See the [design
doc](https://github.com/systemd/systemd/issues/38023#issue-3192567450)
by @ikruglov
This PR introduces the metrics API framework, adds some basic system
wide/per unit/service metrics, and a basic CLI. The PR is broken into
two commits as described below.
### Deviations from the original design
- Introduced top level field "object" for ease of filtering. Instead of
having `fields: { unit: "foo", unit_type: "service" }`, we now have
`object: foo.service` as the top level field.
### First commit
The first commit includes:
- Metrics API definitions
- Code to set up the varlink server
- The describe method which shows all the metrics families
- The list method which lists all the metrics
- Type definitions related to MetricFamily
- Common code to build json objects
### Second commit
The second commit adds some basic metrics, a basic CLI (systemd-report)
which
lists the metrics, and integration tests.
**System wide metrics:**
- units_by_type_total
- units_by_state_total
**Two per unit metrics:**
- unit_active_state
- unit_load_state
**A service state metric:**
- nrestarts
### Sample outputs
**units_by_type_total**:
```
{
"name" : "io.systemd.Manager.units_by_type_total",
"value" : 52,
"fields" : {
"type" : "target"
}
}
{
"name" : "io.systemd.Manager.units_by_type_total",
"value" : 82,
"fields" : {
"type" : "device"
}
}
{
"name" : "io.systemd.Manager.units_by_type_total",
"value" : 2,
"fields" : {
"type" : "automount"
}
}
```
**units_by_state_total**:
```
{
"name" : "io.systemd.Manager.units_by_state_total",
"value" : 216,
"fields" : {
"state" : "active"
}
}
{
"name" : "io.systemd.Manager.units_by_state_total",
"value" : 0,
"fields" : {
"state" : "reloading"
}
}
{
"name" : "io.systemd.Manager.units_by_state_total",
"value" : 120,
"fields" : {
"state" : "inactive"
}
}
```
**unit_active_state**:
```
{
"name" : "io.systemd.Manager.unit_active_state",
"object" : "multi-user.target",
"value" : "active"
}
{
"name" : "io.systemd.Manager.unit_active_state",
"object" : "systemd-sysusers.service",
"value" : "inactive"
}
```
**unit_load_state**:
```
{
"name" : "io.systemd.Manager.unit_load_state",
"object" : "multi-user.target",
"value" : "loaded"
}
```
**nrestarts**:
```
{
"name" : "io.systemd.Manager.nrestarts",
"object" : "user@0.service",
"value" : 0
}
{
"name" : "io.systemd.Manager.nrestarts",
"object" : "user-runtime-dir@0.service",
"value" : 0
}
```
systemdstatedir is created by tmpfiles conf (tmpfiles.d/systemd.conf.in)
already.
By shipping it, systemd prevents the effect of tmpfiles.d/var.conf.in:
q /var 0755 - - -
This should helps implementing the "Hermetic /usr" pattern.
Links: https://0pointer.net/blog/fitting-everything-together.html
This partially reverts commit e58ba80a40. Those
aliases were removed as part of the removal of sysvinit compat, but the removal
was immediately reported as a regression by users. In particular, users can
have one of those set as the default target, or use those names in Wants= or
Requires= dependencies. At least in Fedora, the removal was immediately
reverted.
When looking at our "portability and stability promise", I noticed that those
units are explicitly listed in the promise. Let's restore them upstream. They
are conditionalized under a new -Dcompat-sysv-interfaces=bool option, so that
downstreams like Arch that wasnt to strictly avoid legacy interfaces can easily
disable them. They are dropped from the docs, like we do for all deprecated
interfaces.
This does not yet support directory images properly
as systemd itself does not support unpriv directory
images properly yet.
The user profiles are a copy of the system profiles but without
DynamicUser=yes (can't be used by user managers) and without
ProtectHome=yes (this masks /home which breaks StateDirectory= which
is lcoated inside /home)
A recent commit made the standalone binaries always buildable
on demand, but as a side effect due to how 'meson install' works,
they are always built and installed by 'meson install' even
if the standalone-binaries= option is disabled.
Fix it so that 'meson install' only installs them if the
option is explicitly enabled, while still allowing
building them on demand.
Follow-up for 54492552a1
As announced by a few releases now, finally drop support for
sysvinit scripts.
Keep rc-local generator for now, as it's really a distinct
feature even though from the same era.
The current tree doesn't even compile with libidn(1) after
2c7bdaf9f1, which included
a non-existent call to check_dlopen_blocked() somehow.
Hence, it feels safe to just nuke legacy support from
our repo.
