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.
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.
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.
In https://github.com/systemd/systemd/pull/4004 , a runtime detection
method for seccomp was added. However, it does not detect the case
where CONFIG_SECCOMP=y but CONFIG_SECCOMP_FILTER=n. This is possible
if the architecture does not support filtering yet.
Add a check for that case too.
While at it, change get_proc_field usage to use PR_GET_SECCOMP prctl,
as that should save a few system calls and (unnecessary) allocations.
Previously, reading of /proc/self/stat was done as recommended by
prctl(2) as safer. However, given that we need to do the prctl call
anyway, lets skip opening, reading and parsing the file.
Code for checking inspired by
https://outflux.net/teach-seccomp/autodetect.html
dbus-daemon does NSS name look-ups in order to enforce its bus policy. This
might dead-lock if an NSS module use wants to use D-Bus for the look-up itself,
like our nss-systemd does. Let's work around this by bypassing bus
communication in the NSS module if we run inside of dbus-daemon. To make this
work we keep a bit of extra state in /run/systemd/dynamic-uid/ so that we don't
have to consult the bus, but can still resolve the names.
Note that the normal codepath continues to be via the bus, so that resolving
works from all mount namespaces and is subject to authentication, as before.
This is a bit dirty, but not too dirty, as dbus daemon is kinda special anyway
for PID 1.
This adds the boolean RemoveIPC= setting to service, socket, mount and swap
units (i.e. all unit types that may invoke processes). if turned on, and the
unit's user/group is not root, all IPC objects of the user/group are removed
when the service is shut down. The life-cycle of the IPC objects is hence bound
to the unit life-cycle.
This is particularly relevant for units with dynamic users, as it is essential
that no objects owned by the dynamic users survive the service exiting. In
fact, this patch adds code to imply RemoveIPC= if DynamicUser= is set.
In order to communicate the UID/GID of an executed process back to PID 1 this
adds a new "user lookup" socket pair, that is inherited into the forked
processes, and closed before the exec(). This is needed since we cannot do NSS
from PID 1 due to deadlock risks, However need to know the used UID/GID in
order to clean up IPC owned by it if the unit shuts down.
The ExecParameters structure contains a number of bit-flags, that were so far
exposed as bool:1, change this to a proper, single binary bit flag field. This
makes things a bit more expressive, and is helpful as we add more flags, since
these booleans are passed around in various callers, for example
service_spawn(), whose signature can be made much shorter now.
Not all bit booleans from ExecParameters are moved into the flags field for
now, but this can be added later.
This setting adds minimal user namespacing support to a service. When set the invoked
processes will run in their own user namespace. Only a trivial mapping will be
set up: the root user/group is mapped to root, and the user/group of the
service will be mapped to itself, everything else is mapped to nobody.
If this setting is used the service runs with no capabilities on the host, but
configurable capabilities within the service.
This setting is particularly useful in conjunction with RootDirectory= as the
need to synchronize /etc/passwd and /etc/group between the host and the service
OS tree is reduced, as only three UID/GIDs need to match: root, nobody and the
user of the service itself. But even outside the RootDirectory= case this
setting is useful to substantially reduce the attack surface of a service.
Example command to test this:
systemd-run -p PrivateUsers=1 -p User=foobar -t /bin/sh
This runs a shell as user "foobar". When typing "ps" only processes owned by
"root", by "foobar", and by "nobody" should be visible.
This way, invoking nspawn from a shell in the best case inherits the TERM
setting all the way down into the login shell spawned in the container.
Fixes: #3697
This adds a new boolean setting DynamicUser= to service files. If set, a new
user will be allocated dynamically when the unit is started, and released when
it is stopped. The user ID is allocated from the range 61184..65519. The user
will not be added to /etc/passwd (but an NSS module to be added later should
make it show up in getent passwd).
For now, care should be taken that the service writes no files to disk, since
this might result in files owned by UIDs that might get assigned dynamically to
a different service later on. Later patches will tighten sandboxing in order to
ensure that this cannot happen, except for a few selected directories.
A simple way to test this is:
systemd-run -p DynamicUser=1 /bin/sleep 99999
All other functions in execute.c that need the unit id take a Unit* parameter
as first argument. Let's change connect_logger_as() to follow a similar logic.
This patch renames Read{Write,Only}Directories= and InaccessibleDirectories=
to Read{Write,Only}Paths= and InaccessiblePaths=, previous names are kept
as aliases but they are not advertised in the documentation.
Renamed variables:
`read_write_dirs` --> `read_write_paths`
`read_only_dirs` --> `read_only_paths`
`inaccessible_dirs` --> `inaccessible_paths`
By cleaning up before setting up PAM we maintain control of overriding
behavior in setting variables. Otherwise, pam_putenv is in control.
