Skip to contentSkip to conversation

Make read-only rootfs work with live images

PATCH Superseded [email protected] 4 participants 18 messages Aug 7, 2013 – Aug 8, 2013

v4 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

v4 cover Make read-only rootfs work with live images

From: Chen Qi <[email protected]>

Changes from V3:
1. Take hidden directories and files into consideration when copying before bind mounting.
2. Use 'cp -a' in populate-volatile.sh to avoid potential problems

Change from V2:
1. modify the irda init script so that it doesn't necessarily need to write to /etc/sysconfig/irda
2. drop the lighttpd patch
3. modify the logic for openssh in case of a read-only rootfs

Change from V1:
1. Disable the generating of ssh keys at rootfs time
2. Add a mechanism to read-only-rootfs-hook.sh to check whether a directory is on a read-only partition.
3. Fix a new bug#4937 (https://bugzilla.yoctoproject.org/show_bug.cgi?id=4937)

The following changes since commit f63e7f4323368c0d6fe7a1d44393a7e15652d4f2:

subversion: Add patch to use neon 0.30 (2013-08-07 07:43:47 +0100)

are available in the git repository at:

git://git.pokylinux.org/poky-contrib ChenQi/read-only-rootfs-in-live-images
http://git.pokylinux.org/cgit.cgi/poky-contrib/log/?h=ChenQi/read-only-rootfs-in-live-images

Chen Qi (10):
init-live.sh: make $ROOT_MOUNT/media writable when necessary use a uniform way to determine whether rootfs is read-only
udev: remove implicit dependency on initscripts populate-volatile.sh: use $ROOT_DIR/var/volatile/tmp as TMPDIR populate-volatile.sh: don't spawn background process at rootfs time
runqemu-internal: fix to start X correctly in live images
initscripts: use a uniform way to handle directories in read-only
rootfs
populate-volatile.sh: use 'cp -a' to avoid potential problem
irda-utils: fix for read-only rootfs
openssh: fix for read-only rootfs

 meta/classes/image.bbclass                         |   16 +++++--
 .../irda-utils/irda-utils-0.9.18/init              |   13 +-----
 .../openssh/openssh-6.2p2/init                     |   22 ++++++---
 meta/recipes-connectivity/openssh/openssh_6.2p2.bb |    9 +++-
 meta/recipes-core/initrdscripts/files/init-live.sh |    2 +
 .../initscripts/initscripts-1.0/functions          |   14 ++++++
 .../initscripts-1.0/populate-volatile.sh           |   15 ++++--
 .../initscripts-1.0/read-only-rootfs-hook.sh       |   48 ++++++++++++++++++--
 meta/recipes-core/initscripts/initscripts_1.0.bb   |    5 ++
 meta/recipes-core/sysvinit/sysvinit/rcS-default    |    4 --
 meta/recipes-core/udev/udev/init                   |   21 +++++++--
 meta/recipes-core/udev/udev/udev-cache             |    5 ++
 .../0001-add-is_rootfs_readonly-to-functions.patch |   37 +++++++++++++++
 meta/recipes-extended/lsb/lsbinitscripts_9.48.bb   |    1 +
 scripts/runqemu-internal                           |    2 +-
 15 files changed, 172 insertions(+), 42 deletions(-)
 create mode 100644 meta/recipes-extended/lsb/lsbinitscripts/0001-add-is_rootfs_readonly-to-functions.patch
Signature · 1 line
1.7.9.5
#2 lore raw

PATCH v4 2/10 use a uniform way to determine whether rootfs is read-only

From: Chen Qi <[email protected]>

Previously, we had two ways to check whether the rootfs was read-only. In some part of the system, we determined whether the rootfs is read-only by checking the fstab or /proc/mounts; in other part of the system, we used the value of ROOTFS_READ_ONLY in /etc/default/rcS as a criteria.

Having two ways to check the rootfs is confusing and makes systems inconsistent.

We should drop the use of ROOTFS_READ_ONLY and figure out a uniform and consistent way to determine whether rootfs is read-only.

This patch fixes this problem by using the following strategy.

On target, we use /proc/mounts to check whether / is read-only; on host, we use $ROOT_DIR/etc/fstab to check whether the rootfs is going to be mounted as read-only or not.

[YOCTO #4880]
[YOCTO #4103]

---
 meta/classes/image.bbclass                         |    4 ---
 .../initscripts/initscripts-1.0/functions          |   14 ++++++++
 .../initscripts-1.0/populate-volatile.sh           |    4 +++
 .../initscripts-1.0/read-only-rootfs-hook.sh       |    4 ++-
 meta/recipes-core/sysvinit/sysvinit/rcS-default    |    4 ---
 meta/recipes-core/udev/udev/init                   |    5 +++
 meta/recipes-core/udev/udev/udev-cache             |    5 +++
 .../0001-add-is_rootfs_readonly-to-functions.patch |   37 ++++++++++++++++++++
 meta/recipes-extended/lsb/lsbinitscripts_9.48.bb   |    1 +
 9 files changed, 69 insertions(+), 9 deletions(-)
 create mode 100644 meta/recipes-extended/lsb/lsbinitscripts/0001-add-is_rootfs_readonly-to-functions.patch
meta/classes/image.bbclass +0 -4
index cf02f88..15fe51a 100644
@@ -259,10 +259,6 @@ read_only_rootfs_hook () {
 	if ${@base_contains("DISTRO_FEATURES", "sysvinit", "true", "false", d)}; then
 	        # Tweak the mount option and fs_passno for rootfs in fstab
 		sed -i -e '/^[#[:space:]]*rootfs/{s/defaults/ro/;s/\([[:space:]]*[[:digit:]]\)\([[:space:]]*\)[[:digit:]]$/\1\20/}' ${IMAGE_ROOTFS}/etc/fstab
-	        # Change the value of ROOTFS_READ_ONLY in /etc/default/rcS to yes
-		if [ -e ${IMAGE_ROOTFS}/etc/default/rcS ]; then
-			sed -i 's/ROOTFS_READ_ONLY=no/ROOTFS_READ_ONLY=yes/' ${IMAGE_ROOTFS}/etc/default/rcS
-		fi
 	        # Run populate-volatile.sh at rootfs time to set up basic files
 	        # and directories to support read-only rootfs.
 		if [ -x ${IMAGE_ROOTFS}/etc/init.d/populate-volatile.sh ]; then
meta/recipes-core/initscripts/initscripts-1.0/functions +14 -0
index 8e15762..ca477b7 100644
@@ -58,3 +58,17 @@ status() {
     fi
     return 3
 }
+
+# Determine whether rootfs is read-only or not according to /proc/mounts or /etc/fstab.
+is_rootfs_readonly () {
+    local DIRNAME=`dirname $0`
+    local ROOT_DIR=`echo $DIRNAME | sed -ne 's:/etc/.*::p'`
+    local criteria_file
+    [ -n "$ROOT_DIR" ] && criteria_file="$ROOT_DIR/etc/fstab" || criteria_file="/proc/mounts"
+    local flag
+    for flag in `awk '{ if ($2 == "/") { split($4,FLAGS,",") } }; END { for (f in FLAGS) print FLAGS[f] }' < $criteria_file`; do
+	[ "$flag" = "ro" ] && { echo "yes"; return 0; }
+    done
+    echo "no"
+    return 0
+}
\ No newline at end of file
meta/recipes-core/initscripts/initscripts-1.0/populate-volatile.sh +4 -0
index 91c70efb..a760081 100755
@@ -13,8 +13,12 @@ DIRNAME=`dirname $0`
 ROOT_DIR=`echo $DIRNAME | sed -ne 's:/etc/.*::p'`
 
