This PR allows an option for systemd exec units to enable UTS namespaces
but not restrict changing hostname via seccomp. Thus, units can change
hostname without affecting the host. This is useful for OS-like
containers running as units where they should have freedom to change
their container hostname if they want, but not the host's hostname.
Fixes: #30348
Otherwise, if networkd is failed, e.g. .network files that triggered the
failure will remain, and the next test case will start with previous
.network files. So, most subsequent test will fail.
This adds the following three options:
- detect_invalid_pointer_pairs=2
- handle_ioctl=1
- print_cmdline=1
Note, these options were used in the CentOS CI job.
This is a revised version of cbd5f5d7d6,
which is reverted by 560412148f.
The filter condition for _UID= field was added by
fdcd37df3b. However, after
888e378da2, coredumps for system services are
stored by systemd-coredump user. Hence, _UID=0 does not match anything anymore.
When we are reading the system journal, let's set _UID=<systemd-coredump>,
then we can show coredumps for system services again. When we are reading
an external journal, then the uid of systemd-coredump may be different
from the one on the current system, so let's drop the filter condition
for _UID= in that case.
Currently, mkosi GitHub action complains the following:
===
Could not find 'setfiles' which is required to relabel files.
===
Let's tentatively disable SELinux test.
This effectively reverts fdcd37df3b.
After 888e378da2, coredumps for system
services are stored by systemd-coredump user. Henve, _UID=0 does not
match anything anymore.
Before:
```
$ build/journalctl -u systemd-networkd.service -p2 -n1
-- No entries --
```
After:
```
$ build/journalctl -u systemd-networkd.service -p2 -n1
Dec 06 05:39:55 H systemd-coredump[4187]: [🡕] Process 4145 (systemd-network) of user 192 dumped core.
(snip)
```
- make udevd generate debugging logs for loopback and DM devices,
- insert 'udevadm wait' at several places to make the device processed
by udevd,
- cleanup generated integritysetup service before moving to next
algorithm,
- drop unnecessary exit on command failure,
- also test data splitting mode for all algorithms.
This partially reverts the commit 405be62f05
"tree-wide: refuse enumerated device with ID_PROCESSING=1".
Otherwise, when systemd-udev-trigger.service is (re)started just before
daemon-reexec, which can be easily happen on systemd package update, then
udev database files for many devices may have ID_PROCESSING=1 property,
thus devices may not be enumerated on daemon-reexec. That causes many
units especially mount units being deactivated after daemon-reexec.
Fixes#35329.
After 259125d53d, network interfaces
declared by .netdev files are created after systemd-networkd sends READY
notification. So, even when networkd is started, the netdevs may not
be created yet, and 'ip' command may fail. Let's also check the return
code of the command.
This also
- drops never worked stdout checks,
- makes the test fail if the interface is not created within the timeout.
Otherwise, nexthop ID may contain e.g. 300, then
===
AssertionError: '300' unexpectedly found in
'default nhid 3860882700 via fe80::1034:56ff:fe78:9a99 proto ra metric 512 expires 1798sec pref high\n
default nhid 2639230080 via fe80::1034:56ff:fe78:9a98 proto ra metric 2048 expires 1798sec pref low'
===
Fixes IPv6 Core Conformance test failures reported at #33468.
https://www.ipv6ready.org/docs/Core_Conformance.pdf
Test v6LC.2.2.23 h and j: Processing Router Advertisement with Route
Information Option (Host Only)
When a RA contains route option with ::/0 prefix, then previously that
may contradict with the default route requested with the RA header.
If the route option has zero lifetime, the existing default route should
be removed, and a new route based on the RA header should be configured.
If the route option has non-zero lifetime, the RA header should be
ignored.
So, we first need to process options with zero lifetime (not only
route option, as the similar reasons), then configure the default route
based on the RA, finally process options with non-zero lifetime.
While doing that, even if mknod() failed, we anyway try to fall back to
use bind mount if arg_uid_shift == 0.
Mostly no functional change, just refactoring and preparation for later commit.
Follow-up for dc3223919f.
If nspawn is invoked with DevicePolicy= but DeviceAllow= does not
contain /dev/fuse, nspawn will fail to get fuse version with -EPERM.
