]> git.ipfire.org Git - thirdparty/systemd.git/log
thirdparty/systemd.git
7 years agounits: add system-update-cleanup.service to guard against offline-update loops 4763/head
Zbigniew Jędrzejewski-Szmek [Tue, 29 Nov 2016 06:29:02 +0000 (01:29 -0500)] 
units: add system-update-cleanup.service to guard against offline-update loops

Note: the name is "system-update-cleanup.service" rather than
"system-update-done.service", because it should not run normally, and also
because there's already "systemd-update-done.service", and having them named
so similarly would be confusing.

In https://bugzilla.redhat.com/show_bug.cgi?id=1395686 the system repeatedly
entered system-update.target on boot. Because of a packaging issue, the tool
that created the /system-update symlink could be installed without the service
unit that was supposed to perform the upgrade (and remove the symlink). In
fact, if there are no units in system-update.target, and /system-update symlink
is created, systemd always "hangs" in system-update.target. This is confusing
for users, because there's no feedback what is happening, and fixing this
requires starting an emergency shell somehow, and also knowing that the symlink
must be removed. We should be more resilient in this case, and remove the
symlink automatically ourselves, if there are no upgrade service to handle it.

This adds a service which is started after system-update.target is reached and
the symlink still exists. It nukes the symlink and reboots the machine. It
should subsequently boot into the default default.target.

This is a more general fix for
https://bugzilla.redhat.com/show_bug.cgi?id=1395686 (the packaging issue was
already fixed).

7 years agoman: update the description of offline updates
Zbigniew Jędrzejewski-Szmek [Tue, 29 Nov 2016 06:19:24 +0000 (01:19 -0500)] 
man: update the description of offline updates

- use "service" instead of "script", because various offline updaters that we have
  aren't really scripts, e.g. dnf-plugin-system-upgrade, packagekit-offline-update,
 fwupd-offline-update.
- strongly recommend After=sysinit.target, Wants=sysinit.target
- clarify a bit what should happen when multiple update services are started
- replace links to the wiki with refs to the man page that replaced it.

7 years agoMerge pull request #4761 from fsateler/python3
Martin Pitt [Mon, 28 Nov 2016 20:10:57 +0000 (21:10 +0100)] 
Merge pull request #4761 from fsateler/python3

Explicitly use python3 everywhere

7 years agonetworkd: move event loop handling out of the manager (#4723)
Tom Gundersen [Mon, 28 Nov 2016 19:42:40 +0000 (20:42 +0100)] 
networkd: move event loop handling out of the manager (#4723)

This will allow us to have several managers sharing an event loop
and running in parallel, as if they were running in separate processes.

The long term-aim is to allow networkd to be split into separate
processes, so restructure the code to make this simpler.

For now we drop the exit-on-idle logic, as this was anyway severely
restricted at the moment. Once split, we will revisit this as it may
then make more sense again.

7 years agoUse python3 explicitly in all python scripts 4761/head
Felipe Sateler [Sun, 21 Aug 2016 23:25:37 +0000 (20:25 -0300)] 
Use python3 explicitly in all python scripts

7 years agobuild-sys: explicitly require python3
Felipe Sateler [Sun, 21 Aug 2016 21:20:05 +0000 (18:20 -0300)] 
build-sys: explicitly require python3

Otherwise python programs might be run with python2

7 years agosocket-proxyd: Introduced dynamic connection limit via an option. (#4749)
(GalaxyMaster) [Mon, 28 Nov 2016 17:25:11 +0000 (04:25 +1100)] 
socket-proxyd: Introduced dynamic connection limit via an option. (#4749)

7 years agoudevd: check correct return value of fcntl() (#4758)
Daniel Wagner [Mon, 28 Nov 2016 17:24:26 +0000 (18:24 +0100)] 
udevd: check correct return value of fcntl() (#4758)

This looks like a copy&paste error from the code block above.

7 years agotest: make transient hostname tests fail verbosely (#4754)
Martin Pitt [Mon, 28 Nov 2016 11:35:49 +0000 (12:35 +0100)] 
test: make transient hostname tests fail verbosely (#4754)

This test fails sometimes but it is hard to reproduce, so we need more
information what happens. Set journal log level to "debug" for the entirety of
networkd-test.py, and show networkd's and hostnamed's journals and the DHCP
server log on failure of the two test_transient_hostname* tests. Also sync the
journal before querying it to get more precise output.

This should help with tracking down issue #4753.

7 years agodevice: Avoid calling unit_free(NULL) in device setup logic (#4748)
Dave Reisner [Sun, 27 Nov 2016 22:05:39 +0000 (17:05 -0500)] 
device: Avoid calling unit_free(NULL) in device setup logic (#4748)

Since a581e45ae8f9bb5c, there's a few function calls to
unit_new_for_name which will unit_free on failure. Prior to this commit,
a failure would result in calling unit_free with a NULL unit, and hit an
assertion failure, seen at least via device_setup_unit:

Assertion 'u' failed at src/core/unit.c:519, function unit_free().  Aborting.

