With Oracle's physical servers, both `/sys/class/dmi/id/sys_vendor` and
`/sys/class/dmi/id/board_vendor` contain `Oracle Corporation`, so those
servers are detected as `oracle` (VirtualBox).
VirtualBox has the following values in the latest versions:
Presumably the existing check for `innotek GmbH` is meant to detect
older versions of VirtualBox, while changing the second checked value
from `Oracle Corporation` to `VirtualBox` will reliably detect later and future
versions.
json: do something remotely reasonable when we see NaN/infinity
JSON doesn't have NaN/infinity/-infinity concepts in the spec.
Implementations vary what they do with it. JSON5 + Python simply
generate special words "NAN" and "Inifinity" from it. Others generate
"null" for it.
At this point we never actually want to output this, so let's be
conservative and generate RFC compliant JSON, i.e. convert to null.
One day should JSON5 actually become a thing we can revisit this, but in
that case we should implement things via a flag, and only optinally
process nan/infinity/-infinity.
This patch is extremely simple: whenever accepting a
nan/infinity/-infinity from outside it converts it to NULL. I.e. we
convert on input, not output.
The setting is completely meaningless, as WithoutRA= and UseDelegatedPrefix=
in [DHCPv6] section, and DHCPv6Client= in [IPv6AcceptRA] section control
the behavior.
Yu Watanabe [Wed, 13 Oct 2021 07:26:09 +0000 (16:26 +0900)]
network: dhcp6: introduce UseDelegatedPrefix= setting and enable by default
Previously, the prefix delegation is enabled when at least one
downstream interfaces request it. But, when the DHCPv6 client on the
upstream interface is configured, some downstream interfaces may not
exist yet, nor have .network file assigned.
Also, if a system has thousands of interfaces, then the previous logic
introduce O(n^2) search.
This makes the prefix delegation is always enabled, except when it is
explicitly disabled. Hopefully, that should not break anything, as the
DHCPv6 server should ignore the prefix delegation request if the server
do not have any prefix to delegate.
Yu Watanabe [Mon, 25 Oct 2021 17:29:09 +0000 (02:29 +0900)]
network: delay dropping addresses or so on reloading .network files
When a .network file is updated but its change is not so big, it is not
necessary to first drop all configs and then reassign later again.
This slightly optimize such situation. First foreignize all configs, and
then drop later when it is not requested by the updated .network file.
Apparently memory sanitizer doesn't grok getdents64() properly. Let's
address that by explicitly marken memory initialized by getdents64() as
unpoisoned.
stat-util: optimize dir_is_empty_at() a bit, by using getdents64()
That way we have a single syscall only for it, instead of the multiple
readdir() and friends do. And we can operate entirely on the stack, no
malloc() implicit.
Frantisek Sumsal [Mon, 25 Oct 2021 09:02:22 +0000 (11:02 +0200)]
test: tweak TriggerLimitIntervalSec= when built with coverage
Collecting coverage causes a significant slowdown in general, but since
this test requires certain timing, we need to tweak the defaults to make
it reliably pass.
varlink: don't try to talk to oomd from unit tests
Talking to external daemons we ourselves maintain is a job for the
integration tests, not the unit tests. This communication is likely to
fail hence don#t even bother.
mount-util: move opening of /proc/self/mountinfo into bind_remount_one_with_mountinfo()
Let's move things around a bit, and open /proc/self/mountinfo if needed
inside of bind_remount_one_with_mountinfo(). That way bind_remount_one()
can become a superthin inline wrapper around
bind_remount_one_with_mountinfo(). Main benefit is that we don't even
have to open /p/s/mi in case mount_setattr() actually worked for us.
We should drop caches if we are configured to do so in all cases where
we are done with home dir operations: except if that operation is
activation, because in that case we are not destroying anything, but
leaving it on.
Hence, turn off the flag that reminds us that we should drop caches
before exiting, once activation completed fully,
nspawn: bump RLIMIT_NOFILE for nspawn payload similar to how host PID 1 does it for its payload
We try to pass containers roughly the same rlimits as the host gets from
the kernel. However, this means we'd set the RLIMIT_NOFILE to 4K. Which
is quite limiting though, and is something we actually departed from in
PID1: since 52d620757817bc0fa7de3ddbe43024544ced7ea0 we raise the limit
substantially for all userspace.
