chiark / gitweb /
7 years agosd-dhcp-client: log positive error number
Umut Tezduyar Lindskog [Sun, 27 Apr 2014 20:01:42 +0000 (22:01 +0200)]
sd-dhcp-client: log positive error number

Log error no for such client_stop(client, DHCP_EVENT_STOP)

7 years agolibnetworkd: add link local tests
Umut Tezduyar Lindskog [Sun, 27 Apr 2014 19:58:26 +0000 (21:58 +0200)]
libnetworkd: add link local tests

- 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.]

7 years agonetworkd: do not complain about IFF_RUNNING
Umut Tezduyar Lindskog [Sun, 27 Apr 2014 20:33:42 +0000 (22:33 +0200)]
networkd: do not complain about IFF_RUNNING

eth0: unknown link flags gained: 0x00040 (ignoring)

[tomegun: hiding these messages is ok, as IFF_RUNNING is redundant
          and can be deduced from operstate and IFF_LOWER_UP]

7 years agojob: add waiting jobs to run queue in unit_coldplug
Brandon Philips [Fri, 25 Apr 2014 15:31:59 +0000 (09:31 -0600)]
job: add waiting jobs to run queue in unit_coldplug

When we have job installed and added to run queue for service which is
still in dead state and systemd initiates reload then after reload we
never add deserialized job to the run queue again. This is caused by
check in service_coldplug() where we check if deserialized state is
something else than dead state, which is not the case thus we never call
service_set_state() and finally unit_notify() where we would have added
job to the run queue.

Thanks to Michal Sekletar <> for the original patch.

7 years agobuild-sys: add configure switch for -fsanitize=undefined
Zbigniew Jędrzejewski-Szmek [Thu, 24 Apr 2014 18:49:15 +0000 (20:49 +0200)]
build-sys: add configure switch for -fsanitize=undefined

--enable-undefined-sanitizer mirrors --enable-memory-sanitizer.

7 years agotest-journal-flush: avoid predictable names in /var/tmp
Zbigniew Jędrzejewski-Szmek [Fri, 25 Apr 2014 23:58:21 +0000 (19:58 -0400)]
test-journal-flush: avoid predictable names in /var/tmp

7 years agoman: networkd typo fixes
poma [Sat, 26 Apr 2014 09:09:45 +0000 (11:09 +0200)]
man: networkd typo fixes

7 years agonetworkd-wait-online: refactor a bit
Tom Gundersen [Thu, 24 Apr 2014 19:32:25 +0000 (21:32 +0200)]
networkd-wait-online: refactor a bit

Clarify that we are waiting for any link to reach 'carrier' state, regardless
of who manages it. This will be useful when we add support for waiting for more
operational states.

7 years agonetworkd-wait-online: drop config file and add commandline options instead
Tom Gundersen [Thu, 24 Apr 2014 19:28:46 +0000 (21:28 +0200)]
networkd-wait-online: drop config file and add commandline options instead

7 years agocore: reindent {selinux, ima, smack}-setup.c
Will Woods [Fri, 25 Apr 2014 22:26:33 +0000 (18:26 -0400)]
core: reindent {selinux, ima, smack}-setup.c

7-space indentation is just too weird to leave alone.
Make it 8 spaces, as per CODING_STYLE. No other changes.

7 years agoupdate TODO
Lennart Poettering [Fri, 25 Apr 2014 11:28:02 +0000 (13:28 +0200)]
update TODO

7 years agocore: expose CFS CPU time quota as high-level unit properties
Lennart Poettering [Fri, 25 Apr 2014 11:27:25 +0000 (13:27 +0200)]
core: expose CFS CPU time quota as high-level unit properties

7 years agocore: make sure we always write changed cgroup attributes to the cgroupfs
Lennart Poettering [Fri, 25 Apr 2014 11:26:37 +0000 (13:26 +0200)]
core: make sure we always write changed cgroup attributes to the cgroupfs

7 years agoupdate TODO
Lennart Poettering [Fri, 25 Apr 2014 05:16:01 +0000 (07:16 +0200)]
update TODO

7 years agobootchart: print to stdout and display default values in help
Zbigniew Jędrzejewski-Szmek [Thu, 24 Apr 2014 23:16:37 +0000 (19:16 -0400)]
bootchart: print to stdout and display default values in help