Fixes #4747
https://bugs.archlinux.org/task/51950

7 years agoMerge pull request #4736 from dobyrch/calendar-cleanup
Djalal Harouni [Sun, 27 Nov 2016 10:43:26 +0000 (11:43 +0100)] 
Merge pull request #4736 from dobyrch/calendar-cleanup

calendarspec: miscellaneous parsing and formatting fixes

7 years agocalendarspec: refactor format_chain() 4736/head
Douglas Christman [Fri, 25 Nov 2016 02:44:14 +0000 (21:44 -0500)] 
calendarspec: refactor format_chain()

Factor out repeated references to usec and remove nested ifs.

7 years agofix journald startup problem when code is compiled with -DNDEBUG (#4735)
Waldemar Brodkorb [Fri, 25 Nov 2016 10:24:58 +0000 (11:24 +0100)] 
fix journald startup problem when code is compiled with -DNDEBUG (#4735)

Similar to this patch from here:
http://systemd-devel.freedesktop.narkive.com/AvfCbi6c/patch-0-3-using-assert-se-on-actions-with-side-effects-on-test-cases

If the code is compiled with -DNDEBUG which is the default for
some embedded buildsystems, systemd-journald does not startup
and silently fails.

7 years agoRevert "hwdb/parse_hwdb.py: open files with UTF-8 mode"
Martin Pitt [Fri, 25 Nov 2016 06:59:37 +0000 (07:59 +0100)] 
Revert "hwdb/parse_hwdb.py: open files with UTF-8 mode"

"encoding" is not a valid Python 2 keyword, and despite the hashbang this
script can be called with Python 2.

This reverts commit 115a10c58d343d00b73bd9442f7ce6c294debad8.

7 years agocalendarspec: rename "eom" to "end_of_month"
Douglas Christman [Thu, 24 Nov 2016 22:47:25 +0000 (17:47 -0500)] 
calendarspec: rename "eom" to "end_of_month"

7 years agocalendarspec: make specifications with ranges reversible
Douglas Christman [Thu, 24 Nov 2016 20:41:04 +0000 (15:41 -0500)] 
calendarspec: make specifications with ranges reversible

"*-*-01..03" is now formatted as "*-*-01..03" instead of "*-*-01,02,03"

7 years agocalendarspec: allow whole second ranges
Douglas Christman [Thu, 24 Nov 2016 20:51:07 +0000 (15:51 -0500)] 
calendarspec: allow whole second ranges

Previously a string like "00:00:01..03" would fail to parse due to the
ambiguity between a decimal point and the start of a range.

7 years agocalendarspec: make specifications with seconds wildcard reversible
Douglas Christman [Thu, 24 Nov 2016 19:34:17 +0000 (14:34 -0500)] 
calendarspec: make specifications with seconds wildcard reversible

"*:*:*" is now formatted as "*:*:*" instead of "*:*:00/1"

7 years agocalendarspec: reject strings with spurious spaces and signs
Douglas Christman [Thu, 24 Nov 2016 17:47:55 +0000 (12:47 -0500)] 
calendarspec: reject strings with spurious spaces and signs

strtoul() parses leading whitespace and an optional sign;
check that the first character is a digit to prevent odd
specifications like "00:  00:  00" and "-00:+00/-1".

7 years agocalendarspec: reject open weekday ranges
Douglas Christman [Thu, 24 Nov 2016 19:04:13 +0000 (14:04 -0500)] 
calendarspec: reject open weekday ranges

Forbid open ranges like "Tue.."; trailing commas are still OK.

7 years agocalendarspec: reject strings that only contain a timezone
Douglas Christman [Thu, 24 Nov 2016 17:21:37 +0000 (12:21 -0500)] 
calendarspec: reject strings that only contain a timezone

This makes " UTC" an illegal date specification.

7 years agocalendarspec: always interpret a missing time as 00:00:00
Douglas Christman [Thu, 24 Nov 2016 16:50:06 +0000 (11:50 -0500)] 
calendarspec: always interpret a missing time as 00:00:00

"*-*-*" is now equivalent to "*-*-* 00:00:00" (daily)
rather than "*-*-* *:*:*" (every second).

7 years agoMerge pull request #4734 from whot/hwdb-updates
Peter Hutterer [Thu, 24 Nov 2016 22:46:28 +0000 (08:46 +1000)] 
Merge pull request #4734 from whot/hwdb-updates

hwdb: add axis overrides for HP Pavilion dv7

7 years agohwdb: add axis overrides for HP Pavilion dv7 4734/head
Peter Hutterer [Thu, 24 Nov 2016 21:53:18 +0000 (07:53 +1000)] 
hwdb: add axis overrides for HP Pavilion dv7

Fixes #4731

7 years agocore: make sure initrd-switch-root command survives PID1's killing spree (#4730)
Franck Bui [Thu, 24 Nov 2016 17:52:04 +0000 (18:52 +0100)] 
core: make sure initrd-switch-root command survives PID1's killing spree (#4730)

This is a different way to implement the fix proposed by commit
a4021390fef27f4136497328f suggested by Lennart Poettering.

In this patch we instruct PID1 to not kill "systemctl switch-root" command
started by initrd-switch-root service using the "argv[0][0]='@'" trick.

See: https://www.freedesktop.org/wiki/Software/systemd/RootStorageDaemons/ for
more details.

We had to backup argv[0] because argv is modified by dispatch_verb().

