Introduce support for configuring cpus and mems for processes using
cgroup v2 CPUSET controller. This allows users to limit which cpus
and memory NUMA nodes can be used by processes to better utilize
system resources.
The cgroup v2 interfaces to control it are cpuset.cpus and cpuset.mems
where the requested configuration is written. However, it doesn't mean
that the requested configuration will be actually used as parent cgroup
may limit the cpus or mems as well. In order to reflect the real
configuration cgroup v2 provides read-only files cpuset.cpus.effective
and cpuset.mems.effective which are exported to users as well.
During the rework of unit file loading, commit e8630e6952 dropped the
initialization u->source_mtime. This had the bad side effect that generated
units always needed daemon reloading.
People do have usernames with dots, and it makes them very unhappy that systemd
doesn't like their that. It seems that there is no actual problem with allowing
dots in the username. In particular chown declares ":" as the official
separator, and internally in systemd we never rely on "." as the seperator
between user and group (nor do we call chown directly). Using dots in the name
is probably not a very good idea, but we don't need to care. Debian tools
(adduser) do not allow users with dots to be created.
This patch allows *existing* names with dots to be used in User, Group,
SupplementaryGroups, SocketUser, SocketGroup fields, both in unit files and on
the command line. DynamicUsers and sysusers still follow the strict policy.
user@.service and tmpfiles already allowed arbitrary user names, and this
remains unchanged.
Fixes#12754.
v2:
- do not watch mtime of transient and generated dirs
We'd reload the map after every transient unit we created, which we don't
need to do, since we create those units ourselves and know their fragment
path.
This reworks how we load units from disk. Instead of chasing symlinks every
time we are asked to load a unit by name, we slurp all symlinks from disk
and build two hashmaps:
1. from unit name to either alias target, or fragment on disk
(if an alias, we put just the target name in the hashmap, if a fragment
we put an absolute path, so we can distinguish both).
2. from a unit name to all aliases
Reading all this data can be pretty costly (40 ms) on my machine, so we keep it
around for reuse.
The advantage is that we can reliably know what all the aliases of a given unit
are. This means we can reliably load dropins under all names. This fixes#11972.
I opted to embed the Bitmap structure directly in the ExitStatusSet.
This means that memory usage is a bit higher for units which don't define
this setting:
Service changes:
/* size: 2720, cachelines: 43, members: 73 */
/* sum members: 2680, holes: 9, sum holes: 39 */
/* sum bitfield members: 7 bits, bit holes: 1, sum bit holes: 1 bits */
/* last cacheline: 32 bytes */
/* size: 2816, cachelines: 44, members: 73 */
/* sum members: 2776, holes: 9, sum holes: 39 */
/* sum bitfield members: 7 bits, bit holes: 1, sum bit holes: 1 bits */
But this way the code is simpler and we do less pointer chasing.
Whenever I see EXTRACT_QUOTES, I'm always confused whether it means to
leave the quotes in or to take them out. Let's say "unquote", like we
say "cunescape".
We'd skip any whitespace immediately after "=", but then we'd treat whitespace
that is between "|" or "!" and the value as significant. This is rather
confusing, let's ignore it too.
Takes a single /sys/fs/bpf/pinned_prog string as argument, but may be
specified multiple times. An empty assignment resets all previous filters.
Closes https://github.com/systemd/systemd/issues/10227
Make possible to set NUMA allocation policy for manager. Manager's
policy is by default inherited to all forked off processes. However, it
is possible to override the policy on per-service basis. Currently we
support, these policies: default, prefer, bind, interleave, local.
See man 2 set_mempolicy for details on each policy.
Overall NUMA policy actually consists of two parts. Policy itself and
bitmask representing NUMA nodes where is policy effective. Node mask can
be specified using related option, NUMAMask. Default mask can be
overwritten on per-service level.
The CPU_SET_S api is pretty bad. In particular, it has a parameter for the size
of the array, but operations which take two (CPU_EQUAL_S) or even three arrays
(CPU_{AND,OR,XOR}_S) still take just one size. This means that all arrays must
be of the same size, or buffer overruns will occur. This is exactly what our
code would do, if it received an array of unexpected size over the network.
("Unexpected" here means anything different from what cpu_set_malloc() detects
as the "right" size.)
Let's rework this, and store the size in bytes of the allocated storage area.
The code will now parse any number up to 8191, independently of what the current
kernel supports. This matches the kernel maximum setting for any architecture,
to make things more portable.
Fixes#12605.
These make sense to be explicitly set at 0 (which has a different effect
than the default, since it can affect processing of `DefaultMemoryXXX`).
Without this, it's not easily possible to relinquish memory protection
for a subtree, which is not great.
When shooting down a service with SIGABRT the user might want to have a
much longer stop timeout than on regular stops/shutdowns. Especially in
the face of short stop timeouts the time might not be sufficient to
write huge core dumps before the service is killed.
This commit adds a dedicated (Default)TimeoutAbortSec= timer that is
used when stopping a service via SIGABRT. In all other cases the
existing TimeoutStopSec= is used. The timer value is unset by default
to skip the special handling and use TimeoutStopSec= for state
'stop-watchdog' to keep the old behaviour.
