a) those which are constant as long as an object exists
b) those which can change and PropertiesChange messages with contents are generated
c) those which can change and where the PropertesChange merely includes invalidation
d) those which can change but for which no events are generated
Clients (through code generators run on the introspection XML) can thus
aggressively cache a, b, c, with only d excluded.
Djalal Harouni [Sat, 21 Dec 2013 14:49:46 +0000 (15:49 +0100)]
loginctl: correctly show session IDs on session-status
Commit f8f14b3654bcd introduced a regression that makes
loginctl session-status to not show the correct session ID(s)
In print_session_status_info() the map[] array, element "Seat" receives
the offset of the "id" in "SessionStatusInfo" struct instead of the
offset of the "seat" member.
This will cause prop_map_first_of_struct() function to overwrite the
SessionStatusInfo.id memory with seats if there are any.
Fix this typo by using the "seat" member.
Before:
- tixxdz (1000)
Since: Sat 2013-12-21 10:07:23 CET; 5h 26min ago
Leader: 1265 (sshd)
After:
1 - tixxdz (1000)
Since: Sat 2013-12-21 10:07:23 CET; 5h 26min ago
Leader: 1265 (sshd)
Apparently automake does not include the sources if they are under
a conditional that is disabled when making dist. This means that
everything would have to be enabled to make distcheck work.
bus: when introspecting, turn unprivileged flag into inverse annoation of "privileged"
Internally, it makes sense to have a default of "privileged" for
methods, and a flag to open it up. However, externally in the bus
introspection turn this around since negative options actually suck.
logging: reduce send timeout to something more sensible
For a user, the timeout of 1 min per message seems equivalent to a hang.
If journald cannot process a message from PID1 for 10 ms then something
is significantly wrong. It's better to lose the message and continue.
bus: when watching bus names, we always need to add a change match
Since for the kernel change of ownership of a name between starter and
real client is a change and for userspace thats addition/removal, we
need to unconditionally subscribe to these change events.
1) Instead of checking if we need to print a new line on each iteration,
pass the "new_line" as a pointer to those functions, so they can use
it to check if a new line is needed. This makes the code more consistent
as it is done in other places: machinectl, systemctl...
2) Move the error messages from show_{session|user|seat}() to their
appropriate print_{session|user|seat}_status_info() functions, this will
prevent from logging an error message twice in case show_properties()
fails and it will improve code readability.
3) Also do not ignore error codes on these functions.
Jason St. John [Wed, 18 Dec 2013 02:36:27 +0000 (21:36 -0500)]
journalctl,zsh-completion: fix several issues in --help message text
-- fix grammar and reword some descriptions for clarity
-- add a useful description of what --follow does
-- fix the description for --after-cursor
-- properly introduce the FSS acronym for "Forward Secure Sealing" in
both sections
-- clarify the --disk-usage command
[zj: perform similar changes to zsh completions]
squash! journalctl: fix several issues in --help message text
Remove -i option which would case exit(1) to happen.
Remove some unused code.
Convert to bool where appropriate.
Simplify things a bit.
Always free everything.
Kay Sievers [Wed, 18 Dec 2013 21:01:31 +0000 (22:01 +0100)]
libudev: disable monitor inside a container
Uevents are events of the host, which should not leak into a container.
Containers do not support hotplug at the moment, and devices and uevents
are not namespace aware.
units: run systemd-networkd.service only if CAP_NET_ADMIN capability is around
This has the effect that systemd-networkd won't run in containers
without network namespacing wher CAP_NET_ADMIN is (usually) not
available. It will still run in containers with network namespacing on
(where CAP_NET_ADMIN is usually avilable).
We might remove this condition check again if networkd provides services
to apps that also are useful in containers lacking network namespacing,
however, as long as it doesn't it should be handled like udevd and be
excluded in such containers.
In some circumstances, for example when start-up times out we
immediately jump into the final state, at which point we still should
try to watch the main pid so that the SIGCHLD allows us to quickly
move into dead state.
Thomas Bächler [Sun, 15 Dec 2013 11:06:37 +0000 (12:06 +0100)]
login: Don't stop a running user manager from garbage-collecting the user.
With the current logic, a user will never be garbage-collected, since its
manager will always be around. Change the logic such that a user is
garbage-collected when it has no sessions and linger is disabled.