7 years agobootchart: rewrite usage message more generally
WaLyong Cho [Thu, 24 Apr 2014 15:50:52 +0000 (00:50 +0900)]
bootchart: rewrite usage message more generally

7 years agobootchart: add control group option
WaLyong Cho [Thu, 24 Apr 2014 15:50:51 +0000 (00:50 +0900)]
bootchart: add control group option

7 years agoudev: warn when name_to_handle_at is not implemented
Zbigniew Jędrzejewski-Szmek [Sun, 20 Apr 2014 17:57:26 +0000 (13:57 -0400)]
udev: warn when name_to_handle_at is not implemented

We have a bunch of reports from people who have a custom kernel and
are confused why udev is not running. Issue a warning on
error. Barring an error in the code, the only error that is possible

7 years agoservice: rename StartLimitAction enum to FailureAction
Michael Olbrich [Thu, 24 Apr 2014 07:35:38 +0000 (09:35 +0200)]
service: rename StartLimitAction enum to FailureAction

It's used for the FailureAction property as well.

7 years agoservice: add FailureAction= option
Michael Olbrich [Thu, 24 Apr 2014 07:35:37 +0000 (09:35 +0200)]
service: add FailureAction= option

It has the same possible values as StartLimitAction= and is executed
immediately if a service fails.

7 years agoudev: increase the size of RESULT buffer
Robert Milasan [Thu, 24 Apr 2014 09:23:33 +0000 (11:23 +0200)]
udev: increase the size of RESULT buffer

Under some conditions, in udev_rules_apply_to_event the fact that
result is 1024 bytes, creates problems if the output of the running
command/app is bigger then 1024 bytes.

7 years agobus: update kdbus.h
Kay Sievers [Thu, 24 Apr 2014 08:36:19 +0000 (10:36 +0200)]
bus: update kdbus.h

7 years agotest-resolve.c: Cleanup and make use of sockaddr_pretty()
Daniel Buch [Wed, 26 Mar 2014 13:17:07 +0000 (14:17 +0100)]
test-resolve.c: Cleanup and make use of sockaddr_pretty()

7 years agoFix keysize handling in cryptsetup (bits vs. bytes)
David Härdeman [Tue, 25 Mar 2014 10:05:28 +0000 (11:05 +0100)]
Fix keysize handling in cryptsetup (bits vs. bytes)

The command line key-size is in bits but the libcryptsetup API expects bytes.

Note that the modulo 8 check is in the original cryptsetup binary as well, so
it's no new limitation.

(v2: changed the point at which the /= 8 is performed, rebased, removed tabs)

7 years agoAdd more password agent information
David Härdeman [Tue, 25 Mar 2014 10:05:23 +0000 (11:05 +0100)]
Add more password agent information

Add an (optional) "Id" key in the password agent .ask files. The Id is
supposed to be a simple string in "<subsystem>:<target>" form which
is used to provide more information on what the requested passphrase
is to be used for (which e.g. allows an agent to only react to cryptsetup

(v2: rebased, fixed indentation, escape name, use strappenda)

7 years agoerrno: make sure to handle the 3 errnos that are aliases for others properly
Lennart Poettering [Thu, 24 Apr 2014 06:11:39 +0000 (08:11 +0200)]
errno: make sure to handle the 3 errnos that are aliases for others properly

7 years agoutil: make sure all our name_to_handle_at() code makes use of file_handle_union
Lennart Poettering [Thu, 24 Apr 2014 05:46:31 +0000 (07:46 +0200)]
util: make sure all our name_to_handle_at() code makes use of file_handle_union

7 years agonetworkd-wait-online: improve interoptability and enable by default
Tom Gundersen [Wed, 23 Apr 2014 15:42:55 +0000 (17:42 +0200)]
networkd-wait-online: improve interoptability and enable by default

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 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

7 years agohostnamectl: read kernel name and release from remote
Djalal Harouni [Wed, 23 Apr 2014 21:41:13 +0000 (22:41 +0100)]
hostnamectl: read kernel name and release from remote

7 years agohostnamed: expose KernelName and KernelRelease on the bus
Djalal Harouni [Wed, 23 Apr 2014 21:41:12 +0000 (22:41 +0100)]
hostnamed: expose KernelName and KernelRelease on the bus

7 years agoupdate TODO
Lennart Poettering [Wed, 23 Apr 2014 20:41:44 +0000 (22:41 +0200)]
update TODO

7 years agoanalyze: fix plot with bad y size
Jeffrey Clark [Wed, 23 Apr 2014 20:37:43 +0000 (22:37 +0200)]
analyze: fix plot with bad y size

systemd-analyze plot > test.svg produces output with all y and height
element attributes equal to zero. This of course causes the resulting
svg to appear blank (zero height). Bug does not affect x86. Looks like
a compiler optimization may be the culprit.

