add support for unified image layout #1870

bcressey · 2021-12-16T17:56:36Z

Issue number:
N/A

Description of changes:
Add support for a "unified" image layout, where the OS and data partitions reside on the same disk.

In order to resize the data partition when it's on the same disk as the root filesystem, switch from our homegrown growpart tool to systemd-repart, which knows how to tell the kernel about changes to individual partition entries.

Refactor the unit dependencies for setting up /local to take advantage of the new repart functionality, which lets us an ordinary mount unit for the filesystem.

Testing done:

"unified" and "split" layout AMI registration works
"unified" and "split" layout OVA creation works
/local is correctly resized to fill the disk for both "unified" and "split"
systemd-repart and growpart resize the disk to the same end sector
overlayfs and bind mount directories are created and labeled

Terms of contribution:

By submitting this pull request, I agree that this contribution is dual-licensed under the terms of both the Apache License, version 2.0, and the MIT license.

The repart service is a oneshot, so we can't use a drop-in to replace the `ExecStart` command that runs by default, which attempts to add or grow defined partitions on the same device as the root filesystem. This is OK for "unified" images, where the data partition is at the end of the device, but not OK for "split" images, where it lives on a different device. Signed-off-by: Ben Cressey <bcressey@amazon.com>

Switch from our `growpart` tool to `systemd-repart` to resize the data partition. For a unified root+data image. `growpart` uses the `gptman` crate, which calls the BLKRRPART ioctl to tell the kernel to re-read the partition table. This call fails if the device contains mounted partitions. `systemd-repart` uses the newer BLKPG ioctl, which manipulates the kernel's view of individual partitions. This works even if the root filesystem is present on the same device and already mounted. It also avoids the need to handle the partition symlink going away and coming back, since udev does not get the change event that triggers this. The two tools differ in how much free space is left on the device after the last partition is resized. `growpart` ends the partition one sector before the last 1 MiB boundary, while `systemd-repart` ends it just before the GPT label. Both tools run on every boot. To avoid problems on downgrade after a newer release resizes the data filesystem beyond where the older release will end the partition, we constrain `systemd-repart` to leave the older number of free sectors. Since `/local` can be mounted during the resize operation, we can use a real mount unit for it, which greatly simplifies the dependencies, and allows us to decouple the "prepare" logic from "resize" logic. Signed-off-by: Ben Cressey <bcressey@amazon.com>

For some targets such as bare metal systems, the requirement for a separate block device to hold the data partition is unworkable. Implement a "unified" image layout, which places the data partition after the final OS partition, and is suitable for targets which may only have one disk. The old "split" layout remains the default. Signed-off-by: Ben Cressey <bcressey@amazon.com>

cbgbt · 2021-12-16T22:19:16Z

packages/release/release.spec

-  %{S:1001} %{S:1002} %{S:1003} %{S:1004} %{S:1005} \
-  %{S:1006} %{S:1007} %{S:1008} %{S:1009} %{S:1010} %{S:1011} %{S:1012} \
-  %{S:1015} %{S:1040} %{S:1041} %{S:1060} %{S:1061} %{S:1062} %{S:1080} \
+  %{S:1001} %{S:1002} %{S:1003} %{S:1004} %{S:1005} %{S:1006} %{S:1007} \
+  %{S:1008} %{S:1009} %{S:1010} %{S:1011} %{S:1012} %{S:1013} %{S:1015} \
+  %{S:1040} %{S:1041} %{S:1042} %{S:1060} %{S:1061} %{S:1062} %{S:1080} \


This hurts to review 😰

arnaldo2792 · 2021-12-16T19:22:52Z

Makefile.toml

+      echo "No OVF template or VMDK images, skipping OVA build"
+      exit 0
+   else
+      # Warn the user if a VMDK exists but an OVF template does not.  Assume we do not


Nit: I don't know if this is intended, but there is an extra space here

Suggested change

# Warn the user if a VMDK exists but an OVF template does not. Assume we do not

# Warn the user if a VMDK exists but an OVF template does not. Assume we do not

I tried to preserve the existing "two spaces after a period" convention in this section, even though I belong to the "one space after a period" tribe.

arnaldo2792 · 2021-12-16T21:19:45Z

Makefile.toml

+fi
+
+root_image_size_bytes="$(measure_image "${root_vmdk_path}")"
+root_image_size_gib="$((root_image_size_bytes / 1024 / 1024 / 1024))"


Is it possible for 1024 * 1024 * 1024 and 1024 / 1024 / 1024 to be constants?

