add quota support to VFS graphdriver").
On some filesystems, vfs fails to init with the following error:
> Error starting daemon: error initializing graphdriver: Failed to mknod
> /go/src/github.com/docker/docker/bundles/test-integration/d6bcf6de610e9/root/vfs/backingFsBlockDev:
> function not implemented
As quota is not essential for vfs, let's ignore (but log as a warning) any error
from quota init.
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
Upstream-commit: 1e8a087850aa9f96c5000a3ad90757d2e9c0499f
Component: engine
If mknod() returns ENOSYS, it most probably means quota is not supported
here, so return the appropriate error.
This is a conservative* fix to regression in vfs graph driver introduced
by commit 7a1618ced359a3ac92 ("add quota support to VFS graphdriver").
On some filesystems, vfs fails to init with the following error:
> Error starting daemon: error initializing graphdriver: Failed to mknod
> /go/src/github.com/docker/docker/bundles/test-integration/d6bcf6de610e9/root/vfs/backingFsBlockDev:
> function not implemented
Reported-by: Brian Goff <cpuguy83@gmail.com>
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
Upstream-commit: 2dd39b7841bdb9968884bbedc5db97ff77d4fe3e
Component: engine
Files that are suffixed with `_linux.go` or `_windows.go` are
already only built on Linux / Windows, so these build-tags
were redundant.
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
Upstream-commit: 6ed1163c98703f8dd0693cecbadc84d2cda811c3
Component: engine
The overlay2 driver was not setting up the archive.TarOptions field
properly like other storage backend routes to "applyTarLayer"
functionality. The InUserNS field is populated now for overlay2 using
the same query function used by the other storage drivers.
Signed-off-by: Phil Estes <estesp@linux.vnet.ibm.com>
Upstream-commit: 05b8d59015f8a5ce26c8bbaa8053b5bc7cb1a77b
Component: engine
Right now we only log source and destination (and demsg) if mount operation
fails. fstype and mount options are available easily. It probably is a good
idea to log these as well. Especially sometimes failures can happen due to
mount options.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Upstream-commit: f728d74ac5d185adaa5f1a88eadc71217806859f
Component: engine
Even though it's highly discouraged, there are existing
installs that are running overlay/overlay2 on filesystems
without d_type support.
This patch allows the daemon to start in such cases, instead of
refusing to start without an option to override.
For fresh installs, backing filesystems without d_type support
will still cause the overlay/overlay2 drivers to be marked as
"unsupported", and skipped during the automatic selection.
This feature is only to keep backward compatibility, but
will be removed at some point.
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
Upstream-commit: 0a4e793a3da9ba6d20bccfb83f7c48e20a76d895
Component: engine
Support for running overlay/overlay2 on a backing filesystem
without d_type support (most likely: xfs, as ext4 supports
this by default), was deprecated for some time.
Running without d_type support is problematic, and can
lead to difficult to debug issues ("invalid argument" errors,
or unable to remove files from the container's filesystem).
This patch turns the warning that was previously printed
into an "unsupported" error, so that the overlay/overlay2
drivers are not automatically selected when detecting supported
storage drivers.
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
Upstream-commit: 0abb8dec3f730f3ad2cc9a161c97968a6bfd0631
Component: engine
The fsmagic check was always performed on "data-root" (`/var/lib/docker`),
not on the storage-driver's home directory (e.g. `/var/lib/docker/<somedriver>`).
This caused detection to be done on the wrong filesystem in situations
where `/var/lib/docker/<somedriver>` was a mount, and a different
filesystem than `/var/lib/docker` itself.
This patch checks if the storage-driver's home directory exists, and only
falls back to `/var/lib/docker` if it doesn't exist.
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
Upstream-commit: f9c8fa305e1501d8056f8744cb193a720aab0e13
Component: engine
This commit is a set of fixes and improvement for zfs graph driver,
in particular:
1. Remove mount point after umount in `Get()` error path, as well
as in `Put()` (with `MNT_DETACH` flag). This should solve "failed
to remove root filesystem for <ID> .... dataset is busy" error
reported in Moby issue 35642.
To reproduce the issue:
- start dockerd with zfs
- docker run -d --name c1 --rm busybox top
- docker run -d --name c2 --rm busybox top
- docker stop c1
- docker rm c1
Output when the bug is present:
```
Error response from daemon: driver "zfs" failed to remove root
filesystem for XXX : exit status 1: "/sbin/zfs zfs destroy -r
scratch/docker/YYY" => cannot destroy 'scratch/docker/YYY':
dataset is busy
```
Output when the bug is fixed:
```
Error: No such container: c1
```
(as the container has been successfully autoremoved on stop)
2. Fix/improve error handling in `Get()` -- do not try to umount
if `refcount` > 0
3. Simplifies unmount in `Get()`. Specifically, remove call to
`graphdriver.Mounted()` (which checks if fs is mounted using
`statfs()` and check for fs type) and `mount.Unmount()` (which
parses `/proc/self/mountinfo`). Calling `unix.Unmount()` is
simple and sufficient.
