Hi Magnus, Markus and Javier - thanks for the reports, and special
thanks Magnus for the patch.

I've came up with a different approach [0] - initramfs-tools provides a
retry mechanism in which scripts are able to run multiple times on
local-block stage. The mdadm script that implements such "poor man's
last resource" (i.e., the degrading array assemble) runs in local-block
stage, but cryptroot fails right upfront in local-top, preventing mdadm
scripts from starting the array.

My approach (patch attached here) relies in changing cryptroot in order
it fails gracefully on local-top and allows local-block script to run.
It'll then be executed itself on local-block, and decrypt the volume
successfully.

Test results with my patch and reviews are greatly appreciated!
Cheers,


Guilherme


[0]https://salsa.debian.org/cryptsetup-team/cryptsetup/-/merge_requests/18
From 13666455fe3ef5d2aa1d926969fce431ffe42e4c Mon Sep 17 00:00:00 2001
From: "Guilherme G. Piccoli" <gpicc...@canonical.com>
Date: Wed, 8 Jul 2020 09:54:25 -0300
Subject: [PATCH] d/initramfs/cryptroot-script: Allow some retries on
 local-block stage

Currently, cryptsetup try mounting encrypted rootfs only in local-top,
failing if not possible. Hence cases like LUKS on top of RAID1 (if the
array is degraded) cannot work properly.

This patch allows cryptsetup to retry on local-block stage, relying in
a heuristic based on ROOTDELAY and on initramfs looping at local-block
phase. We also added a script to local-bottom stage in order to clean
some control files used to track progress in local-block.

The tests with Debian Buster were successful; we created RAID1 and a
LUKS volume on top of it, to hold the rootfs. By removing one disk of
RAID1, we fail to boot without the patch (initramfs-tools drops into a
shell), whereas with the patch we have the boot succeeding.

Signed-off-by: Guilherme G. Piccoli <gpicc...@canonical.com>
---
 debian/cryptsetup-initramfs.install           |  1 +
 debian/functions                              |  2 +
 .../initramfs/scripts/local-block/cryptroot   |  4 ++
 .../initramfs/scripts/local-bottom/cryptroot  | 23 ++++++
 debian/initramfs/scripts/local-top/cryptroot  | 70 +++++++++++++------
 5 files changed, 80 insertions(+), 20 deletions(-)
 create mode 100644 debian/initramfs/scripts/local-bottom/cryptroot

diff --git a/debian/cryptsetup-initramfs.install b/debian/cryptsetup-initramfs.install
index 026ea37c0ff4..6780893ac2d2 100644
--- a/debian/cryptsetup-initramfs.install
+++ b/debian/cryptsetup-initramfs.install
@@ -5,5 +5,6 @@ debian/initramfs/hooks/*                            /usr/share/initramfs-tools/h
 debian/initramfs/scripts/local-block/cryptroot      /usr/share/initramfs-tools/scripts/local-block/
 debian/initramfs/scripts/local-bottom/cryptgnupg-sc /usr/share/initramfs-tools/scripts/local-bottom/
 debian/initramfs/scripts/local-bottom/cryptopensc   /usr/share/initramfs-tools/scripts/local-bottom/
+debian/initramfs/scripts/local-bottom/cryptroot     /usr/share/initramfs-tools/scripts/local-bottom/
 debian/initramfs/scripts/local-top/cryptopensc      /usr/share/initramfs-tools/scripts/local-top/
 debian/initramfs/scripts/local-top/cryptroot        /usr/share/initramfs-tools/scripts/local-top/
diff --git a/debian/functions b/debian/functions
index 54c7630b9649..673e94305251 100644
--- a/debian/functions
+++ b/debian/functions
@@ -9,6 +9,8 @@ else
     TABFILE="${TABFILE-/etc/crypttab}"
 fi
 export DM_DEFAULT_NAME_MANGLING_MODE=hex # for dmsetup(8)
+export CRYPTR_LOCAL_BLOCK="/run/cryptroot.local-block"
+export CRYPTR_CNT_FILE="/run/cryptroot.initrd.cnt"
 
 # Logging helpers. Send the argument list to plymouth(1), or fold it
 # and print it to the standard error.
diff --git a/debian/initramfs/scripts/local-block/cryptroot b/debian/initramfs/scripts/local-block/cryptroot
index 8a9b4c02c54d..8719fb6b4801 100644
--- a/debian/initramfs/scripts/local-block/cryptroot
+++ b/debian/initramfs/scripts/local-block/cryptroot
@@ -15,6 +15,10 @@ prereqs)
 	;;
 esac
 
+[ -f /lib/cryptsetup/functions ] || return 0
+. /lib/cryptsetup/functions
+
 if [ -x /scripts/local-top/cryptroot ]; then
+	touch ${CRYPTR_LOCAL_BLOCK}
 	exec /scripts/local-top/cryptroot
 fi
diff --git a/debian/initramfs/scripts/local-bottom/cryptroot b/debian/initramfs/scripts/local-bottom/cryptroot
new file mode 100644
index 000000000000..4f44d38c9cd5
--- /dev/null
+++ b/debian/initramfs/scripts/local-bottom/cryptroot
@@ -0,0 +1,23 @@
+#!/bin/sh
+set +x
+PREREQ=""
+
+prereqs()
+{
+	echo "$PREREQ"
+}
+
+case $1 in
+prereqs)
+	prereqs
+	exit 0
+	;;
+esac
+
+# If we reached this stage, we do have a rootfs mounted
+# so let's clean-up cryptroot setup mess...
+[ -f /lib/cryptsetup/functions ] || return 0
+. /lib/cryptsetup/functions
+
+rm -f ${CRYPTR_LOCAL_BLOCK}
+rm -f ${CRYPTR_CNT_FILE}
diff --git a/debian/initramfs/scripts/local-top/cryptroot b/debian/initramfs/scripts/local-top/cryptroot
index 4f36259f80f4..be3d06a18dec 100644
--- a/debian/initramfs/scripts/local-top/cryptroot
+++ b/debian/initramfs/scripts/local-top/cryptroot
@@ -31,8 +31,8 @@ esac
 
