Skip to contentSkip to conversation

overlayfs.bbclass: generate overlayfs mount units

PATCH In review Vyacheslav Yurkov 5 participants 10 messages Jun 3, 2021 – Jun 9, 2021

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 overlayfs.bbclass: generate overlayfs mount units

It's often desired in Embedded System design to have a read-only rootfs. But a lot of different applications might want to have a read-write access to some parts of a filesystem. It can be especially useful when your update mechanism overwrites the whole rootfs, but you want your application data to be preserved between updates. This class provides a way to achieve that by means of overlayfs and at the same time keeping the base rootfs read-only.

---
 meta/classes/overlayfs.bbclass | 132 +++++++++++++++++++++++++++++++++
 1 file changed, 132 insertions(+)
 create mode 100644 meta/classes/overlayfs.bbclass
meta/classes/overlayfs.bbclass new file +132 -0
new file mode 100644
index 0000000000..7fac3e696e
@@ -0,0 +1,132 @@
+# Class for generation of overlayfs mount units
+#
+# It's often desired in Embedded System design to have a read-only rootfs.
+# But a lot of different applications might want to have a read-write access to
+# some parts of a filesystem. It can be especially useful when your update mechanism
+# overwrites the whole rootfs, but you want your application data to be preserved
+# between updates. This class provides a way to achieve that by means
+# of overlayfs and at the same time keeping the base rootfs read-only.
+#
+# Usage example.
+#
+# Set a mount point for a partition overlayfs is going to use as upper layer
+# in your machine configuration. Underlying file system can be anything that
+# is supported by overlayfs
+#
+#   OVERLAYFS_MOUNT_POINT[data] ?= "/data"
+#
+# The class assumes you have a data.mount systemd unit defined elsewhere in your
+# BSP and installed to the image.
+#
+# Then you can specify writable directories on a recipe base
+#
+#   OVERLAYFS_WRITABLE_PATHS[data] = "/usr/share/my-custom-application"
+#
+# To support several mount points you can use a different variable flag. Assume we
+# what to have a writable location on the file system, but not interested where the data
+# survive a reboot. The we could have a mnt-overlay.mount unit for a tmpfs file system:
+#
+#   OVERLAYFS_MOUNT_POINT[mnt-overlay] = "/mnt/overlay"
+#   OVERLAYFS_WRITABLE_PATHS[mnt-overlay] = "/usr/share/another-application"
+
+OVERLAYFS_WRITABLE_PATHS[data] ?= ""
+
+inherit systemd
+
+def strForBash(s):
+    return s.replace('\\', '\\\\')
+
+def unitFileList(d):
+    fileList = []
+    overlayMountPoints = d.getVarFlags("OVERLAYFS_MOUNT_POINT")
+    for mountPoint in overlayMountPoints:
+        for path in d.getVarFlag('OVERLAYFS_WRITABLE_PATHS', mountPoint).split():
+            fileList.append(mountUnitName(path))
+            fileList.append(helperUnitName(path))
+
+    return fileList
+
+# this function is based on https://github.com/systemd/systemd/blob/main/src/basic/unit-name.c
+def escapeSystemdUnitName(path):
+    escapeMap = {
+        '/': '-',
+        '-': "\\x2d",
+        '\\': "\\x5d"
+    }
+    return "".join([escapeMap.get(c, c) for c in path.strip('/')])
+
+def mountUnitName(unit):
+    return escapeSystemdUnitName(unit) + '.mount'
+
+def helperUnitName(unit):
+    return escapeSystemdUnitName(unit) + '-create-upper-dir.service'
+
+python do_create_overlayfs_units() {
+    CreateDirsUnitTemplate = """[Unit]
+Description=Overlayfs directories setup
+Requires={DATA_MOUNT_UNIT}
+After={DATA_MOUNT_UNIT}
+DefaultDependencies=no
+
+[Service]
+Type=oneshot
+ExecStart=mkdir -p {DATA_MOUNT_POINT}/workdir{LOWERDIR} && mkdir -p {DATA_MOUNT_POINT}/upper{LOWERDIR}
+RemainAfterExit=true
+StandardOutput=journal
+
+[Install]
+WantedBy=multi-user.target
+"""
+    MountUnitTemplate = """[Unit]
+Description=Overlayfs mount unit
+Requires={CREATE_DIRS_SERVICE}
+After={CREATE_DIRS_SERVICE}
+
+[Mount]
+What=overlay
+Where={LOWERDIR}
+Type=overlay
+Options=lowerdir={LOWERDIR},upperdir={DATA_MOUNT_POINT}/upper{LOWERDIR},workdir={DATA_MOUNT_POINT}/workdir{LOWERDIR}
+
+[Install]
+WantedBy=multi-user.target
+"""
+
+    def prepareUnits(data, lower):
+        args = {
+            'DATA_MOUNT_POINT': data,
+            'DATA_MOUNT_UNIT': mountUnitName(data),
+            'CREATE_DIRS_SERVICE': helperUnitName(lower),
+            'LOWERDIR': lower,
+        }
+
+        with open(os.path.join(d.getVar('WORKDIR'), mountUnitName(lower)), 'w') as f:
+            f.write(MountUnitTemplate.format(**args))
+
+        with open(os.path.join(d.getVar('WORKDIR'), helperUnitName(lower)), 'w') as f:
+            f.write(CreateDirsUnitTemplate.format(**args))
+
+    overlayMountPoints = d.getVarFlags("OVERLAYFS_MOUNT_POINT")
+    for mountPoint in overlayMountPoints:
+        for lower in d.getVarFlag('OVERLAYFS_WRITABLE_PATHS', mountPoint).split():
+            prepareUnits(d.getVarFlag('OVERLAYFS_MOUNT_POINT', mountPoint), lower)
+}
+
+# we need to generate file names early during parsing stage
+python () {
+    unitList = unitFileList(d)
+    for unit in unitList:
+        d.appendVar('SYSTEMD_SERVICE_' + d.getVar('PN'), ' ' + unit);
+        d.appendVar('FILES_' + d.getVar('PN'), strForBash(unit))
+
+    d.setVar('OVERLAYFS_UNIT_LIST', ' '.join([strForBash(s) for s in unitList]))
+}
+
+do_install_append() {
+    install -d ${D}${systemd_system_unitdir}
+    for unit in ${OVERLAYFS_UNIT_LIST}; do
+        install -m 0444 ${WORKDIR}/${unit} ${D}${systemd_system_unitdir}
+    done
+}
+
+addtask create_overlayfs_units before do_install
Signature · 1 line
2.28.0
#1 lore raw
Quoted 7 lines · Vyacheslav Yurkov

It's often desired in Embedded System design to have a read-only rootfs. But a lot of different applications might want to have a read-write access to some parts of a filesystem. It can be especially useful when your update mechanism overwrites the whole rootfs, but you want your application data to be preserved between updates. This class provides a way to achieve that by means of overlayfs and at the same time keeping the base rootfs read-only.

Hi Vyacheslav,

This looks like it could be quite useful, but a few comments came to mind on a really quick scan.

Since this is systemd based, it should probably have REQUIRED_DISTRO_FEATURES = "systemd", to ensure that the rest of the system and packages are consistently enabled.

As part of getting this into core, it needs some sort of test cases / test suite (and one that covers the different modes you describe in the class) .. so something like a test image (a user) and the tests themselves.

It really also needs a maintainer entry, since otherwise the maintenance work falls to the same already overloaded folks.

Since this is intended to be used on a per-recipe basis (at least that is my understanding from reading it), what happens if two recipes create conflicting overlay mounts ? or if the mentioned data.mount is not present in the image ?

