systemd/README
2021-10-06 10:52:15 +02:00

364 lines
15 KiB
Plaintext

systemd System and Service Manager
DETAILS:
http://0pointer.de/blog/projects/systemd.html
WEB SITE:
https://www.freedesktop.org/wiki/Software/systemd
GIT:
git@github.com:systemd/systemd.git
https://github.com/systemd/systemd
MAILING LIST:
https://lists.freedesktop.org/mailman/listinfo/systemd-devel
IRC:
#systemd on irc.libera.chat
BUG REPORTS:
https://github.com/systemd/systemd/issues
AUTHOR:
Lennart Poettering
Kay Sievers
...and many others
LICENSE:
LGPL-2.1-or-later for all code, exceptions noted in LICENSES/README.md
REQUIREMENTS:
Linux kernel >= 3.13
Linux kernel >= 4.2 for unified cgroup hierarchy support
Linux kernel >= 4.10 for cgroup-bpf egress and ingress hooks
Linux kernel >= 4.15 for cgroup-bpf device hook
Linux kernel >= 4.17 for cgroup-bpf socket address hooks
Linux kernel >= 5.3 for bounded-loops in BPF program
Linux kernel >= 5.4 for signed Verity images support
Linux kernel >= 5.7 for BPF links and the BPF LSM hook
Kernel Config Options:
CONFIG_DEVTMPFS
CONFIG_CGROUPS (it is OK to disable all controllers)
CONFIG_INOTIFY_USER
CONFIG_SIGNALFD
CONFIG_TIMERFD
CONFIG_EPOLL
CONFIG_UNIX (it requires CONFIG_NET, but every other flag in it is not necessary)
CONFIG_SYSFS
CONFIG_PROC_FS
CONFIG_FHANDLE (libudev, mount and bind mount handling)
Kernel crypto/hash API
CONFIG_CRYPTO_USER_API_HASH
CONFIG_CRYPTO_HMAC
CONFIG_CRYPTO_SHA256
udev will fail to work with the legacy sysfs layout:
CONFIG_SYSFS_DEPRECATED=n
Legacy hotplug slows down the system and confuses udev:
CONFIG_UEVENT_HELPER_PATH=""
Userspace firmware loading is not supported and should
be disabled in the kernel:
CONFIG_FW_LOADER_USER_HELPER=n
Some udev rules and virtualization detection relies on it:
CONFIG_DMIID
Support for some SCSI devices serial number retrieval, to
create additional symlinks in /dev/disk/ and /dev/tape:
CONFIG_BLK_DEV_BSG
Required for PrivateNetwork= in service units:
CONFIG_NET_NS
Note that systemd-localed.service and other systemd units use
PrivateNetwork so this is effectively required.
Required for PrivateUsers= in service units:
CONFIG_USER_NS
Optional but strongly recommended:
CONFIG_IPV6
CONFIG_AUTOFS_FS
CONFIG_TMPFS_XATTR
CONFIG_{TMPFS,EXT4_FS,XFS,BTRFS_FS,...}_POSIX_ACL
CONFIG_SECCOMP
CONFIG_SECCOMP_FILTER (required for seccomp support)
CONFIG_CHECKPOINT_RESTORE (for the kcmp() syscall)
Required for CPUShares= in resource control unit settings
CONFIG_CGROUP_SCHED
CONFIG_FAIR_GROUP_SCHED
Required for CPUQuota= in resource control unit settings
CONFIG_CFS_BANDWIDTH
Required for IPAddressDeny=, IPAddressAllow=, IPIngressFilterPath=,
IPEgressFilterPath= in resource control unit settings
unit settings
CONFIG_BPF
CONFIG_BPF_SYSCALL
CONFIG_BPF_JIT
CONFIG_HAVE_EBPF_JIT
CONFIG_CGROUP_BPF
Required for SocketBind{Allow|Deny}=, RestrictNetworkInterfaces= in
resource control unit settings
CONFIG_BPF
CONFIG_BPF_SYSCALL
CONFIG_BPF_JIT
CONFIG_HAVE_EBPF_JIT
CONFIG_CGROUP_BPF
For UEFI systems:
CONFIG_EFIVAR_FS
CONFIG_EFI_PARTITION
Required for signed Verity images support:
CONFIG_DM_VERITY_VERIFY_ROOTHASH_SIG
Required for RestrictFileSystems= in service units:
CONFIG_BPF
CONFIG_BPF_SYSCALL
CONFIG_BPF_LSM
CONFIG_DEBUG_INFO_BTF
CONFIG_LSM="...,bpf" or kernel booted with lsm="...,bpf".
