Fixes <https://issues.guix.gnu.org/71173>.
* gnu/services/base.scm (network-set-up/linux): Define
‘max-set-up-duration’ and use it.
* gnu/tests/networking.scm (%static-networking-with-nonexistent-device):
New variable.
(run-static-networking-failure-test): New procedure.
(%test-static-networking-failure): New variable.
Change-Id: Idba9b36750aa8c6368c8f6d1bc1358066f7432e4
Running that code in PID 1 was fun but it’s not really beneficial and
somewhat risky: risk of blocking, file descriptor leak, inability to
reload Guile-Netlink in shepherd when it’s upgraded, and so on.
This change runs set-up and tear-down as separate processes, which, for
the price of one fork(1), buys us peace of mind.
* gnu/services/base.scm (network-set-up/hurd, network-tear-down/hurd)
(network-tear-down/linux): Use ‘program-file’ instead of ‘scheme-file’.
(network-set-up/linux): Likewise, and remove #:blocking? argument to
‘wait-for-link’.
Change-Id: Ia41479b50eab31ea40c67243fcb1cffe29ac874a
* gnu/services/base.scm (syslog-shepherd-service): Separate incorrectly
combined arguments which resulted in an argument like "-f
/etc/syslog.conf" being passed to syslogd, leading it to ignore the
argument and execute without a configuration file. Effects of this
included no log files being written, though the Shepherd service ran
successfully.
Ref: https://issues.guix.gnu.org/70677#4-lineno7
Change-Id: I3dbe00eabd4a10804e554c12e1466483c0b185b7
Signed-off-by: Ludovic Courtès <ludo@gnu.org>
* gnu/services/base.scm (syslog-shepherd-service): Change flag for designating
configuration file. The long option is not compatible with rsyslog while the
short is; switch to the short one.
Signed-off-by: Ludovic Courtès <ludo@gnu.org>
* gnu/services/base.scm (<syslog-configuration>): Add extra-options field.
(syslog-shepherd-service): Use it when running the service.
* doc/guix.texi: Document it.
Change-Id: I540d070b9a9678b45ec9fa28d6fdc761f9b3fd9a
Signed-off-by: Ludovic Courtès <ludo@gnu.org>
For auto-login on systems with elogind, dbus-system needs to be started. This
commit adds ability to express that ordering.
* gnu/services/base.scm (<mingetty-configuration>): Add shepherd-requirement
field.
(mingetty-shepherd-service): Use it.
* doc/guix.texi (Base Services)<mingetty-configuration>: Document it.
Change-Id: Iedbdc4375180740379d561aa193d7c63350d2e7b
Modified-by: Maxim Cournoyer <maxim.cournoyer@gmail.com>
Signed-off-by: Maxim Cournoyer <maxim.cournoyer@gmail.com>
Rename the accessors to ensure all start with `mingetty-configuration-'
prefix. Some were named just `mingetty-$FIELD', instead of
`mingetty-configuration-$FIELD'.
The renaming *is* backwards compatible, since in the define-module's #:export
argument the correct (`mingetty-configuration-$FIELD') were used already and
thus the accessors were not accessible.
* gnu/services/base.scm (<mingetty-configuration>): Rename accessors for
auto-login, login-program, login-pause?, clear-on-logout?.
Change-Id: I4557a82498805ade0b341feda9d33eccc305690f
Modified-by: Maxim Cournoyer <maxim.cournoyer@gmail.com>
Signed-off-by: Maxim Cournoyer <maxim.cournoyer@gmail.com>
Not all aspects of mingetty were configurable, so this commit adds the
additional configuration fields to support that.
* gnu/services/base.scm (<mingetty-configuration>): Add delay, print-issue,
print-hostname, nice, working-directory, root-directory fields.
(mingetty-shepherd-service): Use the new fields.
(define-module)<#:export>: Export the new accessors.
* doc/guix.texi (Base Services)<mingetty-configuration>: Document the
additional field.