7 years agojournal-gatewayd: bugfix for invalid level case
Thomas Hindoe Paaboel Andersen [Wed, 23 Apr 2014 18:32:29 +0000 (20:32 +0200)]
journal-gatewayd: bugfix for invalid level case

introduced in cafc7f91306ea17ace4a6c3d76d81c8780c87452

7 years agoman: recommend that XDG_SESSION_DESKTOP and XDG_CURRENT_DESKTOP use the same identifiers
Lennart Poettering [Wed, 23 Apr 2014 18:11:38 +0000 (20:11 +0200)]
man: recommend that XDG_SESSION_DESKTOP and XDG_CURRENT_DESKTOP use the same identifiers

7 years agouse new name DRAW_TREE_VERTICAL
Thomas Hindoe Paaboel Andersen [Wed, 23 Apr 2014 17:22:39 +0000 (19:22 +0200)]

Was renamed in 6b01f1d3911bd7c7eadbb8a3b4375bd3ac05c98f

7 years agodelta: draw arrows with draw_special_char()
Lennart Poettering [Wed, 23 Apr 2014 17:05:46 +0000 (19:05 +0200)]
delta: draw arrows with draw_special_char()

Let's unify generation of unicode chars at one place.

Also, don't add an extra space into chars we print, except for the tree
chars where this is really necessary.

7 years agoinstall: simplification
Lennart Poettering [Wed, 23 Apr 2014 04:57:34 +0000 (06:57 +0200)]
install: simplification

7 years agolabel: there is no label_retest_selinux() call
Lennart Poettering [Wed, 23 Apr 2014 04:57:24 +0000 (06:57 +0200)]
label: there is no label_retest_selinux() call

7 years agobus: simplification
Lennart Poettering [Wed, 23 Apr 2014 04:57:00 +0000 (06:57 +0200)]
bus: simplification

7 years agobacklight: warn if kernel exposes backlight device with bogus max_brightness
Lennart Poettering [Wed, 23 Apr 2014 04:55:54 +0000 (06:55 +0200)]
backlight: warn if kernel exposes backlight device with bogus max_brightness

We shouldn't silently tape over broken kernel drivers.

7 years agoupdate TODO
Lennart Poettering [Tue, 22 Apr 2014 04:49:16 +0000 (06:49 +0200)]
update TODO

7 years agologin: rename session_mute_vt() to session_prepare_vt()
David Herrmann [Wed, 23 Apr 2014 15:38:43 +0000 (17:38 +0200)]
login: rename session_mute_vt() to session_prepare_vt()

This function is no longer just about muteing the VT. We do all kinds of
VT setup for sessions using the controller-API. Rename the function to
something more appropriate.

7 years agologin: give session ownership of VT
Ray Strode [Tue, 22 Apr 2014 17:27:58 +0000 (13:27 -0400)]
login: give session ownership of VT

The tty associated with a VT should be owned by the owner of the session
running on the VT. This is important for supporting a socket activated X
server, since the X server will open the tty itself.

This commit makes sure to chown the tty any time a session is
created and and chown it back to root any time the session
is removed. This behavior is copied from /bin/login.

7 years agobus: remove "on behalf" kdbus logic (ABI break)
Kay Sievers [Tue, 22 Apr 2014 20:52:19 +0000 (22:52 +0200)]
bus: remove "on behalf" kdbus logic (ABI break)

7 years agonetworkd: link - add missing curly brackets
Thomas Hindoe Paaboel Andersen [Tue, 22 Apr 2014 19:45:17 +0000 (21:45 +0200)]
networkd: link - add missing curly brackets

introduced in 1e9be60bbabe179f5233217384f1daec757c17c7

7 years agonetworkd: link - reduce default verbosity a bit
Tom Gundersen [Tue, 22 Apr 2014 17:51:27 +0000 (19:51 +0200)]
networkd: link - reduce default verbosity a bit

7 years agonetworkd: link - support IFLA_OPERSTATE
Tom Gundersen [Tue, 22 Apr 2014 17:40:05 +0000 (19:40 +0200)]
networkd: link - support IFLA_OPERSTATE

This properly detects the state of the link based on both the link flags and the

Moreover, always log state-changes even if we are not yet managing the link.

