Re: [PATCH v3 1/1] expand-on-first-boot: wait for udev to create symlink

["Schaffner, Tobias" <tobias.schaffner@siemens.com>]

On 09.12.22 14:57, Schild, Henning (T CED SES-DE) wrote:
> Am Fri, 9 Dec 2022 12:24:41 +0100
> schrieb "T. Schaffner" <tobias.schaffner@siemens.com>:
> 
>> From: Tobias Schaffner <tobias.schaffner@siemens.com>
>>
>> systemd-growfs depends on a symlink to the partition of the filesystem
>> that should be resized. This symlink is created by udev in
>> /dev/block/.
>>
>> If this symlink is not yet created for example because systemd-udev is
>> not up yet systemd-growfs will fail.
>>
>> We could use Require and After to depend on the systemd-udev service
>> but this could again create a race condition if udev is up but not
>> fast enough after the partx -u.
>>
>> Run systemd-growfs periodically until the symlink appears.
>>
>> Signed-off-by: Tobias Schaffner <tobias.schaffner@siemens.com>
>> ---
>>   .../files/expand-last-partition.sh            | 21
>> ++++++++++++++++++- 1 file changed, 20 insertions(+), 1 deletion(-)
>>
>> diff --git
>> a/meta/recipes-support/expand-on-first-boot/files/expand-last-partition.sh
>> b/meta/recipes-support/expand-on-first-boot/files/expand-last-partition.sh
>> index 57055cc..4d78e4f 100755 ---
>> a/meta/recipes-support/expand-on-first-boot/files/expand-last-partition.sh
>> +++
>> b/meta/recipes-support/expand-on-first-boot/files/expand-last-partition.sh
>> @@ -80,6 +80,25 @@ if [ ! -d "${MOUNT_POINT}" ]; then fi mount
>> "${LAST_PART}" "${MOUNT_POINT}" -/lib/systemd/systemd-growfs
>> "${MOUNT_POINT}" +
>> +EXIT_CODE=1
>> +# If systemd-udevd if not up yet or was not fast enough the symlinks
>> in +# /dev/block/ might be missing. Retry in that case.
>> +# This retry logic is only needed up to systemd-version 252
>> +for run in $(seq 0 50); do
>> +    if GROWFS_OUTPUT=$(/lib/systemd/systemd-growfs "${MOUNT_POINT}"
>> 2>&1); then
>> +        EXIT_CODE=0
>> +        break
>> +    else:
>> +        if ! echo ${GROWFS_OUTPUT} | grep -q "^Failed to open
>> \"/dev/block/.*\": No such file or directory$"; then
>> +            break
>> +        fi
>> +    fi
>> +    sleep 0.1
>> +done
> 
> I wanted to review this and suggest some changes to better deal with
> systemd 252. But i am actually getting the strong feeling that we
> should simply go back to calling the respective filesystem resize tool.
> 
> The whole story with repart did not work out so we thought we could at
> least use some bits (growfs) to support i.e. btrfs. But that thing now
> has this weird mount, and that funny udev problem ... who knows what
> comes next ...
> 
> I propose to go back to resize2fs and add btrfs support with a
> switch-case. Still the whole story told us how repart/growfs can be used
> and we now have the wic patches needed for that and the bail-outs so
> the two methods do not start conflicting. We just did not manage to
> write a generic isar helper package.
> 
> Maybe i will propose to switch one target over to repart/growfs ... but
> honestly, i do not trust that stuff too much at the moment and am not
> sure it has the maturity one might want.
> 
> Henning

I don't think that we can completely blame systemd-growfs for that.

The combination of systemd units coming up in a nondeterministic order
and asynchronous udev events creates an environment that is hard to
test. That why we had problems with this script in several times with
multiple projects.

Reverting the patches that introduced systemd-growfs might anyways be a
sane decision for now as it saves us a lot of complexity.

Best,
Tobias

>> +
>> +echo "${GROWFS_OUTPUT}"
>>   umount "${MOUNT_POINT}"
>>   rmdir "${MOUNT_POINT}"
>> +exit ${EXIT_CODE}
>> +
>