Change-Id: I4557a82498805ade0b341feda9d33eccc305690f
* gnu/services/base.scm (swap-service-type): Do not include 'udev' requirement
for the Hurd. Use system* with "swapon", "swapoff" for the Hurd.
* gnu/system.scm (hurd-default-essential-services): Add swap-services.
* gnu/services/base.scm (swap-service-type):
Change-Id: I1d4d445c614921752dc84aa0dd6ff42cdbf62aa8
Previously, when being stopped, the ‘user-file-systems’ service would
attempt to unmount / and /dev, which was bound to fail. This was
harmless, apart from a couple of lines in /var/log/messages, but it was
wrong.
* gnu/services/base.scm (file-system-shepherd-services)[user-unmount]:
Remove “/” and “/dev” from the list of file systems subject to
unmounting.
Change-Id: Ieb68fe46c114909a64d54f885c94a9d7cd61f5e0
Previously, the “running value” of the ‘host-name’ service would
be #<unspecified>. This change makes it more pleasing to the eye.
* gnu/services/base.scm (host-name-service-type): ‘start’ returns NAME.
Change-Id: I38b8320e43639b6623475871ca6fbad3a459eb59
In the absence of the /etc/guix/channels.scm file, %default-channels is used
anyway. If user manually (or by extra-special-file) created the file, we
should respect it. This commit therefore changes the default to #f, hopefully
having zero impact on people not actively using the `channels' field.
* gnu/services/base.scm (<guix-configuration>)[channels]: Set default to #f.
Change-Id: I516c1735a037a153fabbebfc337051aaf0be2155
Signed-off-by: Ludovic Courtès <ludo@gnu.org>
This is a followup to 7c27bd115b.
* gnu/services/base.scm (file-system-shepherd-service): Catch
'system-error from ‘umount’ call when FILE-SYSTEM is marked as
mount-may-fail.
Change-Id: I2234f8da320b43089f4ee058cad8608ce9c078f8
Add a mechanism to only require mounting a subset of file-system entries
during early Shepherd initialization. Any file-system with additional Shepherd
service requirements (e.g. networking) is not required to provision
'file-systems.
* gnu/services/base.scm (file-system-shepherd-service): Splice
file-system-requirements into the Shepherd service requirement list.
(file-system-shepherd-services): Provision 'file-system only when file system
services without additional Shepherd requirements are started.
* gnu/system/file-systems.scm (file-system): Add shepherd-requirements field
to the file-system record. This field is used for adding additional Shepherd
requirements to a file-system Shepherd service.
* doc/guix.texi: Add documentation for file-system shepherd-requirements.
Change-Id: If0392db03d48e8820aa53df1df482c12ec72e1a5
Signed-off-by: Ludovic Courtès <ludo@gnu.org>
This allows users to specify NSS plugins such as LDAP via the
‘name-services’ field of <nscd-configuration>. Failing that, user code
will dlopen whatever passwd/group plugins are listed in
/etc/nsswitch.conf, which is likely to fail, typically because those are
not in $LD_LIBRARY_PATH.
* gnu/services/base.scm (%nscd-default-caches): Add ‘passwd’ and ‘group’
caches.
Change-Id: I9c03346a1de2710685f7801eccd2e08007427f5d
The ‘action’ method was deprecated in Shepherd 0.10.0.
* gnu/services/base.scm (shepherd-set-http-proxy-action)
(shepherd-discover-action): Use ‘perform-service-action’ instead of the
now-deprecated ‘action’ method.
Change-Id: Ibe1c79a44148596292c2c8907011ec787f8a9ddd
It’s currently difficult to programmatically add a build-machine, because the
`service-extension' mechanism is the only facility which can do that. It
relies on the `guix-service-type', ala:
(service-extension guix-service-type
(guix-extension (build-machines (list ...))))
...but `guix-service-type' is already instantiated as part of
`%base-services', and replacing it may lose other configuration, like
substitute servers and authorized keys.
