[Accton][as9957-32db] Add IMSM RAID1 boot support for ONL

packages/base/all/vendor-config-onl/src/python/onl/mounts/__init__.py

@@ -9,6 +9,24 @@
 import yaml
 import tempfile
 import shutil
+import stat
+
+
+def resolve_block_device(label):
+    """Resolve a label using an optional boot-time device map."""
+    path = '/run/onl/block-devices.json'
+    if not os.path.exists(path):
+        return subprocess.check_output(('blkid', '-L', label)).strip()
+
+    with open(path) as stream:
+        devices = json.load(stream)
+    if label not in devices:
+        # A device map is authoritative. Do not search other disks.
+        raise subprocess.CalledProcessError(2, ['blkid', '-L', label])
+    device = devices[label]
+    if not device.startswith('/dev/') or not stat.S_ISBLK(os.stat(device).st_mode):
+        raise ValueError('Invalid block device for %s: %s' % (label, device))
+    return device
 
 class MountManager(object):
 
@@ -155,7 +173,7 @@ def _discover(k):
             lbl = v.get('label', k)
             useUbiDev = False
             try:
-                v['device'] = subprocess.check_output(('blkid', '-L', lbl,)).strip()
+                v['device'] = resolve_block_device(lbl)
             except subprocess.CalledProcessError:
                 useUbiDev = True
             if useUbiDev == True:
@@ -443,7 +461,7 @@ def __init__(self, label, logger):
 class OnlOnieBootContext(MountContext):
     def __init__(self, mdir="/mnt/onie-boot", mode="rw", label="ONIE-BOOT", logger=None):
         try:
-            device = subprocess.check_output("blkid -L %s" % label, shell=True).strip()
+            device = resolve_block_device(label)
         except subprocess.CalledProcessError:
             self.logger.debug("Block label %s does not yet exist..." % label)
             raise

packages/base/any/kernels/5.4-lts/configs/x86_64-all/x86_64-all.config

@@ -1623,7 +1623,7 @@ CONFIG_BLK_DEV_MD=y
 CONFIG_MD_AUTODETECT=y
 # CONFIG_MD_LINEAR is not set
 # CONFIG_MD_RAID0 is not set
-# CONFIG_MD_RAID1 is not set
+CONFIG_MD_RAID1=y
 # CONFIG_MD_RAID10 is not set
 # CONFIG_MD_RAID456 is not set
 # CONFIG_MD_MULTIPATH is not set

packages/platforms/accton/x86-64/as9957-32db/platform-config/r0/.gitignore

@@ -0,0 +1,8 @@
+src/lib/sbin/mdadm
+src/lib/sbin/mdmon
+src/lib/raid/COPYING.mdadm
+builds/mdadm-*.tar.xz
+builds/mdadm-*.tar.xz.part
+builds/.mdadm-build/
+__pycache__/
+*.pyc

packages/platforms/accton/x86-64/as9957-32db/platform-config/r0/PKG.yml

@@ -1 +1,33 @@
-!include $ONL_TEMPLATES/platform-config-platform.yml ARCH=amd64 VENDOR=accton BASENAME=x86-64-accton-as9957-32db REVISION=r0
+# Keep the RAID boot hooks local to this platform.
+prerequisites:
+  packages:
+    - onl-vendor-config-accton:all
+    - onl-platform-modules-x86-64-accton-as9957-32db:amd64
+
+common:
+  version: 1.0.0
+  arch: amd64
+  copyright: Copyright 2013, 2014, 2015 Big Switch Networks
+  maintainer: support@bigswitch.com
+  support: opennetworklinux@googlegroups.com
+  changelog: None
+  dists: $DISTS
+
+packages:
+  - name: onl-platform-config-x86-64-accton-as9957-32db-r0
+    depends: onl-vendor-config-accton,onl-platform-modules-x86-64-accton-as9957-32db
+    summary: ONL Platform Configuration Package for x86-64-accton-as9957-32db-r0
+    # Copy the platform tree before placing the boot entry inside it.
+    files:
+      - src/lib: /lib/platform-config/x86-64-accton-as9957-32db-r0/onl
+      - src/python: ${PY_INSTALL}/onl/platform/
+      - src/boot/mdraid: /lib/platform-config/x86-64-accton-as9957-32db-r0/onl/boot/x86-64-accton-as9957-32db-r0
+      - src/boot.d/05.mdraid: /etc/boot.d/05.mdraid
+      - src/init.d/onl-mdraid-clean: /etc/init.d/onl-mdraid-clean
+    # Register the shutdown script when the package is configured.
+    after-install: $__DIR__/src/init.d/postinst
+
+  - name: onl-platform-build-x86-64-accton-as9957-32db-r0
+    summary: ONL Platform Build Package for x86-64-accton-as9957-32db-r0
+    optional-files:
+      builds: $$PKG_INSTALL

