We use our cgroup APIs in various contexts, including from our libraries
sd-login, sd-bus. As we don#t control those environments we can't rely
that the unified cgroup setup logic succeeds, and hence really shouldn't
assert on it.
This more or less reverts 415fc41cea.
Currently the hybrid mode mounts cgroup v2 on /sys/fs/cgroup instead of the v1
name=systemd hierarchy. While this works fine for systemd itself, it breaks
tools which expect cgroup v1 hierarchy on /sys/fs/cgroup/systemd.
This patch updates the hybrid mode so that it mounts v2 hierarchy on
/sys/fs/cgroup/unified and keeps v1 "name=systemd" hierarchy on
/sys/fs/cgroup/systemd for compatibility. systemd itself doesn't depend on the
"name=systemd" hierarchy at all. All operations take place on the v2 hierarchy
as before but the v1 hierarchy is kept in sync so that any tools which expect
it to be there can keep doing so. This allows systemd to take advantage of
cgroup v2 process management without requiring other tools to be aware of the
hybrid mode.
The hybrid mode is implemented by mapping the special systemd controller to
/sys/fs/cgroup/unified and making the basic cgroup utility operations -
cg_attach(), cg_create(), cg_rmdir() and cg_trim() - also operate on the
/sys/fs/cgroup/systemd hierarchy whenever the cgroup2 hierarchy is updated.
While a bit messy, this will allow dropping complications from using cgroup v1
for process management a lot sooner than otherwise possible which should make
it a net gain in terms of maintainability.
v2: Fixed !cgns breakage reported by @evverx and renamed the unified mount
point to /sys/fs/cgroup/unified as suggested by @brauner.
v3: chown the compat hierarchy too on delegation. Suggested by @evverx.
v4: [zj]
- drop the change to default, full "legacy" is still the default.
cg_[all_]unified() test whether a specific controller or all controllers are on
the unified hierarchy. While what's being asked is a simple binary question,
the callers must assume that the functions may fail any time, which
unnecessarily complicates their usages. This complication is unnecessary.
Internally, the test result is cached anyway and there are only a few places
where the test actually needs to be performed.
This patch simplifies cg_[all_]unified().
* cg_[all_]unified() are updated to return bool. If the result can't be
decided, assertion failure is triggered. Error handlings from their callers
are dropped.
* cg_unified_flush() is updated to calculate the new result synchrnously and
return whether it succeeded or not. Places which need to flush the test
result are updated to test for failure. This ensures that all the following
cg_[all_]unified() tests succeed.
* Places which expected possible cg_[all_]unified() failures are updated to
call and test cg_unified_flush() before calling cg_[all_]unified(). This
includes functions used while setting up mounts during boot and
manager_setup_cgroup().
cgroup mode detection is broken in two different ways.
* detect_unified_cgroup_hierarchy() is called too nested in outer_child().
sync_cgroup() which is used by run() also needs to know the requested cgroup
mode but it's currently always getting CGROUP_UNIFIED_UNKNOWN. This makes it
skip syncing the inner cgroup hierarchy on some config combinations.
$ cat /proc/self/cgroup | grep systemd
1:name=systemd:/user.slice/user-0.slice/session-c1.scope
$ UNIFIED_CGROUP_HIERARCHY=0 SYSTEMD_NSPAWN_USE_CGNS=0 systemd-nspawn -M container
...
[root@container ~]# cat /proc/self/cgroup | grep systemd
1:name=systemd:/machine.slice/machine-container.x86_64.scope
$ exit
$ UNIFIED_CGROUP_HIERARCHY=1 SYSTEMD_NSPAWN_USE_CGNS=0 systemd-nspawn -M container
[root@container ~]# cat /proc/self/cgroup | grep 0::
0::/
$ exit
Note how the unified hierarchy case's path is not synchronized with the host.
This for example can cause issues when there are multiple such containers.
Fixed by moving detect_unified_cgroup_hierarchy() invocation to main().
* inner_child() was invoking cg_unified_flush(). inner_child() executes fully
scoped and can't determine which cgroup mode the host was in. It doesn't
make sense to keep flushing the detected mode when the host mode can't
change.
