| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Process ADD and CHANGE uevents as prior to the blamed commit. These
can be injected from userspace, and the ADD uevent is important for
devices that get enumerated during early boot (ie. prior to rootfs being
mounted). udev later cold-plugs them into the system via synthetic
ADD uevents (see systemd-udev-trigger.service) to trigger the rule
processing.
As we're not getting synthetic BIND uevents for such devices,
matching on the ADD uevents is needed to get networking set up for
eg. initramfs-less systems.
Note that we're still restricting the matching to devices with the
expected DRIVER attribute.
Fixes: 4b1fc32788a3 ("ccw.udev: match BIND uevent for ccw network devices")
Reported-by: Sven Schnelle <svens@linux.ibm.com>
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Tested-by: Sven Schnelle <svens@linux.ibm.com>
|
|
|
|
|
|
|
| |
As the ccw.udev rule now only matches BIND uevents, $DRIVER is always
set and we don't need to extract it from sysfs.
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
|
|
|
|
|
|
|
| |
There's nothing in the kernel that would raise a CHANGE uevent for ccw
devices, so don't match it until we know what the semantics might be.
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Instead of matching the ADD uevent for a hard-coded list of ccw device
types, wait until the device is actually bound to a driver that follows
the ccwgroup model. This way we don't need to maintain a hard-coded list
of device types, and avoid trying to apply a configuration for devices
that the kernel drivers don't even support.
Also when building the ccwgroup device via /sys/bus/ccwgroup/.../group,
the ccwgroup bus code requires that the ccw devices are bound to the
correct ccw_driver. So matching the ADD uevent seems too early.
This eliminates support of some truly ancient device types (P390, CLAW),
and allows for phasing out support of eg. OSN and OSX.
Note that this requires kernel-side support that was added in 4.14
with commit 1455cf8dbfd0 ("driver core: emit uevents when device is
bound to a driver").
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
|
| |
|
|
|
|
| |
(#1790496, #1790790)
|
| |
|
|
|
|
|
|
|
|
| |
update scripts for https://fedoraproject.org/wiki/Changes/NetworkManager_keyfile_instead_of_ifcfg_rh
that will be implemented in Fedora 33.
In addition to ifcfg files we must look at the NM keyfiles too. New systems
will use keyfiles only, but upgraded systems use ifcfg still.
|
| |
|
| |
|
| |
|
|
|
|
| |
script had moved to /usr/sbin
|
| |
|
| |
|
|
|
|
|
|
|
|
| |
Having the udeavdm settle call in the global wait_on_devices function
causes a 2 minutes delay during normal startup. Until the cause is known
move it just to the wait after freeing aa specified set of devices.
https://bugzilla.redhat.com/show_bug.cgi?id=1054691#c15
|
|
|
|
|
|
| |
Use cio_settle mechanism in both paths where wait_on_devices is used. Also
issue udevadm settle to resolve timing issues when anaconda (or rather blivet)
discovers new zFCP devices.
|
|
|
|
| |
zfcpconf.sh tries to online a zfcp device multiple times (rhbz#1042496)
|
| |
|
|
|
|
|
| |
We pipe the output of this script into a file, so we don't want any
error strings to leak into it.
|
|
|
|
|
|
|
|
|
| |
Format A from CMSCONFFILE or parm file,
<range>(feature),<range>(feature)
Format B from zipl.conf, <dasd>,feature=N
We have to support and "normalize" both.
|
| |
|
|
|
|
| |
This reverts commit 7f5df88d984fcecb086a41bb96cf9cc423c77f19.
|
|
|
|
|
| |
Running the device_cio_free job when rcS is in the stopped state should solve
a race between device_cio_and and /proc possibly not yet mounted (rhbz#752146)
|
| |
|
| |
|
|
|
|
|
|
|
| |
The /proc/cio_settle interface (if available) is used to wait until
devices become online. It should seedup the whole process of unblocking
devices and also solves situations when non-existent devices are
unblocked and timeout is reached when waiting for every such device.
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
| |
grep should be run with a predefined locale to get reproducable results
|
|
|
|
|
|
|
| |
__sed_discard_ignored_files
device_cio_ignore should match the behaviour of the initscripts network setup functions
and ignore the config files that match the expression in __sed_discard_ignored_files
|
| |
|
| |
|
|
|
|
|
| |
When the DEVICE variable exists in the environment that calls device_cio_free
then the devices from the command line are appended to this value (rhbz#595569).
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
| |
Cecause the seq utility lives in /usr/bin and the script must be runnable
without the /usr filesystem mounted, it's replaced by a while cycle.
|
|
|
|
|
| |
use sed for lowercase (and comma to space) substitutions because it lives in /bin
while tr lives in /usr/bin and thus requires /usr to be mounted
|
|
|
|
| |
The scripts can be run in an environment where /usr is not mounted yet.
|
|\ |
|
| | |
|
| | |
|
| |
| |
| |
| |
| | |
This is second version of the merged ccw_init script that is shared
between dracut and udev.
|
| | |
|