7 years agoMerge pull request #4729 from martinpitt/hwdb
Martin Pitt [Thu, 24 Nov 2016 17:19:59 +0000 (18:19 +0100)] 
Merge pull request #4729 from martinpitt/hwdb

fix hwdb syntax error and add test for it

7 years agotests: add hwdb parsing test 4729/head
Martin Pitt [Thu, 24 Nov 2016 10:29:42 +0000 (11:29 +0100)] 
tests: add hwdb parsing test

This calls the built "systemd-hwdb update" on the source tree's hwdb/ in a
temporary directory and verifies that there are no error messages.

7 years agohwdb: fix syntax error in 60-keyboard.hwdb
Martin Pitt [Thu, 24 Nov 2016 10:08:49 +0000 (11:08 +0100)] 
hwdb: fix syntax error in 60-keyboard.hwdb

Fixes #4728

7 years agohwdb/parse_hwdb.py: open files with UTF-8 mode
Martin Pitt [Thu, 24 Nov 2016 09:38:01 +0000 (10:38 +0100)] 
hwdb/parse_hwdb.py: open files with UTF-8 mode

pyparsing uses the system locale by default, which in the case of 'C' (in lots
of build environment) will fail with a UnicodeDecodeError. Explicitly open it
with UTF-8 encoding to guard against this.

7 years agoMerge pull request #4710 from martinpitt/networkd-dbus
Lennart Poettering [Thu, 24 Nov 2016 00:58:33 +0000 (01:58 +0100)] 
Merge pull request #4710 from martinpitt/networkd-dbus

networkd: allow networkd to start in early boot

7 years agoman: document protocol error type for service failures (#4724)
Jouke Witteveen [Wed, 23 Nov 2016 21:51:33 +0000 (22:51 +0100)] 
man: document protocol error type for service failures (#4724)

7 years agonetworkd: handle MTU field in IPv6 RA (#4719)
Susant Sahani [Wed, 23 Nov 2016 21:32:19 +0000 (03:02 +0530)] 
networkd: handle MTU field in IPv6 RA (#4719)

This patch handles the custom MTU field in IPv6 RA.

fixes RFE #4464

7 years agoMerge pull request #4703 from dobyrch/calendar-offset
Lennart Poettering [Wed, 23 Nov 2016 21:31:01 +0000 (22:31 +0100)] 
Merge pull request #4703 from dobyrch/calendar-offset

calendarspec: add support for scheduling timers at the end of the month

7 years agohwdb: use systemd-hwdb instead of obsolete udevadm hwdb (#4722)
Michael Biebl [Wed, 23 Nov 2016 18:21:56 +0000 (19:21 +0100)] 
hwdb: use systemd-hwdb instead of obsolete udevadm hwdb (#4722)

Fixes: #4721
7 years agocalendarspec: add support for scheduling timers at the end of the month 4703/head
Douglas Christman [Tue, 22 Nov 2016 15:05:10 +0000 (10:05 -0500)] 
calendarspec: add support for scheduling timers at the end of the month

"*-*~1"       => The last day of every month
"*-02~3..5"   => The third, fourth, and fifth last days in February
"Mon 05~07/1" => The last Monday in May

Resolves #3861

7 years agocalendarspec: add upper bound on year
Douglas Christman [Tue, 22 Nov 2016 17:57:07 +0000 (12:57 -0500)] 
calendarspec: add upper bound on year

Stop looking for matches after MAX_YEAR so impossible dates like
"*-02-30" and "*-04-31" don't cause an infinite loop.

7 years agonetworkd: allow networkd to start in early boot 4710/head
Martin Pitt [Tue, 22 Nov 2016 07:41:51 +0000 (08:41 +0100)] 
networkd: allow networkd to start in early boot

With the previous improvements, networkd.service's "After=dbus.service" can now
be dropped. That ordering effectively forced networkd.service to run in late
boot only (dbus.service was rejected to run in early boot in
https://bugs.freedesktop.org/show_bug.cgi?id=98254).

Fixes #4504

7 years agonetworkd: set DHCP-acquired timezone and hostname after connecting to D-Bus
Martin Pitt [Tue, 22 Nov 2016 07:36:20 +0000 (08:36 +0100)] 
networkd: set DHCP-acquired timezone and hostname after connecting to D-Bus

If setting the received timezone or transient hostname fails because D-Bus is
not (yet) up, store the data in the Manager object and try again after
connecting to D-Bus.

7 years agoMerge pull request #4259 from joukewitteveen/notify
Lennart Poettering [Wed, 23 Nov 2016 15:45:19 +0000 (16:45 +0100)] 
Merge pull request #4259 from joukewitteveen/notify

service: fix main processes exit behavior for type notify services

7 years agonetworkd: fix size of MTUBytes so that it does not overwrites ARP (#4707)
Susant Sahani [Wed, 23 Nov 2016 15:33:01 +0000 (21:03 +0530)] 
networkd: fix size of MTUBytes so that it does not overwrites ARP (#4707)

config_parse_iec_size overwrites the next varible that is ARP.
Now the mtu is unsigned . Make it size_t .

Fixes #4644

7 years agonetworkd: allow networkd to set the timezone in timedated
Martin Pitt [Tue, 22 Nov 2016 07:05:18 +0000 (08:05 +0100)] 
networkd: allow networkd to set the timezone in timedated

systemd-networkd runs as user "systemd-network" and thus is not privileged to
set the timezone acquired from DHCP:

  systemd-networkd[4167]: test_eth42: Could not set timezone: Interactive authentication required.