We recommend to turn off Real-Time group scheduling in the
kernel when using systemd. RT group scheduling effectively
makes RT scheduling unavailable for most userspace, since it
requires explicit assignment of RT budgets to each unit whose
processes making use of RT. As there's no sensible way to
assign these budgets automatically this cannot really be
fixed, and it's best to disable group scheduling hence.
CONFIG_RT_GROUP_SCHED=n
It's a good idea to disable the implicit creation of networking bonding
devices by the kernel networking bonding module, so that the
automatically created "bond0" interface doesn't conflict with any such
device created by systemd-networkd (or other tools). Ideally there
would be a kernel compile-time option for this, but there currently
isn't. The next best thing is to make this change through a modprobe.d
drop-in. This is shipped by default, see modprobe.d/systemd.conf.
Required for systemd-nspawn:
CONFIG_DEVPTS_MULTIPLE_INSTANCES or Linux kernel >= 4.7
Required for systemd-oomd:
CONFIG_PSI
Note that kernel auditing is broken when used with systemd's
container code. When using systemd in conjunction with
containers, please make sure to either turn off auditing at
runtime using the kernel command line option "audit=0", or
turn it off at kernel compile time using:
CONFIG_AUDIT=n
If systemd is compiled with libseccomp support on
architectures which do not use socketcall() and where seccomp
is supported (this effectively means x86-64 and ARM, but
excludes 32-bit x86!), then nspawn will now install a
work-around seccomp filter that makes containers boot even
with audit being enabled. This works correctly only on kernels
3.14 and newer though. TL;DR: turn audit off, still.
glibc >= 2.16
libcap
libmount >= 2.30 (from util-linux)
(util-linux *must* be built without --enable-libmount-support-mtab)
libseccomp >= 2.3.1 (optional)
libblkid >= 2.24 (from util-linux) (optional)
libkmod >= 15 (optional)
PAM >= 1.1.2 (optional)
libcryptsetup (optional), >= 2.3.0 required for signed Verity images support
libaudit (optional)
libacl (optional)
libbpf >= 0.2.0 (optional)
libfdisk >= 2.33 (from util-linux) (optional)
libselinux (optional)
liblzma (optional)
liblz4 >= 1.3.0 / 130 (optional)
libzstd >= 1.4.0 (optional)
libgcrypt (optional)
libqrencode (optional)
libmicrohttpd (optional)
libpython (optional)
libidn2 or libidn (optional)
gnutls >= 3.1.4 (optional, >= 3.6.0 is required to support DNS-over-TLS with gnutls)
openssl >= 1.1.0 (optional, required to support DNS-over-TLS with openssl)
elfutils >= 158 (optional)
polkit (optional)
tzdata >= 2014f (optional)
pkg-config
gperf
docbook-xsl (optional, required for documentation)
xsltproc (optional, required for documentation)
python-jinja2
python-lxml (optional, required to build the indices)
python >= 3.5
meson >= 0.47 (>= 0.49 is required to build position-independent executables)
ninja
gcc, awk, sed, grep, and similar tools
clang >= 10.0, llvm >= 10.0 (optional, required to build BPF programs
from source code in C)
During runtime, you need the following additional
dependencies:
util-linux >= v2.27.1 required
dbus >= 1.4.0 (strictly speaking optional, but recommended)
NOTE: If using dbus < 1.9.18, you should override the default
policy directory (--with-dbuspolicydir=/etc/dbus-1/system.d).
dracut (optional)
polkit (optional)
To build in directory build/:
meson setup build/ && meson compile -C build/
Any configuration options can be specified as -Darg=value... arguments
to meson. After the build directory is initially configured, meson will
refuse to run again, and options must be changed with:
meson configure -Darg=value build/
meson configure without any arguments will print out available options and
their current values.
Useful commands:
meson compile -v -C build/ some/target
meson test -C build/
sudo meson install -C build/
DESTDIR=... meson install -C build/
A tarball can be created with:
git archive --format=tar --prefix=systemd-222/ v222 | xz > systemd-222.tar.xz
When systemd-hostnamed is used, it is strongly recommended to
install nss-myhostname to ensure that, in a world of
dynamically changing hostnames, the hostname stays resolvable
under all circumstances. In fact, systemd-hostnamed will warn
if nss-myhostname is not installed.
nss-systemd must be enabled on systemd systems, as that's required for
DynamicUser= to work. Note that we ship services out-of-the-box that
make use of DynamicUser= now, hence enabling nss-systemd is not
optional.