 [ -e ${ROOT_DIR}/etc/default/rcS ] && . ${ROOT_DIR}/etc/default/rcS
+FUNCTIONS_FILE="${ROOT_DIR}`readlink -f ${ROOT_DIR}/etc/init.d/functions`"
+. $FUNCTIONS_FILE
 # When running populate-volatile.sh at rootfs time, disable cache.
 [ -n "$ROOT_DIR" ] && VOLATILE_ENABLE_CACHE=no
+# Determine whether the rootfs is read-only
+ROOTFS_READ_ONLY=`is_rootfs_readonly`
 # If rootfs is read-only, disable cache.
 [ "$ROOTFS_READ_ONLY" = "yes" ] && VOLATILE_ENABLE_CACHE=no
 
meta/recipes-core/initscripts/initscripts-1.0/read-only-rootfs-hook.sh +3 -1
index 9cf0921..d523924 100644
@@ -1,6 +1,8 @@
 #!/bin/sh
 
-. /etc/default/rcS
+. /etc/init.d/functions
+
+ROOTFS_READ_ONLY=`is_rootfs_readonly`
 
 [ "$ROOTFS_READ_ONLY" = "no" ] && exit 0
 
meta/recipes-core/sysvinit/sysvinit/rcS-default +0 -4
index 709cdf6..3c9dea9 100644
@@ -23,7 +23,3 @@ FSCKFIX=yes
 #TICKADJ=10000
 # Enable caching in populate-volatile.sh
 VOLATILE_ENABLE_CACHE=yes
-# Indicate whether the rootfs is intended to be read-only or not.
-# Setting ROOTFS_READ_ONLY to yes and rebooting will give you a read-only rootfs.
-# Normally you should not change this value.
-ROOTFS_READ_ONLY=no
meta/recipes-core/udev/udev/init +5 -0
index d90d446..4e5094a 100644
@@ -79,6 +79,11 @@ case "$1" in
 			    echo "$NEWDATA" > /dev/shm/udev.cache
                     fi
 	    else
+		    # Determine whether the rootfs is read-only or not
+		    ROOTFS_READ_ONLY="no"
+		    for flag in `awk '{ if ($2 == "/") { split($4,FLAGS,",") } }; END { for (f in FLAGS) print FLAGS[f] }' < /proc/mounts`; do
+			    [ "$flag" = "ro" ] && ROOTFS_READ_ONLY="yes"
+		    done
 		    if [ "$ROOTFS_READ_ONLY" != "yes" ]; then
 			    # If rootfs is not read-only, it's possible that a new udev cache would be generated;
 			    # otherwise, we do not bother to read files.
meta/recipes-core/udev/udev/udev-cache +5 -0
index db5a513..862e0e6 100644
@@ -18,6 +18,11 @@ export TZ=/etc/localtime
 [ -f /etc/default/rcS ] && . /etc/default/rcS
 [ -f /etc/default/udev-cache ] && . /etc/default/udev-cache
 
+ROOTFS_READ_ONLY="no"
+for flag in `awk '{ if ($2 == "/") { split($4,FLAGS,",") } }; END { for (f in FLAGS) print FLAGS[f] }' < /proc/mounts`; do
+    [ "$flag" = "ro" ] && ROOTFS_READ_ONLY="yes"
+done
+
 if [ "$ROOTFS_READ_ONLY" = "yes" ]; then
     [ "$VERBOSE" != "no" ] && echo "udev-cache: read-only rootfs, skip generating udev-cache"
     exit 0
meta/recipes-extended/lsb/lsbinitscripts/0001-add-is_rootfs_readonly-to-functions.patch new file +37 -0
new file mode 100644
index 0000000..f076b0d
@@ -0,0 +1,37 @@
+From b8e8e7f93aca54561cc1d7945ef3ec7aca5f6f43 Mon Sep 17 00:00:00 2001
+From: Chen Qi <[email protected]>
+Date: Tue, 23 Jul 2013 10:48:32 +0800
+Subject: [PATCH V4 V3] add is_rootfs_readonly to functions
+
+---
+ rc.d/init.d/functions |   14 ++++++++++++++
+ 1 file changed, 14 insertions(+)
+
+diff --git a/rc.d/init.d/functions b/rc.d/init.d/functions
+index a5b2b9e..0663c7f 100644
+--- a/rc.d/init.d/functions
++++ b/rc.d/init.d/functions
+@@ -577,6 +577,20 @@ apply_sysctl() {
+     fi
+ }
+ 
++# Determine whether rootfs is read-only or not according to /proc/mounts or /etc/fstab.
++is_rootfs_readonly () {
++    local DIRNAME=`dirname $0`
++    local ROOT_DIR=`echo $DIRNAME | sed -ne 's:/etc/.*::p'`
++    local criteria_file
++    [ -n "$ROOT_DIR" ] && criteria_file="$ROOT_DIR/etc/fstab" || criteria_file="/proc/mounts"
++    local flag
++    for flag in `awk '{ if ($2 == "/") { split($4,FLAGS,",") } }; END { for (f in FLAGS) print FLAGS[f] }' < $criteria_file`; do
++	[ "$flag" = "ro" ] && { echo "yes"; return 0; }
++    done
++    echo "no"
++    return 0
++}
++
+ # A sed expression to filter out the files that is_ignored_file recognizes
+ __sed_discard_ignored_files='/\(~\|\.bak\|\.orig\|\.rpmnew\|\.rpmorig\|\.rpmsave\)$/d'
+ 
+-- 
+1.7.9.5
+
meta/recipes-extended/lsb/lsbinitscripts_9.48.bb +1 -0
index 3011f98..927c9db 100644
@@ -8,6 +8,7 @@ LIC_FILES_CHKSUM = "file://COPYING;md5=ebf4e8b49780ab187d51bd26aaa022c6"
 S="${WORKDIR}/initscripts-${PV}"
 SRC_URI = "http://pkgs.fedoraproject.org/repo/pkgs/initscripts/initscripts-9.48.tar.bz2/7dfab81a5a8d3f0dea5ba55e391c26f3/initscripts-9.48.tar.bz2 \
            file://functions.patch \
+           file://0001-add-is_rootfs_readonly-to-functions.patch \
           " 
 
