Using --run with incremental assembly ensures that the system will be able to
boot and won't drop to recovery shell when only the minimal number of devices
are present. [1]

[1] https://git.kernel.org/pub/scm/utils/mdadm/mdadm.git/tree/ReadMe.c?h=mdadm-4.1#n600

Handle case when the crypt header is a normal file which is included in
initramfs and not a blockdevice.

if search device is prefixed with ZFS= then we search for a label with
the zpool name in the zfs path. For example, if search device is
"ZFS=tank/alpine/root" then we search for device that is type
"zfs_member" and label "tank".

This makes it work better with grub which creates a boot cmdline with
ZFS=

lemmarathon authored Apr 30, 2018 and Natanael Copa committed Jun 12, 2018

The "cryptkey" boot parameter enables keyfile decryption. By default,
init will look for a keyfile named "/crypto_keyfile.bin". Another file
may be specified like so: "cryptkey=/path/to/keyfile.bin". If keyfile
decryption fails, init will fall back to passphrase mode.

Since we previously specified an explicit request_type for the
crypt_load() function nlplug-findfs couldn't open LUKS2 containers. By
using CRYPT_LUKS crypt_load accepts any known LUKUS container format.

We could add an additional command line flag to nlplug-finds for
specifying the request_type but I guess this is good enough for now.

When enabled allows the use of discard (TRIM) requests for the device.
See cryptsetup(1) for more information.

Without a call to crypt_set_data_device(), the cryptsetup system does
not know where to find the data device. It works whether the header is
deported or not, according to
https://github.com/mbroz/cryptsetup/blob/8f84fb49faa69b0ddde3d534ee9c72119256f4c9/src/cryptsetup.c#L782
so it is fine to call it in all cases.

Natanael Copa authored Jul 05, 2016 and 7heo .tk committed Apr 05, 2017

move the logic to separate function

Natanael Copa authored Jul 05, 2016 and 7heo .tk committed Apr 05, 2017

put all data related cryptsetup in a struct. No changes in
functionality.

We could sucessfully read the password but fail to reset the tty. If
that happens, then just warn, but return success.

we dont need mount and scan the tree if not explicitly told to look for
apkovls or bootrepos.

ref #6473

Likely introduced in the major rehaul done in
e4af128b

makes things a bit cleaner

Increase timeout to 5sec if we have not found anything so we don't get
error too early.

If boot repos are found then reduce the event timeout to 250ms. If
usb_storage is found, then always add 1 second of delay in addition, to
let the usb host settle.

certain platform and usb devices expose things deep down the tree,
increase the recursion limit

raid disks are created zero sized. when mdadm has found all the
disks it finally enables it. this causes md devices to be zero
sized for some time after their creation, and blkid will not
detect them. catch the 'change' notification and rescan block
devices for updated blkid.

Based on patch by donoban. Limit recursion depth for repository
search to 2 levels (shell script had -maxdepth 3 for finding the
file entry, so it's maximum of 2 levels of directories). For sysfs
entries deeper search is allowed. ref #5192

add -n option to allow return success even if nothing was found. This is
useful for net boot, where apkovl and repositories are on network.

kernel does not set any "DRIVER" env var in uevent from netlink so we
use kmod to find usb_storage driver.

We also reduce the timeout to 1 sec, which should be enough.

Avoid printing error message if there are lvm devices but root is not on
any of those.