Bruce

Quoted 151 lines
---
 meta/classes/overlayfs.bbclass | 132 +++++++++++++++++++++++++++++++++
 1 file changed, 132 insertions(+)
 create mode 100644 meta/classes/overlayfs.bbclass
meta/classes/overlayfs.bbclass new file +132 -1
new file mode 100644
index 0000000000..7fac3e696e
@@ -0,0 +1,132 @@
+# Class for generation of overlayfs mount units
+#
+# It's often desired in Embedded System design to have a read-only rootfs.
+# But a lot of different applications might want to have a read-write access to
+# some parts of a filesystem. It can be especially useful when your update mechanism
+# overwrites the whole rootfs, but you want your application data to be preserved
+# between updates. This class provides a way to achieve that by means
+# of overlayfs and at the same time keeping the base rootfs read-only.
+#
+# Usage example.
+#
+# Set a mount point for a partition overlayfs is going to use as upper layer
+# in your machine configuration. Underlying file system can be anything that
+# is supported by overlayfs
+#
+#   OVERLAYFS_MOUNT_POINT[data] ?= "/data"
+#
+# The class assumes you have a data.mount systemd unit defined elsewhere in your
+# BSP and installed to the image.
+#
+# Then you can specify writable directories on a recipe base
+#
+#   OVERLAYFS_WRITABLE_PATHS[data] = "/usr/share/my-custom-application"
+#
+# To support several mount points you can use a different variable flag. Assume we
+# what to have a writable location on the file system, but not interested where the data
+# survive a reboot. The we could have a mnt-overlay.mount unit for a tmpfs file system:
+#
+#   OVERLAYFS_MOUNT_POINT[mnt-overlay] = "/mnt/overlay"
+#   OVERLAYFS_WRITABLE_PATHS[mnt-overlay] = "/usr/share/another-application"
+
+OVERLAYFS_WRITABLE_PATHS[data] ?= ""
+
+inherit systemd
+
+def strForBash(s):
+    return s.replace('\\', '\\\\')
+
+def unitFileList(d):
+    fileList = []
+    overlayMountPoints = d.getVarFlags("OVERLAYFS_MOUNT_POINT")
+    for mountPoint in overlayMountPoints:
+        for path in d.getVarFlag('OVERLAYFS_WRITABLE_PATHS', mountPoint).split():
+            fileList.append(mountUnitName(path))
+            fileList.append(helperUnitName(path))
+
+    return fileList
+
+# this function is based on https://github.com/systemd/systemd/blob/main/src/basic/unit-name.c
+def escapeSystemdUnitName(path):
+    escapeMap = {
+        '/': '-',
+        '-': "\\x2d",
+        '\\': "\\x5d"
+    }
+    return "".join([escapeMap.get(c, c) for c in path.strip('/')])
+
+def mountUnitName(unit):
+    return escapeSystemdUnitName(unit) + '.mount'
+
+def helperUnitName(unit):
+    return escapeSystemdUnitName(unit) + '-create-upper-dir.service'
+
+python do_create_overlayfs_units() {
+    CreateDirsUnitTemplate = """[Unit]
+Description=Overlayfs directories setup
+Requires={DATA_MOUNT_UNIT}
+After={DATA_MOUNT_UNIT}
+DefaultDependencies=no
+
+[Service]
+Type=oneshot
+ExecStart=mkdir -p {DATA_MOUNT_POINT}/workdir{LOWERDIR} && mkdir -p {DATA_MOUNT_POINT}/upper{LOWERDIR}
+RemainAfterExit=true
+StandardOutput=journal
+
+[Install]
+WantedBy=multi-user.target
+"""
+    MountUnitTemplate = """[Unit]
+Description=Overlayfs mount unit
+Requires={CREATE_DIRS_SERVICE}
+After={CREATE_DIRS_SERVICE}
+
+[Mount]
+What=overlay
+Where={LOWERDIR}
+Type=overlay
+Options=lowerdir={LOWERDIR},upperdir={DATA_MOUNT_POINT}/upper{LOWERDIR},workdir={DATA_MOUNT_POINT}/workdir{LOWERDIR}
+
+[Install]
+WantedBy=multi-user.target
+"""
+
+    def prepareUnits(data, lower):
+        args = {
+            'DATA_MOUNT_POINT': data,
+            'DATA_MOUNT_UNIT': mountUnitName(data),
+            'CREATE_DIRS_SERVICE': helperUnitName(lower),
+            'LOWERDIR': lower,
+        }
+
+        with open(os.path.join(d.getVar('WORKDIR'), mountUnitName(lower)), 'w') as f:
+            f.write(MountUnitTemplate.format(**args))
+
+        with open(os.path.join(d.getVar('WORKDIR'), helperUnitName(lower)), 'w') as f:
+            f.write(CreateDirsUnitTemplate.format(**args))
+
+    overlayMountPoints = d.getVarFlags("OVERLAYFS_MOUNT_POINT")
+    for mountPoint in overlayMountPoints:
+        for lower in d.getVarFlag('OVERLAYFS_WRITABLE_PATHS', mountPoint).split():
+            prepareUnits(d.getVarFlag('OVERLAYFS_MOUNT_POINT', mountPoint), lower)
+}
+
+# we need to generate file names early during parsing stage
+python () {
+    unitList = unitFileList(d)
+    for unit in unitList:
+        d.appendVar('SYSTEMD_SERVICE_' + d.getVar('PN'), ' ' + unit);
+        d.appendVar('FILES_' + d.getVar('PN'), strForBash(unit))
+
+    d.setVar('OVERLAYFS_UNIT_LIST', ' '.join([strForBash(s) for s in unitList]))
+}
+
+do_install_append() {
+    install -d ${D}${systemd_system_unitdir}
+    for unit in ${OVERLAYFS_UNIT_LIST}; do
+        install -m 0444 ${WORKDIR}/${unit} ${D}${systemd_system_unitdir}
+    done
+}
+
+addtask create_overlayfs_units before do_install
--

2.28.0

Signature · 3 lines
- Thou shalt not follow the NULL pointer, for chaos and madness await
thee at its end
- "Use the force Harry" - Gandalf, Star Trek II
#2 lore raw

Hi Bruce,
Thanks for a quick feedback.

Quoted 8 lines

Hi Vyacheslav,

This looks like it could be quite useful, but a few comments came to mind on a really quick scan.

Since this is systemd based, it should probably have REQUIRED_DISTRO_FEATURES = "systemd", to ensure that the rest of the system and packages are consistently enabled.

Will include it in my v2.

Quoted 4 lines

As part of getting this into core, it needs some sort of test cases / test suite (and one that covers the different modes you describe in the class) .. so something like a test image (a user) and the tests themselves.

Could you please give me a few hints as to where current tests are stored and how to run them? Is it meta-selftest layer?

Quoted 2 lines

It really also needs a maintainer entry, since otherwise the maintenance work falls to the same already overloaded folks.

You mean just an entry like:
MAINTAINER = "Vyacheslav Yurkov <[email protected]>

in the class itself?
Because meta/conf/distro/include/maintainers.inc mentions only recipe maintainers.

Quoted 4 lines

Since this is intended to be used on a per-recipe basis (at least that is my understanding from reading it), what happens if two recipes create conflicting overlay mounts ? or if the mentioned data.mount is not present in the image ?

These are good questions. I guess if two recipes would try to use the same mounts, then do_rootfs should simply fail the same way when two recipes install the same file.
As for handling a mount.unit, that has to be done on an image level, I guess. The other question is how to do that.. Perhaps a QA-hook in ROOTFS_POSTPROCESS_COMMAND, which checks the file existence (not validity though!)?

