Subtle: dhcpcd.service would call resolvconf during shutdown, which in
turn would start invalidate-nscd.service, causing the shutdown to be
cancelled. Instead, give nscd.service a proper reload action, and do
"systemctl reload --no-block nscd.service". The --no-block is
necessary to prevent that command from waiting until a timeout occurs
(bug in systemd?).
As non-QWERTY keyboards don't feel so warm and cozy if they hug QWERTY LUKS
password prompts, it was on honor for me to serve King Dvorak XV to fight the
glorious keyboard war against... what?! Yes, I'm awake!
We're fighting with loadkeys to spit out busybox binary keymaps against loadkmap
(loadkeys does have a special target -b for that).
And yep, I'm somewhat abusing preLVMCommands, if someone got issues with that,
feel free to introduce a new substitute in stage-i-init.sh.
Sent from my iPhone
We had a "mount -o remount,rw none /" that was setting back 'relatime',
although we had set 'noatime' at initrd mount. Removing the word 'none' fixed
it.
Specifying a device (in this case 'none'), makes mount to forget previous
device options. According to manpage, it says not to read fstab or mtab. But the
effect is that of setting 'relatime', if it was mounted 'noatime.
BusyBox doesn't handle the "auto" filesystem type very well: fsck will
just ignore such filesystems, and mount will only work properly if the
required kernel module is already loaded. Therefore, use blkid to
determine the filesystem type.
Also generate an /etc/fstab in the initrd rootfs on the fly. This is
useful if you're dropped into an emergency shell since it allows you
to say "fsck /dev/sda1" or "mount /dev/sda" and have the right thing
happen.
Using BusyBox instead of Bash plus a bunch of other tools gives us a
much more feature-full, yet smaller initrd. In particular, BusyBox
contains networking commands such as ip and a DHCP client, useful for
NFS boots. It's also much more convenient for rescue situations
because the shell has builtin readline support and there are many more
tools (including vi).
Enabled a bunch of units that ship with systemd. Also added an option
‘boot.systemd.units’ that can be used to define additional units
(e.g. ‘sshd.service’).
Upstart requires /dev/ptmx since its 1.4, and will lock up in case of it missing.
I was hitting this in the fuloong, where I don't use the nixos initrd.
svn path=/nixos/trunk/; revision=34429
* Load scsi_wait_scan after all other kernel modules to ensure that
all SCSI device nodes have been created.
* Increase the timeout for the appearance of the root device to 20
seconds.
* Do a "udevadm settle" just after the root device has appeared to
make sure that udev isn't accessing the device anymore (hopefully).
On EC2 (Xen), I've seen fsck on the root fail randomly with "device
in use" errors.
svn path=/nixos/trunk/; revision=33650
* Add a slash to the end of $MODULE_DIR, as expected by depmod. (Not
that running depmod from the command line is all that useful, since
you can't use it to update the tree in the Nix store. But at least
commands like "depmod -n" work now.) Reported by Kirill Elagin on
IRC.
svn path=/nixos/trunk/; revision=33312
It needs udevd to be running because the modules may require
firmware. Thanks to Mathijs and Arie for pointing this out.
svn path=/nixos/trunk/; revision=33234
modprobe.
* Move the implementation of boot.kernelModules from the udev job to
the activation script. This prevents races with the udev job.
* Drop references to the "capability" kernel module, which no longer
exists.
svn path=/nixos/trunk/; revision=33208
This allows setting a fixed device name per array, thus ensuring that at boot,
the arrays will always be mounted with the same names. I think this allows
solving the problem of grub getting confused about softraid device names
(prefix and root), if the devices always get the same naming at initrd.
svn path=/nixos/trunk/; revision=33033
I set it as default because users can benefit of this without having to prepare
their nixos first, and I don't think it will break any nixos for the initrd
size increase.
It can be disabled with 'boot.initrd.withExtraTools = false'.
svn path=/nixos/trunk/; revision=33000
reiserfs now have separate modules that are conditional on
boot.supportedFilesystems and boot.initrd.supportedFilesystems.
By default, these include the filesystems specified in the fsType
attribute in fileSystems. Ext2/3/4 support is currently
unconditional.
Also unbreak the installer test (http://hydra.nixos.org/build/2272302).
svn path=/nixos/trunk/; revision=32954
pierron recommended the use of types.string over mergeOptionString, as
it is superior but might break things.
For my system the change evaluated to the exactly same.
svn path=/nixos/trunk/; revision=31138
This reverts commit 025f8c40b40fad50086e8761eee61098d8fb2651.
The check was intened for building the initrd of the installer.
svn path=/nixos/trunk/; revision=31137
popt-0.16 and cryptsetup-1.4.1 both generated pkgconfig (in contrast
to older versions). The pkgconfig files (popt.pc and cryptsetup.pc)
contain references into the store that are not removed by patchelf and
stage-1 fails with errors like: "output is not allowed to refer to
path `/nix/store/qccjhn063cfv171rcaxvxh0yk96zf7l2-cryptsetup-1.4.1'".
Now, only the cryptsetup binaries and its dependencies are copied,
determined by ldd. In addition the cryptsetup binary and lvm are
tested after patchelf has adjusted the library paths.
Thanks to Peter Simons and Eelco Dolstra for giving the rights hints.
svn path=/nixos/trunk/; revision=31128
After the change from revision 30103, nixos-rebuild suddenly consumed
freaky amounts of memory. I had to abort the process after it had
allocated well in excess of 30GB(!) of RAM. I'm not sure what is causing
this behavior, but undoing that assignment fixes the problem. The other
two commits needed to be revoked, too, because they depend on 30103.
svn path=/nixos/trunk/; revision=30127
possibility that a udevd process survives, preventing udevd from
starting in stage 2:
machine# udevd[1421]: bind failed: Address already in use
machine# udevd[1421]: error binding udev control socket
svn path=/nixos/trunk/; revision=29434
was already the case on Linux 2.6.32, but in newer kernels the CFQ
scheduler is built as a module, so all block devices got the ‘none’
scheduler instead.
svn path=/nixos/trunk/; revision=28972
be set when udevd calls external programs. (The udev manpage claims
that udevd passes its own environment variables, but this is not the
case.)
* Get rid of some udev rule hacks that no longer seem needed.
svn path=/nixos/trunk/; revision=25991
problem is that configuration values below a mkIf are evaluated
strictly even if the condition is false. Thus "${luksRoot}" causes
an evaluation error. As a workaround, use the empty string instead
of `null' as the default value. However, we should really fix the
laziness of mkIf. It's likely that NixOS evaluation would be much
faster if it didn't have to evaluate disabled configuration values.
svn path=/nixos/trunk/; revision=24477
init script. This removes the need for the `systemConfig' boot
parameter; `init=<stage-2-init>' is enough. However, the GRUB menu
builder still needs to add `systemConfig' to the kernel command line
for compatibility with old configurations.
svn path=/nixos/trunk/; revision=23775
* Moved some scriptlets to the appropriate modules.
* Put the scriptlet that sets the default path at the start, since it
never makes sense not to have it there. It no longer needs to be
declared as a dependency.
* If a scriptlet has no dependencies, it can be denoted as a plain
string (i.e., `noDepEntry' is not needed anymore).
svn path=/nixos/trunk/; revision=23762