Similarly to commit e8c0de912, add a polkit rule to grant
org.freedesktop.timedate1.set-timezone to the "systemd-network" system user.
Move the polkit rules from src/hostname/ to src/network/ to avoid too many
small distributed policy snippets (there might be more in the future), as it's
easier to specify the privileges for a particular subject in this case.

Add NetworkdClientTest.test_dhcp_timezone() test case to verify this (for
all people except those in Pacific/Honolulu, there the test doesn't prove
anything -- sorry ☺ ).

7 years agonetworkd: move setting hostname and timezone to Manager
Martin Pitt [Mon, 21 Nov 2016 22:15:41 +0000 (23:15 +0100)] 
networkd: move setting hostname and timezone to Manager

Hostname and time zone are global settings, not link specific.  Move these
methods from Link into Manager.

7 years agocore: consider SIGTERM as a clean exit status for initrd-switch-root.service (#4713)
Franck Bui [Wed, 23 Nov 2016 15:31:24 +0000 (16:31 +0100)] 
core: consider SIGTERM as a clean exit status for initrd-switch-root.service (#4713)

Since commit 1f0958f640b8717, systemd considers SIGTERM for short-running
services (aka Type=oneshot) as a failure.

This can be an issue with initrd-switch-root.service as the command run by this
service (in order to switch to the new rootfs) may still be running when
systemd does the switch.

However PID1 sends SIGTERM to all remaining processes right before
switching and initrd-switch-root.service can be one of those.

After systemd is reexecuted and its previous state is deserialized, systemd
notices that initrd-switch-root.service was killed with SIGTERM and considers
this as a failure which leads to the emergency shell.

To prevent this, this patch teaches systemd to consider a SIGTERM exit as a
clean one for this service.

It also removes "KillMode=none" since this is pretty useless as the service is
never stopped by systemd but it either exits normally or it's killed by a
SIGTERM as described previously.

7 years agobuild-sys: check for lz4 in the old and new numbering scheme (#4717)
Zbigniew Jędrzejewski-Szmek [Wed, 23 Nov 2016 15:18:30 +0000 (10:18 -0500)] 
build-sys: check for lz4 in the old and new numbering scheme (#4717)

lz4 upstream decided to switch to an incompatible numbering scheme
(1.7.3 follows 131, to match the so version).
PKG_CHECK_MODULES does not allow two version matches for the same package,
so e.g. lz4 < 10 || lz4 >= 125 cannot be used. Check twice, once for
"new" numbers (anything below 10 is assume to be new), once for the "old"
numbers (anything above >= 125). This assumes that the "new" versioning
will not get to 10 to quickly. I think that's a safe assumption, lz4 is a
mature project.

Fixed #4690.

7 years agoCONTRIBUTING: reword the part about unsetting the label (#4716)
Zbigniew Jędrzejewski-Szmek [Wed, 23 Nov 2016 07:52:19 +0000 (02:52 -0500)] 
CONTRIBUTING: reword the part about unsetting the label (#4716)

Only project members can do it.

7 years agoMerge pull request #4692 from poettering/networkd-dhcp
Zbigniew Jędrzejewski-Szmek [Wed, 23 Nov 2016 04:22:04 +0000 (23:22 -0500)] 
Merge pull request #4692 from poettering/networkd-dhcp

Various networkd/DHCP fixes.

7 years agoDocument an edge-case with resume and mounting (#4581)
Janne Heß [Wed, 23 Nov 2016 04:19:56 +0000 (05:19 +0100)] 
Document an edge-case with resume and mounting (#4581)

When trying to read keyfiles from an encrypted partition to unlock the swap,
a cyclic dependency is generated because systemd can not mount the
filesystem before it has checked if there is a swap to resume from.

Closes #3940

7 years agoMerge pull request #4693 from poettering/nspawn-ephemeral
Zbigniew Jędrzejewski-Szmek [Wed, 23 Nov 2016 04:15:07 +0000 (23:15 -0500)] 
Merge pull request #4693 from poettering/nspawn-ephemeral

nspawn: support ephemeral boots from images

7 years agoservice: fix main processes exit behavior for type notify services 4259/head
Jouke Witteveen [Sat, 1 Oct 2016 12:06:48 +0000 (14:06 +0200)] 
service: fix main processes exit behavior for type notify services

Before this commit, when the main process of a Type=notify service exits the
service would enter a running state without passing through the startup post
state. This meant ExecStartPost= from being executed and allowed follow-up
units to start too early (before the ready notification).
Additionally, when RemainAfterExit=yes is used on a Type=notify service, the
exit status of the main process would be disregarded.

After this commit, an unsuccessful exit of the main process of a Type=notify
service puts the unit in a failed state. A successful exit is inconsequential
in case RemainAfterExit=yes. Otherwise, when no ready notification has been
received, the unit is put in a failed state because it has never been active.
When all processes in the cgroup of a Type=notify service are gone and no ready
notification has been received yet, the unit is also put in a failed state.

7 years agoservice: introduce protocol error type
Jouke Witteveen [Tue, 22 Nov 2016 16:39:56 +0000 (17:39 +0100)] 
service: introduce protocol error type

Introduce a SERVICE_FAILURE_PROTOCOL error type for when a service does
not follow the protocol.
This error type is used when a pid file is expected, but not delivered.

