chiark / gitweb /
7 years agocgroup: make sure to properly send SIGCONT to all processes of a cgroup if that's...
Lennart Poettering [Thu, 6 Feb 2014 18:27:59 +0000 (19:27 +0100)]
cgroup: make sure to properly send SIGCONT to all processes of a cgroup if that's requested

7 years agologind: rework session shutdown logic
Lennart Poettering [Thu, 6 Feb 2014 17:32:14 +0000 (18:32 +0100)]
logind: rework session shutdown logic

Simplify the shutdown logic a bit:

- Keep the session FIFO around in the PAM module, even after the session
  shutdown hook has been finished. This allows logind to track precisely
  when the PAM handler goes away.

- In the ReleaseSession() call start a timer, that will stop terminate
  the session when elapsed.

- Never fiddle with the KillMode of scopes to configure whether user
  processes should be killed or not. Instead, simply leave the scope
  units around when we terminate a session whose processes should not be

- When killing is enabled, stop the session scope on FIFO EOF or after
  the ReleaseSession() timeout. When killing is disabled, simply tell
  PID 1 to abandon the scope.

Because the scopes stay around and hence all processes are always member
of a scope, the system shutdown logic should be more robust, as the
scopes can be shutdown as part of the usual shutdown logic.

7 years agocore: watch SIGCHLD more closely to track processes of units with no reliable cgroup...
Lennart Poettering [Thu, 6 Feb 2014 16:17:51 +0000 (17:17 +0100)]
core: watch SIGCHLD more closely to track processes of units with no reliable cgroup empty notifier

When a process dies that we can associate with a specific unit, start
watching all other processes of that unit, so that we can associate
those processes with the unit too.

Also, for service units start doing this as soon as we get the first
SIGCHLD for either control or main process, so that we can follow the
processes of the service from one to the other, as long as process that
remain are processes of the ones we watched that died and got reassigned
to us as parent.

Similar, for scope units start doing this as soon as the scope
controller abandons the unit, and thus management entirely reverts to
systemd. To abandon a unit introduce a new Abandon() scope unit method

7 years agocore: fix warning
Thomas Hindoe Paaboel Andersen [Thu, 6 Feb 2014 23:01:19 +0000 (00:01 +0100)]
core: fix warning

introduced in c7040b5d1c2c148f12b6a5eef3dfce1661805131

7 years agoshell-completion: fix completion of localectl set-locale
Zbigniew Jędrzejewski-Szmek [Thu, 6 Feb 2014 16:59:33 +0000 (11:59 -0500)]
shell-completion: fix completion of localectl set-locale

7 years agonspawn: fix HAVE_SELINUX ifdef
Tom Gundersen [Wed, 5 Feb 2014 16:07:20 +0000 (17:07 +0100)]
nspawn: fix HAVE_SELINUX ifdef

7 years agotransaction: print more information about conflicting jobs
Zbigniew Jędrzejewski-Szmek [Thu, 6 Feb 2014 05:37:18 +0000 (00:37 -0500)]
transaction: print more information about conflicting jobs

Also remove some debug statement that should not have been committed.

7 years agobash-completion: fix completion of complete verbs
Zbigniew Jędrzejewski-Szmek [Thu, 6 Feb 2014 05:31:22 +0000 (00:31 -0500)]
bash-completion: fix completion of complete verbs

When doing 'command verb<TAB>', the arguments for verb would be
proposed, but it is too early. We should complete verb first.

7 years agocore: only send SIGHUP when doing first kill, not when doing final sigkill
Lennart Poettering [Thu, 6 Feb 2014 00:50:41 +0000 (01:50 +0100)]
core: only send SIGHUP when doing first kill, not when doing final sigkill

7 years agoudev: add zram to the list of devices inappropriate for symlinks
Jóhann B. Guðmundsson [Sun, 2 Feb 2014 13:29:19 +0000 (13:29 +0000)]
udev: add zram to the list of devices inappropriate for symlinks

udev seems to have a race condition with swapon to see which can open
/dev/zram0 first, causing swapon to fail. Seems to be most noticeable
on arm devices one out of every 7 times or something.