Vyacheslav

Quoted 152 lines

Bruce

Quoted 150 lines
---
  meta/classes/overlayfs.bbclass | 132 +++++++++++++++++++++++++++++++++
  1 file changed, 132 insertions(+)

create mode 100644 meta/classes/overlayfs.bbclass

meta/classes/overlayfs.bbclass new file +132 -1
new file mode 100644
index 0000000000..7fac3e696e
@@ -0,0 +1,132 @@
+# Class for generation of overlayfs mount units
+#
+# It's often desired in Embedded System design to have a read-only rootfs.
+# But a lot of different applications might want to have a read-write access to
+# some parts of a filesystem. It can be especially useful when your update mechanism
+# overwrites the whole rootfs, but you want your application data to be preserved
+# between updates. This class provides a way to achieve that by means
+# of overlayfs and at the same time keeping the base rootfs read-only.
+#
+# Usage example.
+#
+# Set a mount point for a partition overlayfs is going to use as upper layer
+# in your machine configuration. Underlying file system can be anything that
+# is supported by overlayfs
+#
+#   OVERLAYFS_MOUNT_POINT[data] ?= "/data"
+#
+# The class assumes you have a data.mount systemd unit defined elsewhere in your
+# BSP and installed to the image.
+#
+# Then you can specify writable directories on a recipe base
+#
+#   OVERLAYFS_WRITABLE_PATHS[data] = "/usr/share/my-custom-application"
+#
+# To support several mount points you can use a different variable flag. Assume we
+# what to have a writable location on the file system, but not interested where the data
+# survive a reboot. The we could have a mnt-overlay.mount unit for a tmpfs file system:
+#
+#   OVERLAYFS_MOUNT_POINT[mnt-overlay] = "/mnt/overlay"
+#   OVERLAYFS_WRITABLE_PATHS[mnt-overlay] = "/usr/share/another-application"
+
+OVERLAYFS_WRITABLE_PATHS[data] ?= ""
+
+inherit systemd
+
+def strForBash(s):
+    return s.replace('\\', '\\\\')
+
+def unitFileList(d):
+    fileList = []
+    overlayMountPoints = d.getVarFlags("OVERLAYFS_MOUNT_POINT")
+    for mountPoint in overlayMountPoints:
+        for path in d.getVarFlag('OVERLAYFS_WRITABLE_PATHS', mountPoint).split():
+            fileList.append(mountUnitName(path))
+            fileList.append(helperUnitName(path))
+
+    return fileList
+
+# this function is based on https://github.com/systemd/systemd/blob/main/src/basic/unit-name.c
+def escapeSystemdUnitName(path):
+    escapeMap = {
+        '/': '-',
+        '-': "\\x2d",
+        '\\': "\\x5d"
+    }
+    return "".join([escapeMap.get(c, c) for c in path.strip('/')])
+
+def mountUnitName(unit):
+    return escapeSystemdUnitName(unit) + '.mount'
+
+def helperUnitName(unit):
+    return escapeSystemdUnitName(unit) + '-create-upper-dir.service'
+
+python do_create_overlayfs_units() {
+    CreateDirsUnitTemplate = """[Unit]
+Description=Overlayfs directories setup
+Requires={DATA_MOUNT_UNIT}
+After={DATA_MOUNT_UNIT}
+DefaultDependencies=no
+
+[Service]
+Type=oneshot
+ExecStart=mkdir -p {DATA_MOUNT_POINT}/workdir{LOWERDIR} && mkdir -p {DATA_MOUNT_POINT}/upper{LOWERDIR}
+RemainAfterExit=true
+StandardOutput=journal
+
+[Install]
+WantedBy=multi-user.target
+"""
+    MountUnitTemplate = """[Unit]
+Description=Overlayfs mount unit
+Requires={CREATE_DIRS_SERVICE}
+After={CREATE_DIRS_SERVICE}
+
+[Mount]
+What=overlay
+Where={LOWERDIR}
+Type=overlay
+Options=lowerdir={LOWERDIR},upperdir={DATA_MOUNT_POINT}/upper{LOWERDIR},workdir={DATA_MOUNT_POINT}/workdir{LOWERDIR}
+
+[Install]
+WantedBy=multi-user.target
+"""
+
+    def prepareUnits(data, lower):
+        args = {
+            'DATA_MOUNT_POINT': data,
+            'DATA_MOUNT_UNIT': mountUnitName(data),
+            'CREATE_DIRS_SERVICE': helperUnitName(lower),
+            'LOWERDIR': lower,
+        }
+
+        with open(os.path.join(d.getVar('WORKDIR'), mountUnitName(lower)), 'w') as f:
+            f.write(MountUnitTemplate.format(**args))
+
+        with open(os.path.join(d.getVar('WORKDIR'), helperUnitName(lower)), 'w') as f:
+            f.write(CreateDirsUnitTemplate.format(**args))
+
+    overlayMountPoints = d.getVarFlags("OVERLAYFS_MOUNT_POINT")
+    for mountPoint in overlayMountPoints:
+        for lower in d.getVarFlag('OVERLAYFS_WRITABLE_PATHS', mountPoint).split():
+            prepareUnits(d.getVarFlag('OVERLAYFS_MOUNT_POINT', mountPoint), lower)
+}
+
+# we need to generate file names early during parsing stage
+python () {
+    unitList = unitFileList(d)
+    for unit in unitList:
+        d.appendVar('SYSTEMD_SERVICE_' + d.getVar('PN'), ' ' + unit);
+        d.appendVar('FILES_' + d.getVar('PN'), strForBash(unit))
+
+    d.setVar('OVERLAYFS_UNIT_LIST', ' '.join([strForBash(s) for s in unitList]))
+}
+
+do_install_append() {
+    install -d ${D}${systemd_system_unitdir}
+    for unit in ${OVERLAYFS_UNIT_LIST}; do
+        install -m 0444 ${WORKDIR}/${unit} ${D}${systemd_system_unitdir}
+    done
+}
+
+addtask create_overlayfs_units before do_install
--

2.28.0

#3 lore raw
Quoted 15 lines · Vyacheslav Yurkov

Hi Bruce,
Thanks for a quick feedback.

Quoted 8 lines

Hi Vyacheslav,

This looks like it could be quite useful, but a few comments came to mind on a really quick scan.

Since this is systemd based, it should probably have REQUIRED_DISTRO_FEATURES = "systemd", to ensure that the rest of the system and packages are consistently enabled.

Will include it in my v2.

Quoted 7 lines
Quoted 4 lines

As part of getting this into core, it needs some sort of test cases / test suite (and one that covers the different modes you describe in the class) .. so something like a test image (a user) and the tests themselves.

Could you please give me a few hints as to where current tests are stored and how to run them? Is it meta-selftest layer?

Richard can guide you better on this front than I can, but the tests can be in a number of forms. ptests (i.e. if you create the reference image that uses it, that recipe could have a ptest), and there are also the lib/oeqa/selftest/ tests.

How they are fully hooked in, so they can be tested with core, I haven't looked in a while, and might steer you wrong, So my recommendations are to checkout those components, and follow what is on the wiki and in the docs (https://wiki.yoctoproject.org/wiki/Oe-selftest),
https://docs.yoctoproject.org/test-manual/intro.html, etc

Quoted 6 lines
Quoted 3 lines

It really also needs a maintainer entry, since otherwise the maintenance work falls to the same already overloaded folks.

You mean just an entry like:
MAINTAINER = "Vyacheslav Yurkov <[email protected]>

Yep.

Quoted 4 lines

in the class itself?
Because meta/conf/distro/include/maintainers.inc mentions only recipe maintainers.

Richard could advise better on this, but yes, even if this isn't a recipe we'd want someone to look after it if it merges to core. So if we are breaking into a new category here, that is probably ok.