 SRC_URI[md5sum] = "7dfab81a5a8d3f0dea5ba55e391c26f3"
Signature · 1 line
1.7.9.5
#1 lore raw

PATCH v4 1/10 init-live.sh: make $ROOT_MOUNT/media writable when necessary

In reply to [email protected]

From: Chen Qi <[email protected]>

If the live image is mounted as read-only, we cannot make necessary directories under $ROOT_MOUNT/media, so trying to move the mount points lead to errors.

So in case that no unification filesystem mechanism is available in kernel and the rootfs is mounted as read-only, we mount tmpfs on $ROOT_MOUNT/media so that it's possible to make necessary directories under it.

[YOCTO #4881]
[YOCTO #4103]

---
 meta/recipes-core/initrdscripts/files/init-live.sh |    2 ++
 1 file changed, 2 insertions(+)
meta/recipes-core/initrdscripts/files/init-live.sh +2 -0
index 890c562..861d874 100644
@@ -184,6 +184,8 @@ mount_and_boot() {
 	"")
 	    if ! mount -o rw,loop,noatime,nodiratime /media/$i/$ISOLINUX/$ROOT_IMAGE $ROOT_MOUNT ; then
 		fatal "Could not mount rootfs image"
+	    else
+		mount -t tmpfs -o rw,noatime,mode=755 tmpfs $ROOT_MOUNT/media
 	    fi
 	    ;;
     esac
Signature · 1 line
1.7.9.5
#5 lore raw

PATCH v4 3/10 udev: remove implicit dependency on initscripts

In reply to [email protected]

From: Chen Qi <[email protected]>

At some point, the udev was modified to source the functions from initscripts or lsbinitscripts. This dependency is actually not needed. If we use udev in a system where initscripts from oe-core is not available, there will be errors.

This patch fixes this problem by removing the implicit dependency.

[YOCTO #4882]
[YOCTO #4103]

---
 meta/recipes-core/udev/udev/init |   16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)
meta/recipes-core/udev/udev/init +12 -4
index 4e5094a..52c7925 100644
@@ -9,8 +9,6 @@
 # Short-Description: Start udevd, populate /dev and load drivers.
 ### END INIT INFO
 
-. /etc/init.d/functions
-
 export TZ=/etc/localtime
 
 [ -d /sys/class ] || exit 1
@@ -31,6 +29,11 @@ readfiles () {
    done
 }
 
+kill_udevd () {
+    pid=`pidof -x udevd`
+    [ -n "$pid" ] && kill $pid
+}
+
 case "$1" in
   start)
     export ACTION=add
@@ -94,7 +97,7 @@ case "$1" in
     fi
 
     # make_extra_nodes
-    killproc udevd > "/dev/null" 2>&1
+    kill_udevd > "/dev/null" 2>&1
 
     # trigger the sorted events
     echo -e '\000\000\000\000' > /proc/sys/kernel/hotplug
@@ -119,7 +122,12 @@ case "$1" in
     $0 start
     ;;
   status)
-    status udevd
+    pid=`pidof -x udevd`
+    if [ -n "$pid" ]; then
+	echo "udevd (pid $pid) is running ..."
+    else
+	echo "udevd is stopped"
+    fi
     ;;
   *)
     echo "Usage: $0 {start|stop|status|restart}"
Signature · 1 line
1.7.9.5
#6 lore raw

PATCH v4 4/10 populate-volatile.sh: use $ROOT_DIR/var/volatile/tmp as TMPDIR

In reply to [email protected]

From: Chen Qi <[email protected]>

It's possible that a 'No such directory' error occurs when doing check_requirement in populate-volatile.sh at rootfs time. This is because the $ROOT_DIR/var/tmp might be a dead link.

Use $ROOT_DIR/var/volatile/tmp as the TMPDIR instead to avoid this error.

[YOCTO #4883]
[YOCTO #4103]

---
 .../initscripts-1.0/populate-volatile.sh           |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
meta/recipes-core/initscripts/initscripts-1.0/populate-volatile.sh +1 -1
index a760081..d8ae14e 100755
@@ -23,7 +23,7 @@ ROOTFS_READ_ONLY=`is_rootfs_readonly`
 [ "$ROOTFS_READ_ONLY" = "yes" ] && VOLATILE_ENABLE_CACHE=no
 
 CFGDIR="${ROOT_DIR}/etc/default/volatiles"
-TMPROOT="${ROOT_DIR}/var/tmp"
+TMPROOT="${ROOT_DIR}/var/volatile/tmp"
 COREDEF="00_core"
 
 [ "${VERBOSE}" != "no" ] && echo "Populating volatile Filesystems."
Signature · 1 line
1.7.9.5
#7 lore raw

PATCH v4 5/10 populate-volatile.sh: don't spawn background process at rootfs time

In reply to [email protected]

From: Chen Qi <[email protected]>

If we're building a read-only rootfs, we'll get the following error now and then.

tar: .: file changed as we read it

The root cause is that we spawn background process at rootfs time. When the tar command is running, it's possible that files under rootfs are changed by background processes, thus this error.