7 years agoUpdate some message formats
Zbigniew Jędrzejewski-Szmek [Tue, 4 Feb 2014 00:31:53 +0000 (19:31 -0500)]
Update some message formats

Use PID_FMT/USEC_FMT/... in more places.

Also update logind error messages to print the full path to a file that
failed. This should make debugging easier for people who do not know
off the top of their head where logind stores it state.

7 years agonspawn: add --quiet switch for turning off any output noise
Lennart Poettering [Wed, 5 Feb 2014 23:43:14 +0000 (00:43 +0100)]
nspawn: add --quiet switch for turning off any output noise

7 years agonspawn: always use default bus
Lennart Poettering [Wed, 5 Feb 2014 22:06:34 +0000 (23:06 +0100)]
nspawn: always use default bus

7 years agobus: properly unset default bus pointer when destroying last reference
Lennart Poettering [Wed, 5 Feb 2014 22:06:13 +0000 (23:06 +0100)]
bus: properly unset default bus pointer when destroying last reference

7 years agoman: introduce new "Desktop" property for sessions
Lennart Poettering [Wed, 5 Feb 2014 19:34:11 +0000 (20:34 +0100)]
man: introduce new "Desktop" property for sessions

This is initialized from XDG_SESSION_DESKTOP and is useful for GNOME
to recognize its own sessions. It's supposed to be set to a short string
identifying the session, such as "kde" or "gnome".

7 years agologind: make session type and class settable via the same ways
Lennart Poettering [Wed, 5 Feb 2014 17:55:18 +0000 (18:55 +0100)]
logind: make session type and class settable via the same ways

If the session type/class is set via environment variables, use that,
and otherwise fallback to something that is set via the PAM module
command line.

7 years agologind: add new "wayland" session type
Lennart Poettering [Wed, 5 Feb 2014 17:27:43 +0000 (18:27 +0100)]
logind: add new "wayland" session type

7 years agocore: don't wait for non-control/non-main processes when killing processes on the...
Lennart Poettering [Wed, 5 Feb 2014 01:22:11 +0000 (02:22 +0100)]
core: don't wait for non-control/non-main processes when killing processes on the host either

Since the current kernel cgroup notification logic is easily confused by
existing subgroups, let's do the same thing as in containers before. and
just not wait for non-control and non-main processes.

This should be corrected as soon as we have sane cgroup notifications
from the kernel.

7 years agokill: fix error return
Lennart Poettering [Wed, 5 Feb 2014 01:11:18 +0000 (02:11 +0100)]
kill: fix error return

7 years agocore: allow User=, Group=, Nice=, Environment=, Type= to be passed when creating...
Lennart Poettering [Wed, 5 Feb 2014 01:02:00 +0000 (02:02 +0100)]
core: allow User=, Group=, Nice=, Environment=, Type= to be passed when creating a transient service

7 years agoAdded attribute support for sd-rtnl
Susant Sahani [Tue, 4 Feb 2014 08:49:20 +0000 (14:19 +0530)]
Added attribute support for sd-rtnl

  Added sd_rtnl_message_append_u8 and
  few attribute support in sd_rtnl_message_append_u32

7 years agonspawn: various fixes in selinux hookup
Lennart Poettering [Tue, 4 Feb 2014 21:56:07 +0000 (22:56 +0100)]
nspawn: various fixes in selinux hookup

- As suggested, prefix argument variables with "arg_" how we do this

- As suggested, don't involve memory allocations when storing command
  line arguments.

- Break --help text at 80 chars

- man: explain that this is about SELinux

- don't do unnecessary memory allocations when putting together mount
  option string

7 years agoAdd SELinux support to systemd-nspawn
Dan Walsh [Thu, 30 Jan 2014 21:28:02 +0000 (16:28 -0500)]
Add SELinux support to systemd-nspawn

This patch adds to new options:


This specifies the process label to run on processes run within the container.


The file label to assign to memory file systems created within the container.