Note, this drops logging only test case for crypt_preferred_method(),
as that requires explicitly dlopen() the library. But, we should test
that make_salt() and friends automatically dlopen() it.
libcrypt was no longer built by default since glibc-2.38, and it has been
completely removed since glibc-2.39.
Let's always use libxcrypt, unless when building with musl. As already
major distribution already have libxcrypt-4.4.x, hence let's also bump
the required minimum version to 4.4.0.
libxcrypt cannot be built with musl, hence the previous fallback logic
in libcrypt-util.c are moved to musl/crypt.c.
Note, libxcrypt-4.4.0 was released on 2018-11-20.
See also #38608.
Major distributions already have libseccomp 2.5.x or newer.
Let's bump to the required minimum version to 2.4.0, which provides
SCMP_ACT_KILL_PROCESS, SCMP_ACT_LOG, SCMP_ARCH_PARISC, and
SCMP_ARCH_PARISC64.
Note, libseccomp 2.4.0 was released on 2019-03-15.
See also #38608.
Major distributions already have cryptsetup newer than 2.4.0.
Let's bump the minimal required version.
Note, cryptsetup 2.4.0 was released on 2021-08-18.
See also #38608.
Major distributions already have elfutils >= 0.190.
Let's bump the required minimum version.
Note, elfutils 0.177 was released on 2019-08-14.
See also #38608.
Major distributions already have blkid >= 2.37.
Let's bump the minimal required version.
Note, util-linux (which provides blkid) 2.37 was released on 2021-06-01.
See also #38608.
All major distributions have switched to OpenSSL version 3.x.
Let's drop support of OpenSSL version 1.x.
Note, OpenSSL 3.0 was released on 2021-09-07 (and will be EOL on 2026-09-07).
See also #38608.
Let's make sure we don't define the same clang-tidy
test multiple times by deduplicating sources beforehand.
We end up with the same sources in a few cases because
e.g. we reuse the same fuzz-main.c for each fuzz target.
So far we compiled the normal and standalone versions completely
independently. Let's use the 'extract' template pattern to avoid any
additional compilation and only require an single link to produce the
.standalone variants.
Unfortunately, as designed, the 'extract' framework only allows one set of
object files to be extracted. Since we need all the files for the
.standalone version, we cannot use 'extract' for other purposes. Thus, in
the two cases where 'extract' was used for the test binaries, this is now
changed to compile the files a second time. But the number of files in that
list is small, so this seems like a better option.
(If we weren't using the template system, we could easily extract just the
objects we need. But with the current system, at the point of the
definition, the binaries are not defined yet. We'd need to handle all of
this through sets of dictionaries, and that just seems like too much
trouble to avoid double compilation of a few small files.)
Allow .standalone version to be built on-demand, even if -Dstandalone=false
is configured. In other words, this changes the meson option from a hard
disablement to a soft "build is on/off by default".
The meson config was originally written in this way but we lost this feature
after the transition to templates. It is nice to build additional targets on
demand during development, so add this back.
The coreutils release that included this was in 2012. Since then,
systemd requires a newer meson, mostly uses install_symlink, and
so on. Additionally, where it does use this feature, it uses the
'-r' short flag. There are implementations of 'ln' that do have
support for '-r' but no longopts. Things will still work on those
but this check will fail.
Previously, we looked for scripts in the tools/ directory ad hoc,
wherever they were needed. Let's do those checks in one place.
The main meson.build file is shrunk somewhat, which is always nice.
Quite often we need to specify include_directories('.'). Normally, meson
does this automatically, but we specify implicit_include_directories : false,
see 30d20907bd for an explanation.
Passing the include_directories argument through the template layers was rather
cumbersome. Let's simplify things by always including the directory of the
first source file. This makes the definitions simpler, at the cost of having an
unnecessary include directory in the list in some cases.
(Tests are excluded from this change, because they happen to not need this, but
also because some tests have source[0] which is a custom target, and we'd need
newer meson to support that.)
Currently, if the libmount feature is disabled, we don't build
libshared and as a result skip building every other executable as
well. Among other things, this makes our nodeps CI builds kind of
pointless since hardly any code will be compiled.
Let's improve on the situation by making libmount properly optional
in libshared. Then, we only skip building the executables that
actually need libmount.
In all cases, everything that we list in 'extract', we also list in
'sources'. We can simplify things by automatically appending the first
list to the second.
In the listings, move 'extract' key right below 'sources', since now
they are both "sources", just with slightly different meanings.
Symbols exported by libshared can't get pruned by the linker, so
every unused exported symbol is effectively dead code we ship to users
for no good reason. Let's add a script to analyze how many such symbols
we have.