[YOCTO #4937]

---
 .../initscripts-1.0/populate-volatile.sh           |    6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
meta/recipes-core/initscripts/initscripts-1.0/populate-volatile.sh +3 -3
index d8ae14e..827fd7d 100755
@@ -46,7 +46,7 @@ create_file() {
 			# but these failures should not be logged to make sure the do_rootfs
 			# process doesn't fail. This does no harm, as this script will
 			# run on target to set up the correct files and directories.
-			eval $EXEC > /dev/null 2>&1 &
+			eval $EXEC > /dev/null 2>&1
 		fi
 	}
 }
@@ -91,7 +91,7 @@ link_file() {
 	else
 		# For the same reason with create_file(), failures should
 		# not be logged.
-		eval $EXEC > /dev/null 2>&1 &
+		eval $EXEC > /dev/null 2>&1
 	fi
 }
 
@@ -160,7 +160,7 @@ apply_cfgfile() {
 		[ "${TTYPE}" = "l" ] && {
 			TSOURCE="$TLTARGET"
 			[ "${VERBOSE}" != "no" ] && echo "Creating link -${TNAME}- pointing to -${TSOURCE}-."
-			link_file "${TSOURCE}" "${TNAME}" &
+			link_file "${TSOURCE}" "${TNAME}"
 			continue
 		}
 
Signature · 1 line
1.7.9.5
#8 lore raw

PATCH v4 6/10 runqemu-internal: fix to start X correctly in live images

In reply to [email protected]

From: Chen Qi <[email protected]>

The QEMUOPTIONS for ISOFS was not complete, leading to failures when trying to start X in live images.

This patch fixes this problem.

[YOCTO #4103]
[YOCTO #4884]

---
 scripts/runqemu-internal |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
scripts/runqemu-internal +1 -1
index 9619bec..8a6e551 100755
@@ -526,7 +526,7 @@ if [ "x$RAMFS" = "xtrue" ]; then
 fi
 
 if [ "x$ISOFS" = "xtrue" ]; then
-    QEMUOPTIONS="-cdrom $ROOTFS"
+    QEMUOPTIONS="$QEMU_NETWORK_CMD -cdrom $ROOTFS $QEMU_UI_OPTIONS"
 fi
 
 if [ "x$QEMUOPTIONS" = "x" ]; then
Signature · 1 line
1.7.9.5
#9 lore raw

PATCH v4 7/10 initscripts: use a uniform way to handle directories in read-only rootfs

In reply to [email protected]

From: Chen Qi <[email protected]>

Packages in our system may need to write to some directories to function correctly. In read-only rootfs, these directories should be made writable.

This patch uses a convenient and uniform way to handle such situations. The read-only-rootfs-hook.sh script searches the /etc/default/readonly diretory for config files and then apply them one by one.

The config files simply have the following format. <original diretory> <corresponding directory in volatile story>

For example, /etc/default/readonly/initscripts have the following content. /var/lib /var/volatile/lib

This patch only has effect for systems with read-only rootfs.

[YOCTO #4103]
[YOCTO #4888]

---
 .../initscripts-1.0/read-only-rootfs-hook.sh       |   44 ++++++++++++++++++--
 meta/recipes-core/initscripts/initscripts_1.0.bb   |    5 +++
 2 files changed, 45 insertions(+), 4 deletions(-)
meta/recipes-core/initscripts/initscripts-1.0/read-only-rootfs-hook.sh +40 -4
index d523924..eb89e18 100644
@@ -1,15 +1,51 @@
 #!/bin/sh
 
 . /etc/init.d/functions
+READONLY_CFGDIR="/etc/default/readonly"
 
 ROOTFS_READ_ONLY=`is_rootfs_readonly`
 
 [ "$ROOTFS_READ_ONLY" = "no" ] && exit 0
 
+is_on_read_only_partition () {
+    DIRECTORY=$1
+    dir=`readlink -f $DIRECTORY`
+    while true; do
+	if [ ! -d "$dir" ]; then
+	    echo "ERROR: $dir is not a directory"
+	    exit 1
+	else
+	    for flag in `awk -v dir=$dir '{ if ($2 == dir) { print "FOUND"; split($4,FLAGS,",") } }; END { for (f in FLAGS) print FLAGS[f] }' < /proc/mounts`; do
+		[ "$flag" = "FOUND" ] && partition="read-write"
+		[ "$flag" = "ro" ] && { partition="read-only"; break; }
+	    done
+	fi
+	if [ "$dir" = "/" -o -n "$partition" ]; then
+	    break
+	else
+	    dir=`dirname $dir`
+	fi
+    done
+    [ "$partition" = "read-only" ] && echo "yes" || echo "no"
+}
+
+apply_conf () {
+    cfgfile=$1
+    cat $cfgfile | while read line; do
+	eval `echo "$line" | sed -n "s/\(.*\)\ \(.*\)/DIR_READONLY=\1; DIR_VOLATILE=\2;/p"`
+	if [ `is_on_read_only_partition $DIR_READONLY` = "yes" ]; then
+	    mkdir -p $DIR_VOLATILE
+	    cp -a $DIR_READONLY/* $DIR_VOLATILE 2>/dev/null
+	    cp -a $DIR_READONLY/.[!.]* $DIR_VOLATILE 2>/dev/null
+	    mount --bind $DIR_VOLATILE $DIR_READONLY
+	fi
+    done
+}
+
 if [ "$1" = "start" ] ; then
-	grep -q "tmpfs /var/volatile" /proc/mounts || mount /var/volatile
-	mkdir -p /var/volatile/lib
-	cp -a /var/lib/* /var/volatile/lib
-	mount --bind /var/volatile/lib /var/lib
+    grep -q "tmpfs /var/volatile" /proc/mounts || mount /var/volatile
+    for file in `ls -1 "$READONLY_CFGDIR"`; do
+	apply_conf "$READONLY_CFGDIR/$file"
+    done
 fi
 