arnaldo2792 · 2021-12-17T00:08:53Z

packages/release/release.spec

@@ -101,6 +104,9 @@ install -p -m 0644 %{S:11} %{buildroot}%{_cross_factorydir}%{_cross_sysconfdir}
 install -d %{buildroot}%{_cross_factorydir}%{_cross_sysconfdir}/wicked/ifconfig
 install -p -m 0644 %{S:1000} %{buildroot}%{_cross_factorydir}%{_cross_sysconfdir}/wicked/ifconfig

+install -d %{buildroot}%{_cross_libdir}/repart.d
+install -p -m 0644 %{S:96} %{buildroot}%{_cross_libdir}/repart.d/80-local.conf


Why is this a drop-in, and not part of the unit itself?

systemd-repart looks for its config files in /usr/lib/repart.d, like howsystemd-tmpfiles looks in /usr/lib/tmpfiles.d.

Although the existing AWS and VMware variants use the "split" image layout, custom variants for these platforms might use the "unified" layout instead. Adapt the AMI registration and OVA creation logic to account for the possibility that we only build a single disk image. Signed-off-by: Ben Cressey <bcressey@amazon.com>

Using `state_t` as the label makes the directories read-only for all unprivileged containers, even if they have access via a host mount. Signed-off-by: Ben Cressey <bcressey@amazon.com>

bcressey · 2021-12-17T02:15:39Z

Added bytes_in_gib constant to make the calculations more readable.

arnaldo2792

⛵

webern · 2021-12-17T02:11:21Z

Makefile.toml

+fi
+
+root_image_size_gib="$(($(stat -c %s "${root_image}") / 1024 / 1024 / 1024))"
+if [ "${is_split}" == "yes" ] ; then


It feels like we are at the limit of what should be done with scriptlets inside of a Makefile.toml!

Consider the gauntlet thrown down!

(But yes - we definitely need a different approach.)

zmrow

🥇

bcressey added 3 commits December 16, 2021 14:45

bcressey requested review from cbgbt, zmrow, jpculp and arnaldo2792 December 16, 2021 17:56

bcressey force-pushed the unified-layout branch from b706d30 to 9e66877 Compare December 16, 2021 18:31

jpculp approved these changes Dec 16, 2021

View reviewed changes

cbgbt approved these changes Dec 16, 2021

View reviewed changes

arnaldo2792 reviewed Dec 17, 2021

View reviewed changes

bcressey added 2 commits December 17, 2021 02:14

release: label overlayfs state directories

29de89f

Using `state_t` as the label makes the directories read-only for all unprivileged containers, even if they have access via a host mount. Signed-off-by: Ben Cressey <bcressey@amazon.com>

bcressey force-pushed the unified-layout branch from 9e66877 to 29de89f Compare December 17, 2021 02:15

arnaldo2792 approved these changes Dec 17, 2021

View reviewed changes

webern approved these changes Dec 17, 2021

View reviewed changes

zmrow approved these changes Dec 17, 2021

View reviewed changes

bcressey merged commit 0710193 into bottlerocket-os:develop Dec 17, 2021

bcressey deleted the unified-layout branch December 17, 2021 03:13

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

add support for unified image layout #1870

add support for unified image layout #1870

bcressey commented Dec 16, 2021 •

edited

Loading

cbgbt Dec 16, 2021

arnaldo2792 Dec 16, 2021

bcressey Dec 17, 2021

arnaldo2792 Dec 16, 2021

arnaldo2792 Dec 17, 2021

bcressey Dec 17, 2021

bcressey commented Dec 17, 2021

arnaldo2792 left a comment

webern Dec 17, 2021

bcressey Dec 17, 2021 •

edited

Loading

zmrow left a comment

	# Warn the user if a VMDK exists but an OVF template does not. Assume we do not
	# Warn the user if a VMDK exists but an OVF template does not. Assume we do not

add support for unified image layout #1870

add support for unified image layout #1870

Conversation

bcressey commented Dec 16, 2021 • edited Loading

cbgbt Dec 16, 2021

Choose a reason for hiding this comment

arnaldo2792 Dec 16, 2021

Choose a reason for hiding this comment

bcressey Dec 17, 2021

Choose a reason for hiding this comment

arnaldo2792 Dec 16, 2021

Choose a reason for hiding this comment

arnaldo2792 Dec 17, 2021

Choose a reason for hiding this comment

bcressey Dec 17, 2021

Choose a reason for hiding this comment

bcressey commented Dec 17, 2021

arnaldo2792 left a comment

Choose a reason for hiding this comment

webern Dec 17, 2021

Choose a reason for hiding this comment

bcressey Dec 17, 2021 • edited Loading

Choose a reason for hiding this comment

zmrow left a comment

Choose a reason for hiding this comment

bcressey commented Dec 16, 2021 •

edited

Loading

bcressey Dec 17, 2021 •

edited

Loading