7 years agonspawn: don't require chown() if userns is not on 4693/head
Lennart Poettering [Tue, 22 Nov 2016 12:35:24 +0000 (13:35 +0100)] 
nspawn: don't require chown() if userns is not on

Fixes: #4711
7 years agonspawn: add fallback top normal copy/reflink when we cannot btrfs snapshot
Lennart Poettering [Mon, 21 Nov 2016 19:02:43 +0000 (20:02 +0100)] 
nspawn: add fallback top normal copy/reflink when we cannot btrfs snapshot

Given that other file systems (notably: xfs) support reflinks these days, let's
extend the file system snapshotting logic to fall back to plan copies or
reflinks when full btrfs subvolume snapshots are not available.

This essentially makes "systemd-nspawn --ephemeral" and "systemd-nspawn
--template=" available on non-btrfs subvolumes. Of course, both operations will
still be slower on non-btrfs than on btrfs (simply because reflinking each file
individually in a directory tree is still slower than doing this in one step
for a whole subvolume), but it's probably good enough for many cases, and we
should provide the users with the tools, they have to figure out what's good
for them.

Note that "machinectl clone" already had a fallback like this in place, this
patch generalizes this, and adds similar support to our other cases.

7 years agonspawn: remove temporary root directory on exit
Lennart Poettering [Fri, 18 Nov 2016 23:00:16 +0000 (00:00 +0100)] 
nspawn: remove temporary root directory on exit

When mountint a loopback image, we need a temporary root directory we can mount
stuff to. Make sure to actually remove it when exiting, so that we don't leave
stuff around in /tmp unnecessarily.

See: #4664

7 years agonspawn: try to wait for the container PID 1 to exit, before we exit
Lennart Poettering [Fri, 18 Nov 2016 22:47:09 +0000 (23:47 +0100)] 
nspawn: try to wait for the container PID 1 to exit, before we exit

Let's make the shutdown logic synchronous, so that there's a better chance to
detach the loopback device after use.

7 years agonspawn: add ability to run nspawn without container locks applied
Lennart Poettering [Fri, 18 Nov 2016 22:34:39 +0000 (23:34 +0100)] 
nspawn: add ability to run nspawn without container locks applied

This adds a new undocumented env var $SYSTEMD_NSPAWN_LOCK. When set to "0",
nspawn will not attempt to lock the image.

Fixes: #4037
7 years agoshared: make sure image_path_lock() return parameters are always initialized on success
Lennart Poettering [Fri, 18 Nov 2016 22:31:01 +0000 (23:31 +0100)] 
shared: make sure image_path_lock() return parameters are always initialized on success

We forgot to initialize the "global" return parameter in one case. Fix that.

7 years agonspawn: support ephemeral boots from images
Lennart Poettering [Fri, 18 Nov 2016 17:38:06 +0000 (18:38 +0100)] 
nspawn: support ephemeral boots from images

Previously --ephemeral was only supported with container trees in btrfs
subvolumes (i.e. in combination with --directory=). This adds support for
--ephemeral in conjunction with disk images (i.e. --image=) too.

As side effect this fixes that --ephemeral was accepted but ignored when using
-M on a container that turned out to be an image.

Fixes: #4664
7 years agoseccomp: add @filesystem syscall group (#4537)
Lennart Poettering [Tue, 22 Nov 2016 00:29:12 +0000 (01:29 +0100)] 
seccomp: add @filesystem syscall group (#4537)

@filesystem groups various file system operations, such as opening files and
directories for read/write and stat()ing them, plus renaming, deleting,
symlinking, hardlinking.

7 years agosd-event: fix sd_event_source_get_priority() (#4712)
Martin Ejdestig [Tue, 22 Nov 2016 00:21:00 +0000 (01:21 +0100)] 
sd-event: fix sd_event_source_get_priority() (#4712)

To properly store priority in passed in pointer and return 0 for success.
Also add a test for verifying that it works correctly.

7 years agosd-dhcp-client: use free_and_strdup 4692/head
Zbigniew Jędrzejewski-Szmek [Sat, 19 Nov 2016 17:53:29 +0000 (12:53 -0500)] 
sd-dhcp-client: use free_and_strdup

This changes the return value a bit: 1 will be returned if the value is
changed. But the return value was not documented, and the change should
be for the good anyway. Current callers don't care.

7 years agoman: make /etc/nsswitch.conf documentation for nss-resolve match example
Lennart Poettering [Fri, 18 Nov 2016 16:31:49 +0000 (17:31 +0100)] 
man: make /etc/nsswitch.conf documentation for nss-resolve match example

Fixes: #4683
7 years agodhcp4: filter bogus DNS/NTP server addresses silently
Lennart Poettering [Fri, 18 Nov 2016 16:19:44 +0000 (17:19 +0100)] 
dhcp4: filter bogus DNS/NTP server addresses silently

if we receive a bogus lease with a DNS/NTP server within local scope let's
politely ignore.

Fixes: #4524
7 years agonetworkd: do not automatically propagate bogus DNS/NTP servers
Lennart Poettering [Fri, 18 Nov 2016 16:11:12 +0000 (17:11 +0100)] 
networkd: do not automatically propagate bogus DNS/NTP servers

Never propagate DNS/NTP servers on the local link via the DHCP server. The
DNS/NTP servers 0.0.0.0 and 127.0.0.1 only make sense in the local context,
hence never propagate them automatically to other hosts.