meta/recipes-core/initscripts/initscripts_1.0.bb +5 -0
index 52e1c9c..46c4c99 100644
@@ -105,6 +105,11 @@ do_install () {
 	install -m 0755 ${WORKDIR}/umountfs	${D}${sysconfdir}/init.d/umountfs
 	install -m 0755		${WORKDIR}/device_table.txt		${D}${sysconfdir}/device_table
 #
+# Create config files for read-only rootfs
+#
+	install -d ${D}${sysconfdir}/default/readonly
+	echo "/var/lib /var/volatile/lib" > ${D}${sysconfdir}/default/readonly/initscripts
+#
 # Create runlevel links
 #
 	update-rc.d -r ${D} rmnologin.sh start 99 2 3 4 5 .
Signature · 1 line
1.7.9.5
#10 lore raw

PATCH v4 8/10 populate-volatile.sh: use 'cp -a' to avoid potential problem

In reply to [email protected]

From: Chen Qi <[email protected]>

Previously, dead links in target directory will not be copied. This is incorrect as dead links are not uncommon in our rootfs. So we use '-a' option instead.

---
 .../initscripts-1.0/populate-volatile.sh           |    3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
meta/recipes-core/initscripts/initscripts-1.0/populate-volatile.sh +2 -1
index 827fd7d..b1c0245 100755
@@ -76,7 +76,8 @@ link_file() {
 	if [ -L \"$2\" ]; then
 		[ \"\$(readlink -f \"$2\")\" != \"\$(readlink -f \"$1\")\" ] && { rm -f \"$2\"; ln -sf \"$1\" \"$2\"; };
 	elif [ -d \"$2\" ]; then
-		for f in $2/* $2/.[^.]*; do [ -e \$f ] && cp -rf \$f $1; done;
+		cp -a $2/* $1
+		cp -a $2/.[!.]* $1
 		rm -rf \"$2\";
 		ln -sf \"$1\" \"$2\";
 	else
Signature · 1 line
1.7.9.5
#12 lore raw

PATCH v4 9/10 irda-utils: fix for read-only rootfs

In reply to [email protected]

From: Chen Qi <[email protected]>

The init script for irda writes configuration items to /etc/sysconfig/irda if that file is not available in system. But it's actually not necessary, the behavior doesn't change whether the init script writes to the file or not.

Considering it issues error messages in case of a read-only rootfs, I delete the writing process.

[YOCTO #4103]
[YOCTO #4886]

---
 .../irda-utils/irda-utils-0.9.18/init              |   13 ++-----------
 1 file changed, 2 insertions(+), 11 deletions(-)
meta/recipes-connectivity/irda-utils/irda-utils-0.9.18/init +2 -11
index 63750f1..14efb94 100755
@@ -13,7 +13,6 @@ module_id() {
 }
 
 if [ ! -f /etc/sysconfig/irda ]; then
-
     case `module_id` in
 	"HP iPAQ H2200" | "HP iPAQ HX4700" | "HTC Universal")
 	    IRDA=yes
@@ -28,18 +27,10 @@ if [ ! -f /etc/sysconfig/irda ]; then
 	    DISCOVERY=
 	    ;;
     esac
-
-    mkdir -p /etc/sysconfig
-    echo "IRDA=$IRDA" > /etc/sysconfig/irda
-    if [ $IRDA = "yes" ]; then
-	echo "DEVICE=$DEVICE" >> /etc/sysconfig/irda
-	echo "DONGLE=$DONGLE" >> /etc/sysconfig/irda
-	echo "DISCOVERY=$DISCOVERY" >> /etc/sysconfig/irda
-    fi
+else
+    . /etc/sysconfig/irda
 fi
 
-. /etc/sysconfig/irda
-
 # Check that irda is up.
 [ ${IRDA} = "no" ] && exit 0
 
Signature · 1 line
1.7.9.5
#17 lore raw

PATCH v4 10/10 openssh: fix for read-only rootfs

In reply to [email protected]

From: Chen Qi <[email protected]>

If the rootfs is read-only and the ssh keys are not available at system start-up, the init script will generate ssh keys into /etc/ssh, thus causing a 'read-only file system' error.

In order for Yocto based image to work correctly for read-only rootfs, we use the following logic for openssh.

If the rootfs is read-only and there are pre-generated keys under /etc/ssh, we use the pre-generated keys. Note the pre-generated keys are mainly for debugging or development purpose.
If the rootfs is read-only and there are no pre-generated keys under /etc/ssh, we use /var/run/ssh as the location for ssh keys. That is, at system boot-up, the generated ssh keys will put into /var/run/ssh.

[YOCTO #4887]

---
 meta/classes/image.bbclass                         |   12 +++++++++++
 .../openssh/openssh-6.2p2/init                     |   22 +++++++++++++-------
 meta/recipes-connectivity/openssh/openssh_6.2p2.bb |    9 +++++++-
 3 files changed, 35 insertions(+), 8 deletions(-)
meta/classes/image.bbclass +12 -0
index 15fe51a..769cf81 100644
@@ -264,6 +264,18 @@ read_only_rootfs_hook () {
 		if [ -x ${IMAGE_ROOTFS}/etc/init.d/populate-volatile.sh ]; then
 			${IMAGE_ROOTFS}/etc/init.d/populate-volatile.sh
 		fi
+		# If we're using openssh and the /etc/ssh directory has no pre-generated keys,
+		# we should configure openssh to use the configuration file /etc/ssh/sshd_config_readonly
+		# and the keys under /var/run/ssh.
+		if [ -d ${IMAGE_ROOTFS}/etc/ssh ]; then
+			if [ -e ${IMAGE_ROOTFS}/etc/ssh/ssh_host_rsa_key ]; then
+				echo "SYSCONFDIR=/etc/ssh" >> ${IMAGE_ROOTFS}/etc/default/ssh
+				echo "SSHD_OPTS=" >> ${IMAGE_ROOTFS}/etc/default/ssh
+			else
+				echo "SYSCONFDIR=/var/run/ssh" >> ${IMAGE_ROOTFS}/etc/default/ssh
+				echo "SSHD_OPTS='-f /etc/ssh/sshd_config_readonly'" >> ${IMAGE_ROOTFS}/etc/default/ssh
+			fi
+		fi
 	fi
 }
 
meta/recipes-connectivity/openssh/openssh-6.2p2/init +15 -7
index 6beec84..12fb79b 100644
@@ -6,14 +6,22 @@ set -e
 test -x /usr/sbin/sshd || exit 0
 ( /usr/sbin/sshd -\? 2>&1 | grep -q OpenSSH ) 2>/dev/null || exit 0
 
