If device access is restricted via PrivateDevices=, let's also block the
various low-level I/O syscalls at the same time, so that we know that the
minimal set of devices in our virtualized /dev are really everything the unit
can access.
Let's make this an excercise in dogfooding: let's turn on more security
features for all our long-running services.
Specifically:
- Turn on RestrictRealtime=yes for all of them
- Turn on ProtectKernelTunables=yes and ProtectControlGroups=yes for most of
them
- Turn on RestrictAddressFamilies= for all of them, but different sets of
address families for each
Also, always order settings in the unit files, that the various sandboxing
features are close together.
Add a couple of missing, older settings for a numbre of unit files.
Note that this change turns off AF_INET/AF_INET6 from udevd, thus effectively
turning of networking from udev rule commands. Since this might break stuff
(that is already broken I'd argue) this is documented in NEWS.
Let's merge a couple of columns, to make the table a bit shorter. This
effectively just drops whitespace, not contents, but makes the currently
humungous table much much more compact.
This reworks the documentation for ReadOnlyPaths=, ReadWritePaths=,
InaccessiblePaths=. It no longer claims that we'd follow symlinks relative to
the host file system. (Which wasn't true actually, as we didn't follow symlinks
at all in the most recent releases, and we know do follow them, but relative to
RootDirectory=).
This also replaces all references to the fact that all fs namespacing options
can be undone with enough privileges and disable propagation by a single one in
the documentation of ReadOnlyPaths= and friends, and then directs the read to
this in all other places.
Moreover a hint is added to the documentation of SystemCallFilter=, suggesting
usage of ~@mount in case any of the fs namespacing related options are used.
Let's drop the reference to the cap_from_name() function in the documentation
for the capabilities setting, as it is hardly helpful. Our readers are not
necessarily C hackers knowing the semantics of cap_from_name(). Moreover, the
strings we accept are just the plain capability names as listed in
capabilities(7) hence there's really no point in confusing the user with
anything else.
This adds logic to chase symlinks for all mount points that shall be created in
a namespace environment in userspace, instead of leaving this to the kernel.
This has the advantage that we can correctly handle absolute symlinks that
shall be taken relative to a specific root directory. Moreover, we can properly
handle mounts created on symlinked files or directories as we can merge their
mounts as necessary.
(This also drops the "done" flag in the namespace logic, which was never
actually working, but was supposed to permit a partial rollback of the
namespace logic, which however is only mildly useful as it wasn't clear in
which case it would or would not be able to roll back.)
Fixes: #3867
Let's create the new namespace only after we validated and processed all
parameters, right before we start with actually mounting things.
This way, the window where we can roll back is larger (not that it matters
IRL...)
If PrivateDevices=yes is set, the namespace code creates device nodes in /dev
that should be owned by the host's root, hence let's make sure we set up the
namespace before dropping group privileges.
Let's make sure that services that use DynamicUser=1 cannot leave files in the
file system should the system accidentally have a world-writable directory
somewhere.
This effectively ensures that directories need to be whitelisted rather than
blacklisted for access when DynamicUser=1 is set.
Let's tighten our sandbox a bit more: with this change ProtectSystem= gains a
new setting "strict". If set, the entire directory tree of the system is
mounted read-only, but the API file systems /proc, /dev, /sys are excluded
(they may be managed with PrivateDevices= and ProtectKernelTunables=). Also,
/home and /root are excluded as those are left for ProtectHome= to manage.
In this mode, all "real" file systems (i.e. non-API file systems) are mounted
read-only, and specific directories may only be excluded via
ReadWriteDirectories=, thus implementing an effective whitelist instead of
blacklist of writable directories.
While we are at, also add /efi to the list of paths always affected by
ProtectSystem=. This is a follow-up for
b52a109ad3 which added /efi as alternative for
/boot. Our namespacing logic should respect that too.
Previously, if ReadWritePaths= was nested inside a ReadOnlyPaths=
specification, then we'd first recursively apply the ReadOnlyPaths= paths, and
make everything below read-only, only in order to then flip the read-only bit
again for the subdirs listed in ReadWritePaths= below it.
