RootImageOptions=/ExtensionImages=/MountImages= all support custom
mount options, but mountfsd does not support it. Add varlink
parameters to allow callers to specify mount options so that
those directives can work as expected.
Given the security implications of some mount options (e.g.: ACLs,
SELinux), require additional privileges to use custom mount options.
Pass the list of requested options to polkit details, so that the
policy can check the exact details if needed, for example to only
allow nosuid and reject the rest.
The verity user certs weren't looked up in the given --root= for
systemd-sysext which made it fail to set up extensions with a strict
image policy.
Look up verity user certs from inside the --root= when we operate on
images in it. The main use case where this matters is when the initrd
sets up the extensions for the final system and thus systemd-sysext
should do the same thing as it would do in the final system.
Let's accept O_NONBLOCK on image fds. It's usually wise to open block
device with O_NONBLOCK as it means that drivers with removable media
won't block until a medium is inserted. Otherwise it has no effect. It
also has no effect when opening regular files. Since block devices and
raw devices are the only two things we accept in mountfsd, it's hence
safe to allow it.
So far we expected the client to access the access mode themselves, and
set a restrictive 0700 server-side.
However, in some scenarios it's handy if the inode is created
server-side already with the right mode, in particular when the client
doesn't do anything with the dir just yet, and just needs it to be there
with the right mode. If it has to adjust the mode client-side it might
otherwise need to fork something off, join the selected userns, adjust
the mode, and kill the child off again.
Usually we gracefully ignore unknown configuration parameters, so that
service files can be written by upstreams and used across a variegated
range of distributions with various versions of systemd, to avoid
forcing users to the minimum common denominator and only adding settings
that are supported by the oldest distro supported.
Image policies do not behave like this, and any unknown partition or
policy designator causes the whole unit to fail to parse and a hard
error.
Change it so that parsing RootImagePolicy and friends via unit file or
D-Bus logs but otherwise ignores unknown specifiers, like other options
do.
This allows us to add new specifiers in the future, and users to adopt
them immediately.
Follow-up for d452335aa4
This is useful when we start to call mountfsd from root, for example
from the tests where we just use a simple squashfs/erofs.
Note that this requires the caller to be root, and it will be rejected
otherwise, as such images are classified as 'unprotected' and the
enforced policy does not accept them for unprivileged users.
https://github.com/systemd/systemd/pull/39168 made verity sharing
opt-in, and enabled it for system services.
Also enable it for user services for RootImage/etc, and for nspawn, for
the same reasons.
Previously, we'd take the image policy only into consideration when
dissecting the mage, but for the unlock/verity step we'd go via best
effort. Change that. This means we can now enforce policies such as
activating by root hash only even if a signature exists and similar.
Also, introduce a separate error code if we try to unlock a Verity
volume but have no root hash. Previously we'd return ENOKEY for that,
exactly like we do for encrypted volumes where we have no passparse. The
interctive unlock loop dissected_image_decrypt_interactively() is
otherwise very confused and will ask for a root hash, which makes no
sense. Hence use two distinct errors for this.
When establishing a idmapped mount for a directory with foreign mapping
we so far insisted in the dir being properly opened (i.e. via a
non-O_PATH fd) being passed to mountfsd. This is problematic however,
since the client might not actually be able to open the dir (which after
all is owned by the foreign UID, not by the user). Hence, let's relax
the rules, and accept an O_PATH fd too (which the client can get even
without privs). This should be safe, since the load-bearing security
check is whether the dir has a parent owned by the client's UID, and
for that check O_PATH or not O_PATH is not relevant.
Since c5de7b14ae
file searching implies a new mount api syscall by default,
to trigger automounts.
But, this is not necessary in most cases, e.g. when chasing
syspath in sd-device (actually this causes regression in umockdev,
see https://github.com/martinpitt/umockdev/issues/271).
Another example is reading unit files, especially .network files,
as automount may trigger mounting network filesystems...