Quoted 9 lines
Quoted 5 lines

Since this is intended to be used on a per-recipe basis (at least that is my understanding from reading it), what happens if two recipes create conflicting overlay mounts ? or if the mentioned data.mount is not present in the image ?

These are good questions. I guess if two recipes would try to use the same mounts, then do_rootfs should simply fail the same way when two recipes install the same file.

Yes, the image would definitely conflict during package install in that case. It wasn't entirely clear to me that they'd have to have exactly the same file.

Quoted 4 lines

As for handling a mount.unit, that has to be done on an image level, I guess. The other question is how to do that.. Perhaps a QA-hook in ROOTFS_POSTPROCESS_COMMAND, which checks the file existence (not validity though!)?

That might be feasible, assuming that it doesn't always run, or if it always run, that it doesn't take much time.

Cheers,

Bruce

Quoted 147 lines

Vyacheslav

Quoted 144 lines

Bruce

Quoted 142 lines
---
  meta/classes/overlayfs.bbclass | 132 +++++++++++++++++++++++++++++++++
  1 file changed, 132 insertions(+)

create mode 100644 meta/classes/overlayfs.bbclass

meta/classes/overlayfs.bbclass new file +130 -0
new file mode 100644
index 0000000000..7fac3e696e
@@ -0,0 +1,132 @@
+# Class for generation of overlayfs mount units
+#
+# It's often desired in Embedded System design to have a read-only rootfs.
+# But a lot of different applications might want to have a read-write access to
+# some parts of a filesystem. It can be especially useful when your update mechanism
+# overwrites the whole rootfs, but you want your application data to be preserved
+# between updates. This class provides a way to achieve that by means
+# of overlayfs and at the same time keeping the base rootfs read-only.
+#
+# Usage example.
+#
+# Set a mount point for a partition overlayfs is going to use as upper layer
+# in your machine configuration. Underlying file system can be anything that
+# is supported by overlayfs
+#
+#   OVERLAYFS_MOUNT_POINT[data] ?= "/data"
+#
+# The class assumes you have a data.mount systemd unit defined elsewhere in your
+# BSP and installed to the image.
+#
+# Then you can specify writable directories on a recipe base
+#
+#   OVERLAYFS_WRITABLE_PATHS[data] = "/usr/share/my-custom-application"
+#
+# To support several mount points you can use a different variable flag. Assume we
+# what to have a writable location on the file system, but not interested where the data
+# survive a reboot. The we could have a mnt-overlay.mount unit for a tmpfs file system:
+#
+#   OVERLAYFS_MOUNT_POINT[mnt-overlay] = "/mnt/overlay"
+#   OVERLAYFS_WRITABLE_PATHS[mnt-overlay] = "/usr/share/another-application"
+
+OVERLAYFS_WRITABLE_PATHS[data] ?= ""
+
+inherit systemd
+
+def strForBash(s):
+    return s.replace('\\', '\\\\')
+
+def unitFileList(d):
+    fileList = []
+    overlayMountPoints = d.getVarFlags("OVERLAYFS_MOUNT_POINT")
+    for mountPoint in overlayMountPoints:
+        for path in d.getVarFlag('OVERLAYFS_WRITABLE_PATHS', mountPoint).split():
+            fileList.append(mountUnitName(path))
+            fileList.append(helperUnitName(path))
+
+    return fileList
+
+# this function is based on https://github.com/systemd/systemd/blob/main/src/basic/unit-name.c
+def escapeSystemdUnitName(path):
+    escapeMap = {
+        '/': '-',
+        '-': "\\x2d",
+        '\\': "\\x5d"
+    }
+    return "".join([escapeMap.get(c, c) for c in path.strip('/')])
+
+def mountUnitName(unit):
+    return escapeSystemdUnitName(unit) + '.mount'
+
+def helperUnitName(unit):
+    return escapeSystemdUnitName(unit) + '-create-upper-dir.service'
+
+python do_create_overlayfs_units() {
+    CreateDirsUnitTemplate = """[Unit]
+Description=Overlayfs directories setup
+Requires={DATA_MOUNT_UNIT}
+After={DATA_MOUNT_UNIT}
+DefaultDependencies=no
+
+[Service]
+Type=oneshot
+ExecStart=mkdir -p {DATA_MOUNT_POINT}/workdir{LOWERDIR} && mkdir -p {DATA_MOUNT_POINT}/upper{LOWERDIR}
+RemainAfterExit=true
+StandardOutput=journal
+
+[Install]
+WantedBy=multi-user.target
+"""
+    MountUnitTemplate = """[Unit]
+Description=Overlayfs mount unit
+Requires={CREATE_DIRS_SERVICE}
+After={CREATE_DIRS_SERVICE}
+
+[Mount]
+What=overlay
+Where={LOWERDIR}
+Type=overlay
+Options=lowerdir={LOWERDIR},upperdir={DATA_MOUNT_POINT}/upper{LOWERDIR},workdir={DATA_MOUNT_POINT}/workdir{LOWERDIR}
+
+[Install]
+WantedBy=multi-user.target
+"""
+
+    def prepareUnits(data, lower):
+        args = {
+            'DATA_MOUNT_POINT': data,
+            'DATA_MOUNT_UNIT': mountUnitName(data),
+            'CREATE_DIRS_SERVICE': helperUnitName(lower),
+            'LOWERDIR': lower,
+        }
+
+        with open(os.path.join(d.getVar('WORKDIR'), mountUnitName(lower)), 'w') as f:
+            f.write(MountUnitTemplate.format(**args))
+
+        with open(os.path.join(d.getVar('WORKDIR'), helperUnitName(lower)), 'w') as f:
+            f.write(CreateDirsUnitTemplate.format(**args))
+
+    overlayMountPoints = d.getVarFlags("OVERLAYFS_MOUNT_POINT")
+    for mountPoint in overlayMountPoints:
+        for lower in d.getVarFlag('OVERLAYFS_WRITABLE_PATHS', mountPoint).split():
+            prepareUnits(d.getVarFlag('OVERLAYFS_MOUNT_POINT', mountPoint), lower)
+}
+
+# we need to generate file names early during parsing stage
+python () {
+    unitList = unitFileList(d)
+    for unit in unitList:
+        d.appendVar('SYSTEMD_SERVICE_' + d.getVar('PN'), ' ' + unit);
+        d.appendVar('FILES_' + d.getVar('PN'), strForBash(unit))
+
+    d.setVar('OVERLAYFS_UNIT_LIST', ' '.join([strForBash(s) for s in unitList]))
+}
+
+do_install_append() {
+    install -d ${D}${systemd_system_unitdir}
+    for unit in ${OVERLAYFS_UNIT_LIST}; do
+        install -m 0444 ${WORKDIR}/${unit} ${D}${systemd_system_unitdir}
+    done
+}
Quoted 9 lines
Quoted 8 lines
Quoted 8 lines
meta/classes/overlayfs.bbclass new file +2 -1
+
+addtask create_overlayfs_units before do_install
--

2.28.0

--
- Thou shalt not follow the NULL pointer, for chaos and madness await thee at its end
- "Use the force Harry" - Gandalf, Star Trek II

#4 lore raw
In reply to Vyacheslav Yurkov

Hi,

volatile-binds (maybe the *volatile* naming is misleading) recipe in yocto is already doing that, just to be appended .ie:

VOLATILE_BINDS += "\
/data/lib/systemd/network /lib/systemd/network\n\
/data/home/root /home/root\n\
"

It will generate for each overlay a mount point service.