Given that nspawn is quite often invoked without proper PID1, let's raise the
limits for container payloads the same way as we do from the real PID1
to its service payloads.
Jan Janssen [Wed, 20 Oct 2021 10:15:03 +0000 (12:15 +0200)]
sd-boot: Add keys to reboot into firmware interface
This is useful if the auto-firmware setting has been disabled. The
keys used here are based on what the majority of firmware employ in
the wild.
This also ensures there's a chance for the user to discover this in
case they were too slow during POST or simply used the wrong ones.
We are using this for creating userns namespaces, and we really
shouldn't try to sync there. Moreover the use of free() in shutdown code
doesn't need it anyway, since it just sync()ed right before anyway. Only
the third user of freeze() we have actually needs the syc(), hence do it
there and nowhere else.
namespace-util: introduce userns_acquire() as helper for allocating new unbound userns
This returns a namespace fd, and takes a uidmap/gidmap as string. This
is split out out mount-util.c's remount_idmap() logic, so that we can
allocate a userns independently.
Previously the call did two things, and the second thing was optional
(depending on first arg being NULL). Let's simplify this and just make
it two distinct functions, where one calls the other.
This should make things a bit more readable, given that we called a
function called "…and_mount()" which didn't actually mount...
fd-util: when re-opening a directory with fd_reopen() go via openat(…, ".", …)
This adds a tiny shortcut to fd_reopen(): if we are about to reopen the
fd via O_DIRECTORY then we know it#s a directory and we might as well
reopen it via opening "." using the fd as "at fd" in openat().
This has the benefit that we don't need /proc/self/fd/ around for this
special case: fewer sources of errors.
The clock to use internally is clock_boottime_or_monotonic(), but the
test used CLOCK_MONOTONIC. After one system suspend the test thus likely
starts to fail.
Frantisek Sumsal [Thu, 21 Oct 2021 16:06:36 +0000 (18:06 +0200)]
test: loosen sandbox restrictions for integration tests as well
Otherwise we miss quite a lot of coverage (mainly from logind,
hostnamed, networkd, and possibly others), since they can't write their
reports with `ProtectSystem=strict`.
Frantisek Sumsal [Thu, 21 Oct 2021 15:34:43 +0000 (17:34 +0200)]
test: loosen certain sandbox restrictions when collecting coverage
With `ProtectSystem=strict` gcov is unable to write the *.gcda files
with collected coverage. Let's add a yet another switch to make such
restriction less strict to make gcov happy.
This addresses following errors:
```
...
systemd-networkd[272469]: profiling:/systemd-meson-build/src/shared/libsystemd-shared-249.a.p/binfmt-util.c.gcda:Cannot open
systemd-networkd[272469]: profiling:/systemd-meson-build/src/shared/libsystemd-shared-249.a.p/base-filesystem.c.gcda:Cannot open
systemd-networkd[272469]: profiling:/systemd-meson-build/src/shared/libsystemd-shared-249.a.p/barrier.c.gcda:Cannot open
systemd-networkd[272469]: profiling:/systemd-meson-build/src/shared/libsystemd-shared-249.a.p/ask-password-api.c.gcda:Cannot open
systemd-networkd[272469]: profiling:/systemd-meson-build/src/shared/libsystemd-shared-249.a.p/apparmor-util.c.gcda:Cannot open
systemd-networkd[272469]: profiling:/systemd-meson-build/src/shared/libsystemd-shared-249.a.p/acpi-fpdt.c.gcda:Cannot open
...
```
Frantisek Sumsal [Thu, 21 Oct 2021 13:59:57 +0000 (15:59 +0200)]
test: wait a bit for the given PID to die if it's still alive
When playing around with the coverage-enabled build I kept hitting
an issue where dnsmasq failed to start because the previous instance was
still shutting down. This should, hopefully, help to mitigate that.