7 years agonetworkd: link - don't react to successfull link up
Tom Gundersen [Tue, 22 Apr 2014 17:36:40 +0000 (19:36 +0200)]
networkd: link - don't react to successfull link up

Avoid having two code-paths racing with eacother to do the same thing. The change
of flags will be detected in the normal way, so only use the link_up_handler
to detect if the 'up' failed and in that case fail the link.

7 years agonetworkd: link - log when starting clients fail
Tom Gundersen [Tue, 22 Apr 2014 17:29:14 +0000 (19:29 +0200)]
networkd: link - log when starting clients fail

7 years agonetworkd: link - avoid cycle when failing
Tom Gundersen [Tue, 22 Apr 2014 17:26:04 +0000 (19:26 +0200)]
networkd: link - avoid cycle when failing

7 years agonetworkd: link - stop clients when failing
Tom Gundersen [Tue, 22 Apr 2014 17:25:31 +0000 (19:25 +0200)]
networkd: link - stop clients when failing

7 years agoremove bus-driverd, the interface is now handled natively by bus-proxyd
Kay Sievers [Tue, 22 Apr 2014 17:24:41 +0000 (19:24 +0200)]
remove bus-driverd, the interface is now handled natively by bus-proxyd

7 years agobus-proxyd: handle org.freedesktob.DBus calls natively
Kay Sievers [Mon, 21 Apr 2014 19:16:26 +0000 (21:16 +0200)]
bus-proxyd: handle org.freedesktob.DBus calls natively

7 years agoRevert "networkd: netdev - set predictable mac address when creating netdev"
Tom Gundersen [Mon, 21 Apr 2014 20:08:04 +0000 (22:08 +0200)]
Revert "networkd: netdev - set predictable mac address when creating netdev"

This reverts commit cdc85c875b842b9309f72caefc51c262f521cf92.

There appears to be a kernel bug that (among other things) cause bridges
not to get a bridge id set when supplying IFLA_ADDRESS when creating the

Simply revert the whole thing until we sort this out in the kernel.


Reported-by: C. R. Oldham <>
7 years agoservice: add support for reboot argument when triggered by StartLimitAction=
Michael Olbrich [Tue, 25 Mar 2014 13:15:45 +0000 (14:15 +0100)]
service: add support for reboot argument when triggered by StartLimitAction=

When rebooting with systemctl, an optional argument can be passed to the
reboot system call. This makes it possible the specify the argument in a
service file and use it when the service triggers a restart.
This is useful to distinguish between manual reboots and reboots caused by
failing services.

7 years agosystemctl: delete REBOOT_PARAM_FILE if no parameter is specified
Michael Olbrich [Tue, 25 Mar 2014 13:15:44 +0000 (14:15 +0100)]
systemctl: delete REBOOT_PARAM_FILE if no parameter is specified

And move it to sperate function.

7 years agoimplement a union to pad out file_handle
Dave Reisner [Sat, 19 Apr 2014 17:22:35 +0000 (13:22 -0400)]
implement a union to pad out file_handle

Cases where name_to_handle_at is used allocated the full struct to be
MAX_HANDLE_SZ, and assigned this size to handle_bytes. This is wrong
since handle_bytes should describe the length of the flexible array
member and not the whole struct.

Define a union type which includes sufficient padding to allow
assignment of MAX_HANDLE_SZ to be correct.