For example if you wanted to wrap an container with SELinux sandbox labels, you could execute a command line the following

chcon system_u:object_r:svirt_sandbox_file_t:s0:c0,c1 -R /srv/container
systemd-nspawn -L system_u:object_r:svirt_sandbox_file_t:s0:c0,c1 -Z system_u:system_r:svirt_lxc_net_t:s0:c0,c1 -D /srv/container /bin/sh

7 years agojournal: Drop pkgconfig reference to libsystemd-id128.
Colin Guthrie [Thu, 30 Jan 2014 12:58:24 +0000 (13:58 +0100)]
journal: Drop pkgconfig reference to libsystemd-id128.

This is now part of libsystemd.

7 years agobus: when closing the bus don't end up in a recursive destruction deadlock
Lennart Poettering [Mon, 3 Feb 2014 12:26:24 +0000 (13:26 +0100)]
bus: when closing the bus don't end up in a recursive destruction deadlock

7 years agoconf-parser: warn when we open configuration files with weird access bits
Lennart Poettering [Mon, 3 Feb 2014 11:52:16 +0000 (12:52 +0100)]
conf-parser: warn when we open configuration files with weird access bits

7 years agobuild-sys: move -lresolv out of CFLAGS
Tom Gundersen [Mon, 3 Feb 2014 14:25:45 +0000 (15:25 +0100)]
build-sys: move -lresolv out of CFLAGS

Thomas H.P. Andersen <> wrote:
> Does -lresolv belong in libsystemd_la_CFLAGS? I would have thought
> that it should be in LIBADD for the lib and LDADD for the test.

7 years agobus: update kdbus.h (ABI break)
Kay Sievers [Sat, 1 Feb 2014 13:41:13 +0000 (14:41 +0100)]
bus: update kdbus.h (ABI break)

7 years agonetworkd: remove unused variable
Tom Gundersen [Sat, 1 Feb 2014 15:27:41 +0000 (16:27 +0100)]
networkd: remove unused variable

7 years agonetworkd: don't fail on DHCP errors
Tom Gundersen [Sat, 1 Feb 2014 15:27:13 +0000 (16:27 +0100)]
networkd: don't fail on DHCP errors

7 years agos390/getty-generator: initialize essential system terminals/consoles
Hendrik Brueckner [Fri, 31 Jan 2014 16:08:37 +0000 (17:08 +0100)]
s390/getty-generator: initialize essential system terminals/consoles

Ensure to start getty programs on all essential system consoles on Linux on
System z.  Add these essential devices to the list of virtualization_consoles
to always generate getty configurations.

For the sake of completion, the list of essential consoles is:

  /dev/sclp_line0 - Operating system messages applet (LPAR)
  /dev/ttysclp0 - Integrated ASCII console applet (z/VM and LPAR)
  /dev/ttyS0 - Already handled by systemd (3215 console on z/VM)
  /dev/hvc0  - Already handled by systemd (IUCV HVC terminal on z/VM)

Depending on the environment, z/VM or LPAR, only a subset of these terminals
are available.

See also RH BZ 860158[1] "Cannot login via Operating System Console into RHEL7
instance installed on a LPAR".  This bugzilla actually blocks the installation
of Linux on System z instances in LPAR mode.


7 years agorun: drop mistakenly committed test code
Lennart Poettering [Fri, 31 Jan 2014 17:03:15 +0000 (18:03 +0100)]
run: drop mistakenly committed test code

7 years agocore: fix oom check
Lennart Poettering [Fri, 31 Jan 2014 16:47:22 +0000 (17:47 +0100)]
core: fix oom check

7 years agocore: introduce new stop protocol for unit scopes
Lennart Poettering [Fri, 31 Jan 2014 16:45:13 +0000 (17:45 +0100)]
core: introduce new stop protocol for unit scopes

By specifiy a Controller property when creating the scope a client can
specify a bus name that will be notified with a RequestStop bus signal
when the scope has been asked to shut down, instead of sending SIGTERM
to the scope processes themselves.