4. Add unmounting of driver's home to `Cleanup()`.
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
Upstream-commit: a450a575a672b90f02a4b1d3d9300ce1d70a6311
Component: engine
Previously, the code would set the mtime on the directories before
creating files in the directory itself. This was problematic
because it resulted in the mtimes on the directories being
incorrectly set. This change makes it so that the mtime is
set only _after_ all of the files have been created.
Signed-off-by: Sargun Dhillon <sargun@sargun.me>
Upstream-commit: 77a2bc3e5bbc9be3fe166ed8321b7cd04e7bd097
Component: engine
There was a small issue here, where it copied the data using
traditional mechanisms, even when copy_file_range was successful.
Signed-off-by: Sargun Dhillon <sargun@sargun.me>
Upstream-commit: 0eac562281782257e6f69d58bcbc13fa889f1759
Component: engine
This change makes the VFS graphdriver use the kernel-accelerated
(copy_file_range) mechanism of copying files, which is able to
leverage reflinks.
Signed-off-by: Sargun Dhillon <sargun@sargun.me>
Upstream-commit: d2b71b26604370620630d8d3f35aba75ae474f3f
Component: engine
Previously, graphdriver/copy would improperly copy hardlinks as just regular
files. This patch changes that behaviour, and instead the code now keeps
track of inode numbers, and if it sees the same inode number again
during the copy loop, it hardlinks it, instead of copying it.
Signed-off-by: Sargun Dhillon <sargun@sargun.me>
Upstream-commit: b467f8b2ef21dc2239dcd136a29283ea6c3a0aee
Component: engine
The overlay2 storage-driver requires multiple lower dir
support for overlayFs. Support for this feature was added
in kernel 4.x, but some distros (RHEL 7.4, CentOS 7.4) ship with
an older kernel with this feature backported.
This patch adds feature-detection for multiple lower dirs,
and will perform this feature-detection on pre-4.x kernels
with overlayFS support.
With this patch applied, daemons running on a kernel
with multiple lower dir support will now select "overlay2"
as storage-driver, instead of falling back to "overlay".
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
Upstream-commit: 955c1f881ac94af19c99f0f7d5635e6a574789f2
Component: engine
This subtle bug keeps lurking in because error checking for `Mkdir()`
and `MkdirAll()` is slightly different wrt to `EEXIST`/`IsExist`:
- for `Mkdir()`, `IsExist` error should (usually) be ignored
(unless you want to make sure directory was not there before)
as it means "the destination directory was already there"
- for `MkdirAll()`, `IsExist` error should NEVER be ignored.
Mostly, this commit just removes ignoring the IsExist error, as it
should not be ignored.
Also, there are a couple of cases then IsExist is handled as
"directory already exist" which is wrong. As a result, some code
that never worked as intended is now removed.
NOTE that `idtools.MkdirAndChown()` behaves like `os.MkdirAll()`
rather than `os.Mkdir()` -- so its description is amended accordingly,
and its usage is handled as such (i.e. IsExist error is not ignored).
For more details, a quote from my runc commit 6f82d4b (July 2015):
TL;DR: check for IsExist(err) after a failed MkdirAll() is both
redundant and wrong -- so two reasons to remove it.
Quoting MkdirAll documentation:
> MkdirAll creates a directory named path, along with any necessary
> parents, and returns nil, or else returns an error. If path
> is already a directory, MkdirAll does nothing and returns nil.
This means two things:
1. If a directory to be created already exists, no error is
returned.
2. If the error returned is IsExist (EEXIST), it means there exists
a non-directory with the same name as MkdirAll need to use for
directory. Example: we want to MkdirAll("a/b"), but file "a"
(or "a/b") already exists, so MkdirAll fails.
The above is a theory, based on quoted documentation and my UNIX
knowledge.
3. In practice, though, current MkdirAll implementation [1] returns
ENOTDIR in most of cases described in #2, with the exception when
there is a race between MkdirAll and someone else creating the
last component of MkdirAll argument as a file. In this very case
MkdirAll() will indeed return EEXIST.
Because of #1, IsExist check after MkdirAll is not needed.
Because of #2 and #3, ignoring IsExist error is just plain wrong,
as directory we require is not created. It's cleaner to report
the error now.
Note this error is all over the tree, I guess due to copy-paste,
or trying to follow the same usage pattern as for Mkdir(),
or some not quite correct examples on the Internet.
[1] https://github.com/golang/go/blob/f9ed2f75/src/os/path.go
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
Upstream-commit: 516010e92d56cfcd6d1e343bdc02b6f04bc43039
Component: engine
This removes and recreates the merged dir with each umount/mount
respectively.
This is done to make the impact of leaking mountpoints have less
user-visible impact.
It's fairly easy to accidentally leak mountpoints (even if moby doesn't,
other tools on linux like 'unshare' are quite able to incidentally do
so).
As of recently, overlayfs reacts to these mounts being leaked (see
One trick to force an unmount is to remove the mounted directory and
recreate it. Devicemapper now does this, overlay can follow suit.
Signed-off-by: Euan Kemp <euan.kemp@coreos.com>
Upstream-commit: af0d589623eff9f8cefced8b527dbd7cf221ce61
Component: engine
