We have a number of components these days that are split into multiple
binaries, i.e. a primary one, and a worker callout usually. These
binaries are closely related, they typically speak a protocol that is
internal, and not safe to mix and match. Examples for this:
- homed and its worker binary homework
- userdbd and its worker binary userwork
- import and the various pull/import/export handlers
- sysupdate the same
- the service manager and the executor binary
Running any of these daemons directly from the meson build tree is
messy, since the implementations will typically invoke the installed
callout binaries, not the ones from the build tree. This is very
annoying, and not obvious at first.
Now, we could always invoke relevant binaries from $(dirname
/proc/self/exe) first, before using the OS installed ones. But that's
typically not what is desired, because this means in the installed case
(i.e. the usual one) we'll look for these callout binaries at a place
they typically will not be found (because these callouts generally are
located in libexecdir, not bindir when installed).
Hence, let's try to do things a bit smarter, and follow what build
systems such as meson have already been doing to make sure dynamic
library discovery works correctly when binaries are run from a build
directory: let's start looking at rpath/runpath in the main binary that
is executed: if there's an rpath/runpath set, then we'll look for the
callout binaries next to the main binary, otherwise we won't. This
should generally be the right thing to do as meson strips the rpath
during installation, and thus we'll look for the callouts in the build
dir if in build dir mode, and in the OS otherwise.
Rather than adding more and more parameters to ask_password_auto(), let's
pass a structure of the fields that often are constant anyway.
This way, callers can fill in what they need, and we take the filled
structure which we can pass around internally as one.
This is in particular preparation for adding one more field in one of
the next commits.
The varlink_collect_full function did not set varlink client's state
when the reply was an error. The state was stuck in "collecting-reply".
I discovered that while hacking on network varlink interface (adding a
new varlink method). The debug logs shows the process of performing the
first query which replies with an error:
varlink: Setting state idle-client
network: Sending message: {"method":"io.systemd.Network.LLDPNeighbors","parameters":{"ifindex":1},"more":true}
network: Changing state idle-client → collecting
network: Received message: {"error":"org.varlink.service.MethodNotFound","parameters":{"method":"io.systemd.Network.LLDPNeighbors"}}
network: Changing state collecting → collecting-reply
Now another varlink_collect call is being made, but
network: Connection busy.
Failed to execute varlink call: Device or resource busy
This was not caught by the tests because there were no varlink_collect
calls that resulted in error reply.
This refactors the Unit structure a bit: all cgroup-related state fields
are moved to a new structure CGroupRuntime, which is only allocated as
we realize a cgroup.
This is both a nice cleanup and should make unit structures considerably
smaller that have no cgroup associated, because never realized or
because they belong to a unit type that doesn#t have cgroups anyway.
This makes things nicely symmetric:
ExecContext → static user configuration about execution
ExecRuntime → dynamic user state of execution
CGroupContext → static user configuration about cgroups
CGroupRuntime → dynamic user state of cgroups
And each time the XyzContext is part of the unit type structures such as
Service or Slice that need it, but the runtime object is only allocated
when a unit is started.
From readlinkat(2):
Since Linux 2.6.39, pathname can be an empty string, in which case the
call operates on the symbolic link referred to by dirfd (which should
have been obtained using open(2) with the O_PATH and O_NOFOLLOW flags).
Currently, hashmap_dump_sorted sorts by key and then returns the values
in order sorted by key. This commit adds another helper that does the
same but returns the sorted keys instead
ADD_SAFE/SUB_SAFE/MUL_SAFE do addition/subtraction/multiplication
respectively with an overflow check. If an overflow occurs these return
false, otherwise true. Example: (c = a + b) would become ADD_SAFE(&c, a,
b)
INC_SAFE/DEC_SAFE/MUL_ASSIGN_SAFE are like above but they also reassign
the first argument. Example: (a += b) would become INC_SAFE(&a, b)
This reworks varlink_collect() so that it is not just a wrapper around
varlink_observe(), varlink_bind_reply() and others. It becomes a first
class operation.
This has various benefits:
1. Memory management is normalized: the reply json variant is now
tracked as part of the varlink object, and thus we do not pass
ownership to the caller. This is just like we do it for simple method
calls and removes a lot of confusion.
2. The bind reply/user data pointer can be used for user stuff, we'll
not silently override this.
3. We enforce an overall time-out operation on the whole thing, so that
this synchronous operation does no longer block forever.
Since 2e106312e2 the test unit fails with 'resources' result instead of
'exit-code', which the test didn't account for when running unprivileged.
Before 2e106312e2:
$ /root/systemd/build/test-path
Failed to start transient scope unit: Interactive authentication required.
Couldn't allocate a scope unit for this test, proceeding without.
...
-.slice: Failed to enable/disable controllers on cgroup /user.slice/user-1000.slice/session-1.scope, ignoring: Permission denied
app.slice: Failed to create cgroup /user.slice/user-1000.slice/session-1.scope/app.slice: Permission denied
-.slice: Failed to enable/disable controllers on cgroup /user.slice/user-1000.slice/session-1.scope, ignoring: Permission denied
app.slice: Failed to create cgroup /user.slice/user-1000.slice/session-1.scope/app.slice: Permission denied
...
line 151: path-exists.path: state = running; result = success (left: 29986250)
line 151: path-exists.service: state = start; result = success
path-exists.service: Main process exited, code=exited, status=219/CGROUP
path-exists.service: Failed with result 'exit-code'.
line 151: path-exists.path: state = running; result = success (left: 29985948)
line 151: path-exists.service: state = failed; result = exit-code
Failed to start service path-exists.service, aborting test: failed/exit-code
After 2e106312e2:
$ /root/systemd/build/test-path
Failed to start transient scope unit: Interactive authentication required.