BR

Quoted 144 lines · Vyacheslav Yurkov

It's often desired in Embedded System design to have a read-only rootfs. But a lot of different applications might want to have a read-write access to some parts of a filesystem. It can be especially useful when your update mechanism overwrites the whole rootfs, but you want your application data to be preserved between updates. This class provides a way to achieve that by means of overlayfs and at the same time keeping the base rootfs read-only.

---
  meta/classes/overlayfs.bbclass | 132 +++++++++++++++++++++++++++++++++
  1 file changed, 132 insertions(+)

create mode 100644 meta/classes/overlayfs.bbclass

meta/classes/overlayfs.bbclass new file +122 -0
new file mode 100644
index 0000000000..7fac3e696e
@@ -0,0 +1,132 @@
+# Class for generation of overlayfs mount units
+#
+# It's often desired in Embedded System design to have a read-only rootfs.
+# But a lot of different applications might want to have a read-write access to
+# some parts of a filesystem. It can be especially useful when your update mechanism
+# overwrites the whole rootfs, but you want your application data to be preserved
+# between updates. This class provides a way to achieve that by means
+# of overlayfs and at the same time keeping the base rootfs read-only.
+#
+# Usage example.
+#
+# Set a mount point for a partition overlayfs is going to use as upper layer
+# in your machine configuration. Underlying file system can be anything that
+# is supported by overlayfs
+#
+#   OVERLAYFS_MOUNT_POINT[data] ?= "/data"
+#
+# The class assumes you have a data.mount systemd unit defined elsewhere in your
+# BSP and installed to the image.
+#
+# Then you can specify writable directories on a recipe base
+#
+#   OVERLAYFS_WRITABLE_PATHS[data] = "/usr/share/my-custom-application"
+#
+# To support several mount points you can use a different variable flag. Assume we
+# what to have a writable location on the file system, but not interested where the data
+# survive a reboot. The we could have a mnt-overlay.mount unit for a tmpfs file system:
+#
+#   OVERLAYFS_MOUNT_POINT[mnt-overlay] = "/mnt/overlay"
+#   OVERLAYFS_WRITABLE_PATHS[mnt-overlay] = "/usr/share/another-application"
+
+OVERLAYFS_WRITABLE_PATHS[data] ?= ""
+
+inherit systemd
+
+def strForBash(s):
+    return s.replace('\\', '\\\\')
+
+def unitFileList(d):
+    fileList = []
+    overlayMountPoints = d.getVarFlags("OVERLAYFS_MOUNT_POINT")
+    for mountPoint in overlayMountPoints:
+        for path in d.getVarFlag('OVERLAYFS_WRITABLE_PATHS', mountPoint).split():
+            fileList.append(mountUnitName(path))
+            fileList.append(helperUnitName(path))
+
+    return fileList
+
+# this function is based on https://github.com/systemd/systemd/blob/main/src/basic/unit-name.c
+def escapeSystemdUnitName(path):
+    escapeMap = {
+        '/': '-',
+        '-': "\\x2d",
+        '\\': "\\x5d"
+    }
+    return "".join([escapeMap.get(c, c) for c in path.strip('/')])
+
+def mountUnitName(unit):
+    return escapeSystemdUnitName(unit) + '.mount'
+
+def helperUnitName(unit):
+    return escapeSystemdUnitName(unit) + '-create-upper-dir.service'
+
+python do_create_overlayfs_units() {
+    CreateDirsUnitTemplate = """[Unit]
+Description=Overlayfs directories setup
+Requires={DATA_MOUNT_UNIT}
+After={DATA_MOUNT_UNIT}
+DefaultDependencies=no
+
+[Service]
+Type=oneshot
+ExecStart=mkdir -p {DATA_MOUNT_POINT}/workdir{LOWERDIR} && mkdir -p {DATA_MOUNT_POINT}/upper{LOWERDIR}
+RemainAfterExit=true
+StandardOutput=journal
+
+[Install]
+WantedBy=multi-user.target
+"""
+    MountUnitTemplate = """[Unit]
+Description=Overlayfs mount unit
+Requires={CREATE_DIRS_SERVICE}
+After={CREATE_DIRS_SERVICE}
+
+[Mount]
+What=overlay
+Where={LOWERDIR}
+Type=overlay
+Options=lowerdir={LOWERDIR},upperdir={DATA_MOUNT_POINT}/upper{LOWERDIR},workdir={DATA_MOUNT_POINT}/workdir{LOWERDIR}
+
+[Install]
+WantedBy=multi-user.target
+"""
+
+    def prepareUnits(data, lower):
+        args = {
+            'DATA_MOUNT_POINT': data,
+            'DATA_MOUNT_UNIT': mountUnitName(data),
+            'CREATE_DIRS_SERVICE': helperUnitName(lower),
+            'LOWERDIR': lower,
+        }
+
+        with open(os.path.join(d.getVar('WORKDIR'), mountUnitName(lower)), 'w') as f:
+            f.write(MountUnitTemplate.format(**args))
+
+        with open(os.path.join(d.getVar('WORKDIR'), helperUnitName(lower)), 'w') as f:
+            f.write(CreateDirsUnitTemplate.format(**args))
+
+    overlayMountPoints = d.getVarFlags("OVERLAYFS_MOUNT_POINT")
+    for mountPoint in overlayMountPoints:
+        for lower in d.getVarFlag('OVERLAYFS_WRITABLE_PATHS', mountPoint).split():
+            prepareUnits(d.getVarFlag('OVERLAYFS_MOUNT_POINT', mountPoint), lower)
+}
+
+# we need to generate file names early during parsing stage
+python () {
+    unitList = unitFileList(d)
+    for unit in unitList:
+        d.appendVar('SYSTEMD_SERVICE_' + d.getVar('PN'), ' ' + unit);
+        d.appendVar('FILES_' + d.getVar('PN'), strForBash(unit))
+
+    d.setVar('OVERLAYFS_UNIT_LIST', ' '.join([strForBash(s) for s in unitList]))

meta/classes/overlayfs.bbclass line 132

Quoted 10 lines
meta/classes/overlayfs.bbclass new file +10 -0
+}
+
+do_install_append() {
+    install -d ${D}${systemd_system_unitdir}
+    for unit in ${OVERLAYFS_UNIT_LIST}; do
+        install -m 0444 ${WORKDIR}/${unit} ${D}${systemd_system_unitdir}
+    done
+}
+
+addtask create_overlayfs_units before do_install

Mit freundlichen Grüßen / Kind regards

Signature · 10 lines
Ayoub Zaki
Embedded Systems Consultant

Vaihinger Straße 2/1
D-71634 Ludwigsburg

Mobile   : +4917662901545
Email    : [email protected]
Homepage : https://embexus.com
VAT No.  : DE313902634
#5 lore raw

Hi Ayoub,
Thanks for the hint. I knew the recipe exists, but I didn't know it uses overlay.

I did have a closer look at the mount-copybind script, and it uses overlay indeed, but it seems it's used for the other purpose. Originally the recipe provided only bind-mount functionality, they added overlayfs support later to save on space.

I tried to use the recipe with my use case, but it turns out that reverses the options of mount command in mount-copybind. I'm also not sure how the dependencies should be injected into the mount unit.

Vyacheslav

Quoted 184 lines · Ayoub Zaki

Hi,

volatile-binds (maybe the *volatile* naming is misleading) recipe in yocto is already doing that, just to be appended .ie:

VOLATILE_BINDS += "\
/data/lib/systemd/network /lib/systemd/network\n\
/data/home/root /home/root\n\
"

It will generate for each overlay a mount point service.