Fixes: #4524
7 years agonetworkd: rename link_push_dns_to_dhcp_server() → link_push_uplink_dns_to_dhcp_server()
Lennart Poettering [Fri, 18 Nov 2016 16:07:52 +0000 (17:07 +0100)] 
networkd: rename link_push_dns_to_dhcp_server() → link_push_uplink_dns_to_dhcp_server()

Let's clarify that this is only relevant for pushing uplink DNS config data,
not in any other context.

(Same for NTP)

7 years agonetworkd: store DNS servers configured per-network as parsed addresses
Lennart Poettering [Fri, 18 Nov 2016 16:04:26 +0000 (17:04 +0100)] 
networkd: store DNS servers configured per-network as parsed addresses

DNS servers must be specified as IP addresses, hence let's store them as that
internally, so that they are guaranteed to be fully normalized always, and
invalid data cannot be stored.

7 years agonetworkd: minor word extraction fixes
Lennart Poettering [Fri, 18 Nov 2016 16:03:13 +0000 (17:03 +0100)] 
networkd: minor word extraction fixes

Let's make the loops around extract_first_word() be more alike.

7 years agonetworkd: reorder fields in Network struct a bit
Lennart Poettering [Fri, 18 Nov 2016 16:01:32 +0000 (17:01 +0100)] 
networkd: reorder fields in Network struct a bit

Let's reorder them a bit, so that stuff that belongs together semantically is
placed together (in particular, move the various DHCP "use" booleans together).

7 years agonetworkd: validate NTP server strings configured in .network files
Lennart Poettering [Fri, 18 Nov 2016 16:00:25 +0000 (17:00 +0100)] 
networkd: validate NTP server strings configured in .network files

Let's be a bit stricter, and make sure users only configure proper DNS names or
IP addresses, but not any other strings.

7 years agoshared: add new API to validate a string as hostname or IP address
Lennart Poettering [Fri, 18 Nov 2016 15:55:01 +0000 (16:55 +0100)] 
shared: add new API to validate a string as hostname or IP address

7 years agobasic: make sure in_addr_from_string() can take NULL arguments
Lennart Poettering [Fri, 18 Nov 2016 15:48:19 +0000 (16:48 +0100)] 
basic: make sure in_addr_from_string() can take NULL arguments

This way, we can use it for validating IP address strings.

7 years agobasic: add explicit ipv4-specific in_addr classification calls
Lennart Poettering [Fri, 18 Nov 2016 12:13:28 +0000 (13:13 +0100)] 
basic: add explicit ipv4-specific in_addr classification calls

This adds in4_addr_is_localhost() and in4_addr_is_link_local() that only take
an IPv4 "struct in_addr", to match in_addr_is_localhost() and
in_addr_is_link_local() that that a "union in_addr_union".

This matches the existing in4_addr_is_null() call that already exists.

For IPv6 glibc already exports a set of macros, hence we don't add similar
functions in6_addr_is_localhost(). We also drop in6_addr_is_null() as
IN6_IS_ADDR_UNSPECIFIED() already provides that.

7 years agonetworkd: rework how we set the hostname on the dhcp4 client
Lennart Poettering [Fri, 18 Nov 2016 12:02:57 +0000 (13:02 +0100)] 
networkd: rework how we set the hostname on the dhcp4 client

Let's split out the code into its own little function and let's use
gethostname_strict() which already filters out localhost for us.

7 years agosd-dhcp: permit unsetting the hostname again
Lennart Poettering [Fri, 18 Nov 2016 11:33:14 +0000 (12:33 +0100)] 
sd-dhcp: permit unsetting the hostname again

Let's handle NULL hostnames (for unsetting it) before we validate the name.

7 years agoMerge pull request #4701 from martinpitt/networkd-polkit
Lennart Poettering [Mon, 21 Nov 2016 17:24:51 +0000 (18:24 +0100)] 
Merge pull request #4701 from martinpitt/networkd-polkit

hostnamed: allow networkd to set the transient hostname

7 years agotests: check transient hostname with and without static hostname 4701/head
Martin Pitt [Mon, 21 Nov 2016 11:53:24 +0000 (12:53 +0100)] 
tests: check transient hostname with and without static hostname

We expect a static hostname to win over a transient one, so test both cases.

7 years agoMerge pull request #4691 from keszybz/test-suite-fixes
Evgeny Vereshchagin [Sun, 20 Nov 2016 16:38:58 +0000 (19:38 +0300)] 
Merge pull request #4691 from keszybz/test-suite-fixes

Test suite fixes

7 years agohostnamed: allow networkd to set the transient hostname
Martin Pitt [Fri, 18 Nov 2016 15:17:01 +0000 (16:17 +0100)] 
hostnamed: allow networkd to set the transient hostname

systemd-networkd runs as user "systemd-network" and thus is not privileged to
set the transient hostname:

  systemd-networkd[516]: ens3: Could not set hostname: Interactive authentication required.

Standard polkit *.policy files do not have a syntax for granting privileges to
a user, so ship a pklocalauthority (for polkit < 106) and a JavaScript rules
file (for polkit >= 106) that grants the "systemd-network" system user that
privilege.