packages/platforms/accton/x86-64/as9957-32db/platform-config/r0/builds/Makefile

@@ -0,0 +1,60 @@
+############################################################
+#
+# Static mdadm and mdmon for the RAID boot helper.
+#
+# Both tools are built from one pinned upstream release and are
+# linked statically. The loader and the runtime system then use
+# the same files, on every Debian release.
+#
+############################################################
+MDADM_VERSION := 4.3
+MDADM_ARCHIVE := mdadm-$(MDADM_VERSION).tar.xz
+# From https://www.kernel.org/pub/linux/utils/raid/mdadm/sha256sums.asc
+MDADM_SHA256 := 416727ae1f1080ea6e3090cea36dd076826fc369151e36ab736557ba92196f9f
+MDADM_URL := https://www.kernel.org/pub/linux/utils/raid/mdadm/$(MDADM_ARCHIVE)
+# Keep the upstream warnings, but do not stop on warnings from newer
+# compilers. Build without libudev: a static libudev is usually not
+# available, and only "mdadm --monitor" and waiting for udev use it.
+# mdadm 4.3 still needs the libudev.h header to compile.
+MDADM_CXFLAGS := -O2 -Wno-error -DNO_LIBUDEV
+
+# ONL adds the sbin directory of the running platform to PATH, so
+# "mdadm" and "mdmon" can be typed at the shell after login.
+SBIN := ../src/lib/sbin
+LICENSE := ../src/lib/raid/COPYING.mdadm
+WORK := .mdadm-build
+SOURCE := $(WORK)/mdadm-$(MDADM_VERSION)
+
+.PHONY: all clean
+
+all: $(SBIN)/mdmon
+
+# Download the source if it is not here. A copy placed here by hand
+# works without network access. Both are checked before the build.
+$(MDADM_ARCHIVE):
+	wget -O $@.part $(MDADM_URL)
+	echo "$(MDADM_SHA256)  $@.part" | sha256sum -c -
+	mv $@.part $@
+
+# The loader has no C library for these tools, so they must be static.
+# mdmon is installed last: it appears only after every check passed.
+$(SBIN)/mdmon: $(MDADM_ARCHIVE) Makefile
+	echo "$(MDADM_SHA256)  $(MDADM_ARCHIVE)" | sha256sum -c -
+	rm -rf $(WORK)
+	rm -f $(SBIN)/mdadm $(SBIN)/mdmon $(LICENSE)
+	mkdir -p $(WORK) $(SBIN)
+	tar -xf $(MDADM_ARCHIVE) -C $(WORK)
+	$(MAKE) -C $(SOURCE) CC=gcc CXFLAGS="$(MDADM_CXFLAGS)" LDFLAGS=-static RUN_DIR=/run/mdadm mdadm mdmon
+	strip $(SOURCE)/mdadm $(SOURCE)/mdmon
+	for tool in mdadm mdmon; do readelf -d $(SOURCE)/$$tool | grep -q "no dynamic section" || exit 1; done
+	$(SOURCE)/mdadm --version 2>&1 | grep -q "mdadm - v$(MDADM_VERSION) "
+	$(SOURCE)/mdmon --help 2>&1 | grep -q -- "--takeover"
+	cp $(SOURCE)/COPYING $(LICENSE)
+	install -m 0755 $(SOURCE)/mdadm $(SBIN)/mdadm
+	install -m 0755 $(SOURCE)/mdmon $(SBIN)/mdmon
+	rm -rf $(WORK)
+
+# Other files in the sbin directory are not touched.
+clean:
+	rm -rf $(WORK) $(MDADM_ARCHIVE).part
+	rm -f $(SBIN)/mdadm $(SBIN)/mdmon $(LICENSE)

packages/platforms/accton/x86-64/as9957-32db/platform-config/r0/src/boot.d/05.mdraid