Note that the build prefix for systemd must be /usr. (Moreover,
packages systemd relies on — such as D-Bus — really should use the same
prefix, otherwise you are on your own.) -Dsplit-usr=false (which is the
default and does not need to be specified) is the recommended setting.
-Dsplit-usr=true can be used to give a semblance of support for systems
with programs installed split between / and /usr. Moving everything
under /usr is strongly encouraged.
Additional packages are necessary to run some tests:
- busybox (used by test/TEST-13-NSPAWN-SMOKE)
- nc (used by test/TEST-12-ISSUE-3171)
- python3-pyparsing
- python3-evdev (used by hwdb parsing tests)
- strace (used by test/test-functions)
- capsh (optional, used by test-execute)
USERS AND GROUPS:
Default udev rules use the following standard system group
names, which need to be resolvable by getgrnam() at any time,
even in the very early boot stages, where no other databases
and network are available:
audio, cdrom, dialout, disk, input, kmem, kvm, lp, render, tape, tty, video
During runtime, the journal daemon requires the
"systemd-journal" system group to exist. New journal files will
be readable by this group (but not writable), which may be used
to grant specific users read access. In addition, system
groups "wheel" and "adm" will be given read-only access to
journal files using systemd-tmpfiles.service.
The journal remote daemon requires the
"systemd-journal-remote" system user and group to
exist. During execution this network facing service will drop
privileges and assume this uid/gid for security reasons.
Similarly, the network management daemon requires the
"systemd-network" system user and group to exist.
Similarly, the name resolution daemon requires the
"systemd-resolve" system user and group to exist.
Similarly, the coredump support requires the
"systemd-coredump" system user and group to exist.
NSS:
systemd ships with four glibc NSS modules:
nss-myhostname resolves the local hostname to locally configured IP
addresses, as well as "localhost" to 127.0.0.1/::1.
nss-resolve enables DNS resolution via the systemd-resolved DNS/LLMNR
caching stub resolver "systemd-resolved".
nss-mymachines enables resolution of all local containers registered
with machined to their respective IP addresses.
nss-systemd enables resolution of users/group registered via the
User/Group Record Lookup API (https://systemd.io/USER_GROUP_API),
including all dynamically allocated service users. (See the
DynamicUser= setting in unit files.)
To make use of these NSS modules, please add them to the "hosts:",
"passwd:" and "group:" lines in /etc/nsswitch.conf. The "resolve"
module should replace the glibc "dns" module in this file (and don't
worry, it chain-loads the "dns" module if it can't talk to resolved).
The four modules should be used in the following order:
passwd: compat systemd
group: compat systemd
hosts: files mymachines resolve [!UNAVAIL=return] dns myhostname
SYSV INIT.D SCRIPTS:
When calling "systemctl enable/disable/is-enabled" on a unit which is a
SysV init.d script, it calls /usr/lib/systemd/systemd-sysv-install;
this needs to translate the action into the distribution specific
mechanism such as chkconfig or update-rc.d. Packagers need to provide
this script if you need this functionality (you don't if you disabled
SysV init support).
Please see src/systemctl/systemd-sysv-install.SKELETON for how this
needs to look like, and provide an implementation at the marked places.
WARNINGS:
systemd will warn during early boot if /usr is not already mounted at
this point (that means: either located on the same file system as / or
already mounted in the initrd). While in systemd itself very little
will break if /usr is on a separate, late-mounted partition, many of
its dependencies very likely will break sooner or later in one form or
another. For example, udev rules tend to refer to binaries in /usr,
binaries that link to libraries in /usr or binaries that refer to data
files in /usr. Since these breakages are not always directly visible,
systemd will warn about this, since this kind of file system setup is
not really supported anymore by the basic set of Linux OS components.
systemd requires that the /run mount point exists. systemd also
requires that /var/run is a symlink to /run.
For more information on this issue consult
https://www.freedesktop.org/wiki/Software/systemd/separate-usr-is-broken
To run systemd under valgrind, compile with meson option
-Dvalgrind=true and have valgrind development headers installed
(i.e. valgrind-devel or equivalent). Otherwise, false positives will be
triggered by code which violates some rules but is actually safe. Note
that valgrind generates nice output only on exit(), hence on shutdown
we don't execve() systemd-shutdown.
STABLE BRANCHES AND BACKPORTS:
Stable branches with backported patches are available in the
systemd-stable repo at https://github.com/systemd/systemd-stable.
Stable branches are started for certain releases of systemd and named
after them, e.g. v238-stable. Stable branches are managed by
distribution maintainers on an as needed basis. See
https://www.freedesktop.org/wiki/Software/systemd/Backports/ for some
more information and examples.