This is not only ugly (as for the dirs in question we first turn on the RO bit,
only to turn it off again immediately after), but also problematic in
containers, where a container manager might have marked a set of dirs read-only
and this code will undo this is ReadWritePaths= is set for any.
With this patch behaviour in this regard is altered: ReadOnlyPaths= will not be
applied to the children listed in ReadWritePaths= in the first place, so that
we do not need to turn off the RO bit for those after all.
This means that ReadWritePaths=/ReadOnlyPaths= may only be used to turn on the
RO bit, but never to turn it off again. Or to say this differently: if some
dirs are marked read-only via some external tool, then ReadWritePaths= will not
undo it.
This is not only the safer option, but also more in-line with what the man page
currently claims:
"Entries (files or directories) listed in ReadWritePaths= are
accessible from within the namespace with the same access rights as
from outside."
To implement this change bind_remount_recursive() gained a new "blacklist"
string list parameter, which when passed may contain subdirs that shall be
excluded from the read-only mounting.
A number of functions are updated to add more debug logging to make this more
digestable.
This adds a new call get_user_creds_clean(), which is just like
get_user_creds() but returns NULL in the home/shell parameters if they contain
no useful information. This code previously lived in execute.c, but by
generalizing this we can reuse it in run.c.
While an interface is still being processed by udev, it is in state "pending",
instead of "unmanaged". We must not flush device configuration then.
Further fixes commit 3104883ddc after commit c436d55397.
Fixes#4186
Replace move-to-rootlibdir calls in post-install hooks with explicitly
used ${rootlibdir} where needed.
Signed-off-by: Maciek Borzecki <maciek.borzecki@gmail.com>
Currently, systemd-sysctl command configures kernel parameters in each sysctl
configuration files in random order due to characteristics of iterator of
Hashmap.
However, kernel parameters need to be configured in the order they occur in
each sysctl configuration files.
- For example, consider fs.suid_coredump and kernel.core_pattern. If
fs.suid_coredump=2 is configured before kernel.core_pattern= whose default
value is "core", then kernel outputs the following message:
Unsafe core_pattern used with suid_dumpable=2. Pipe handler or fully qualified core dump path required.
Note that the security issue mentioned in this message has already been fixed
on recent kernels, so this is just a warning message on such kernels. But
it's still confusing to users that this message is output on some boot and
not output on another boot.
- I don't know but there could be other kernel parameters that are significant
in the order they are configured.
- The legacy sysctl command configures kernel parameters in the order they
occur in each sysctl configuration files. Although I didn't find any official
specification explaining this behavior of sysctl command, I don't think there
is any meaningful reason to change this behavior, in particular, to the
random one.
This commit does the change by simply using OrderedHashmap instead of Hashmap.
This commit is a minor tweak after the split of `--share-system`, decoupling the `--boot`
option from IPC namespacing.
Historically there has been a single `--share-system` option for sharing IPC/PID/UTS with the
host, which was incompatible with boot/pid1 mode. After the split, it is now possible to express
the requirements with better granularity.
For reference, this is a followup to #4023 which contains references to previous discussions.
I realized too late that CLONE_NEWIPC is not strictly needed for boot mode.
Since commit 5996c7c295 (v190 !), the
calculation of the HMAC is broken because the hash for a data object
including a field is done in the wrong order: the field object is
hashed before the data object is.
However during verification, the hash is done in the opposite order as
objects are scanned sequentially.
NSS modules (libnss_*.so.*) need to be installed into
${rootlibdir} (typically /lib) in order to be used. Previously, the
modules were installed into ${libdir}, thus usually ending up in
/usr/lib, even on systems where split usr is enabled, or ${libdir} is
passed explicitly.
Signed-off-by: Maciek Borzecki <maciek.borzecki@gmail.com>
If Kind is not specied, the message about "Invalid Kind" was misleading.
If Kind was specified in an invalid way, we get a message in the parsing
phase anyway. Reword the message to cover both cases better.
Also update the description of drop-ins in systemd.unit(5) to say that .d
directories, not .conf files, are in /etc/system/system, /run/systemd/system,
etc.
Lennart Poettering