[OE-core] Question about multi rootfs UIDs when using wic

Tue May 8 13:38:44 UTC 2018

Hi Ricardo,

On 07.05.2018 15:22, Ricardo Ribalda Delgado wrote:
> In my case this patch does the trick
Unfortunately your patch results in having the recovery image's UID and 
GID settings to be set to my current user instead of root. So no, your 
patch does not seem to work for me.
I didn't have the time to check in detail what's wrong with your version.

> 
> commit 269334f2400c6ee0c7b4fec7d9bf0f701e50c329 (HEAD -> europa, origin/europa)
> Author: Ricardo Ribalda Delgado <ricardo.ribalda at gmail.com>
> Date:   Mon May 7 13:54:15 2018 +0200
> 
>      wic: Fix partition files UIDs on multi rootfs images
> 
> diff --git a/scripts/lib/wic/partition.py b/scripts/lib/wic/partition.py
> index 84fe85d62b..49921e7494 100644
> --- a/scripts/lib/wic/partition.py
> +++ b/scripts/lib/wic/partition.py
> @@ -206,7 +206,7 @@ class Partition():
>           """
>           p_prefix = os.environ.get("PSEUDO_PREFIX", "%s/usr" % native_sysroot)
>           p_localstatedir = os.environ.get("PSEUDO_LOCALSTATEDIR",
> -                                         "%s/../pseudo" %
> get_bitbake_var("IMAGE_ROOTFS"))
> +                                         "%s/../pseudo" %  rootfs_dir)
>           p_passwd = os.environ.get("PSEUDO_PASSWD", rootfs_dir)
>           p_nosymlinkexp = os.environ.get("PSEUDO_NOSYMLINKEXP", "1")
>           pseudo = "export PSEUDO_PREFIX=%s;" % p_prefix
> diff --git a/scripts/lib/wic/plugins/imager/direct.py
> b/scripts/lib/wic/plugins/imager/direct.py
> index da1c061063..a90a47c926 100644
> --- a/scripts/lib/wic/plugins/imager/direct.py
> +++ b/scripts/lib/wic/plugins/imager/direct.py
> @@ -121,6 +121,10 @@ class DirectPlugin(ImagerPlugin):
>           if self._update_fstab(fstab_lines, self.parts):
>               # copy rootfs dir to workdir to update fstab
>               # as rootfs can be used by other tasks and can't be modified
> +            new_pseudo = os.path.realpath(os.path.join(self.workdir, "pseudo"))
> +            from_dir = os.path.join(os.path.join(image_rootfs, ".."), "pseudo")
> +            from_dir = os.path.realpath(from_dir)
> +            copyhardlinktree(from_dir, new_pseudo)
>               new_rootfs = os.path.realpath(os.path.join(self.workdir,
> "rootfs_copy"))
>               copyhardlinktree(image_rootfs, new_rootfs)
>               fstab_path = os.path.join(new_rootfs, 'etc/fstab')
> 
> 
> 
> Can you try it on your setup? If it also works for you I will work on
> getting it merged.
> 
> 
> Cheers!
> 
> On Wed, May 2, 2018 at 8:25 PM, Ricardo Ribalda Delgado
> <ricardo.ribalda at gmail.com> wrote:
>> cc: ed
>>
>> Is it a very bad idea to revert your patch?
>>
>> Thanks!
>>
>> On Wed, May 2, 2018 at 8:17 PM, Ricardo Ribalda Delgado
>> <ricardo.ribalda at gmail.com> wrote:
>>> Hi
>>>
>>> I do not know how many people is actively using wic. We have started
>>> using it from last week, it is hard to replace old scripts that work
>>> :)
>>>
>>> Try sending it as a proper patch, I can help you if you need it.