Couldn't allocate a scope unit for this test, proceeding without.
...
-.slice: Failed to enable/disable controllers on cgroup /user.slice/user-1000.slice/session-1.scope, ignoring: Permission denied
app.slice: Failed to create cgroup /user.slice/user-1000.slice/session-1.scope/app.slice: Permission denied
-.slice: Failed to enable/disable controllers on cgroup /user.slice/user-1000.slice/session-1.scope, ignoring: Permission denied
app.slice: Failed to create cgroup /user.slice/user-1000.slice/session-1.scope/app.slice: Permission denied
path-exists.service: Failed to spawn executor: No such file or directory
path-exists.service: Failed to spawn 'start' task: No such file or directory
path-exists.service: Failed with result 'resources'.
The personality() syscall returns a 32-bit value where the top three
bytes are reserved for flags that emulate historical or architectural
quirks, and only the least significant byte reflects the actual
personality we're interested in (in opinionated_personality()).
Use the newly defined mask in the corresponding test as well, otherwise
the test fails on some more "exotic" architectures that set some of the
"quirk" flags:
~# uname -m
armv7l
~# build/test-seccomp
...
/* test_lock_personality */
current personality=0x0
safe_personality(PERSONALITY_INVALID)=0x800000
Assertion '(unsigned long) safe_personality(current) == current' failed at src/test/test-seccomp.c:970, function test_lock_personality(). Aborting.
lockpersonalityseccomp terminated by signal ABRT.
Assertion 'wait_for_terminate_and_check("lockpersonalityseccomp", pid, WAIT_LOG) == EXIT_SUCCESS' failed at src/test/test-seccomp.c:996, function test_lock_personality(). Aborting.
Aborted (core dumped)
See: personality(2) and comments in sys/personality.h
In various contexts it's a bit icky to allow paths below /proc/, /sys/,
/dev/ i.e. file hierarchies where API VFS are placed. Let's add a new
flag for path_simplify_and_warn() to check for this and refuse a path if
in these paths.
Enable this when parsing WorkingDirectory=.
This is inspired by CVE-2024-21626, which uses trickery around the cwd
and /proc/self/fd/.
AFAICS we are not actually vulnerable to the same issue as explained in
the CVE since we execute the WorkingDirectory= setting very late, i.e.
long after we set up the new mount namespace. But let's filter out icky
stuff better earlier than later, as extra safety precaution.
So far credentials are a concept for system services only: to encrypt or
decrypt credential you must be privileged, as only then you can access
the TPM and the host key.
Let's break this up a bit: let's add a "user-scoped" credential, that
are specific to users. Internally this works by adding another step to
the acquisition of the symmetric encryption key for the credential: if a
"user-scoped" credential is used we'll generate an symmetric encryption
key K as usual, but then we'll use it to calculate
K' = HMAC(K, flags || uid || machine-id || username)
and then use the resulting K' as encryption key instead. This basically
includes the (public) user's identity in the encryption key, ensuring
that only if the right user credentials are specified the correct key
can be acquired.
This makes rtnl_resolve_interface() always check the existence of the
resolved interface, which previously did not when a decimal formatted
ifindex is provided, e.g. "1" or "42".
Some tests have hard-coded results that need to match, and change if
the caller has a timezone set via the TZ= environment variable, as it
is the case during reproducible build tests. Unset it.
Let's get rid of the confusing goto so that the flow is more
straightforward. Note that the behavior is slightly changed:
previously, ret_filtered would be an empty string even if
the original opts passed in is NULL, but after this commit
it returns NULL too. But this shouldn't matter, as all our
code handles NULL opts gracefully.
This file is a bit misnamed. What it actually implements is one specific
BPF LSM module, that restricts file systems. As such it really should be
named after that, and not primarily by the mechanism it uses for that.
With this our glue code is now named the same way as the actual bpf code
files in src/core/bpf/, thus things become a bit more symmetric.
This is particular relevant as we'll soon have another BPF LSM in our
tree, see #26826, and we should be able to distinguish them by name.
This commit just renames the files and does some dumb search/replace of
the string. A follow-up commit will name some functions more expressively
inside the files.
so that /usr/lib/systemd/tests/unit-tests/test-libsystemd-sym
builds in a reproducible way
in spite of non-deterministic filesystem readdir order
See https://reproducible-builds.org/ for why this is good.
This patch was done while working on reproducible builds for openSUSE.
With newer versions of AppArmor, unprivileged user namespace creation
may be restricted by default, in which case user manager instances will
not be able to apply PrivateUsers=yes (or the settings which require it).
Additionally, if a kernel has the kernel.unprivileged_userns_clone
sysctl patch, and that sysctl is 0, then unprivileged userns creation
will always fail.
If a test unit is going to be run in a user manager, and that unit
requires PrivateUsers=yes (explicitly or implicitly), then skip it if
we do not have user namespace privileges.
Let's make things nice for desktops, and provide a mime type for
credential files.
This uses the 128bit header identifier that our credential files start
with. However, the files are always base64 encoded, hence we have to
match the base64 string, hence add a small test case that generates them
properly for us, and truncates them at the right place (since 128 is not
evently divisable by 6).
Lennart Poettering
Luca Boccassi
Tomáš Pecka
Yu Watanabe
Adrian Vovk
Frantisek Sumsal
Mike Yuan
Daan De Meyer
Bernhard M. Wiedemann
Nick Rosbrook