Also, when this is used in NSS plugins, programs that load the
plugins may fail because of spuriously configured seccomp. See #38565.
Let's not trigger automount by default, and do only when explicitly
requested.
This introduces CHASE_TRIGGER_AUTOFS, and use it in
- service manager,
- bootctl and finding ESP/xbootldr,
- sysupdate,
- mountfsd,
- systemd-mount.
There may be several more places we should trigger automount, but let's
do that later.
Follow-up for c5de7b14ae.
Fixes#38565.
Replaces #38569.
Co-authored-by: Luca Boccassi <luca.boccassi@gmail.com>
This reverts commit 490aa05ca1.
As commented https://github.com/systemd/systemd/pull/38569#discussion_r2284978273,
the commit makes autofs check bypassed. Before the commit, when
CHASE_NO_AUTOFS is set, we did not shortcut chasing paths, and refused
any autofs mount points in the path. However, with the commit, the flag
was swapped but even when CHASE_AUTOFS is unset, the autofs check may be
skipped.
To fix the issue, rather than swapping the flag, we should introduce
another flag, say CHASE_TRIGGER_AUTOFS. This revert the commit, and in a
later commit, the new flag will be introduced.
Since https://github.com/systemd/systemd/commit/c5de7b14ae2e08d267d8d75bc88934ac6aa7dcd6
file searching implies a new mount api syscall by default,
to trigger automounts.
This is problematic in NSS plugins, as they are dlopen'ed inside
processes by glibc, for two reasons.
First of all, potentially searching on a networked filesystem
automount could lead to nasty surprises, such as the process
responsible for setting up the network filesystem trying to
search on that same filesystem.
More importantly, the new mount api syscall was never part of
the filesystem seccomp filter that we provide by default, and
given mounting/remounting/bind mounting is one of the possible
ways to bypass sandboxing it is very likely not allowed when
custom filters are used in sandboxed processes, if they don't
need to do these operations otherwise.
The filesystem seccomp mask we provide has been updated, however
this only takes effect on the next restart of a service. When
systemd is upgraded via a package upgrade, the new nss plugin is
installed and will be immediately dlopen'ed by glibc when needed,
without waiting for the process to restart, which means the existing
seccomp filter applies, causing the filter to trigger.
Given it's not really possible for any arbitrary program to
predict which NSS modules glibc will load, given programs do not
configure that and instead nsswitch is set up by the sysadmin,
it's impossible to handle at each process level. It's also not
possible to know when it will be triggered, given the plugin
is not linked in each binary tools like need-restart cannot
even pre-emptively restart services that may be affected.
This means in practice, upgrading from systemd << v258 to >= v258
requires a reboot to avoid either subtle or catastrophic system
failures.
By avoiding to trigger automounts in nss-systemd we can avoid
both issues.
userdb drop-ins are searched for in:
/etc/userdb/
/run/userdb/
/run/host/userdb/
/usr/local/lib/userdb/
/usr/lib/userdb/
none of which are supported as automounts anyway.
Note that this happens only when the userdbd service is not running,
as otherwise nss-systemd will go through the varlink IPC, rather than
doing the searches in-process.
So invert CHASE_NO_AUTOFS to CHASE_AUTOFS and set it in the places where
we do want to trigger automounts, like looking for the ESP.
Follow-up for https://github.com/systemd/systemd/commit/c5de7b14ae2e08d267d8d75bc88934ac6aa7dcd6
Fixes https://github.com/systemd/systemd/issues/38565
We use symbols provided by unistd.h without including it. E.g.
open(), close(), read(), write(), access(), symlink(), unlink(), rmdir(),
fsync(), syncfs(), lseek(), ftruncate(), fchown(), dup2(), pipe2(),
getuid(), getgid(), gettid(), getppid(), pipe2(), execv(), _exit(),
environ, STDIN_FILENO, STDOUT_FILENO, STDERR_FILENO, F_OK, and their
friends and variants, so on.
Currently, unistd.h is indirectly included mainly in the following two paths:
- through missing_syscall.h, which is planned to covert to .c file.