Additionally, a default value of `#f' for the build-machines field requires
guarding uses of the field with:
(or (guix-build-machines config) '())
Changing the default to be the empty list avoids that. One can now add
build-machines with code such as:
(modify-services %base-services
(guix-service-type
config =>
(guix-configuration
(inherit config)
(authorized-keys
(cons %build-machine-key
(guix-configuration-authorized-keys config)))
(build-machines (cons #~(build-machine ...)
(guix-configuration-build-machines config))))))
* gnu/services/base.scm (guix-configuration): Rename `guix-build-machines' to
`guix-configuration-build-machines' and export it. Change the default from
`#f' to the empty list.
* gnu/services/base.scm (guix-activation): Update the build-machines test and
reverse the conditions.
Change-Id: I6780c6a5579fd9d4b4f22ee2b2bf7ba7a0454407
Signed-off-by: Ludovic Courtès <ludo@gnu.org>
Previously, if /etc/guix/machines.scm was a dangling symlink, it would
be kept (because ‘file-exists?’ would return #f) and thus the following
‘symlink’ call would throw with EEXIST.
* gnu/services/base.scm (guix-machines-files-installation): Use ‘lstat’
rather than ‘file-exists?’.
Change-Id: I07c7eed842dacabbd19ae2a17ac3e59cf26e46b2
The #:lazy-start? parameter was added in the Shepherd 0.10.0, which is
required since commit 477d4f7d24.
* gnu/services/base.scm (guix-publish-shepherd-service): Use
‘make-systemd-constructor’ unconditionally and pass #:lazy-start?.
Change-Id: I95aa079732cd21e32091c7deea2ed2f1bc50f2f2
Having a timeout seems generally preferable as it makes sure build slots
are not kept busy for no good reason (few package builds, if any, are
expected to exceed these values).
* nix/libstore/globals.cc (Settings::Settings): Change ‘maxSilentTime’
and ‘buildTimeout’.
* gnu/services/base.scm (<guix-configuration>)[max-silent-time]
[timeout]: Change default values.
* doc/guix.texi (Invoking guix-daemon, Base Services): Adjust
accordingly.
Change-Id: I25c50893f3f7fcca451b8f093d9d4d1a15fa58d8
* gnu/services/base.scm (guix-machines-files-installation): Handle
machines being a mixed list of build-machines and lists of
build-machines.
* doc/guix.texi (Base Services): Document it.
Change-Id: Ie404562ca0b564413233c3a624046da831893dc3
Co-authored-by: Ludovic Courtès <ludo@gnu.org>
The udev-configuration record now has a hardware field.
The contents of the /etc/udev directory now includes hwdb.bin, which is
computed when the system is instanciated (prior to system activation). The
hardware description files used to generate hwdb.bin are not installed in
/etc, because they are not required at run-time.
The documentation has been reworked so as to explain why creating udev rules
or hardware needs helper functions for configuration or extension.
* gnu/services/base.scm (udev-hardware): New function.
(file->udev-hardware): New function.
(udev-hardware-service): New function.
(udev-etc): Add hwdb.d and hwdb.bin.
(module): Export udev-hardware, file->udev-hardware, and udev-hardware-service.
(<udev-configuration>): Add the native-udev field.
(udev-service-type) [extend]: Populate the hardware field.
* doc/guix.texi (Base Services)[udev-service-type]: Explain configuration and
extension values.
* doc/guix.texi (Base Services)[udev-hardware]: Document it.
[udev-hardware-service]: Same.
* doc/guix.texi (Base Services)[udev-configuration]: Document the native-udev
field.
* gnu/services/base.scm (udev-configurations-union): New function.
(udev-configuration-file): New function, use file->udev-configuration-file.
(file->udev-configuration-file): New function.
(udev-rules-union): Use udev-configurations-union.
(udev-rule): Use udev-configuration-file.
(file->udev-rule): Use file->udev-configuration-file.
* gnu/services/base.scm (guix-machines-files-installation): New
procedure.
(<guix-configuration>)[build-machines]: New field.
(guix-activation): Call ‘ guix-machines-files-installation’.
(<guix-extension>)[build-machines]: New field.
(guix-extension-merge): Handle it.
