kpcyrd authored Jun 26, 2021 and Natanael Copa committed Aug 06, 2021

See https://twitter.com/sn0int/status/1408853993774845952

omni authored Jan 15, 2021 and Natanael Copa committed Jun 07, 2021

- add zstd to mkinitfs usage output
- add flags -C and -s to man page
- update bugtracker URL
- proofreading

Import the pool without mounting any file systems.

See merge request !78

Output init progress to /dev/kmsg

See merge request !61

features: fix addition of af_packet kernel module for dhcp

See merge request !64

initramfs-init: fix importing the root pool

See merge request !63

Thermi authored Jan 24, 2020 and Natanael Copa committed Dec 22, 2020

Run mkdir -p before copying the kernel modules. Makes sure the dirs exist before calling cp on them.

Fix for issue #4 "mkinitfs fails if feature list is empty"

Closes #4

See merge request !66

initramfs-init: add cmdline consoles to inittab

See merge request !69

Allow embedding custom files into initfs

See merge request !70

Current kernels use ext4 for ext2/3, update config

Closes #1

See merge request !67

mkinitfs: add zFCP SCSI device as root disk on s390x

See merge request !71

mkinitfs: enable nvme by default

See merge request !73

Fix typo in prepare_zfs_root().

See merge request !72

features: add phy drivers to network

Closes aports#11715

See merge request !74

Add zstd compression support

See merge request !75

Now that 5.10 is out and supports zstd compression for the kernel and
initramfs, add support for zstd to mkinitfs. The benefits have been
discussed at length on lkml[1][2].

In my personal experience, the compression (mkinitfs) seems to be
slower, but decompression (booting) seems to be faster. I think most
people boot more than they mkinitfs.

The default is still gzip, so no change should be noticed unless
someone goes looking for trouble.

To test this functionality, just add the following to mkinitfs.conf:

```text
initfscomp=zstd
```

[1] https://lkml.org/lkml/2020/7/23/1010
[2] https://lkml.org/lkml/2020/7/1/722

Some ethernet drivers (e.g. Realtek r8169) require a corresponding
phy driver to function.

Fixes aports#11715

This typo causes $_root_pool to be an empty string, which causes the
subsequent `zpool list` to output encryption status for every pool
rather than just one, breaking the comparison.

On systems with a zfs root and more than one pool present, this
manifests as an error message on boot just before openrc is invoked:

    sh: enabled: unknown operand

Similar to DASD and QETH devices.

Example of kernel argument for 2 device paths:
zfcp=0.0.1905:0x50050763070845e3:0x4082400400000000,0.0.1945:0x50050763071845e3:0x4082400400000000

Each device path is separated by a comma (,). Adapter bus ID and WWPNs and
LUNs are separated by a colon (:).

Since Alpine has the advantage of using root=UUID= for root disk, device
order and device naming don't matter.

Sometimes a console is specified on the kernel command line but doesn't
get initialized by the kernel because for example the driver is built as
a module. The init should still spawn getty on it.

Follow-up to 5a6f829e so `mkinitfs -F dhcp` works as expected.

Without this change, `ip=dhcp` will not work on Raspberry Pi images:

    $ sha256sum alpine-rpi-3.11.3-armv7.tar.gz
    438648629838b02b2eead75ede2137dd480e1ae748b632b312a0b7b13924e6c2  alpine-rpi-3.11.3-armv7.tar.gz

    $ for P in rpi2 rpi4; do
        tar -O -xzf alpine-rpi-3.11.3-armv7.tar.gz ./boot/initramfs-$P | gunzip | cpio -t | grep "af.*ko"
    done

    lib/modules/5.4.12-0-rpi2/kernel/drivers/usb/storage/ums-datafab.ko
    15320 blocks
    lib/modules/5.4.12-0-rpi4/kernel/drivers/usb/storage/ums-datafab.ko
    15552 blocks

ref https://gitlab.alpinelinux.org/alpine/aports/issues/10750

We accidentally used $_root_vol instead of $_root_pol for force importing.
Also import (but without -f) if we don't force import as this is required
on some setups. Add -d /dev as this is required for zpool to find the
pool if you have defined some subvolume as partition (e.g. for VMs)

Without this if you run setup-alpine with ROOTFS set to ext2/3, you're
rewarded with a recovery prompt on first boot, as the correct kernel
modules are missing.

pigz(1) is a parallel gzip implementation. As such, it significantly
faster than gzip(1). This patch uses pigz(1) instead of gzip(1),
if it is installed, otherwise it just falls back to gzip(1). This is
similar to the current pigz support in abuild.

Without this patch:

	$ time mkinitfs
	==> initramfs: creating /boot/initramfs-lts
	real	0m 38.19s
	user	0m 35.47s
	sys	0m 0.49s

With this patch applied and pigz installed:

	==> initramfs: creating /boot/initramfs-lts
	real	0m 13.24s
	user	0m 38.30s
	sys	0m 0.56s

This change was discussed on IRC with _ikke_ and mps.

This enables debug of early-init problems via the use of the earlyprintk kernel parameter.

The LUKS2 cryptsetup format uses a locking mechanism, see the "LUKS2
header locking" section in cryptsetup(8) for further information on this
mechanism. The mechanism requires a locking directory, if it is not
present it will be created automatically. Unfortunately, cryptsetup
emits a warning when creating it.

To silence this warning message make init create the directory.

Without this change this may result in a warning message to be printed
during boot if the rootfstype kernel parameter is not set.