>>>
>>> Regards
>>>
>>> On Wed, May 2, 2018 at 6:31 PM, Volker Vogelhuber
>>> <v.vogelhuber at digitalendoscopy.de> wrote:
>>>> Hi,
>>>>
>>>> I never got any response to my message (until now). So AFAIK I solved the
>>>> problem with the following patch:
>>>>
>>>> diff --git a/scripts/lib/wic/partition.py b/scripts/lib/wic/partition.py
>>>> index 84fe85d62b..66ab757aec 100644
>>>> --- a/scripts/lib/wic/partition.py
>>>> +++ b/scripts/lib/wic/partition.py
>>>> @@ -204,15 +204,10 @@ class Partition():
>>>>
>>>>           Currently handles ext2/3/4, btrfs and vfat.
>>>>           """
>>>> -        p_prefix = os.environ.get("PSEUDO_PREFIX", "%s/usr" %
>>>> native_sysroot)
>>>> -        p_localstatedir = os.environ.get("PSEUDO_LOCALSTATEDIR",
>>>> -                                         "%s/../pseudo" %
>>>> get_bitbake_var("IMAGE_ROOTFS"))
>>>> -        p_passwd = os.environ.get("PSEUDO_PASSWD", rootfs_dir)
>>>> +        pseudo = "export PSEUDO_PREFIX=%s/usr;" % native_sysroot
>>>> +        pseudo += "export PSEUDO_LOCALSTATEDIR=%s/../pseudo;" %
>>>> self.rootfs_dir
>>>> +        pseudo += "export PSEUDO_PASSWD=%s;" % rootfs_dir
>>>>           p_nosymlinkexp = os.environ.get("PSEUDO_NOSYMLINKEXP", "1")
>>>> -        pseudo = "export PSEUDO_PREFIX=%s;" % p_prefix
>>>> -        pseudo += "export PSEUDO_LOCALSTATEDIR=%s;" % p_localstatedir
>>>> -        pseudo += "export PSEUDO_PASSWD=%s;" % p_passwd
>>>> -        pseudo += "export PSEUDO_NOSYMLINKEXP=%s;" % p_nosymlinkexp
>>>>           pseudo += "%s " % get_bitbake_var("FAKEROOTCMD")
>>>>
>>>>           rootfs = "%s/rootfs_%s.%s.%s" % (cr_workdir, self.label,
>>>>
>>>>
>>>>
>>>> On 02.05.2018 17:51, Ricardo Ribalda Delgado wrote:
>>>>>
>>>>> Hi
>>>>>
>>>>> I just got hit by this one. It is specially nasty because nfsroot
>>>>> fails to boot if the uids are wrong.
>>>>>
>>>>> What is the status on this?
>>>>>
>>>>> On Mon, Jan 22, 2018 at 11:39 AM, Volker Vogelhuber
>>>>> <v.vogelhuber at digitalendoscopy.de> wrote:
>>>>>>
>>>>>> I finally found out, what's the reason why the included recovery rootfs
>>>>>> has
>>>>>> the wrong UIDs, while the main image has the correct one. The reason
>>>>>> seems
>>>>>> to be the PSEUDO_LOCALSTATEDIR variable that is set to the state dir of
>>>>>> the
>>>>>> main image in both cases.
>>>>>> I debugged all the calls up to the environment setup in partition.py's
>>>>>> prepare_rootfs method where a check for an existing PSEUDO_LOCALSTATEDIR
>>>>>> environment variable is done. That seems to be the problem.
>>>>>> Instead of using the correct value passed to the prepare_rootfs method,
>>>>>> an
>>>>>> existing ENV value is used that points to the state dir of the main image
>>>>>> instead of the recovery one's. I guess the reason is that in bitbake.conf
>>>>>> the PSEUDO_LOCALSTATEDIR is set to the image currently build (which is
>>>>>> the
>>>>>> main image and not the recovery image only referenced by the main image).
>>>>>> So
>>>>>> because that environment variable is already set, the call to mkfs.ext4
>>>>>> for
>>>>>> the recovery image uses the wrong PSEUDO_LOCALSTATEDIR and does not apply
>>>>>> the correct UID.