@@ -0,0 +1,7 @@
+#!/bin/sh
+# Finish the handover before ONL upgrades or mounts runtime filesystems.
+[ -r /run/onl/mdraid.state ] || exit 0
+while ! python "/lib/platform-config/$(cat /etc/onl/platform)/onl/raid/onl-mdraid.py" takeover; do
+    echo 'ONL mdraid: mdmon handover failed. Exit this shell after correcting the error.' >&2
+    /bin/sh -i
+done

packages/platforms/accton/x86-64/as9957-32db/platform-config/r0/src/boot/mdraid

@@ -0,0 +1,15 @@
+#!/bin/sh
+# This hook runs before the loader checks or mounts ONL filesystems.
+# Keep early RAID logs available before switching rootfs.
+mdraid_log=/etc/onl/mdraid.log
+if python "/lib/platform-config/$(cat /etc/onl/platform)/onl/raid/onl-mdraid.py" loader >"$mdraid_log" 2>&1; then
+    cat "$mdraid_log"
+else
+    # Block raw-member label lookup by later programs such as onl-pki.
+    mkdir -p /run/onl && printf '{}\n' > /run/onl/block-devices.json
+    {
+        echo 'ONL mdraid: RAID setup failed.'
+        cat "$mdraid_log"
+    } > /etc/onl/abort
+    cat /etc/onl/abort >&2
+fi

packages/platforms/accton/x86-64/as9957-32db/platform-config/r0/src/init.d/onl-mdraid-clean

@@ -0,0 +1,40 @@
+#!/bin/sh
+### BEGIN INIT INFO
+# Provides:          onl-mdraid-clean
+# Required-Start:
+# Required-Stop:
+# Should-Stop:       halt reboot kexec
+# X-Stop-After:      umountroot
+# Default-Start:
+# Default-Stop:      0 6
+# Short-Description: Mark the boot RAID clean before halt or reboot
+# Description:       Only mdmon can write the RAID metadata. When the
+#                    root filesystem is read-only, ask mdmon to mark the
+#                    array clean, so the next boot needs no resync.
+### END INIT INFO
+
+[ "$1" = stop ] || exit 0
+# This file exists only when the system was booted from the RAID.
+[ -r /run/onl/mdraid.state ] || exit 0
+
+mdadm="/lib/platform-config/$(cat /etc/onl/platform)/onl/sbin/mdadm"
+if [ ! -x "$mdadm" ]; then
+    echo "ONL mdraid: $mdadm is missing. The next boot will resync the RAID." >&2
+    exit 0
+fi
+
+sync
+# Stop a running check or resync, so that its position is saved too.
+for action in /sys/block/md*/md/sync_action; do
+    if [ -w "$action" ]; then
+        echo idle > "$action" 2> /dev/null
+    fi
+done
+# mdadm waits for at most five seconds for each array.
+if "$mdadm" --wait-clean --scan; then
+    echo 'ONL mdraid: RAID marked clean.'
+else
+    echo 'ONL mdraid: RAID was not marked clean. The next boot will resync it.' >&2
+    sleep 1
+fi
+exit 0

packages/platforms/accton/x86-64/as9957-32db/platform-config/r0/src/init.d/postinst

@@ -0,0 +1,6 @@
+#!/bin/sh
+# Register the shutdown script by its LSB header. The "init:" key of
+# onlpm is not used here: its script skips update-rc.d while the rootfs
+# is built.
+set -e
+update-rc.d onl-mdraid-clean defaults > /dev/null

packages/platforms/accton/x86-64/as9957-32db/platform-config/r0/src/lib/raid/boot-custom

@@ -0,0 +1,3 @@
+#!/bin/sh
+# The existing loader boot command sources this hook before switch_root.
+python "/lib/platform-config/$(cat /etc/onl/platform)/onl/raid/onl-mdraid.py" prepare-root || exit 1