BR

Quoted 142 lines · Vyacheslav Yurkov

It's often desired in Embedded System design to have a read-only rootfs. But a lot of different applications might want to have a read-write access to some parts of a filesystem. It can be especially useful when your update mechanism overwrites the whole rootfs, but you want your application data to be preserved between updates. This class provides a way to achieve that by means of overlayfs and at the same time keeping the base rootfs read-only.

---
  meta/classes/overlayfs.bbclass | 132 +++++++++++++++++++++++++++++++++
  1 file changed, 132 insertions(+)

create mode 100644 meta/classes/overlayfs.bbclass

meta/classes/overlayfs.bbclass new file +122 -0
new file mode 100644
index 0000000000..7fac3e696e
@@ -0,0 +1,132 @@
+# Class for generation of overlayfs mount units
+#
+# It's often desired in Embedded System design to have a read-only rootfs.
+# But a lot of different applications might want to have a read-write access to
+# some parts of a filesystem. It can be especially useful when your update mechanism
+# overwrites the whole rootfs, but you want your application data to be preserved
+# between updates. This class provides a way to achieve that by means
+# of overlayfs and at the same time keeping the base rootfs read-only.
+#
+# Usage example.
+#
+# Set a mount point for a partition overlayfs is going to use as upper layer
+# in your machine configuration. Underlying file system can be anything that
+# is supported by overlayfs
+#
+#   OVERLAYFS_MOUNT_POINT[data] ?= "/data"
+#
+# The class assumes you have a data.mount systemd unit defined elsewhere in your
+# BSP and installed to the image.
+#
+# Then you can specify writable directories on a recipe base
+#
+#   OVERLAYFS_WRITABLE_PATHS[data] = "/usr/share/my-custom-application"
+#
+# To support several mount points you can use a different variable flag. Assume we
+# what to have a writable location on the file system, but not interested where the data
+# survive a reboot. The we could have a mnt-overlay.mount unit for a tmpfs file system:
+#
+#   OVERLAYFS_MOUNT_POINT[mnt-overlay] = "/mnt/overlay"
+#   OVERLAYFS_WRITABLE_PATHS[mnt-overlay] = "/usr/share/another-application"
+
+OVERLAYFS_WRITABLE_PATHS[data] ?= ""
+
+inherit systemd
+
+def strForBash(s):
+    return s.replace('\\', '\\\\')
+
+def unitFileList(d):
+    fileList = []
+    overlayMountPoints = d.getVarFlags("OVERLAYFS_MOUNT_POINT")
+    for mountPoint in overlayMountPoints:
+        for path in d.getVarFlag('OVERLAYFS_WRITABLE_PATHS', mountPoint).split():
+            fileList.append(mountUnitName(path))
+            fileList.append(helperUnitName(path))
+
+    return fileList
+
+# this function is based onhttps://github.com/systemd/systemd/blob/main/src/basic/unit-name.c
+def escapeSystemdUnitName(path):
+    escapeMap = {
+        '/': '-',
+        '-': "\\x2d",
+        '\\': "\\x5d"
+    }
+    return "".join([escapeMap.get(c, c) for c in path.strip('/')])
+
+def mountUnitName(unit):
+    return escapeSystemdUnitName(unit) + '.mount'
+
+def helperUnitName(unit):
+    return escapeSystemdUnitName(unit) + '-create-upper-dir.service'
+
+python do_create_overlayfs_units() {
+    CreateDirsUnitTemplate = """[Unit]
+Description=Overlayfs directories setup
+Requires={DATA_MOUNT_UNIT}
+After={DATA_MOUNT_UNIT}
+DefaultDependencies=no
+
+[Service]
+Type=oneshot
+ExecStart=mkdir -p {DATA_MOUNT_POINT}/workdir{LOWERDIR} && mkdir -p {DATA_MOUNT_POINT}/upper{LOWERDIR}
+RemainAfterExit=true
+StandardOutput=journal
+
+[Install]
+WantedBy=multi-user.target
+"""
+    MountUnitTemplate = """[Unit]
+Description=Overlayfs mount unit
+Requires={CREATE_DIRS_SERVICE}
+After={CREATE_DIRS_SERVICE}
+
+[Mount]
+What=overlay
+Where={LOWERDIR}
+Type=overlay
+Options=lowerdir={LOWERDIR},upperdir={DATA_MOUNT_POINT}/upper{LOWERDIR},workdir={DATA_MOUNT_POINT}/workdir{LOWERDIR}
+
+[Install]
+WantedBy=multi-user.target
+"""
+
+    def prepareUnits(data, lower):
+        args = {
+            'DATA_MOUNT_POINT': data,
+            'DATA_MOUNT_UNIT': mountUnitName(data),
+            'CREATE_DIRS_SERVICE': helperUnitName(lower),
+            'LOWERDIR': lower,
+        }
+
+        with open(os.path.join(d.getVar('WORKDIR'), mountUnitName(lower)), 'w') as f:
+            f.write(MountUnitTemplate.format(**args))
+
+        with open(os.path.join(d.getVar('WORKDIR'), helperUnitName(lower)), 'w') as f:
+            f.write(CreateDirsUnitTemplate.format(**args))
+
+    overlayMountPoints = d.getVarFlags("OVERLAYFS_MOUNT_POINT")
+    for mountPoint in overlayMountPoints:
+        for lower in d.getVarFlag('OVERLAYFS_WRITABLE_PATHS', mountPoint).split():
+            prepareUnits(d.getVarFlag('OVERLAYFS_MOUNT_POINT', mountPoint), lower)
+}
+
+# we need to generate file names early during parsing stage
+python () {
+    unitList = unitFileList(d)
+    for unit in unitList:
+        d.appendVar('SYSTEMD_SERVICE_' + d.getVar('PN'), ' ' + unit);
+        d.appendVar('FILES_' + d.getVar('PN'), strForBash(unit))
+
+    d.setVar('OVERLAYFS_UNIT_LIST', ' '.join([strForBash(s) for s in unitList]))

meta/classes/overlayfs.bbclass line 132

Quoted 10 lines
meta/classes/overlayfs.bbclass new file +10 -0
+}
+
+do_install_append() {
+    install -d ${D}${systemd_system_unitdir}
+    for unit in ${OVERLAYFS_UNIT_LIST}; do
+        install -m 0444 ${WORKDIR}/${unit} ${D}${systemd_system_unitdir}
+    done
+}
+
+addtask create_overlayfs_units before do_install

Mit freundlichen Grüßen / Kind regards

Signature · 10 lines
Ayoub Zaki
Embedded Systems Consultant

Vaihinger Straße 2/1
D-71634 Ludwigsburg

Mobile   : +4917662901545
Email    :[email protected]
Homepage :https://embexus.com
VAT No.  : DE313902634
#6 lore raw
In reply to Vyacheslav Yurkov

Hi Vyacheslav

It would be very nice to have such a feature in Yocto.

If I remember correctly I tried to do that once with an overlay on /etc. However, that did not work with systemd. Did you ever try to add /etc as a mount point?

Quoted 87 lines · Vyacheslav Yurkov

It's often desired in Embedded System design to have a read-only rootfs. But a lot of different applications might want to have a read-write access to some parts of a filesystem. It can be especially useful when your update mechanism overwrites the whole rootfs, but you want your application data to be preserved between updates. This class provides a way to achieve that by means of overlayfs and at the same time keeping the base rootfs read-only.