This also makes sure we use a cleaned up environment in replacing
variables in argv.
This permits services to detect whether their stdout/stderr is connected to the
journal, and if so talk to the journal directly, thus permitting carrying of
metadata.
As requested by the gtk folks: #2473
Move the merger of environment variables before setting up the PAM
session and pass the aggregate environment to PAM setup. This allows
control over the PAM session hooks through environment variables.
PAM session initiation may update the environment. On successful
initiation of a PAM session, we adopt the environment of the
PAM context.
This patch implements the new magic character '!'. By putting '!' in front
of a command, systemd executes it with full privileges ignoring paramters
such as User, Group, SupplementaryGroups, CapabilityBoundingSet,
AmbientCapabilities, SecureBits, SystemCallFilter, SELinuxContext,
AppArmorProfile, SmackProcessLabel, and RestrictAddressFamilies.
Fixes partially https://github.com/systemd/systemd/issues/3414
Related to https://github.com/coreos/rkt/issues/2482
Testing:
1. Create a user 'bob'
2. Create the unit file /etc/systemd/system/exec-perm.service
(You can use the example below)
3. sudo systemctl start ext-perm.service
4. Verify that the commands starting with '!' were not executed as bob,
4.1 Looking to the output of ls -l /tmp/exec-perm
4.2 Each file contains the result of the id command.
`````````````````````````````````````````````````````````````````
[Unit]
Description=ext-perm
[Service]
Type=oneshot
TimeoutStartSec=0
User=bob
ExecStartPre=!/usr/bin/sh -c "/usr/bin/rm /tmp/exec-perm*" ;
/usr/bin/sh -c "/usr/bin/id > /tmp/exec-perm-start-pre"
ExecStart=/usr/bin/sh -c "/usr/bin/id > /tmp/exec-perm-start" ;
!/usr/bin/sh -c "/usr/bin/id > /tmp/exec-perm-star-2"
ExecStartPost=/usr/bin/sh -c "/usr/bin/id > /tmp/exec-perm-start-post"
ExecReload=/usr/bin/sh -c "/usr/bin/id > /tmp/exec-perm-reload"
ExecStop=!/usr/bin/sh -c "/usr/bin/id > /tmp/exec-perm-stop"
ExecStopPost=/usr/bin/sh -c "/usr/bin/id > /tmp/exec-perm-stop-post"
[Install]
WantedBy=multi-user.target]
`````````````````````````````````````````````````````````````````
Let's add an extra safety check before we chmod/chown a TTY to the right user,
as we might end up having connected something to STDIN/STDOUT that is actually
not a TTY, even though this might have been requested, due to permissive
StandardInput= settings or transient service activation with fds passed in.
Fixes:
https://bugs.freedesktop.org/show_bug.cgi?id=85255
New exec boolean MemoryDenyWriteExecute, when set, installs
a seccomp filter to reject mmap(2) with PAGE_WRITE|PAGE_EXEC
and mprotect(2) with PAGE_EXEC.
The macro determines the right length of a AF_UNIX "struct sockaddr_un" to pass to
connect() or bind(). It automatically figures out if the socket refers to an
abstract namespace socket, or a socket in the file system, and properly handles
the full length of the path field.
This macro is not only safer, but also simpler to use, than the usual
offsetof() + strlen() logic.
The manpage of seccomp specify that using seccomp with
SECCOMP_SET_MODE_FILTER will return EACCES if the caller do not have
CAP_SYS_ADMIN set, or if the no_new_privileges bit is not set. Hence,
without NoNewPrivilege set, it is impossible to use a SystemCall*
directive with a User directive set in system mode.
Now, NoNewPrivileges is set if we are in user mode, or if we are in
system mode and we don't have CAP_SYS_ADMIN, and SystemCall*
directives are used.
Throughout the tree there's spurious use of spaces separating ++ and --
operators from their respective operands. Make ++ and -- operator
consistent with the majority of existing uses; discard the spaces.
The setting is hardly useful (since its effect is generally reduced to zero due
to file system caps), and with the advent of ambient caps an actually useful
replacement exists, hence let's get rid of this.
I am pretty sure this was unused and our man page already recommended against
its use, hence this should be a safe thing to remove.
Assign errno-style errors to a variable called "r" when they happen, the same way we do this in most other calls. It's
bad enough that the error handling part of the function deals with two different error variables (pam_code and r) now,
but before this fix it was even three!
gcc is confused by the common idiom of
return errno ? -errno : -ESOMETHING
and thinks a positive value may be returned. Replace this condition
with errno > 0 to help gcc and avoid many spurious warnings. I filed
a gcc rfe a long time ago, but it hard to say if it will ever be
implemented [1].
Both conventions were used in the codebase, this change makes things
more consistent. This is a follow up to bcb161b023.
[1] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61846
