Either become uid/gid of the client we have been forked for, or become
the "systemd-bus-proxy" user if the client was root. We retain
CAP_IPC_OWNER so that we can tell kdbus we are actually our own client.
Rely on modules being built-in or autoloaded on-demand.
As networkd is a network facing service, we want to limits its capabilities,
as much as possible. Also, we may not have CAP_SYS_MODULE in a container,
and we want networkd to work the same there.
Module autoloading does not always work, but should be fixed by the kernel
patch f98f89a0104454f35a: 'net: tunnels - enable module autoloading', which
is currently in net-next and which people may consider backporting if they
want tunneling support without compiling in the modules.
Early adopters may also use a module-load.d snippet and order
systemd-modules-load.service before networkd to force the module
loading of tunneling modules.
This sholud fix the various build issues people have reported.
This allows us to run networkd mostly unpriviliged with the exception of
CAP_NET_* and CAP_SYS_MODULE. I'd really like to get rid of the latter
though...
Instead of accessing /proc/1/environ directly, trying to read the
$container variable from it, let's make PID 1 save the contents of that
variable to /run/systemd/container. This allows us to detect containers
without the need for CAP_SYS_PTRACE, which allows us to drop it from a
number of daemons and from the file capabilities of systemd-detect-virt.
Also, don't consider chroot a container technology anymore. After all,
we don't consider file system namespaces container technology anymore,
and hence chroot() should be considered a container even less.
nspawn and the container child use eventfd to wait and notify each other
that they are ready so the container setup can be completed.
However in its current form the wait/notify event ignore errors that
may especially affect the child (container).
On errors the child will jump to the "child_fail" label and terminate
with _exit(EXIT_FAILURE) without notifying the parent. Since the eventfd
is created without the "EFD_NONBLOCK" flag, this leaves the parent
blocking on the eventfd_read() call. The container can also be killed
at any moment before execv() and the parent will not receive
notifications.
We can fix this by using cheap mechanisms, the new high level eventfd
API and handle SIGCHLD signals:
* Keep the cheap eventfd and EFD_NONBLOCK flag.
* Introduce eventfd states for parent and child to sync.
Child notifies parent with EVENTFD_CHILD_SUCCEEDED on success or
EVENTFD_CHILD_FAILED on failure and before _exit(). This prevents the
parent from waiting on an event that will never come.
* If the child is killed before execv() or before notifying the parent,
we install a NOP handler for SIGCHLD which will interrupt blocking calls
with EINTR. This gives a chance to the parent to call wait() and
terminate in main().
* If there are no errors, parent will block SIGCHLD, restore default
handler and notify child which will do execv(), then parent will pass
control to process_pty() to do its magic.
This was exposed in part by:
https://bugs.freedesktop.org/show_bug.cgi?id=76193
Reported-by: Tobias Hunger tobias.hunger@gmail.com
On Thu, May 22, 2014 at 9:53 AM, Jan Engelhardt <jengelh@inai.de> wrote:
>
> If libsystemd-network.la is relying on that udev function, it ought
> to specify libudev(-internal).la in libsystemd_network_la_LIBADD.
The verbose link-time deprecation warnings are annoying. These libs
will never change or be extended; there is no need to test the list
of exported symbols.
./.libs/libsystemd-network.a(libsystemd_network_la-network-internal.o):
network-internal.c:function net_get_unique_predictable_data:
error: undefined reference to 'udev_device_get_property_value'
collect2: error: ld returned 1 exit status
We shouldn't destroy IPC objects of system users on logout.
http://lists.freedesktop.org/archives/systemd-devel/2014-April/018373.html
This introduces SYSTEM_UID_MAX defined to the maximum UID of system
users. This value is determined compile-time, either as configure switch
or from /etc/login.defs. (We don't read that file at runtime, since this
is really a choice for a system builder, not the end user.)
While we are at it we then also update journald to use SYSTEM_UID_MAX
when we decide whether to split out log data for a specific client.
This makes callback behaviour more like sd-event or sd-resolve, and
creates proper object for unregistering callbacks.
Taking the refernce to the slot is optional. If not taken life time of
the slot will be bound to the underlying bus object (or in the case of
an async call until the reply has been recieved).
New "struct ring" object that implements a basic ring buffer for arbitrary
byte-streams. A new basic runtime test is also added.
This will be needed for our pty helpers for systemd-console and friends.
As the operational state detection in sd-network is still too primitive, timesyncd
will likely try to connect a bit early, so the first attempt will fail.
- Also only allow positive ifindex on both dhcp and ipv4ll
[tomegun: the kernel always sets a positive ifindex, but some APIs accept
ifindex=0 with various meanings, so we should protect against
accidentally passing ifindex=0 along.]
To make sure we don't delay boot on systems where (some) network links are managed by someone else
we don't block if something else has successfully brought up a link.
We will still block until all links we are aware of that are managed by networkd have been
configured, but if no such links exist, and someone else have configured a link sufficiently
that it has a carrier, it may be that the link is ready so we should no longer block.
Note that in all likelyhood the link is not ready (no addresses/routes configured),
so whatever network managment daemon configured it should provide a similar wait-online
service to block network-online.target until it is ready.
The aim is to block as long as we know networking is not fully configured, but no longer. This
will allow systemd-networkd-wait-online.service to be enabled on any system, even if we don't
know whether networkd is the main/only network manager.
Even in the case networking is fully configured by networkd, the default behavior may not be
sufficient: if two links need to be configured, but the first is fully configured before the
second one appears we will assume the network is up. To work around that, we allow specifying
specific devices to wait for before considering the network up.
This unit is enabled by default, just like systemd-networkd, but will only be pulled in if
anyone pulls in network-online.target.
The way the kernel namespaces have been implemented breaks assumptions
udev made regarding uevent sequence numbers. Creating devices in a
namespace "steals" uevents and its sequence numbers from the host. It
confuses the "udevadmin settle" logic, which might block until util a
timeout is reached, even when no uevent is pending.
Remove any assumptions about sequence numbers and deprecate libudev's
API exposing these numbers; none of that can reliably be used anymore
when namespaces are involved.
Use a static table with all the typing information, rather than repeated
switch statements. This should make it a lot simpler to add new types.
We need to keep all the type info to be able to create containers
without exposing their implementation details to the users of the library.
As a freebee we verify the types of appended/read attributes.
The API is extended to nicely deal with unions of container types.
<kay> ssuominen: and drop --no-as-needed from the linkcheck?
<kay> ssuominen: i expect it all triggers without the gc-sections thing alone
<ssuominen> if the intention is to make it strict as possible,
to catch undefined references caused by missing -lfoo in linker line, then
LDFLAGS="-Wl,-fuse-ld=gold -Wl,--as-needed -Wl,--no-gc-sections"