Let's silence the warning in that case.
Previously, when multiple routers send RAs with the same preference,
then the kernel merges routes with the same gateway address:
===
default proto ra metric 1024 expires 595sec pref medium
nexthop via fe80::200:10ff:fe10:1060 dev enp0s9 weight 1
nexthop via fe80::200:10ff:fe10:1061 dev enp0s9 weight 1
===
This causes IPv6 Conformance Test v6LC.2.2.11 failure, as reported in #33470.
To avoid the coalescing issue, we can use nexthop, as suggested by Ido Schimmel:
https://lore.kernel.org/netdev/ZytjEINNRmtpadr_@shredder/
> BTW, you can avoid the coalescing problem by using the nexthop API.
> # ip nexthop add id 1 via fe80::200:10ff:fe10:1060 dev enp0s9
> # ip -6 route add default nhid 1 expires 600 proto ra
> # ip nexthop add id 2 via fe80::200:10ff:fe10:1061 dev enp0s9
> # ip -6 route append default nhid 2 expires 600 proto ra
> # ip -6 route
> fe80::/64 dev enp0s9 proto kernel metric 256 pref medium
> default nhid 1 via fe80::200:10ff:fe10:1060 dev enp0s9 proto ra metric 1024 expires 563sec pref medium
> default nhid 2 via fe80::200:10ff:fe10:1061 dev enp0s9 proto ra metric 1024 expires 594sec pref medium
Fixes#33470.
Suggested-by: Ido Schimmel <idosch@idosch.org>
By the previous commit, configuration source of addresses and routes are
saved on stop and restored on start. Hence, we can keep dynamic
configurations on stop.
Co-authored-by: Jian Zhang <zhangjian.3032@bytedance.com>
Currently, only configuration sources and providers of addresses and
routes are serialized/deserialized.
This should mostly not change behavior, as dynamic (except for DHCPv4)
configurations will be dropped before stopping networkd, and for DHCPv4
protocol, we have already had another logic to handle DHCPv4
configurations.
Preparation for later commits.
Follow-up for 1003093604.
If a netdev is detached for some reasons, then previously the request
was simply cancelled, and the underlying interface never enter the
configured state, as the 'stacked_netdevs_created' flag never set.
This makes the counter decremented manually by the function, and set the
flag. So, the underlying interface can eter the configured state.
After PR #34909, networkd tries to update an existing netdev interface if
possible. But, when .netdev files are loaded on start, we have not
enumerate interfaces, so we do not know if the corresponding interface
exists or not. Let's delay processing request a bit.
Follow-up for PR #34909.
This fixes an issue that network interfaces cannot join a master netdev,
like bond or bridge, when the corresponding .netdev is reloaded.
With PR #34909, networkd supports reloading .netdev files. However,
When a .netdev file is modified and reloaded, ifindex is copied from
the old NetDev object to the new one. Thus, even if the interface is
successfully updated, netdev_set_ifindex_impl() will return 0 and
netdev_enter_ready() will never called. If the netdev is a kind of
master netdev, then port interfaces cannot join the master netdev,
as REQUEST_TYPE_SET_LINK_MASTER requires that the master netdev is
in the ready state.
Follow-up for 17c5337f7b.
Older kernels (older than v6.5) refuse RTM_NEWLINK messages with IFLA_ADDRESS
attribute when the netdev already exists and is running, even if the MAC
address is unchanged.
So, let's not set IFLA_ADDRESS or IFLA_MTU if they are unchanged, and
set the attributes only when we can update them.
The auditing subsystem is still not virtualized for containers, hence
the two values don't really make sense inside them, they will just leak
information from outside into the container. Hence don't make use of the
data if we detect we are run inside of a container.
This has visible effects: logind will no longer try to reuse the
auditing session ids as its own session ids when run inside a container.
While are at it, modernize the calls in more ways:
1. switch to pidref behaviour, all but one of our uses are using pidref
anyway already.
2. use read_virtual_file() + proc_mounted()
3. reasonably distinguish ENOENT errors when reading the process proc
files: distinguish the case where /proc is not mounted, from the case
where the process is already gone, from where auditing is not enabled in
the kernel build.
Yu Watanabe