- through signal.h -> bits/sigstksz.h, which is new since glibc-2.34.
Note, signal.h is included by sd-eevent.h. So, many source files
indirectly include unistd.h if newer glibc is used.
Currently, our baseline on glibc is 2.31. We need to support glibc older
than 2.34, but unfortunately, we do not have any CI environments with
such old glibc. CIFuzz uses glibc-2.31, but it builds only fuzzers, and
many files are even not compiled.
In order to fully support unpriv containers placed in directories owned
by the foreign UID range, below some unpriv user $HOME, we need to away
to actually create these hierarchies in the first place.
Let's add a method call for that. It takes a dir fd, then validates that
its ownership matches the client's identity, and then creates a subdir,
chown()ing it to the foreign UID range. It then returns an fd to the
result.
The result could then be passed to MountDirectory() in order to get a
mount which can then be populated by some code running in a dynamic
userns.
On musl, O_ACCMODE is defined as (03|O_SEARCH), unlike glibc which
defines it as (O_RDONLY|O_WRONLY|O_RDWR). Additionally, O_SEARCH is
simply defined as O_PATH.
This causes problems for systemd on musl, as it changes the
behaviour of open_mkdir_at_full() to return -EINVAL if O_PATH is
included in flags due to the fact that O_ACCMODE includes O_SEARCH
(i.e. O_PATH). Consequently, this makes the test-fs-util test fail.
Upstream musl seems content with this behaviour and doesn't seem
interested in matching glibc's behaviour due to that defining it this
way allows for O_SEARCH to match POSIX better by allowing it to open
directories where read permission is missing. Apparently musl does some
emulation in other places to make this work more consistently as well.
Initially I took the approach of working around this by redefining
O_SEARCH as O_RDONLY if O_SEARCH == O_PATH. This fixes the test and is
the approach taken by both XZ[1] and Gzip[2][3], but was not taken as
redefining system headers potentially could be problematic.
Instead, introduce O_ACCMODE_STRICT which just is a copy of glibc's
O_ACCMODE and use it everywhere. This way we don't have to deal with
unusual definitions of O_ACCMODE from C standard libraries other than
glibc.
[1]: https://git.tukaani.org/?p=xz.git;a=blob;f=src/xz/file_io.c;h=8c83269b13fa31284f7ea5f3627a1dfbce7d6e14;hb=HEAD#l72
[2]: https://git.savannah.gnu.org/cgit/gnulib.git/tree/lib/fcntl.in.h
(lines 380 and 396, commit d7f551b30f3f2a0fa57c1b10c12f4eea41a9b89e)
[3]: https://lists.gnu.org/archive/html/bug-gzip/2025-01/msg00000.html
DDIs may contain multiple versions of the same OS, or even multiple
OSes. Hence it makes sense to not just pick the "newest", whatever that
might be, but only partitions associated with specific images, or in a
specific version.
Let's a concept for such filtering: a per-designator glob expression
that can be applied to the partition label string, and can be used for
such filtering.
Usecase: when picking UKI belonging to OS image X in version Y, make
sure we only pick a /usr/ partition belonging to X in version Y, and a
root and home partition belonging to X in any version.
This only adds the basic infrastructure, but doesn't actually expose it
anywhere.
When dissecting an image, let's make use of the Verity data even if we
got told no root hash explicitly: we can simply determine it by
concatenating the data partition uuid with the verity partition uuid.
Of course, on first thought this doesn't really add much: if the root
hash is not pinned from somewhere, this does not guarantee trust in
the image.
However, this is very useful for attestation: if we have the root hash
we can measure it before mounting things, even if we don't actually
authenticate it.
Hence, at best this helps us with attestation, at worst it doesn't improve
security but certainly doesn't hurt it.
Unprivileged namespaces are only allowed if the "setgroups" file is set
to "deny" for processes. And we need to write it before writing the
gidmap. Hence add a parameter for that.