Add DnsmasqClientTest.test_transient_hostname() test to networkd-test.py to
cover this. Make do_test() a bit more flexible by interpreting "coldplug==None"
as "test sets up the interface by itself". Change DnsmasqClientTest to set up
test_eth42 with a fixed MAC address so that we can configure dnsmasq to send a
special host name for that.

Fixes #4646

7 years agotest: clean up "dummy0" interface after each test case in networkd-test.py
Martin Pitt [Sun, 20 Nov 2016 09:08:23 +0000 (10:08 +0100)] 
test: clean up "dummy0" interface after each test case in networkd-test.py

7 years agotest-functions: also add qemu define for ppc64 4691/head
Zbigniew Jędrzejewski-Szmek [Sun, 20 Nov 2016 03:45:23 +0000 (22:45 -0500)] 
test-functions: also add qemu define for ppc64

7 years agohwdb: keyboard: Ignore brightness hotkey presses on MSI VR420 (model MS-1422) (#4696)
Hans de Goede [Sat, 19 Nov 2016 15:07:41 +0000 (16:07 +0100)] 
hwdb: keyboard: Ignore brightness hotkey presses on MSI VR420 (model MS-1422) (#4696)

Add MSI VR420 (model MS-1422) to the list of MSI models which need to
ignore brightness hotkey presses, as these are already reported through
the acpi-video interface.

link: https://bugzilla.redhat.com/show_bug.cgi?id=1270124
Reported-by: David Dreggors <dadreggors@gmail.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
7 years agoMerge pull request #4395 from s-urbaniak/rw-support
Lennart Poettering [Fri, 18 Nov 2016 11:36:46 +0000 (12:36 +0100)] 
Merge pull request #4395 from s-urbaniak/rw-support

nspawn: R/W support for /sysfs, /proc, and /proc/sys/net

7 years agoMerge pull request #4538 from fbuihuu/confirm-spawn-fixes
Lennart Poettering [Fri, 18 Nov 2016 10:08:06 +0000 (11:08 +0100)] 
Merge pull request #4538 from fbuihuu/confirm-spawn-fixes

Confirm spawn fixes/enhancements

7 years agotests/TEST-13-NSPAWN-SMOKE: add SYSTEMD_NSPAWN_API_VFS_WRITABLE 4395/head
Evgeny Vereshchagin [Sat, 5 Nov 2016 02:46:22 +0000 (02:46 +0000)] 
tests/TEST-13-NSPAWN-SMOKE: add SYSTEMD_NSPAWN_API_VFS_WRITABLE

https://github.com/systemd/systemd/pull/4395

7 years agonspawn: R/W support for /sys, and /proc/sys
Sergiusz Urbaniak [Fri, 14 Oct 2016 12:00:15 +0000 (14:00 +0200)] 
nspawn: R/W support for /sys, and /proc/sys

This commit adds the possibility to leave /sys, and /proc/sys read-write.
It introduces a new (undocumented) env var SYSTEMD_NSPAWN_API_VFS_WRITABLE
to enable this feature.

If set to "yes", /sys, and /proc/sys will be read-write.
If set to "no", /sys, and /proc/sys will be read-only.
If set to "network" /proc/sys/net will be read-write. This is useful in
use-cases, where systemd-nspawn is used in an external network
namespace.

This adds the possibility to start privileged containers which need more
control over settings in the /proc, and /sys filesystem.

This is also a follow-up on the discussion from
https://github.com/systemd/systemd/pull/4018#r76971862 where an
introduction of a simple env var to enable R/W support for those
directories was already discussed.

7 years agotest-ipcrm: skip test if nfsnobody is missing
Zbigniew Jędrzejewski-Szmek [Fri, 18 Nov 2016 01:57:22 +0000 (20:57 -0500)] 
test-ipcrm: skip test if nfsnobody is missing

7 years agobasic/process-util: we need to take the shorter of two strings
Zbigniew Jędrzejewski-Szmek [Fri, 18 Nov 2016 01:57:22 +0000 (20:57 -0500)] 
basic/process-util: we need to take the shorter of two strings

==30496== Conditional jump or move depends on uninitialised value(s)
==30496==    at 0x489F654: memcmp (vg_replace_strmem.c:1091)
==30496==    by 0x49BF203: getenv_for_pid (process-util.c:678)
==30496==    by 0x4993ACB: detect_container (virt.c:442)
==30496==    by 0x182DFF: test_get_process_comm (test-process-util.c:98)
==30496==    by 0x185847: main (test-process-util.c:368)
==30496==

7 years agotest-process-util: bind mount fails under selinux, skip test
Zbigniew Jędrzejewski-Szmek [Fri, 18 Nov 2016 01:57:22 +0000 (20:57 -0500)] 
test-process-util: bind mount fails under selinux, skip test

7 years agoMerge pull request #4671 from poettering/namespace-bind
Zbigniew Jędrzejewski-Szmek [Fri, 18 Nov 2016 00:40:57 +0000 (19:40 -0500)] 
Merge pull request #4671 from poettering/namespace-bind

rework service namespace handling a bit

7 years agobasic/env-uil: fix assertion failure in strv_env_replace (#4688)
Zbigniew Jędrzejewski-Szmek [Thu, 17 Nov 2016 21:25:01 +0000 (16:25 -0500)] 
basic/env-uil: fix assertion failure in strv_env_replace (#4688)

free_and_replace sets the setcond argument to NULL (it's designed
to be used with _clenaup_ macros), and we don't want that here.