+# /etc/default/ssh may set SYSCONFDIR and SSHD_OPTS
 if test -f /etc/default/ssh; then
     . /etc/default/ssh
 fi
 
+[ -z "$SYSCONFDIR" ] && SYSCONFDIR=/etc/ssh
+mkdir -p $SYSCONFDIR
+
+HOST_KEY_RSA=$SYSCONFDIR/ssh_host_rsa_key
+HOST_KEY_DSA=$SYSCONFDIR/ssh_host_dsa_key
+HOST_KEY_ECDSA=$SYSCONFDIR/ssh_host_ecdsa_key
+
 check_for_no_start() {
     # forget it if we're trying to start, and /etc/ssh/sshd_not_to_be_run exists
-    if [ -e /etc/ssh/sshd_not_to_be_run ]; then 
-	echo "OpenBSD Secure Shell server not in use (/etc/ssh/sshd_not_to_be_run)"
+    if [ -e $SYSCONFDIR/sshd_not_to_be_run ]; then
+	echo "OpenBSD Secure Shell server not in use ($SYSCONFDIR/sshd_not_to_be_run)"
 	exit 0
     fi
 }
@@ -32,17 +40,17 @@ check_config() {
 
 check_keys() {
 	# create keys if necessary
-	if [ ! -f /etc/ssh/ssh_host_rsa_key ]; then
+	if [ ! -f $HOST_KEY_RSA ]; then
 		echo "  generating ssh RSA key..."
-		ssh-keygen -q -f /etc/ssh/ssh_host_rsa_key -N '' -t rsa
+		ssh-keygen -q -f $HOST_KEY_RSA -N '' -t rsa
 	fi
-	if [ ! -f /etc/ssh/ssh_host_ecdsa_key ]; then
+	if [ ! -f $HOST_KEY_ECDSA ]; then
 		echo "  generating ssh ECDSA key..."
-		ssh-keygen -q -f /etc/ssh/ssh_host_ecdsa_key -N '' -t ecdsa
+		ssh-keygen -q -f $HOST_KEY_ECDSA -N '' -t ecdsa
 	fi
 	if [ ! -f /etc/ssh/ssh_host_dsa_key ]; then
 		echo "  generating ssh DSA key..."
-		ssh-keygen -q -f /etc/ssh/ssh_host_dsa_key -N '' -t dsa
+		ssh-keygen -q -f $HOST_KEY_DSA -N '' -t dsa
 	fi
 }
 
meta/recipes-connectivity/openssh/openssh_6.2p2.bb +8 -1
index ab2eefb..c76f9ac 100644
@@ -86,6 +86,13 @@ do_install_append () {
 	install -m 0755 ${WORKDIR}/init ${D}${sysconfdir}/init.d/sshd
 	rm -f ${D}${bindir}/slogin ${D}${datadir}/Ssh.bin
 	rmdir ${D}${localstatedir}/run/sshd ${D}${localstatedir}/run ${D}${localstatedir}
+        # Create config files for read-only rootfs
+	install -d ${D}${sysconfdir}/ssh
+	install -m 644 ${WORKDIR}/sshd_config ${D}${sysconfdir}/ssh/sshd_config_readonly
+	sed -i '/HostKey/d' ${D}${sysconfdir}/ssh/sshd_config_readonly
+	echo "HostKey /var/run/ssh/ssh_host_rsa_key" >> ${D}${sysconfdir}/ssh/sshd_config_readonly
+	echo "HostKey /var/run/ssh/ssh_host_dsa_key" >> ${D}${sysconfdir}/ssh/sshd_config_readonly
+	echo "HostKey /var/run/ssh/ssh_host_ecdsa_key" >> ${D}${sysconfdir}/ssh/sshd_config_readonly
 }
 
 ALLOW_EMPTY_${PN} = "1"
@@ -94,7 +101,7 @@ PACKAGES =+ "${PN}-keygen ${PN}-scp ${PN}-ssh ${PN}-sshd ${PN}-sftp ${PN}-misc $
 FILES_${PN}-scp = "${bindir}/scp.${BPN}"
 FILES_${PN}-ssh = "${bindir}/ssh.${BPN} ${sysconfdir}/ssh/ssh_config"
 FILES_${PN}-sshd = "${sbindir}/sshd ${sysconfdir}/init.d/sshd"
-FILES_${PN}-sshd += "${sysconfdir}/ssh/moduli ${sysconfdir}/ssh/sshd_config"
+FILES_${PN}-sshd += "${sysconfdir}/ssh/moduli ${sysconfdir}/ssh/sshd_config ${sysconfdir}/ssh/sshd_config_readonly"
 FILES_${PN}-sftp = "${bindir}/sftp"
 FILES_${PN}-sftp-server = "${libexecdir}/sftp-server"
 FILES_${PN}-misc = "${bindir}/ssh* ${libexecdir}/ssh*"
Signature · 1 line
1.7.9.5
#13 lore raw

Re: [PATCH V4 09/10] irda-utils: fix for read-only rootfs

In reply to [email protected]
Quoted 7 lines · On Wed, Aug 7, 2013 at 1:08 AM, <[email protected]> wrote:

From: Chen Qi <[email protected]>

The init script for irda writes configuration items to /etc/sysconfig/irda if that file is not available in system. But it's actually not necessary, the behavior doesn't change whether the init script writes to the file or not.

Quoted 9 lines

Considering it issues error messages in case of a read-only rootfs, I delete
the writing process.