7 years agoutil: use alloca0() intead of alloca() + memzero()
Lennart Poettering [Fri, 31 Jan 2014 11:27:35 +0000 (12:27 +0100)]
util: use alloca0() intead of alloca() + memzero()

7 years agonetworkd: dhcp - refactor handler
Tom Gundersen [Fri, 31 Jan 2014 02:10:27 +0000 (03:10 +0100)]
networkd: dhcp - refactor handler

7 years agonetworkd: don't check for ifindex in answer to RTM_NEWLINK
Tom Gundersen [Fri, 31 Jan 2014 01:03:39 +0000 (02:03 +0100)]
networkd: don't check for ifindex in answer to RTM_NEWLINK

The kernel will not be changed to support this, so drop the code. Listening
for all RTM_NEWLINK messages and filtering on name is reliable, so it is not
a problem.

7 years agolibsystemd-dhcp: Update secs field only when sending Discover
Patrik Flykt [Fri, 31 Jan 2014 09:31:25 +0000 (11:31 +0200)]
libsystemd-dhcp: Update secs field only when sending Discover

Compute the 'secs' field as seconds since start of lease acquisition
procedure. Start off with a value of zero and increase it only for
each resent DHCP discover message. See the discussion before and
and Section 3.1 of RFC 2131.

7 years agolibsystemd-dhcp: Fix stopping of DHCP client
Patrik Flykt [Fri, 31 Jan 2014 09:31:24 +0000 (11:31 +0200)]
libsystemd-dhcp: Fix stopping of DHCP client

Go back to Init state independent of the current state the client
is in.

7 years agolibsystemd-dhcp: Compute UDP checksum only if set
Patrik Flykt [Fri, 31 Jan 2014 09:31:23 +0000 (11:31 +0200)]
libsystemd-dhcp: Compute UDP checksum only if set

A checksum field with value zero means no UDP checksum has been
computed for the packet.

7 years agolibsystemd-dhcp: Rename function to be clearer that options are parsed
Patrik Flykt [Fri, 31 Jan 2014 09:31:21 +0000 (11:31 +0200)]
libsystemd-dhcp: Rename function to be clearer that options are parsed

7 years agolibsystemd-dhcp: DNS name option must be a multiple of 4 bytes
Patrik Flykt [Fri, 31 Jan 2014 09:31:20 +0000 (11:31 +0200)]
libsystemd-dhcp: DNS name option must be a multiple of 4 bytes

7 years agouse memzero(foo, length); for all memset(foo, 0, length); calls
Greg KH [Fri, 31 Jan 2014 05:51:32 +0000 (06:51 +0100)]
use memzero(foo, length); for all memset(foo, 0, length); calls

In trying to track down a stupid linker bug, I noticed a bunch of
memset() calls that should be using memzero() to make it more "obvious"
that the options are correct (i.e. 0 is not the length, but the data to
set).  So fix up all current calls to memset(foo, 0, length) to
memzero(foo, length).

7 years agoanalyze: fix plot issues when using gummiboot
Thomas Hindoe Paaboel Andersen [Fri, 31 Jan 2014 06:07:20 +0000 (07:07 +0100)]
analyze: fix plot issues when using gummiboot

It would crash and the legend in the bottom followed the time 0.0.

7 years agotmpfiles: fix memory leak of exclude_prefixes
Zbigniew Jędrzejewski-Szmek [Fri, 31 Jan 2014 02:40:27 +0000 (21:40 -0500)]
tmpfiles: fix memory leak of exclude_prefixes

Missed in 5c795114.

7 years agoshared: net - use u32ctz to compute prefixlen
Tom Gundersen [Thu, 30 Jan 2014 16:23:34 +0000 (17:23 +0100)]
shared: net - use u32ctz to compute prefixlen

7 years agoTODO: update sd-rtnl section
Tom Gundersen [Thu, 30 Jan 2014 15:46:27 +0000 (16:46 +0100)]
TODO: update sd-rtnl section

7 years agoutil: add u32ctz() call for determining ctz of uint32_t
Lennart Poettering [Thu, 30 Jan 2014 15:46:48 +0000 (16:46 +0100)]
util: add u32ctz() call for determining ctz of uint32_t

7 years agonetworkd: netdev - reduce chance of race when receiving netdev's ifindex
Tom Gundersen [Wed, 29 Jan 2014 20:24:44 +0000 (21:24 +0100)]
networkd: netdev - reduce chance of race when receiving netdev's ifindex

When creating a new link, the kernel will not inform us about the new ifindex
in its ack. We have to listen for newly created devices and deduce the new
ifindex by matching on the ifname.