7 years agocondense assignment and path_kill_slashes calls
Dave Reisner [Wed, 16 Apr 2014 02:16:40 +0000 (22:16 -0400)]
condense assignment and path_kill_slashes calls

7 years agodoc: fix items' names in PORTING-DBUS1
Lukasz Skalski [Thu, 10 Apr 2014 12:50:23 +0000 (14:50 +0200)]
doc: fix items' names in PORTING-DBUS1

7 years agoRemove duplicate includes
Bas van den Berg [Tue, 8 Apr 2014 09:14:16 +0000 (11:14 +0200)]
Remove duplicate includes

7 years agosilence warnings
Thomas Hindoe Paaboel Andersen [Sun, 20 Apr 2014 22:02:35 +0000 (00:02 +0200)]
silence warnings

7 years agonetworkd: link - add explicit unmanaged state
Tom Gundersen [Sun, 20 Apr 2014 17:49:00 +0000 (19:49 +0200)]
networkd: link - add explicit unmanaged state

7 years agonetworkd: link - set mac addresses when starting clients
Tom Gundersen [Sun, 20 Apr 2014 17:34:51 +0000 (19:34 +0200)]
networkd: link - set mac addresses when starting clients

MAC addreses are also set when NEWLINK messages are receieved, but only if the clients have
already been initialized.

7 years agosd-dhcp-client: log ifindex and mac address when starting
Tom Gundersen [Sun, 20 Apr 2014 17:10:02 +0000 (19:10 +0200)]
sd-dhcp-client: log ifindex and mac address when starting

7 years agonetworkd: link - log changes in some common link flags
Tom Gundersen [Sun, 20 Apr 2014 11:08:59 +0000 (13:08 +0200)]
networkd: link - log changes in some common link flags

7 years agonetworkd: link - use IFF_DORMANT instead of IFF_RUNNING
Tom Gundersen [Sun, 20 Apr 2014 08:28:19 +0000 (10:28 +0200)]
networkd: link - use IFF_DORMANT instead of IFF_RUNNING

Should be no functional change, but avoids spurious log
messages when in operstate UNKNOWN and !IFF_LOWER_UP.

7 years agonetworkd: link - make debug logging a bit more readable
Tom Gundersen [Sun, 20 Apr 2014 10:59:24 +0000 (12:59 +0200)]
networkd: link - make debug logging a bit more readable

Drop leading zeros which will never be used when printing the link flags.

7 years agoudev: link-config - treat any non-random mac address as persistent
Tom Gundersen [Sun, 20 Apr 2014 05:53:16 +0000 (07:53 +0200)]
udev: link-config - treat any non-random mac address as persistent

The address may be already changed by other processes, or set
when creating the netdev from userspace.

7 years agosd-rtnl: log when queues are exhausted
Tom Gundersen [Sun, 20 Apr 2014 05:51:28 +0000 (07:51 +0200)]
sd-rtnl: log when queues are exhausted

7 years agocryptsetup: copy value, not key for (rd.)luks.key
Matthew Monaco [Sat, 19 Apr 2014 18:56:45 +0000 (12:56 -0600)]
cryptsetup: copy value, not key for (rd.)luks.key

7 years agoHandle Unix domain socket connections from outside our namespace v2
Zbigniew Jędrzejewski-Szmek [Sun, 20 Apr 2014 01:58:03 +0000 (21:58 -0400)]
Handle Unix domain socket connections from outside our namespace v2

This is a second attempt at 9754d56, reverted in 2f20a8e, because
I lost a 'break;' when moving chunks around.

7 years agonetworkd: order after udev kernel socket
Tom Gundersen [Sat, 19 Apr 2014 20:13:49 +0000 (22:13 +0200)]
networkd: order after udev kernel socket

Otherwise we will not be able to queuery whether devices are initialized on kdbus enabled systems.

7 years agonetworkd: manager - add a bit of debug output to udev ADD events
Tom Gundersen [Sat, 19 Apr 2014 18:49:06 +0000 (20:49 +0200)]
networkd: manager - add a bit of debug output to udev ADD events

7 years agonetworkd: link - track interface renames
Tom Gundersen [Sat, 19 Apr 2014 18:39:17 +0000 (20:39 +0200)]
networkd: link - track interface renames

7 years agosd-rtnl: socet_read_message - make sure we only shrink the partial rqueue when proces...
Tom Gundersen [Sat, 19 Apr 2014 18:32:01 +0000 (20:32 +0200)]
sd-rtnl: socet_read_message - make sure we only shrink the partial rqueue when processing multi-part message

7 years agonetwork: implement masking of .link, .network and .netdev files
Tom Gundersen [Sat, 19 Apr 2014 17:05:37 +0000 (19:05 +0200)]
network: implement masking of .link, .network and .netdev files

7 years agonetworkd: netdev - give up early when setting already set ifindex
Tom Gundersen [Sat, 19 Apr 2014 15:14:36 +0000 (17:14 +0200)]
networkd: netdev - give up early when setting already set ifindex

We may receive RTM_NEWLINK messages with missing LINKINFO after the initial NEWLINK message,
don't bother verifying these, just drop out early after checking that the ifindex is not in conflict.