Fixes #4684.

7 years agocore: in confirm spawn, suggest 'f' when user selects 'n' choice 4538/head
Franck Bui [Thu, 17 Nov 2016 17:22:43 +0000 (18:22 +0100)] 
core: in confirm spawn, suggest 'f' when user selects 'n' choice

7 years agocore: confirm_spawn: always accept units with same_pgrp set for now
Franck Bui [Mon, 14 Nov 2016 16:37:40 +0000 (17:37 +0100)] 
core: confirm_spawn: always accept units with same_pgrp set for now

For some reasons units remaining in the same process group as PID 1
(same_pgrp=true) fail to acquire the console even if it's not taken by anyone.

So always accept for units with same_pgrp set for now.

7 years agocore: include the unit name when notifying that a confirmation question timed out
Franck Bui [Thu, 10 Nov 2016 09:07:42 +0000 (10:07 +0100)] 
core: include the unit name when notifying that a confirmation question timed out

7 years agocore: add 'c' in confirmation_spawn to resume the boot process
Franck Bui [Tue, 15 Nov 2016 08:29:04 +0000 (09:29 +0100)] 
core: add 'c' in confirmation_spawn to resume the boot process

7 years agocore: add 'j' in confirmation_spawn to list the jobs that are in progress
Franck Bui [Sun, 13 Nov 2016 15:28:04 +0000 (16:28 +0100)] 
core: add 'j' in confirmation_spawn to list the jobs that are in progress

7 years agocore: add 'D' in confirmat spawn to show a full dump of the unit to spawn
Franck Bui [Sat, 12 Nov 2016 14:08:29 +0000 (15:08 +0100)] 
core: add 'D' in confirmat spawn to show a full dump of the unit to spawn

7 years agocore: add 'i' in confirm spawn to give a short summary of the unit to spawn
Franck Bui [Sat, 12 Nov 2016 13:55:12 +0000 (14:55 +0100)] 
core: add 'i' in confirm spawn to give a short summary of the unit to spawn

7 years agocore: rework the confirmation spawn prompt
Franck Bui [Mon, 7 Nov 2016 16:14:59 +0000 (17:14 +0100)] 
core: rework the confirmation spawn prompt

Previously it was "[Yes, Fail, Skip]" which is pretty misleading because it
suggests that the whole word needs to be entered instead of a single char.

Also this won't fit well when we'll extend the number of choices.

This patch addresses this by changing the choice hint with "[y, f, s – h for help]"
so it's now clear that a single letter has to be entered.

It also introduces a new choice 'h' which describes all possible choices since
a single letter can be not descriptive enough for new users.

It also allow to stick with the same hint string regardless of how
many choices we will support.

7 years agocore: limit the length of the confirmation question
Franck Bui [Mon, 7 Nov 2016 16:14:59 +0000 (17:14 +0100)] 
core: limit the length of the confirmation question

When "confirmation_spawn=1", the confirmation question can look like:

  Execute /usr/bin/kmod static-nodes --format=tmpfiles --output=/run/tmpfiles.d/kmod.conf? [Yes, No, Skip]

which is pretty verbose and might not fit in the console width size (which is
usually 80 chars) and thus question will be splitted into 2 consecutive lines.

However since the question is now refreshed every 2 secs, the reprinted
question will overwrite the second line of the previous one...

To prevent this, this patch makes sure that the command line won't be longer
than 60 chars by ellipsizing it if the command is longer:

  Execute /usr/bin/kmod static-nodes --format=tmpfiles --output=/ru…nf? [Yes, No, View, Skip]

A following patch will introduce a new choice that will allow the user to get
details on the command to be executed so it will still be possible to see the
full command line.

7 years agocore: reprint the question every 2 sec in ask_char()
Franck Bui [Mon, 7 Nov 2016 16:14:59 +0000 (17:14 +0100)] 
core: reprint the question every 2 sec in ask_char()

ask_char() now reprints the question every 2sec automatically.

It prefixes its output with '\r' to to bring the cursor to the
beginning of the terminal line, and then print the message, redoing it
every 2sec.

As long as nothing interferes with out output this logic will have no
visible effect as we constantly overprint the visible text with the
exact same text.

However, if something is dumped in the middle, then our question won't
get lost, as we'll ask soon again.

This is useful if the question is asked to a terminal that is also
used to dump some other status messages/logs. For example when
confirmation messages are enabled during the boot
(systemd.confirm_spawn=1), the question can easily be lost if the
kernel logs are also enabled and both use the same console.

Idea suggested by Lennart Poettering.

7 years agocore: in confirm_spawn, the meaning of 'n' and 's' choices are confusing
Franck Bui [Sun, 13 Nov 2016 08:32:52 +0000 (09:32 +0100)] 
core: in confirm_spawn, the meaning of 'n' and 's' choices are confusing

Before this patch we had:

 - "no" which gives "failing execution" but the command is actually assumed as
   succeed.

 - "skip" which gives "skipping", but the command is assumed to have failed,
   which ends up with "Failed to start ..." on the console.

Now we have:

 - "fail" which gives "failing execution" and the command is indeed assumed as
   failed.

 - "skip" which gives "skipping execution" and the command is assumed as
   succeed.