We used to do this by waiting for a new device from libudev, but that is asking
for trouble, as udev will happily rename the device before handing it to us.
Listen on rtnl instead, the chance of the name being changed before reaching us
is much smaller (if not nil).

Kernel patch in the works to make this unneccessary.

7 years agosd-rtnl: beef up rtnl-util a bit
Tom Gundersen [Wed, 29 Jan 2014 20:20:30 +0000 (21:20 +0100)]
sd-rtnl: beef up rtnl-util a bit

7 years agosd-dhcp-client/net-util: make netmask_to_prefixlen generic
Tom Gundersen [Tue, 28 Jan 2014 22:23:31 +0000 (23:23 +0100)]
sd-dhcp-client/net-util: make netmask_to_prefixlen generic

This was originally included in the dhcp-client at my request, but it is not
really dhcp-specific and useful outside of it, so let's pull it out.

7 years agonetworkd: address - add support for broadcast
Tom Gundersen [Tue, 28 Jan 2014 19:01:37 +0000 (20:01 +0100)]
networkd: address - add support for broadcast

7 years agonet-util: verify the address family
Tom Gundersen [Tue, 28 Jan 2014 19:00:47 +0000 (20:00 +0100)]
net-util: verify the address family

Error out if the address family is already set to something incompatible with the
address being parsed.

7 years agonetworkd: dhcpv4 - add notion of 'CriticalConnection'
Tom Gundersen [Mon, 27 Jan 2014 19:52:07 +0000 (20:52 +0100)]
networkd: dhcpv4 - add notion of 'CriticalConnection'

These connections are never torn down, even when the DHCP specifications say that
they should be. This is useful/necessary when the rootfs (or another critical fs)
is mounted over this network connection, and dataloss would result if the connection
is lost.

This option defaults to off, but our initrd generator (TBD) will enable it when

7 years agoutmp: make sure we don't write the utmp reboot record twice on each boot
Lennart Poettering [Thu, 30 Jan 2014 13:18:46 +0000 (14:18 +0100)]
utmp: make sure we don't write the utmp reboot record twice on each boot

(Also, only send the audit msg once, too)

7 years agoupdate-utmp: code modernizations
Lennart Poettering [Thu, 30 Jan 2014 12:28:56 +0000 (13:28 +0100)]
update-utmp: code modernizations

7 years agoupdate TODO
Lennart Poettering [Wed, 29 Jan 2014 20:10:41 +0000 (21:10 +0100)]
update TODO

7 years agobus: when proxying messages from the bus driver patch the driver's well-known name...
Lennart Poettering [Wed, 29 Jan 2014 20:23:41 +0000 (21:23 +0100)]
bus: when proxying messages from the bus driver patch the driver's well-known name into the sender

7 years agobus: the owner of the bus driver name as reported by the driver's
Lennart Poettering [Wed, 29 Jan 2014 20:07:50 +0000 (21:07 +0100)]
bus: the owner of the bus driver name as reported by the driver's
GetNameOwner() bus call is the bus driver name itself, for compatibility
with dbus1

7 years agonspawn: fix reboot event fd reuse
Lennart Poettering [Wed, 29 Jan 2014 19:58:50 +0000 (20:58 +0100)]
nspawn: fix reboot event fd reuse

7 years agocore: in containers, don't wait for cgroup empty notifications which will never come
Lennart Poettering [Wed, 29 Jan 2014 19:12:18 +0000 (20:12 +0100)]
core: in containers, don't wait for cgroup empty notifications which will never come

7 years agocore: use a bit more PID_FMT
Lennart Poettering [Wed, 29 Jan 2014 13:59:45 +0000 (14:59 +0100)]
core: use a bit more PID_FMT

7 years agocore: make sure to always go through both SIGTERM and SIGKILL states of units
Lennart Poettering [Wed, 29 Jan 2014 13:58:04 +0000 (14:58 +0100)]
core: make sure to always go through both SIGTERM and SIGKILL states of units

Given that we now have KillMode=mixed where SIGTERM might kill a smaller
set than SIGKILL we need to make sure to always go explicitly throught
the SIGKILL state to get the right end result.