If the service is in state 'stop-watchdog' and the service should be
stopped explicitly we still go to 'stop-sigterm' and re-apply the usual
TimeoutStopSec= timeout.
In cgroup v2 we have protection tunables -- currently MemoryLow and
MemoryMin (there will be more in future for other resources, too). The
design of these protection tunables requires not only intermediate
cgroups to propagate protections, but also the units at the leaf of that
resource's operation to accept it (by setting MemoryLow or MemoryMin).
This makes sense from an low-level API design perspective, but it's a
good idea to also have a higher-level abstraction that can, by default,
propagate these resources to children recursively. In this patch, this
happens by having descendants set memory.low to N if their ancestor has
DefaultMemoryLow=N -- assuming they don't set a separate MemoryLow
value.
Any affected unit can opt out of this propagation by manually setting
`MemoryLow` to some value in its unit configuration. A unit can also
stop further propagation by setting `DefaultMemoryLow=` with no
argument. This removes further propagation in the subtree, but has no
effect on the unit itself (for that, use `MemoryLow=0`).
Our use case in production is simplifying the configuration of machines
which heavily rely on memory protection tunables, but currently require
tweaking a huge number of unit files to make that a reality. This
directive makes that significantly less fragile, and decreases the risk
of misconfiguration.
After this patch is merged, I will implement DefaultMemoryMin= using the
same principles.
This adds a new per-service OOMPolicy= (along with a global
DefaultOOMPolicy=) that controls what to do if a process of the service
is killed by the kernel's OOM killer. It has three different values:
"continue" (old behaviour), "stop" (terminate the service), "kill" (let
the kernel kill all the service's processes).
On top of that, track OOM killer events per unit: generate a per-unit
structured, recognizable log message when we see an OOM killer event,
and put the service in a failure state if an OOM killer event was seen
and the selected policy was not "continue". A new "result" is defined
for this case: "oom-kill".
All of this relies on new cgroupv2 kernel functionality: the
"memory.events" notification interface and the "memory.oom.group"
attribute (which makes the kernel kill all cgroup processes
automatically).
Building on previous commit, let's pass the unit name when parsing
dbus message or builtin whitelist, which is better than nothing.
seccomp_parse_syscall_filter() is not needed anymore, so it is removed,
and seccomp_parse_syscall_filter_full() is renamed to take its place.
We are about to add system calls (rseq()) not available on old
libseccomp/old kernels, and hence we need to be permissive when parsing
our definitions.
This new setting allows configuration of CFS period on the CPU cgroup, instead
of using a hardcoded default of 100ms.
Tested:
- Legacy cgroup + Unified cgroup
- systemctl set-property
- systemctl show
- Confirmed that the cgroup settings (such as cpu.cfs_period_ns) were set
appropriately, including updating the CPU quota (cpu.cfs_quota_ns) when
CPUQuotaPeriodSec= is updated.
- Checked that clamping works properly when either period or (quota * period)
are below the resolution of 1ms, or if period is above the max of 1s.
ensure that mfree() on filename is called after the logging function
which uses the string pointed by filename
Signed-off-by: Khem Raj <raj.khem@gmail.com>
fdopen doesn't accept "e", it's ignored. Let's not mislead people into
believing that it actually sets O_CLOEXEC.
From `man 3 fdopen`:
> e (since glibc 2.7):
> Open the file with the O_CLOEXEC flag. See open(2) for more information. This flag is ignored for fdopen()
As mentioned by @jlebon in #11131.
Some controllers (like the CPU controller) have a performance cost that
is non-trivial on certain workloads. While this can be mitigated and
improved to an extent, there will for some controllers always be some
overheads associated with the benefits gained from the controller.
Inside Facebook, the fix applied has been to disable the CPU controller
forcibly with `cgroup_disable=cpu` on the kernel command line.
This presents a problem: to disable or reenable the controller, a reboot
is required, but this is quite cumbersome and slow to do for many
thousands of machines, especially machines where disabling/enabling a
stateful service on a machine is a matter of several minutes.
Currently systemd provides some configuration knobs for these in the
form of `[Default]CPUAccounting`, `[Default]MemoryAccounting`, and the
like. The limitation of these is that Default*Accounting is overrideable
by individual services, of which any one could decide to reenable a
controller within the hierarchy at any point just by using a controller
feature implicitly (eg. `CPUWeight`), even if the use of that CPU
feature could just be opportunistic. Since many services are provided by
the distribution, or by upstream teams at a particular organisation,
it's not a sustainable solution to simply try to find and remove
offending directives from these units.
This commit presents a more direct solution -- a DisableControllers=
directive that forcibly disallows a controller from being enabled within
a subtree.
add new "systemd-run-generator" for running arbitrary commands from the kernel command line as system services using the "systemd.run=" kernel command line switch
This adds SuccessActionExitStatus= and FailureActionExitStatus= that may
be used to configure the exit status to propagate in when
SuccessAction=exit or FailureAction=exit is used.
When not specified let's also propagate the exit status of the main
process we fork off for the unit.