 
 # wait_for_source()
-#   Wait for encrypted $CRYPTTAB_SOURCE for up to 180s.  Set
-#   $CRYPTTAB_SOURCE to its normalized device name when it shows up;
+#   Wait for encrypted $CRYPTTAB_SOURCE . Set $CRYPTTAB_SOURCE
+#   to its normalized device name when it shows up;
 #   return 1 if timeout.
 wait_for_source() {
     wait_for_udev 10
@@ -42,17 +42,27 @@ wait_for_source() {
         return 0
     fi
 
-    # The lines below has been taken from
-    # /usr/share/initramfs-tools/scripts/local's local_device_setup(),
-    # as suggested per https://launchpad.net/bugs/164044
-
     # If the source device hasn't shown up yet, give it a little while
     # to allow for asynchronous device discovery (e.g. USB).
+    #
+    # We also need to take into account RAID or other devices that may
+    # only be available on local-block stage. So, wait 5 seconds upfront,
+    # in local-top; if that fails, end execution relying on local-block
+    # invocations. Allow $ROOTDELAY/4 invocations with 1s sleep times (with
+    # a minimum of 20 invocations), and if after that we still fail, then it's
+    # really time to give-up. Variable $initrd_cnt tracks the re-invocations.
+    #
+    # Part of the lines below has been taken from initramfs-tools
+    # scripts/local's local_device_setup(), as suggested per
+    # https://launchpad.net/bugs/164044 .
+
+    local slumber=1
+    if [ ! -f "${CRYPTR_LOCAL_BLOCK}" ]; then # we are running on local-top
+         slumber=5
+    fi
 
     cryptsetup_message "Waiting for encrypted source device $CRYPTTAB_SOURCE..."
 
-    # Default delay is 180s, cf. initramfs-tools(8)
-    local slumber="${ROOTDELAY:-180}"
     while [ $slumber -gt 0 ]; do
         sleep 1
 
@@ -75,7 +85,21 @@ wait_for_source() {
 #   Set up a crypttab(5) mapping defined by $CRYPTTAB_NAME,
 #   $CRYPTTAB_SOURCE, $CRYPTTAB_KEY, $CRYPTTAB_OPTIONS.
 setup_mapping() {
-    local dev
+    local dev initrd_cnt
+
+    # We control here the number of re-invocations of this script from
+    # local-block - the heuristic is $ROOTDELAY/4, with a minimum of 20.
+
+    if [ -f "${CRYPTR_CNT_FILE}" ]; then
+	initrd_cnt=$(cat ${CRYPTR_CNT_FILE})
+else
+	initrd_cnt=${ROOTDELAY:-80}
+	initrd_cnt=$((initrd_cnt/4))
+	if [ "${initrd_cnt}" -lt 20 ]; then
+		initrd_cnt=20
+	fi
+	echo ${initrd_cnt} > "${CRYPTR_CNT_FILE}"
+fi
 
     # The same target can be specified multiple times
     # e.g. root and resume lvs-on-lvm-on-crypto
@@ -86,17 +110,23 @@ setup_mapping() {
     crypttab_parse_options --export --missing-path=fail || return 1
 
     if ! wait_for_source; then
-        # we've given up
-        if [ -n "$panic" ]; then
-            panic "ALERT! encrypted source device $CRYPTTAB_SOURCE does not exist, can't unlock $CRYPTTAB_NAME."
-        else
-            # let the user fix matters if they can
-            echo "	ALERT! encrypted source device $CRYPTTAB_SOURCE does not exist, can't unlock $CRYPTTAB_NAME."
-            echo "	Check cryptopts=source= bootarg: cat /proc/cmdline"
-            echo "	or missing modules, devices: cat /proc/modules; ls /dev"
-            panic "Dropping to a shell."
-        fi
-        return 1 # can't continue because environment is lost
+	if [ ${initrd_cnt} -eq 0 ]; then
+		# we've given up
+		if [ -n "$panic" ]; then
+			panic "ALERT! encrypted source device $CRYPTTAB_SOURCE does not exist, can't unlock $CRYPTTAB_NAME."
+		else
+			# let the user fix matters if they can
+			echo "	ALERT! encrypted source device $CRYPTTAB_SOURCE does not exist, can't unlock $CRYPTTAB_NAME."
+			echo "	Check cryptopts=source= bootarg: cat /proc/cmdline"
+			echo "	or missing modules, devices: cat /proc/modules; ls /dev"
+			panic "Dropping to a shell."
+		fi
+		return 1 # can't continue because environment is lost
+	else
+		initrd_cnt=$((initrd_cnt - 1))
+		echo ${initrd_cnt} > "${CRYPTR_CNT_FILE}"
+		return 0 # allow some attempts on local-block stage
+	fi
     fi
 
     # our `cryptroot-unlock` script searches for cryptsetup processes
-- 
2.27.0

Reply via email to