(guix-service-type)[extend]: Likewise.
* doc/guix.texi (Daemon Offload Setup): Add note linking to
‘guix-configuration’.
(Base Services): Document ‘build-machines’ field of <guix-configuration>
and of <guix-extension>.
(Virtualization Services): Add ‘hurd-vm’ anchor.
Fixes a bug introduced in 0dd293b4d9
whereby guix-daemon on GNU/Hurd would have ‘GUIX_LOCPATH’ set to the
“wrong” locale data (2.35 instead of 2.37).
Consequently, it would fail to setlocale(3) and calls to
‘std::stoi’ (when reading the output of ‘guix authenticate’) would
throw, leading to this error message of guix-daemon:
unexpected build daemon error: stoi
This would manifest when sending store items to a childhurd:
$ guix copy --to=localhost:10022 sed
guix copy: sending 1 store item (1 MiB) to 'localhost'...
guix copy: error: unknown error while sending files over SSH
The “unknown error” is the ‘stoi’ exception.
This commit fixes that, but for the ‘guix-daemon’ service only.
* gnu/services/base.scm (guix-shepherd-service)[locales]: New variable.
Use it instead of ‘glibc-utf8-locales’.
* gnu/services/base.scm (<network-link>): Add mac-address field. Set
type field to #f by default, so it won't be mandatory. network-link
without a type will be used for existing interfaces.
(assert-network-link-mac-address, mac-address?): Add sanitizer. Allow
valid mac-address or #f.
(assert-network-link-type): Add sanitizer. Allow symbol or #f.
* gnu/services/base.scm (network-set-up/linux,
network-tear-down/linux): Adapt to new structure.
* doc/guix.texi (Networking Setup): Document it.
* gnu/tests/networking.scm (run-static-networking-advanced-test): New
variable.
Signed-off-by: Ludovic Courtès <ludo@gnu.org>
* gnu/services/base.scm (pam-limits-service-type)[pam-extension]: Wrap
into a 'lambda' that takes 'limits-file'. Pass that in the <pam-entry>
'arguments' field. Define 'make-limits-file' and use it.
Remove ETC-SERVICE-TYPE extension.
Signed-off-by: Ludovic Courtès <ludo@gnu.org>
As suggested by Sergey Bugaev on bug-hurd
https://lists.gnu.org/archive/html/bug-hurd/2023-05//msg00455.html
* gnu/services/base.scm (static-networking->hurd-pfinet-options): Use
/DEV/ethX. Use long options for settrans.
Signed-off-by: Josselin Poiret <dev@jpoiret.xyz>
Without this a user's pam-limits-service-type configuration does not have an
effect when using lightdm.
* gnu/services/base.scm (pam-limits-service-type): Add "lightdm" to the list.
Previously, with shepherd 0.10.0, guix-daemon would start quickly,
possibly before avahi-daemon is running. Consequently, its "guix
discover" child process would exit immediately with a warning saying
"Avahi daemon is not running".
* gnu/services/base.scm (guix-shepherd-service): When DISCOVER? is true,
add 'avahi-daemon' to 'requirement'.
* gnu/services/base.scm (%root-file-system-shepherd-service): In 'stop'
method, remove reference to 'stop-logging'. That procedure is gone in
Shepherd 0.10.0, leading to an unbound-variable exception.
Additionally, calling is unnecessary since 0.4.0, where shepherd logs to
syslog (thus the open file descriptor is not backed by an on-disk file).
This is a followup to 57e731c358.
* gnu/services/base.scm (agetty-shepherd-service): Change 'stop' method
to return #f immediately when the running value is 'idle.
Fixes a race condition when starting services in parallel with shepherd
0.10.x whereby a service might create files and directories with umask #o137.
An example is the bitlbee service with its least-authority wrapper: the
wrapper would create a tree with directories set to #o640, thereby
making the whole directory tree inaccessible.
* gnu/services/base.scm (syslog-shepherd-service): Pass #:file-creation-mask
to 'make-forkexec-constructor' instead of calling 'umask' in PID 1.