Fixed by replacing cg_unified_flush() invocations in outer_child() and
inner_child() with one in main().
Let's check D-Bus instead of files in /run to see if resolved is
running. This is a bit nicer as bus names are automatically cleaned up
when resolved dies, which is not the case for files in /run.
See: #4649
This adds a unified "copy_flags" parameter to all copy_xyz() function
calls, replacing the various boolean flags so far used. This should make
many invocations more readable as it is clear what behaviour is
precisely requested. This also prepares ground for adding support for
more modes later on.
Add a new --pivot-root argument to systemd-nspawn, which specifies a
directory to pivot to / inside the container; while the original / is
pivoted to another specified directory (if provided). This adds
support for booting container images which may contain several bootable
sysroots, as is common with OSTree disk images. When these disk images
are booted on real hardware, ostree-prepare-root is run in conjunction
with sysroot.mount in the initramfs to achieve the same results.
This makes nspawn's logic of automatically discovering the root hash of
an image file generic, and then reuses it in systemd-dissect and in
PID1's RootImage= logic, so that verity is automatically set up whenever
we can.
There's no point in updating exec_target for each binary we try to
execute, if we override it right-away anyway... Let's just do this once,
and include all binaries we try each time.
Follow-up for 1a68e1e543.
The failure message is typically currently:
execv() failed: No such file or directory
which is not very useful because it doesn’t tell you which file or
directory it was trying to exec.
gcc 7 adds -Wimplicit-fallthrough=3 to -Wextra. There are a few ways
we could deal with that. After we take into account the need to stay compatible
with older versions of the compiler (and other compilers), I don't think adding
__attribute__((fallthrough)), even as a macro, is worth the trouble. It sticks
out too much, a comment is just as good. But gcc has some very specific
requiremnts how the comment should look. Adjust it the specific form that it
likes. I don't think the extra stuff we had in those comments was adding much
value.
(Note: the documentation seems to be wrong, and seems to describe a different
pattern from the one that is actually used. I guess either the docs or the code
will have to change before gcc 7 is finalized.)
This substantially reworks the seccomp code, to ensure better
compatibility with some architectures, including i386.
So far we relied on libseccomp's internal handling of the multiple
syscall ABIs supported on Linux. This is problematic however, as it does
not define clear semantics if an ABI is not able to support specific
seccomp rules we install.
This rework hence changes a couple of things:
- We no longer use seccomp_rule_add(), but only
seccomp_rule_add_exact(), and fail the installation of a filter if the
architecture doesn't support it.
- We no longer rely on adding multiple syscall architectures to a single filter,
but instead install a separate filter for each syscall architecture
supported. This way, we can install a strict filter for x86-64, while
permitting a less strict filter for i386.
- All high-level filter additions are now moved from execute.c to
seccomp-util.c, so that we can test them independently of the service
execution logic.
- Tests have been added for all types of our seccomp filters.
- SystemCallFilters= and SystemCallArchitectures= are now implemented in
independent filters and installation logic, as they semantically are
very much independent of each other.
Fixes: #4575
gperf-3.1 generates lookup functions that take a size_t length
parameter instead of unsigned int. Test for this at configure time.
Fixes: https://github.com/systemd/systemd/issues/5039
Handle properly if /etc is a symlink (i.e. make sure we don't follow the
symlink outside the image). Also follow /etc/resolv.conf if it is a
symlink, and use the resolved path when creating a mount point and
mounting (as both of these operations follow symlinks and rally
shouldn't).
Handle more types of read-only errors as debug-level issues.
In preparation for reusing the image dissector in the GPT auto-discovery
logic, only optionally fail the dissection when we can't identify a root
partition.
In the GPT auto-discovery we are completely fine with any kind of root,
given that we run when it is already mounted and all we do is find some
additional auxiliary partitions on the same disk.
This was broken by 19caffac75 which remounted the
root directory to MS_SHARED before applying the volatile mount logic. This
broke things as MS_MOVE is incompatible with MS_SHARED directory trees, and we
need MS_MOVE in the volatile mount logic to rearrange the directory tree.