---
 meta/classes/overlayfs.bbclass | 132 +++++++++++++++++++++++++++++++++

 1 file changed, 132 insertions(+)

 create mode 100644 meta/classes/overlayfs.bbclass

meta/classes/overlayfs.bbclass new file +68 -0
new file mode 100644
index 0000000000..7fac3e696e
@@ -0,0 +1,132 @@
+# Class for generation of overlayfs mount units
+#
+# It's often desired in Embedded System design to have a read-only rootfs.
+# But a lot of different applications might want to have a read-write access to
+# some parts of a filesystem. It can be especially useful when your update mechanism
+# overwrites the whole rootfs, but you want your application data to be preserved
+# between updates. This class provides a way to achieve that by means
+# of overlayfs and at the same time keeping the base rootfs read-only.
+#
+# Usage example.
+#
+# Set a mount point for a partition overlayfs is going to use as upper layer
+# in your machine configuration. Underlying file system can be anything that
+# is supported by overlayfs
+#
+#   OVERLAYFS_MOUNT_POINT[data] ?= "/data"
+#
+# The class assumes you have a data.mount systemd unit defined elsewhere in your
+# BSP and installed to the image.
+#
+# Then you can specify writable directories on a recipe base
+#
+#   OVERLAYFS_WRITABLE_PATHS[data] = "/usr/share/my-custom-application"
+#
+# To support several mount points you can use a different variable flag. Assume we
+# what to have a writable location on the file system, but not interested where the data
+# survive a reboot. The we could have a mnt-overlay.mount unit for a tmpfs file system:
+#
+#   OVERLAYFS_MOUNT_POINT[mnt-overlay] = "/mnt/overlay"
+#   OVERLAYFS_WRITABLE_PATHS[mnt-overlay] = "/usr/share/another-application"
+
+OVERLAYFS_WRITABLE_PATHS[data] ?= ""
+
+inherit systemd
+
+def strForBash(s):
+    return s.replace('\\', '\\\\')
+
+def unitFileList(d):
+    fileList = []
+    overlayMountPoints = d.getVarFlags("OVERLAYFS_MOUNT_POINT")
+    for mountPoint in overlayMountPoints:
+        for path in d.getVarFlag('OVERLAYFS_WRITABLE_PATHS', mountPoint).split():
+            fileList.append(mountUnitName(path))
+            fileList.append(helperUnitName(path))
+
+    return fileList
+
+# this function is based on https://github.com/systemd/systemd/blob/main/src/basic/unit-name.c
+def escapeSystemdUnitName(path):
+    escapeMap = {
+        '/': '-',
+        '-': "\\x2d",
+        '\\': "\\x5d"
+    }
+    return "".join([escapeMap.get(c, c) for c in path.strip('/')])
+
+def mountUnitName(unit):
+    return escapeSystemdUnitName(unit) + '.mount'
+
+def helperUnitName(unit):
+    return escapeSystemdUnitName(unit) + '-create-upper-dir.service'
+
+python do_create_overlayfs_units() {
+    CreateDirsUnitTemplate = """[Unit]
+Description=Overlayfs directories setup
+Requires={DATA_MOUNT_UNIT}
+After={DATA_MOUNT_UNIT}

meta/classes/overlayfs.bbclass line 78

Quoted 10 lines
meta/classes/overlayfs.bbclass new file +10 -0
+DefaultDependencies=no
+
+[Service]
+Type=oneshot
+ExecStart=mkdir -p {DATA_MOUNT_POINT}/workdir{LOWERDIR} && mkdir -p {DATA_MOUNT_POINT}/upper{LOWERDIR}
+RemainAfterExit=true
+StandardOutput=journal
+
+[Install]
+WantedBy=multi-user.target

The mounting of overlays should probably happen very early at boot, before services start using the desired folders.

Regards,
Adrian

Quoted 56 lines
meta/classes/overlayfs.bbclass new file +54 -0
+"""
+    MountUnitTemplate = """[Unit]
+Description=Overlayfs mount unit
+Requires={CREATE_DIRS_SERVICE}
+After={CREATE_DIRS_SERVICE}
+
+[Mount]
+What=overlay
+Where={LOWERDIR}
+Type=overlay
+Options=lowerdir={LOWERDIR},upperdir={DATA_MOUNT_POINT}/upper{LOWERDIR},workdir={DATA_MOUNT_POINT}/workdir{LOWERDIR}
+
+[Install]
+WantedBy=multi-user.target
+"""
+
+    def prepareUnits(data, lower):
+        args = {
+            'DATA_MOUNT_POINT': data,
+            'DATA_MOUNT_UNIT': mountUnitName(data),
+            'CREATE_DIRS_SERVICE': helperUnitName(lower),
+            'LOWERDIR': lower,
+        }
+
+        with open(os.path.join(d.getVar('WORKDIR'), mountUnitName(lower)), 'w') as f:
+            f.write(MountUnitTemplate.format(**args))
+
+        with open(os.path.join(d.getVar('WORKDIR'), helperUnitName(lower)), 'w') as f:
+            f.write(CreateDirsUnitTemplate.format(**args))
+
+    overlayMountPoints = d.getVarFlags("OVERLAYFS_MOUNT_POINT")
+    for mountPoint in overlayMountPoints:
+        for lower in d.getVarFlag('OVERLAYFS_WRITABLE_PATHS', mountPoint).split():
+            prepareUnits(d.getVarFlag('OVERLAYFS_MOUNT_POINT', mountPoint), lower)
+}
+
+# we need to generate file names early during parsing stage
+python () {
+    unitList = unitFileList(d)
+    for unit in unitList:
+        d.appendVar('SYSTEMD_SERVICE_' + d.getVar('PN'), ' ' + unit);
+        d.appendVar('FILES_' + d.getVar('PN'), strForBash(unit))
+
+    d.setVar('OVERLAYFS_UNIT_LIST', ' '.join([strForBash(s) for s in unitList]))
+}
+
+do_install_append() {
+    install -d ${D}${systemd_system_unitdir}
+    for unit in ${OVERLAYFS_UNIT_LIST}; do
+        install -m 0444 ${WORKDIR}/${unit} ${D}${systemd_system_unitdir}
+    done
+}
+
+addtask create_overlayfs_units before do_install
#7 lore raw

Hi Adrian,
I don't think using /etc with overlay is a wise idea. You'd like to keep system configuration updatable, which wouldn't be simple if the whole /etc is an overlay (TBH, I'm not even sure it's possible that way. You'd probably need to write custom scripts and add initramfs here). But you can export to overlay individual user-specific configuration, for example /etc/wpa_supplicant.

Vyacheslav

Quoted 169 lines · Adrian Freihofer

Hi Vyacheslav

It would be very nice to have such a feature in Yocto.

If I remember correctly I tried to do that once with an overlay on /etc. However, that did not work with systemd. Did you ever try to add /etc as a mount point?

Quoted 87 lines · Vyacheslav Yurkov

It's often desired in Embedded System design to have a read-only rootfs. But a lot of different applications might want to have a read-write access to some parts of a filesystem. It can be especially useful when your update mechanism overwrites the whole rootfs, but you want your application data to be preserved between updates. This class provides a way to achieve that by means of overlayfs and at the same time keeping the base rootfs read-only.