7 years agokeymap: Add Toshiba EQUIUM
Martin Pitt [Wed, 29 Jan 2014 15:13:22 +0000 (16:13 +0100)]
keymap: Add Toshiba EQUIUM

Thanks Aleksander Kowalski <>!

7 years agoservice: allow KillMode=mixed in conjunction with PAMName=
Lennart Poettering [Wed, 29 Jan 2014 12:49:54 +0000 (13:49 +0100)]
service: allow KillMode=mixed in conjunction with PAMName=

7 years agocore: introduce new KillMode=mixed which sends SIGTERM only to the main process,...
Lennart Poettering [Wed, 29 Jan 2014 12:38:55 +0000 (13:38 +0100)]
core: introduce new KillMode=mixed which sends SIGTERM only to the main process, but SIGKILL to all daemon processes

This should fix some race with terminating systemd --user, where the
system systemd instance might race against the user systemd instance
when sending SIGTERM.

7 years agobus: fake security labels again
Lennart Poettering [Tue, 28 Jan 2014 20:38:14 +0000 (21:38 +0100)]
bus: fake security labels again

(The kernel module got fixed, so let's reenable this again)

7 years agoman: remove erroneous history section
Zbigniew Jędrzejewski-Szmek [Wed, 29 Jan 2014 03:02:12 +0000 (22:02 -0500)]
man: remove erroneous history section

7 years agomanager: remove "debugging" "feature"
Zbigniew Jędrzejewski-Szmek [Wed, 29 Jan 2014 02:57:31 +0000 (21:57 -0500)]
manager: remove "debugging" "feature"

7 years agomanager: requeue the cylon eye for 5s later when a job finishes
Zbigniew Jędrzejewski-Szmek [Tue, 28 Jan 2014 23:26:06 +0000 (18:26 -0500)]
manager: requeue the cylon eye for 5s later when a job finishes

We'd reqeue the next status update very soon after. Change it so that we wait
for full 5s without any job status changes until we print anything.

7 years agomanager: print ephemeral information about running jobs' timeouts (v2)
Zbigniew Jędrzejewski-Szmek [Tue, 28 Jan 2014 23:25:39 +0000 (18:25 -0500)]
manager: print ephemeral information about running jobs' timeouts (v2)

This reverts commit 28c758de94bc8ba97b89d9dab3f517cf466978d0
but makes job_coldplug smarter.

In (v1) I changed the job start timestamp to be always set, so the
start time can be reported in the cylon eye message.  The bug was that
when deserializing jobs, they would be ignored if their start
timestamp was unset which was synonymous with no timeout. But after
the change, jobs would have a start timestamp set despite having no
timeout. After deserialization they would be considered immediately
expired. Fix this by checking if the timeout is not zero when
considering jobs for expiration.

7 years agoBase mkostemp_safe on mkostemp
Zbigniew Jędrzejewski-Szmek [Tue, 28 Jan 2014 23:25:31 +0000 (18:25 -0500)]
Base mkostemp_safe on mkostemp

It is nice to wrap umask handling and return convention,
but glibc's mkostemp is async-signal-safe already.

7 years agoGet rid of write_safe
Zbigniew Jędrzejewski-Szmek [Tue, 28 Jan 2014 23:23:38 +0000 (18:23 -0500)]
Get rid of write_safe

Current glibc implementation is safe. Kernel does this atomically,
and write is actually implemented through writev. So if write is
async-signal-safe, than writev pretty much must be too.

7 years agokeymap: Add Sony Vaio VGN-FW250
Martin Pitt [Tue, 28 Jan 2014 17:14:18 +0000 (18:14 +0100)]
keymap: Add Sony Vaio VGN-FW250

7 years agokeymap: Add release quirk for Acer AOA switchvideomode key
Martin Pitt [Tue, 28 Jan 2014 17:10:12 +0000 (18:10 +0100)]
keymap: Add release quirk for Acer AOA switchvideomode key

7 years agobuild-sys: disable lto also for libsystemd-id128
Zbigniew Jędrzejewski-Szmek [Tue, 28 Jan 2014 15:36:11 +0000 (10:36 -0500)]
build-sys: disable lto also for libsystemd-id128

Another instance of

7 years agoalways use the same code for creating temporary files
Lennart Poettering [Tue, 28 Jan 2014 12:47:35 +0000 (13:47 +0100)]
always use the same code for creating temporary files

Let's unify our code here, and also always specifiy O_CLOEXEC.