>>>>>>
>>>>>> Any ideas how to fix that? I tend to just remove the patch introduced by
>>>>>> Ed
>>>>>> Bartosh three years ago (https://patchwork.openembedded.org/patch/90419/)
>>>>>> because I don't need a custom PSEUDO_PREFIX. But maybe not everyone
>>>>>> agrees.
>>>>>>
>>>>>> On 19.01.2018 17:00, Volker Vogelhuber wrote:
>>>>>>>
>>>>>>>
>>>>>>> I'm currently trying to create a multi RootFS WIC image as mentioned in
>>>>>>> directdisk-multi-rootfs.wks. For that I have two image recipes. One that
>>>>>>> is creating only an ext4 image (image-recovery), and the second that is
>>>>>>> also
>>>>>>> creating a WIC image (image-main). I used the IMAGE_FSTYPES variable for
>>>>>>> that. The WKS file for the second image is integrating the recovery
>>>>>>> image by
>>>>>>> specifying --rootfs-dir=image-recovery in it's part description.
>>>>>>>
>>>>>>> # primary / recovery image
>>>>>>> part / --source rootfs --rootfs-dir=image-main --exclude-path mnt/data/
>>>>>>> mnt/data2/ --fstype=ext4 --label primary_rootfs --align 1024 --size 700
>>>>>>> --overhead-factor=1.0
>>>>>>> part /recovery --source rootfs --rootfs-dir=image-recovery --fstype=ext4
>>>>>>> --label recovery_rootfs --align 1024 --size 640 --overhead-factor=1.0
>>>>>>>
>>>>>>> The UIDs of the second rootfs (image-main) are correctly set to 0 within
>>>>>>> the file system when calling mkfs.ext4 during prepare_rootfs_ext. For
>>>>>>> the
>>>>>>> recovery rootfs the UID is always set to my own (host) one which is of
>>>>>>> course not valid for the image where that UID does not exist.
>>>>>>> I tried calling the mkfs.ext4 command myself from a terminal and for
>>>>>>> whatever reason an image created out of the rootfs folder of the second
>>>>>>> image (image-main) recipe is deployed with the correct UID 0, while the
>>>>>>> rootfs folder of the first image (image-recovery) recipe always uses the
>>>>>>> UID
>>>>>>> of the source folder/files.
>>>>>>>
>>>>>>> I search the code of e2fsprogs for the line that sets the UID and added
>>>>>>> a
>>>>>>> printf in set_inode_extra. There I can see clearly that the source UID
>>>>>>> for
>>>>>>> the file is 0 for the rootfs of the image-main rootfs folder while it is
>>>>>>> 10000 (my own UID) for the image-recovery. I wonder how the UID of the
>>>>>>> image-main rootfs folder can be zero when I don't call any command with
>>>>>>> root
>>>>>>> permissions. I searched for a preparation step where the UIDs are
>>>>>>> managed in
>>>>>>> the scripts folder of Yocto, but didn't found any hint for the whole
>>>>>>> behavior. So while it is good that the rootfs partition of the main
>>>>>>> rootfs
>>>>>>> has the UID set correctly to zero, I can't understand why it happens. On
>>>>>>> the
>>>>>>> other side I can understand why the UID of the recovery rootfs is set to
>>>>>>> my
>>>>>>> own one, but it stops me from booting that rootfs because the UIDs of
>>>>>>> the
>>>>>>> files and folders are set to a user that does not exist on the target
>>>>>>> system.
>>>>>>>
>>>>>>> Can someone please explain to me, how that UID handling is meant to be
>>>>>>> done?
>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>>
>>>>
>>>
>>>
>>>
>>> --
>>> Ricardo Ribalda
>>
>>
>>
>> --
>> Ricardo Ribalda
> 
> 
> 