---
  meta/classes/overlayfs.bbclass | 132 +++++++++++++++++++++++++++++++++
  1 file changed, 132 insertions(+)

 create mode 100644 meta/classes/overlayfs.bbclass

meta/classes/overlayfs.bbclass new file +68 -0
new file mode 100644
index 0000000000..7fac3e696e
@@ -0,0 +1,132 @@
+# Class for generation of overlayfs mount units
+#
+# It's often desired in Embedded System design to have a read-only rootfs.
+# But a lot of different applications might want to have a read-write access to
+# some parts of a filesystem. It can be especially useful when your update mechanism
+# overwrites the whole rootfs, but you want your application data to be preserved
+# between updates. This class provides a way to achieve that by means
+# of overlayfs and at the same time keeping the base rootfs read-only.
+#
+# Usage example.
+#
+# Set a mount point for a partition overlayfs is going to use as upper layer
+# in your machine configuration. Underlying file system can be anything that
+# is supported by overlayfs
+#
+#   OVERLAYFS_MOUNT_POINT[data] ?= "/data"
+#
+# The class assumes you have a data.mount systemd unit defined elsewhere in your
+# BSP and installed to the image.
+#
+# Then you can specify writable directories on a recipe base
+#
+#   OVERLAYFS_WRITABLE_PATHS[data] = "/usr/share/my-custom-application"
+#
+# To support several mount points you can use a different variable flag. Assume we
+# what to have a writable location on the file system, but not interested where the data
+# survive a reboot. The we could have a mnt-overlay.mount unit for a tmpfs file system:
+#
+#   OVERLAYFS_MOUNT_POINT[mnt-overlay] = "/mnt/overlay"
+#   OVERLAYFS_WRITABLE_PATHS[mnt-overlay] = "/usr/share/another-application"
+
+OVERLAYFS_WRITABLE_PATHS[data] ?= ""
+
+inherit systemd
+
+def strForBash(s):
+    return s.replace('\\', '\\\\')
+
+def unitFileList(d):
+    fileList = []
+    overlayMountPoints = d.getVarFlags("OVERLAYFS_MOUNT_POINT")
+    for mountPoint in overlayMountPoints:
+        for path in d.getVarFlag('OVERLAYFS_WRITABLE_PATHS', mountPoint).split():
+            fileList.append(mountUnitName(path))
+            fileList.append(helperUnitName(path))
+
+    return fileList
+
+# this function is based on https://github.com/systemd/systemd/blob/main/src/basic/unit-name.c
+def escapeSystemdUnitName(path):
+    escapeMap = {
+        '/': '-',
+        '-': "\\x2d",
+        '\\': "\\x5d"
+    }
+    return "".join([escapeMap.get(c, c) for c in path.strip('/')])
+
+def mountUnitName(unit):
+    return escapeSystemdUnitName(unit) + '.mount'
+
+def helperUnitName(unit):
+    return escapeSystemdUnitName(unit) + '-create-upper-dir.service'
+
+python do_create_overlayfs_units() {
+    CreateDirsUnitTemplate = """[Unit]
+Description=Overlayfs directories setup
+Requires={DATA_MOUNT_UNIT}
+After={DATA_MOUNT_UNIT}

meta/classes/overlayfs.bbclass line 78

Quoted 10 lines
meta/classes/overlayfs.bbclass new file +10 -0
+DefaultDependencies=no
+
+[Service]
+Type=oneshot
+ExecStart=mkdir -p {DATA_MOUNT_POINT}/workdir{LOWERDIR} && mkdir -p {DATA_MOUNT_POINT}/upper{LOWERDIR}
+RemainAfterExit=true
+StandardOutput=journal
+
+[Install]
+WantedBy=multi-user.target

The mounting of overlays should probably happen very early at boot, before services start using the desired folders.

Regards,
Adrian

Quoted 56 lines
meta/classes/overlayfs.bbclass new file +54 -0
+"""
+    MountUnitTemplate = """[Unit]
+Description=Overlayfs mount unit
+Requires={CREATE_DIRS_SERVICE}
+After={CREATE_DIRS_SERVICE}
+
+[Mount]
+What=overlay
+Where={LOWERDIR}
+Type=overlay
+Options=lowerdir={LOWERDIR},upperdir={DATA_MOUNT_POINT}/upper{LOWERDIR},workdir={DATA_MOUNT_POINT}/workdir{LOWERDIR}
+
+[Install]
+WantedBy=multi-user.target
+"""
+
+    def prepareUnits(data, lower):
+        args = {
+            'DATA_MOUNT_POINT': data,
+            'DATA_MOUNT_UNIT': mountUnitName(data),
+            'CREATE_DIRS_SERVICE': helperUnitName(lower),
+            'LOWERDIR': lower,
+        }
+
+        with open(os.path.join(d.getVar('WORKDIR'), mountUnitName(lower)), 'w') as f:
+            f.write(MountUnitTemplate.format(**args))
+
+        with open(os.path.join(d.getVar('WORKDIR'), helperUnitName(lower)), 'w') as f:
+            f.write(CreateDirsUnitTemplate.format(**args))
+
+    overlayMountPoints = d.getVarFlags("OVERLAYFS_MOUNT_POINT")
+    for mountPoint in overlayMountPoints:
+        for lower in d.getVarFlag('OVERLAYFS_WRITABLE_PATHS', mountPoint).split():
+            prepareUnits(d.getVarFlag('OVERLAYFS_MOUNT_POINT', mountPoint), lower)
+}
+
+# we need to generate file names early during parsing stage
+python () {
+    unitList = unitFileList(d)
+    for unit in unitList:
+        d.appendVar('SYSTEMD_SERVICE_' + d.getVar('PN'), ' ' + unit);
+        d.appendVar('FILES_' + d.getVar('PN'), strForBash(unit))
+
+    d.setVar('OVERLAYFS_UNIT_LIST', ' '.join([strForBash(s) for s in unitList]))
+}
+
+do_install_append() {
+    install -d ${D}${systemd_system_unitdir}
+    for unit in ${OVERLAYFS_UNIT_LIST}; do
+        install -m 0444 ${WORKDIR}/${unit} ${D}${systemd_system_unitdir}
+    done
+}
+
+addtask create_overlayfs_units before do_install
#8 lore raw
In reply to Vyacheslav Yurkov

Hi,

Quoted 6 lines · Vyacheslav Yurkov

It's often desired in Embedded System design to have a read-only rootfs. But a lot of different applications might want to have a read-write access to some parts of a filesystem. It can be especially useful when your update mechanism overwrites the whole rootfs, but you want your application data to be preserved between updates. This class provides a way to achieve that by means of overlayfs and at the same time keeping the base rootfs read-only.

I was looking into something like this lately. There are actually two kinds of write access - volatile (tmpfs) and persistent (to disk). So my approach would have been to use bind mounts from a tmpfs/ext4 based on configuration per recipe. Is there any advantage to working with overlayfs in this case? The main idea is to have state that doesn't come from the ro part of the OS because otherwise, updates will be shadowed by the overlayfs. So I fail to see the advantage of using a set of overlayfs mounts instead of bind mounts. At least with bind mounts, it will always shadow the rootfs bits so you can catch early this case.

Depending on your use case, most of the times, you'd want this "state" setup happening pre init (in initramfs, for example), especially if you want to handle persistent systemd logging or machine id, etc.

Signature · 2 lines
Andrei Gherzan 
gpg: rsa4096/D4D94F67AD0E9640
#9 lore raw

Hi Andrei,
Like you mentioned, it's application specific. The advantage of overlayfs here is that you preserve the data that was changed by user (in case of persistent storage), which of course would hide the data from rootfs, or your just see the defaults from rootfs if the data was not changed. With bind mounts you would need to copy default data first into your tmpfs/ext4 bind mount if you wanna have defaults.

Vyacheslav

Quoted 4 lines

I was looking into something like this lately. There are actually two kinds of write access - volatile (tmpfs) and persistent (to disk). So my approach would have been to use bind mounts from a tmpfs/ext4 based on configuration per recipe. Is there any advantage to working with overlayfs in this case? The main idea is to have state that doesn't come from the ro part of the OS because otherwise, updates will be shadowed by the overlayfs. So I fail to see the advantage of using a set of overlayfs mounts instead of bind mounts. At least with bind mounts, it will always shadow the rootfs bits so you can catch early this case.

Depending on your use case, most of the times, you'd want this "state" setup happening pre init (in initramfs, for example), especially if you want to handle persistent systemd logging or machine id, etc.