7 years agoutil: pick slightly safer open() flags when creating temporary files
Lennart Poettering [Tue, 28 Jan 2014 12:26:48 +0000 (13:26 +0100)]
util: pick slightly safer open() flags when creating temporary files

7 years agoutil: define O_TMPFILE on x86/x86-64, where the generic value is used
Lennart Poettering [Tue, 28 Jan 2014 12:25:51 +0000 (13:25 +0100)]
util: define O_TMPFILE on x86/x86-64, where the generic value is used

On other archs we'll not define it so that open_tmpfile() falls back to
unguessable name + unlink.

7 years agoutil: pass original flags value to mkostemp(), in open_tmpfile()
Lennart Poettering [Tue, 28 Jan 2014 12:09:14 +0000 (13:09 +0100)]
util: pass original flags value to mkostemp(), in open_tmpfile()

7 years agoutil: simplify mkostemp_safe()
Lennart Poettering [Tue, 28 Jan 2014 12:08:34 +0000 (13:08 +0100)]
util: simplify mkostemp_safe()

Make it use dev_urandom() and endswith().

7 years agoutil: introduce new dev_urandom() call that is like random_bytes() but doesn't fall...
Lennart Poettering [Tue, 28 Jan 2014 12:07:28 +0000 (13:07 +0100)]
util: introduce new dev_urandom() call that is like random_bytes() but doesn't fall back to PRNG

7 years agoutil: modernize loop_read() and loop_write() a bit
Lennart Poettering [Tue, 28 Jan 2014 12:06:44 +0000 (13:06 +0100)]
util: modernize loop_read() and loop_write() a bit

Let's make use of fd_wait_for_event() here, instead of rolling our own.

7 years agobus: reorder a few things
Lennart Poettering [Tue, 28 Jan 2014 11:33:32 +0000 (12:33 +0100)]
bus: reorder a few things

7 years agobus: update kdbus.h (ABI break)
Kay Sievers [Tue, 28 Jan 2014 11:07:20 +0000 (12:07 +0100)]
bus: update kdbus.h (ABI break)

7 years agoRevert "manager: print ephemeral information about running jobs' timeouts"
Kay Sievers [Tue, 28 Jan 2014 11:03:23 +0000 (12:03 +0100)]
Revert "manager: print ephemeral information about running jobs' timeouts"

This reverts commit 2cba2e03524ec0922ddc70f933e8a89b7d23b4ec.

It breaks bootup with dracut, the transition to the real rootfs fails.

7 years agomanager: also turn on output on unit failure
Zbigniew Jędrzejewski-Szmek [Tue, 28 Jan 2014 03:48:18 +0000 (22:48 -0500)]
manager: also turn on output on unit failure

7 years agomanager: add systemd.show_status=auto mode
Zbigniew Jędrzejewski-Szmek [Tue, 28 Jan 2014 03:27:07 +0000 (22:27 -0500)]
manager: add systemd.show_status=auto mode

When set to auto, status will shown when the first ephemeral message
is shown (a job has been running for five seconds). Then until the
boot or shutdown ends, status messages will be shown.

No indication about the switch is done: I think it should be clear
for the user that first the cylon eye and the ephemeral messages appear,
and afterwards messages are displayed.

The initial arming of the event source was still wrong, but now should
really be fixed.

7 years agojournal: guarantee async-signal-safety in sd_journald_sendv
Zbigniew Jędrzejewski-Szmek [Sun, 26 Jan 2014 04:35:28 +0000 (23:35 -0500)]
journal: guarantee async-signal-safety in sd_journald_sendv

signal(7) provides a list of functions which may be called from a
signal handler. Other functions, which only call those functions and
don't access global memory and are reentrant are also safe.
sd_j_sendv was mostly OK, but would call mkostemp and writev in a
fallback path, which are unsafe.