[YOCTO #4103]
[YOCTO #4886]

This is a start, but is incomplete, from what I can tell. As mentioned in the other thread, the startup script shouldn't be using /etc/sysconfig/ at all, in any form. We use /etc/default/ for our startup script config files, not /etc/sysconfig/. Also, the purpose of this block was clearly to implement a default configuration, yet the recipe isn't altered to ship a default configuration to provide equivalent functionality.

Signature · 5 lines
Christopher Larson
clarson at kergoth dot com
Founder - BitBake, OpenEmbedded, OpenZaurus
Maintainer - Tslib
Senior Software Engineer, Mentor Graphics
#3 lore raw

Re: [PATCH V4 02/10] use a uniform way to determine whether rootfs is read-only

In reply to [email protected]

On Wed, Aug 7, 2013 at 5:08 AM, <[email protected]> wrote:
...

Quoted 11 lines
meta/recipes-core/initscripts/initscripts-1.0/functions +5 -0
@@ -58,3 +58,17 @@ status() {
     fi
     return 3
 }
+
+# Determine whether rootfs is read-only or not according to /proc/mounts or /etc/fstab.
+is_rootfs_readonly () {
+    local DIRNAME=`dirname $0`
+    local ROOT_DIR=`echo $DIRNAME | sed -ne 's:/etc/.*::p'`

meta/recipes-core/initscripts/initscripts-1.0/functions status() { · line 74

Quoted 10 lines
meta/recipes-core/initscripts/initscripts-1.0/functions +9 -0
+    local criteria_file
+    [ -n "$ROOT_DIR" ] && criteria_file="$ROOT_DIR/etc/fstab" || criteria_file="/proc/mounts"
+    local flag
+    for flag in `awk '{ if ($2 == "/") { split($4,FLAGS,",") } }; END { for (f in FLAGS) print FLAGS[f] }' < $criteria_file`; do
+       [ "$flag" = "ro" ] && { echo "yes"; return 0; }
+    done
+    echo "no"
+    return 0
+}
\ No newline at end of file

Please fix this.

Signature · 3 lines
Otavio Salvador                             O.S. Systems
http://www.ossystems.com.br        http://projetos.ossystems.com.br
Mobile: +55 (53) 9981-7854            Mobile: +1 (347) 903-9750
author
#14 lore raw

Re: [PATCH V4 09/10] irda-utils: fix for read-only rootfs

In reply to Chris Larson
Quoted 16 lines · Chris Larson

On Wed, Aug 7, 2013 at 1:08 AM, <[email protected] <mailto:[email protected]>> wrote:

From: Chen Qi <[email protected] <mailto:[email protected]>>

The init script for irda writes configuration items to /etc/sysconfig/irda
if that file is not available in system. But it's actually not necessary,
the behavior doesn't change whether the init script writes to the file or not.

Considering it issues error messages in case of a read-only rootfs, I delete
the writing process.

Quoted 8 lines

[YOCTO #4103]
[YOCTO #4886]

<mailto:[email protected]>>

This is a start, but is incomplete, from what I can tell. As mentioned in the other thread, the startup script shouldn't be using /etc/sysconfig/ at all, in any form. We use /etc/default/ for our startup script config files, not /etc/sysconfig/.

Yeah, I agree with you.
But when I checked the irda source code, I saw that it actually could ship its own init script (irda-utils-xxx/etc/irda.rc). And I think our init script is derived from this one. In its own init script, /etc/sysconfig is used. (Of course, I still think /etc/default is a better location for its configuration file.)

Quoted 1 line

Also, the purpose of this block was clearly to implement a default configuration, yet the recipe isn't altered to ship a default configuration to provide equivalent functionality.

In its source code, there's a default configuration file, and we don't use that. I think there might be a reason. So I'm not sure about this one.

If you have a patch to fix the irda issue properly, send it out and I'll rebase my remote branch and drop this one.

Best Regards,
Chen Qi

Quoted 6 lines
Signature · 5 lines
Christopher Larson
clarson at kergoth dot com
Founder - BitBake, OpenEmbedded, OpenZaurus
Maintainer - Tslib
Senior Software Engineer, Mentor Graphics
author
#4 lore raw

Re: [PATCH V4 02/10] use a uniform way to determine whether rootfs is read-only

In reply to Otavio Salvador
Quoted 16 lines · Otavio Salvador

On Wed, Aug 7, 2013 at 5:08 AM, <[email protected]> wrote:
...

Quoted 14 lines
meta/recipes-core/initscripts/initscripts-1.0/functions +8 -0
@@ -58,3 +58,17 @@ status() {
      fi
      return 3
  }
+
+# Determine whether rootfs is read-only or not according to /proc/mounts or /etc/fstab.
+is_rootfs_readonly () {
+    local DIRNAME=`dirname $0`
+    local ROOT_DIR=`echo $DIRNAME | sed -ne 's:/etc/.*::p'`
+    local criteria_file
+    [ -n "$ROOT_DIR" ] && criteria_file="$ROOT_DIR/etc/fstab" || criteria_file="/proc/mounts"
+    local flag
Quoted 9 lines

meta/recipes-core/initscripts/initscripts-1.0/functions status() { · line 74

Quoted 7 lines
meta/recipes-core/initscripts/initscripts-1.0/functions +6 -0
+    for flag in `awk '{ if ($2 == "/") { split($4,FLAGS,",") } }; END { for (f in FLAGS) print FLAGS[f] }' < $criteria_file`; do
+       [ "$flag" = "ro" ] && { echo "yes"; return 0; }
+    done
+    echo "no"
+    return 0
+}
\ No newline at end of file

Please fix this.

Thanks for pointing it out.

Remote branch has been updated :)

Best Regards,
Chen Qi

#15 lore raw

Re: [PATCH V4 09/10] irda-utils: fix for read-only rootfs

In reply to ChenQi
Quoted 7 lines · ChenQi

On 08/07/2013 10:18 PM, Chris Larson wrote:

Quoted 4 lines · On Wed, Aug 7, 2013 at 1:08 AM, <[email protected]> wrote:

From: Chen Qi <[email protected]>

The init script for irda writes configuration items to

/etc/sysconfig/irda

Quoted 2 lines
Quoted 2 lines

if that file is not available in system. But it's actually not necessary, the behavior doesn't change whether the init script writes to the file

or not.