7 years agonetworkd: netdev - verify name of newlink messages
Tom Gundersen [Fri, 18 Apr 2014 08:28:42 +0000 (10:28 +0200)]
networkd: netdev - verify name of newlink messages

7 years agonetworkd: netdev - set predictable mac address when creating netdev
Tom Gundersen [Thu, 17 Apr 2014 22:50:40 +0000 (00:50 +0200)]
networkd: netdev - set predictable mac address when creating netdev

7 years agonetworkd: tie links to rtnl rather than udev
Tom Gundersen [Tue, 15 Apr 2014 12:21:44 +0000 (14:21 +0200)]
networkd: tie links to rtnl rather than udev

This essentially swaps the roles of rtnl and udev in networkd. After this
change libudev is only used for waiting for udev to initialize devices and
to get udev-specific information needed for some [Match] attributes.

This in particular simplifies the code in containers where udev is not really
useful, but also simplifies things and reduces round-trips in the non-container

7 years agonss-myhostname: port to sd-rtnl
Tom Gundersen [Mon, 14 Apr 2014 10:46:09 +0000 (12:46 +0200)]
nss-myhostname: port to sd-rtnl

7 years agosd-rtnl: message - expose DUMP flag in the api
Tom Gundersen [Mon, 14 Apr 2014 16:07:34 +0000 (18:07 +0200)]
sd-rtnl: message - expose DUMP flag in the api

7 years agosd-rtnl: add multi-part message test
Tom Gundersen [Thu, 17 Apr 2014 09:12:41 +0000 (11:12 +0200)]
sd-rtnl: add multi-part message test

7 years agosd-rtnl: message - concatenate multi-part messages from different packets
Tom Gundersen [Thu, 17 Apr 2014 19:32:25 +0000 (21:32 +0200)]
sd-rtnl: message - concatenate multi-part messages from different packets

7 years agosd-rtnl: message - don't put NULL message on rqueue
Tom Gundersen [Wed, 16 Apr 2014 16:17:24 +0000 (18:17 +0200)]
sd-rtnl: message - don't put NULL message on rqueue

If nothing interesting was receieved we should not put anything on
the queue.

7 years agosd-rtnl: message - make room for all pending messages, not just the first
Tom Gundersen [Tue, 15 Apr 2014 10:01:39 +0000 (12:01 +0200)]
sd-rtnl: message - make room for all pending messages, not just the first

Also, don't actually read any of the message when peeking, just get its length.

7 years agoRevert "Handle Unix domain socket connections from outside our namespace"
Kay Sievers [Sat, 19 Apr 2014 11:53:11 +0000 (13:53 +0200)]
Revert "Handle Unix domain socket connections from outside our namespace"

This reverts commit 9754d56e9b21bfe89fc18f47987d6bef491b8521.

It causes a crash in PID1:
  Apr 19 13:49:32 lon systemd[1]: Code should not be reached 'Unhandled socket type.'
                                  at src/core/socket.c:684, function instance_from_socket(). Aborting.
  Apr 19 13:49:32 lon systemd[1]: Caught <ABRT>, dumped core as pid 336.
  Apr 19 13:49:32 lon systemd[1]: Freezing execution.

7 years agotmpfiles: fix permissions on new journal files
Greg KH [Tue, 15 Apr 2014 21:12:01 +0000 (14:12 -0700)]
tmpfiles: fix permissions on new journal files

When starting up journald on a new system, set the proper permissions on
the system.journal files, not only on the journal directory.