We also add a meson test to run the script on all of our binaries.
Since it detects unused symbols and still has a few false positives,
don't enable the test by default similar to the clang-tidy tests.
The script was 100% vibe coded by Github Copilot with Claude Sonnet 4.5
as the model.
Current results are (without the unused symbols list):
Analysis of libsystemd-shared-259.so
======================================================================
Total exported symbols: 4830
(excluding public API symbols starting with 'sd_')
Used symbols: 4672
Unused symbols: 158
Usage rate: 96.7%
This reverts commit bf9bc5beb0.
libutmps does not support utmpxname(), the function always fails
with ENOSYS, and always uses their own file.
However, our code relies on the funtion needs to succeed.
Let's revert the change now, and revisit later when musl users
request to support libutmps.
So far the idea was that the default is 'auto', and if appropriate, the
distribution will create /var/log/journal/ to tell journald to use persistent
mode. This doesn't work well with factory resets, because after a factory reset
obviously /var/log is gone. That old default was useful when journald was new
and people were reluctant to enable persistent mode and instead relied on
rsyslog and such for the persistent storage. But nowadays that is rarer, and
anyway various features like user journals only work with persistent storage,
so we want people to enable this by default. Add an option to flip the default
and distributions can opt in. The default default value remains unchanged.
(I also tested using tmpfiles to instead change this, since we already set
access mode for /var/log/journal through tmpfiles. Unfortunately, tmpfiles runs
too late, after journald has already started, so if tmpfiles creates the
directory, it'll only be used after a reboot. This probably could be made to
work by adding a new service to flush the journal, but that becomes complicated
and we lose the main advantage of simplicity.)
Resolves https://bugzilla.redhat.com/show_bug.cgi?id=1387796.
- musl does not support gshadow, and does not provide gshadow.h,
- musl does not support nss, and does not provide nss.h which is necessary
for each nss modules,
- musl does not provide NI_IDN.
glibc sets it when _GNU_SOURCE is defined, however, musl does not.
Let's explicitly define it to make getdents64() and struct dirent64
available even when building with musl.
A --empower session is effectively root without being UID 0, so it
doesn't make sense to enforce polkit authentication in those. Let's
add the empower group, add --empower sessions to that group and ship
a polkit rule to skip authentication for all users in the empower
group.
(As a side-effect this will also allow users to add themselves to this
group outside of 'run0 --empower' to mimick NOPASSWD from sudo)
- use loop for checking existence of functions,
- rename HAVE_LIBARCHIVE_XYZ -> HAVE_ARCHIVE_XYZ to make them match with
the function name,
- do not conditionally include user-util.h in libarchive-util.h,
- sort library function symbols.
We'd like to measure various additional things into PCRs, but all
available ones to the OS are already used for various purposes. Hence,
let's introduce a new concept of "NV Index based PCRs", i.e. let's use
TPM2 nv indexes of type TPM2_NT_EXTEND that mostly behave like real
PCRs, but which we can allocate relatively freely from the nv index
space. Let's call these "fake" PCRs "NvPCRs".
My original intention was to get a fixed NV index range assigned from
the TCG, either for Linux or for systemd as a project, but this stalled
with no further updates from the TCG for more than a year and a half
now. I was told an NV index range to use though, even if it never was
officially assigned, hence this PR uses this by default. But the range
is configurable at build time, on purpose, so that downstreams have some
flexibility to change this if they want. To abstract the actual nvindex
number away we introduce a naming concept, so that nvindexes are
referenced by name string rather than number.
NvPCRs are defined in little JSON snippets in /usr/lib/nvpcr/*.nvpcr,
that match up index number and name, as well as pick a hash algorithm.
There's one complication: these nvindex (like any nvindex) can be
deleted by anyone with access to the TPM, and then be recreated. This
could be used to reset the NvPCRs to zero during runtime, which defeats
the whole point of them. Our way out: we measure a secret as first thing
after creation into the NvPCRs. (Or actually, we measure a per-NvPCR
secret we derive from a system secret via an HMAC of the NvPCR name) and
the nvindex handle). This "anchoring" secret is stored in /run/ +
/var/lib/ + ESP/XBOOTLDR (the latter encrypted as credential, locked to
the TPM), to make it available at the whole runtime of the OS.
Zbigniew Jędrzejewski-Szmek
Yaping Li
Max Gautier
Yu Watanabe
Daan De Meyer
Luca Boccassi
Mike Yuan
Haiyue Wang
q66
Lennart Poettering