Simply swap the order here, apply the volatile logic before we switch to
MS_SHARED.
This is useful for reusing the dissector logic in the gpt-auto-discovery logic:
there we really don't want to use MBR or naked file systems as root device.
This moves the VolatileMode enum and its helper functions to src/shared/. This
is useful to then reuse them to implement systemd.volatile= in a later commit.
The container detection code in virt.c we ship checks for /proc/1/environ,
looking for "container=" in it. Let's make sure our "-a" init stub exposes that
correctly.
Without this "systemd-detect-virt" run in a "-a" container won't detect that it
is being run in a container.
When getting SIGCHLD we should not assume that it was the first
child forked from system-nspawn that has died as it may also be coming
from an orphan process. This change adds a signal handler that ignores
SIGCHLD unless it came from the first containerized child - the real
child.
Before this change the problem can be reproduced as follows:
$ sudo systemd-nspawn --directory=/container-root --share-system
Press ^] three times within 1s to kill container.
[root@andreyu-coreos ~]# { true & } &
[1] 22201
[root@andreyu-coreos ~]#
Container root-fedora-latest terminated by signal KILL
This might happen that resolv.conf is missing in a minimal rootfs and in this
case the following warning is emitted:
Failed to mount n/a on /mnt/etc/resolv.conf (MS_BIND ""): No such file or directory
This patch fixes this case.
This adds support for discovering and making use of properly tagged dm-verity
data integrity partitions. This extends both systemd-nspawn and systemd-dissect
with a new --root-hash= switch that takes the root hash to use for the root
partition, and is otherwise fully automatic.
Verity partitions are discovered automatically by GPT table type UUIDs, as
listed in
https://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/
(which I updated prior to this change, to include new UUIDs for this purpose.
mkosi with https://github.com/systemd/mkosi/pull/39 applied may generate images
that carry the necessary integrity data. With that PR and this commit, the
following simply lines suffice to boot up an integrity-protected container image:
```
# mkdir test
# cd test
# mkosi --verity
# systemd-nspawn -i ./image.raw -bn
```
Note that mkosi writes the image file to "image.raw" next to a a file
"image.roothash" that contains the root hash. systemd-nspawn will look for that
file and use it if it exists, in case --root-hash= is not specified explicitly.
This adds support to the image dissector to deal with encrypted images (only
LUKS). Given that we now have a neatly isolated image dissector codebase, let's
add a new feature to it: support for automatically dealing with encrypted
images. This is then exposed in systemd-dissect and nspawn.
It's pretty basic: only support for passphrase-based encryption.
In order to ensure that "systemd-dissect --mount" results in mount points whose
backing LUKS DM devices are cleaned up automatically we use the DM_DEV_REMOVE
ioctl() directly on the device (in DM_DEFERRED_REMOVE mode). libgcryptsetup at
the moment doesn't provide a proper API for this. Thankfully, the ioctl() API
is pretty easy to use.
Let's make use of the new internal API. This mostly doesn't change anything for
the caller, however, "systemd-nspawn --image=/dev/sda7" works now as the new
code can handle disk images with no partition tables, and make any detected
images directly the root.
This extends the --bind= and --overlay= syntax so that an empty string as source/upper
directory is taken as request to automatically allocate a temporary directory
below /var/tmp, whose lifetime is bound to the nspawn runtime. In combination
with the "+" path extension this permits a switch "--overlay=+/var::/var" in
order to use the container's shipped /var, combine it with a writable temporary
directory and mount it to the runtime /var of the container.
If a source path is prefixed with "+" it is taken relative to the container's
root directory instead of the host. This permits easily establishing bind and
overlay mounts based on data from the container rather than the host.
This also reworks custom_mounts_prepare(), and turns it into two functions: one
custom_mount_check_all() that remains in nspawn.c but purely verifies the
validity of the custom mounts configured. And one called
custom_mount_prepare_all() that actually does the preparation step, sorts the
custom mounts, resolves relative paths, and allocates temporary directories as
necessary.