Quoted 2 lines
Quoted 2 lines

Considering it issues error messages in case of a read-only rootfs, I

delete

Quoted 8 lines
Quoted 6 lines

the writing process.

[YOCTO #4103]
[YOCTO #4886]

This is a start, but is incomplete, from what I can tell. As mentioned in

the other thread, the startup script shouldn't be using /etc/sysconfig/ at all, in any form. We use /etc/default/ for our startup script config files, not /etc/sysconfig/.

Quoted 3 lines

Yeah, I agree with you.
But when I checked the irda source code, I saw that it actually could

ship its own init script (irda-utils-xxx/etc/irda.rc). And I think our init script is derived from this one. In its own init script, /etc/sysconfig is used.

Quoted 1 line

(Of course, I still think /etc/default is a better location for its

configuration file.)

Quoted 3 lines

Also, the purpose of this block was clearly to implement a default

configuration, yet the recipe isn't altered to ship a default configuration to provide equivalent functionality.

Quoted 2 lines

In its source code, there's a default configuration file, and we don't

use that. I think there might be a reason.

Quoted 3 lines

So I'm not sure about this one.

If you have a patch to fix the irda issue properly, send it out and I'll

rebase my remote branch and drop this one.

Whether Chris has a patch or not this patch in its form should be dropped since there is a better way to do it

Quoted 12 lines

Best Regards,
Chen Qi

--
Christopher Larson
clarson at kergoth dot com
Founder - BitBake, OpenEmbedded, OpenZaurus
Maintainer - Tslib
Senior Software Engineer, Mentor Graphics

author
#16 lore raw

Re: [PATCH V4 09/10] irda-utils: fix for read-only rootfs

Quoted 40 lines · Khem Raj
Quoted 3 lines · On Wednesday, August 7, 2013, ChenQi <[email protected]…

On 08/07/2013 10:18 PM, Chris Larson wrote:

On Wed, Aug 7, 2013 at 1:08 AM, <[email protected]

Quoted 4 lines · <mailto:[email protected]>> wrote:
Quoted 4 lines

From: Chen Qi <[email protected] <mailto:[email protected]>>

The init script for irda writes configuration items to

/etc/sysconfig/irda

Quoted 1 line
Quoted 1 line

if that file is not available in system. But it's actually not

necessary,

Quoted 1 line
Quoted 1 line

the behavior doesn't change whether the init script writes to the

file or not.

Quoted 2 lines
Quoted 2 lines

Considering it issues error messages in case of a read-only rootfs,

I delete

Quoted 6 lines
Quoted 6 lines

the writing process.

[YOCTO #4103]
[YOCTO #4886]

<mailto:[email protected]>>

Quoted 2 lines

This is a start, but is incomplete, from what I can tell. As

mentioned in the other thread, the startup script shouldn't be using /etc/sysconfig/ at all, in any form. We use /etc/default/ for our startup script config files, not /etc/sysconfig/.

Quoted 3 lines

Yeah, I agree with you.
But when I checked the irda source code, I saw that it actually

could ship its own init script (irda-utils-xxx/etc/irda.rc). And I think our init script is derived from this one. In its own init script, /etc/sysconfig is used.

Quoted 1 line

(Of course, I still think /etc/default is a better location for its

configuration file.)

Quoted 3 lines

Also, the purpose of this block was clearly to implement a default

configuration, yet the recipe isn't altered to ship a default configuration to provide equivalent functionality.

Quoted 1 line
Quoted 9 lines
Quoted 1 line

In its source code, there's a default configuration file, and we

don't use that. I think there might be a reason.

Quoted 3 lines

So I'm not sure about this one.

If you have a patch to fix the irda issue properly, send it out and

I'll rebase my remote branch and drop this one.

Whether Chris has a patch or not this patch in its form should be dropped since there is a better way to do it

Shipping a default configuration file or using /etc/default as the configuration directory? Or something else?

//Chen Qi

Quoted 12 lines
Quoted 12 lines

Best Regards,
Chen Qi

--
Christopher Larson
clarson at kergoth dot com
Founder - BitBake, OpenEmbedded, OpenZaurus
Maintainer - Tslib
Senior Software Engineer, Mentor Graphics

author
#11 lore raw

Re: [PATCH V4 08/10] populate-volatile.sh: use 'cp -a' to avoid potential problem

In reply to [email protected]
Quoted 16 lines · On 08/07/2013 04:08 PM, [email protected] wrote:

From: Chen Qi <[email protected]>

Previously, dead links in target directory will not be copied. This is incorrect as dead links are not uncommon in our rootfs. So we use '-a' option instead.

---
  .../initscripts-1.0/populate-volatile.sh           |    3 ++-
  1 file changed, 2 insertions(+), 1 deletion(-)
meta/recipes-core/initscripts/initscripts-1.0/populate-volatile.sh +0 -0
index 827fd7d..b1c0245 100755

meta/recipes-core/initscripts/initscripts-1.0/populate-volatile.sh link_file() { · line 80

Quoted 6 lines
meta/recipes-core/initscripts/initscripts-1.0/populate-volatile.sh +2 -1
@@ -76,7 +76,8 @@ link_file() {
  	if [ -L \"$2\" ]; then
  		[ \"\$(readlink -f \"$2\")\" != \"\$(readlink -f \"$1\")\" ] && { rm -f \"$2\"; ln -sf \"$1\" \"$2\"; };
  	elif [ -d \"$2\" ]; then
-		for f in $2/* $2/.[^.]*; do [ -e \$f ] && cp -rf \$f $1; done;
+		cp -a $2/* $1
+		cp -a $2/.[!.]* $1

My apologies.
The above two lines are missing colons.
They should be:
cp -a $2/* $1;
cp -a $2/.[!.]* $1;

Remote branch has been updated to fix this error.
Sorry for the inconvenience.

Best Regards,
Chen Qi

Quoted 3 lines
meta/recipes-core/initscripts/initscripts-1.0/populate-volatile.sh +0 -0
  		rm -rf \"$2\";
  		ln -sf \"$1\" \"$2\";
  	else