7 years agobuild-sys: do not try to build test-rtnl-manual w/o kmod
Zbigniew Jędrzejewski-Szmek [Fri, 18 Apr 2014 03:39:40 +0000 (23:39 -0400)]
build-sys: do not try to build test-rtnl-manual w/o kmod

7 years agobuild-sys: add libkmod flags for test_rtnl_manual
Łukasz Stelmach [Mon, 14 Apr 2014 13:14:14 +0000 (15:14 +0200)]
build-sys: add libkmod flags for test_rtnl_manual

7 years agocore: minor typo fix
Ali H. Caliskan [Mon, 14 Apr 2014 17:30:16 +0000 (19:30 +0200)]
core: minor typo fix

7 years agodelta: do not use unicode chars in C locale
Zbigniew Jędrzejewski-Szmek [Thu, 17 Apr 2014 03:33:41 +0000 (23:33 -0400)]
delta: do not use unicode chars in C locale

7 years agoinstall: create_symlink() check unlink() return value
Djalal Harouni [Thu, 17 Apr 2014 00:47:11 +0000 (01:47 +0100)]
install: create_symlink() check unlink() return value

create_symlink() do not check the return value of unlink(), this may
confuse the user.

Before the unlink() call we check the 'force' argument. If it is not set
we fail with -EEXIST, otherwise we unlink() the file, therefore the next
symlink() should not fail with -EEXIST (do not count races...).

However since callers may not have appropriate privileges to unlink()
the file we lose the -EPERM or any other errno code of unlink(), and
return the -EEXIST of the next symlink(). Fix this by checking unlink()

$ systemctl --force --root=~/container-03 set-default
Failed to set default target: File exists

$ systemctl --force --root=~/container-03 set-default
Failed to set default target: Permission denied

7 years agosystemctl: allow to change the default target without the --force switch
Djalal Harouni [Mon, 14 Apr 2014 00:07:52 +0000 (01:07 +0100)]
systemctl: allow to change the default target without the --force switch

Currently "systemctl set-default" will fail to change the default target
due to the '' being a symlink which is always the case.

To work around this, the user must specify the "--force" switch to be
able to overwrite the existing symlink.

This is clearly a regression that was introduced by commit  718db96199e
since it worked before without the "--force" switch and the man pages do
not mention that you need to specify it. It is expected that this is a

So just explicity set the force flag to make it work again.

Reported-by: <>
7 years agoHandle Unix domain socket connections from outside our namespace
Eelco Dolstra [Wed, 16 Apr 2014 16:39:07 +0000 (18:39 +0200)]
Handle Unix domain socket connections from outside our namespace

NixOS uses Unix domain sockets for certain host <-> container
interaction; i.e. the host connects to a socket visible in the
container's directory tree, where the container uses a .socket unit to
spawn the handler program on demand. This worked in systemd 203, but
in 212 fails with "foo.socket failed to queue service startup job
(Maybe the service file is missing or not a template unit?): No data

The reason is that getpeercred() now returns ENODATA if it can't get
the PID of the client, which happens in this case because the client
is not in the same PID namespace. Since getpeercred() is only used to
generate the instance name, this patch simply handles ENODATA by
creating an instance name "<nr>-unknown".

[zj: reorder clauses and remove (unsigned long) casts.]

7 years agoman: document relationship between RequiresMountsFor and noauto
Zbigniew Jędrzejewski-Szmek [Thu, 17 Apr 2014 02:15:42 +0000 (22:15 -0400)]
man: document relationship between RequiresMountsFor and noauto

7 years agotest-path-util: fix running with separate build dir
Zbigniew Jędrzejewski-Szmek [Thu, 17 Apr 2014 02:04:31 +0000 (22:04 -0400)]
test-path-util: fix running with separate build dir

test-path-utils attempts to find itself, but if the binary is not
in current directory, the test would fail.

7 years agosysctl: replaces some slashes with dots
Zbigniew Jędrzejewski-Szmek [Thu, 17 Apr 2014 01:33:46 +0000 (21:33 -0400)]
sysctl: replaces some slashes with dots

It turns out that plain sysctl understands a.b/c syntax to write to
/proc/sys/a/b.c. Support this for compatibility.

7 years agosystemctl: add --recursive for list-timers
Thomas Hindoe Paaboel Andersen [Wed, 16 Apr 2014 21:50:19 +0000 (23:50 +0200)]
systemctl: add --recursive for list-timers

7 years agosystemctl: add --recursive for list-sockets
Thomas Hindoe Paaboel Andersen [Tue, 15 Apr 2014 22:37:17 +0000 (00:37 +0200)]
systemctl: add --recursive for list-sockets