Being able to call sd_j_sendv in a async-signal-safe way is important
because it allows it be used in signal handlers.

Safety is achieved by replacing mkostemp with open(O_TMPFILE) and an
open-coded writev replacement which uses write. Unfortunately,
O_TMPFILE is only available on kernels >= 3.11. When O_TMPFILE is
unavailable, an open-coded mkostemp is used.

7 years agoReplace mkostemp+unlink with open(O_TMPFILE)
Zbigniew Jędrzejewski-Szmek [Sun, 26 Jan 2014 01:48:01 +0000 (20:48 -0500)]
Replace mkostemp+unlink with open(O_TMPFILE)

This will only work on Linux >= 3.11, and probably not on all
filesystems. Fallback code is provided.

7 years agobus: update bloom filter description a bit
Lennart Poettering [Tue, 28 Jan 2014 00:05:32 +0000 (01:05 +0100)]
bus: update bloom filter description a bit

7 years agobus: rework bloom filter logic to operate with variable bloom filter
Lennart Poettering [Mon, 27 Jan 2014 23:57:38 +0000 (00:57 +0100)]
bus: rework bloom filter logic to operate with variable bloom filter
sizes and numbers of hash functions

In order to make the bloom filter logic more future proof communicate
bloom filter parameters from the original bus creator to the clients,
and allow them to be variable within certain ranges.

7 years agobus: add API calls for connecting to starter bus
Lennart Poettering [Mon, 27 Jan 2014 20:34:54 +0000 (21:34 +0100)]
bus: add API calls for connecting to starter bus

Add new calls sd_bus_open() and sd_bus_default() for connecting to the
starter bus a service was invoked for, or -- if the process is not a
bus-activated service -- the appropriate bus for the scope the process
has been started in.

7 years agoresolve: update sd-resolve to match the other APIs in style and functionality
Lennart Poettering [Mon, 27 Jan 2014 19:12:14 +0000 (20:12 +0100)]
resolve: update sd-resolve to match the other APIs in style and functionality

7 years agobus: when we cannot bus activate a service because we go down, drop one
Lennart Poettering [Wed, 22 Jan 2014 22:36:06 +0000 (23:36 +0100)]
bus: when we cannot bus activate a service because we go down, drop one
message from the queue

7 years agokeymap: Add HP Chromebook 14 (Falco)
Martin Pitt [Mon, 27 Jan 2014 08:04:22 +0000 (09:04 +0100)]
keymap: Add HP Chromebook 14 (Falco)

Thanks to Stefan Nagy <>.

7 years agocore: require mounts for the root and working directories
Zbigniew Jędrzejewski-Szmek [Mon, 27 Jan 2014 06:23:20 +0000 (01:23 -0500)]
core: require mounts for the root and working directories

We know that launching a unit will fail if some required directories
haven't been mounted yet. There's no point in starting it just to
have it fail even before it gets a chance to run.

7 years agocore: simplify check for path absoluteness
Zbigniew Jędrzejewski-Szmek [Mon, 27 Jan 2014 06:23:16 +0000 (01:23 -0500)]
core: simplify check for path absoluteness

Just a microopt.

7 years agomanager: rearm jobs timer
Zbigniew Jędrzejewski-Szmek [Mon, 27 Jan 2014 06:15:27 +0000 (01:15 -0500)]
manager: rearm jobs timer

It would fire just once.

Also fix units from sec to usec as appropriate.

Decrease the switching interval to 1/3 s, so that when the time
remaining is displayed with 1s precision, it doesn't jump by 2s every
once in a while. Also, the system is feels noticably faster when the
status changes couple of times per second instead of every few

7 years agomanager: print ephemeral information about running jobs' timeouts
Zbigniew Jędrzejewski-Szmek [Mon, 27 Jan 2014 05:59:08 +0000 (00:59 -0500)]
manager: print ephemeral information about running jobs' timeouts

Produces output like:
[  *** ] (1 of 2) A start job is running for slow.service (33s / 1min 30s)

The first nubmer is the time since job start, the second is the job timeout.