Then, also patch all current users to actually enable this. The usecase
generally don't need it (because they don't care about unprivileged
userns), but it doesn't hurt to enable the concept anyway in all current
users (none of them actually runs complex userspace in them, but they
mostly use userns_acquire() for idmapped mounts and similar).
Let's anyway make this option explicit in the function call, to indicate
that the concept exists and is applied.
When mounting a disk image we return a bunch of mount fds referencing
the various partitions in the disk, along with some metadata about them.
One key metadata field is the "designator" which is supposed to tell
clients what is what, and where to mount it.
Let's make this more explicit: let's also include the literal relative
path where each mount shall be placed, to simplify implementations of
clients that do not care about the concept of designators.
To resolve conflict with sys/mount.h and linux/mount.h or linux/fs.h.
The conflict between sys/mount.h and linux/mount.h is resolved in
glibc-2.37 (774058d72942249f71d74e7f2b639f77184160a6), but our baseline
is still glibc-2.31. Also, even with the version or newer, still
sys/mount.h conflicts with linux/fs.h, which is included by
linux/btrfs.h.
This introduces our own implementation of sys/mount.h, that can be
simultaneously included with linux/mount.h and linux/fs.h. This also
imports linux/fs.h, linux/mount.h, and several other dependent headers.
The introduced sys/mount.h header itself may not be enough simple, but
by using the header, we can drop most of workarounds in other source files.
systemd-mountfsd so far provided a MountImage() API call for mounting a
disk image and returning a set of mount fds. This complements the API
with a new MountDirectory() API call, that operates on a directory
instead of an image file. Now, what makes this interesting is that it
applies an idmapping from the foreign UID range to the provided target
userns – and in which case unpriveleged operation is allowed (well,
under some conditions: in particular the client must own a parent dir of
the provided path).
This allows container managers to run fully unprivileged from
directories – as long as those directories are owned by the foreign UID
range. Basic operation is like this:
1. acquire a transient userns from systemd-nsresourced with 64K users
2. ask systemd-mountfsd for an idmapped mount of the container dir
matching that userns
3. join the userns and bind the mount fd as root.
Note that we have to drop various sandboxing knobs from the mountfsd
service file for this to work, since the kernel's security checks that
try to ensure than an obstructed /proc/ cannot be circumvented via
mounting a new procfs will otherwise prohibit mountfsd to duplicate the
mounts properly.
…ith fd passing enabled
Let's add a simple flag that enables fd passing for all connections of a
server. It's much easier to use this than to install a connect handler
which manually enables this for each connection.
- Make fd_is_namespace() take NamespaceType
- Drop support for kernel without NS_GET_NSTYPE (< 4.11)
- Port is_our_namespace() to namespace_open_by_type()
(preparation for later commits, where the latter
would go by pidfd if available, avoiding procfs)
The values that were previously hardcoded in sd-varlink.c are now defined
in new varlink_set_info_systemd() and that function is called everywhere
where we create a server.
It's time. sd-json was already done earlier in this cycle, let's now
make sd-varlink public too.
This is mostly just a search/replace job of epical proportions.
I left some functions internal (mostly IDL handling), and I turned some
static inline calls into regular calls.
This is preparation for making our Varlink API a public API. Since our
Varlink API is built on top of our JSON API we need to make that public
first (it's a nice API, but JSON APIs there are already enough, this is
purely about the Varlink angle).
I made most of the json.h APIs public, and just placed them in
sd-json.h. Sometimes I wasn't so sure however, since the underlying data
structures would have to be made public too. If in doubt I didn#t risk
it, and moved the relevant API to src/libsystemd/sd-json/json-util.h
instead (without any sd_* symbol prefixes).
This is mostly a giant search/replace patch.
Luca Boccassi
Kai Lueke
Lennart Poettering
Luca Boccassi
Jelle van der Waa
Yu WatanabeandLuca Boccassi
Daan De Meyer
Stefan Hansson
Mike Yuan
Zbigniew Jędrzejewski-Szmek