ubifs: regression since "ubifs: xattr: Don't operate on deleted inodes"

ubifs: regression since "ubifs: xattr: Don't operate on deleted inodes"

[Koen Vandeputte]

Hi all,

I'm currently in the process of updating the kernel version within 
OpenWrt. (4.14.68 to 4.14.69)

Testing shows some issues on devices using specifically UBIFS.
Altering a perfect valid writable file shows weird errors:


[ Node 2 | node-2 ] ls -l /root/custom/scripts/banner.sh
-rwxr-xr-x    1 root     root           283 Sep 11 09:52 
/root/custom/scripts/banner.sh

[ Node 2 | node-2 ] cat /root/custom/scripts/banner.sh
#!/bin/sh

if [ ! -f /root/.banner_ok ]
then
     RELEASE=$(cat /root/build_date)
     VERSION=$(cat /root/version)

     echo "Generating banner: $VERSION $RELEASE"
     sed s/VERSION/$VERSION/g /root/custom/banner > /etc/banner
     sed -i s/RELEASE/$RELEASE/g /etc/banner

     touch /root/.banner_ok
fi

[ Node 2 | node-2 ] echo "test" > /root/custom/scripts/banner.sh
-ash: can't create /root/custom/scripts/banner.sh: nonexistent directory



I'm also noticing other apps fail because /etc doesn't exists yet after 
UBIFS boot loading.
these 2 issues were not seen on 4.14.68.

The bootlog doesn't show any error:
https://pastebin.com/raw/dJx47uBp


I'm only seeing these issues on UBIFS enabled volumes.
Reverting ("ubifs: xattr: Don't operate on deleted inodes") fixes these 
weird issues.


Thanks,

Koen

Re: ubifs: regression since "ubifs: xattr: Don't operate on deleted inodes"

[gregkh]

adding stable@ for stable kernel issues...

On Thu, Sep 13, 2018 at 09:58:35AM +0200, Koen Vandeputte wrote:
> Hi all,
> 
> I'm currently in the process of updating the kernel version within OpenWrt.
> (4.14.68 to 4.14.69)
> 
> Testing shows some issues on devices using specifically UBIFS.
> Altering a perfect valid writable file shows weird errors:
> 
> 
> [ Node 2 | node-2 ] ls -l /root/custom/scripts/banner.sh
> -rwxr-xr-x��� 1 root���� root���������� 283 Sep 11 09:52
> /root/custom/scripts/banner.sh
> 
> [ Node 2 | node-2 ] cat /root/custom/scripts/banner.sh
> #!/bin/sh
> 
> if [ ! -f /root/.banner_ok ]
> then
> ��� RELEASE=$(cat /root/build_date)
> ��� VERSION=$(cat /root/version)
> 
> ��� echo "Generating banner: $VERSION $RELEASE"
> ��� sed s/VERSION/$VERSION/g /root/custom/banner > /etc/banner
> ��� sed -i s/RELEASE/$RELEASE/g /etc/banner
> 
> ��� touch /root/.banner_ok
> fi
> 
> [ Node 2 | node-2 ] echo "test" > /root/custom/scripts/banner.sh
> -ash: can't create /root/custom/scripts/banner.sh: nonexistent directory
> 
> 
> 
> I'm also noticing other apps fail because /etc doesn't exists yet after
> UBIFS boot loading.
> these 2 issues were not seen on 4.14.68.
> 
> The bootlog doesn't show any error:
> https://pastebin.com/raw/dJx47uBp
> 
> 
> I'm only seeing these issues on UBIFS enabled volumes.
> Reverting ("ubifs: xattr: Don't operate on deleted inodes") fixes these
> weird issues.

Do you also have the same problem on Linus's latest tree, or 4.18, with
this change in the tree?

thanks,

greg k-h

Re: ubifs: regression since "ubifs: xattr: Don't operate on deleted inodes"

[gregkh]

adding stable@ for stable kernel issues...

On Thu, Sep 13, 2018 at 09:58:35AM +0200, Koen Vandeputte wrote:
> Hi all,
> 
> I'm currently in the process of updating the kernel version within OpenWrt.
> (4.14.68 to 4.14.69)
> 
> Testing shows some issues on devices using specifically UBIFS.
> Altering a perfect valid writable file shows weird errors:
> 
> 
> [ Node 2 | node-2 ] ls -l /root/custom/scripts/banner.sh
> -rwxr-xr-x    1 root     root           283 Sep 11 09:52
> /root/custom/scripts/banner.sh
> 
> [ Node 2 | node-2 ] cat /root/custom/scripts/banner.sh
> #!/bin/sh
> 
> if [ ! -f /root/.banner_ok ]
> then
>     RELEASE=$(cat /root/build_date)
>     VERSION=$(cat /root/version)
> 
>     echo "Generating banner: $VERSION $RELEASE"
>     sed s/VERSION/$VERSION/g /root/custom/banner > /etc/banner
>     sed -i s/RELEASE/$RELEASE/g /etc/banner
> 
>     touch /root/.banner_ok
> fi
> 
> [ Node 2 | node-2 ] echo "test" > /root/custom/scripts/banner.sh
> -ash: can't create /root/custom/scripts/banner.sh: nonexistent directory
> 
> 
> 
> I'm also noticing other apps fail because /etc doesn't exists yet after
> UBIFS boot loading.
> these 2 issues were not seen on 4.14.68.
> 
> The bootlog doesn't show any error:
> https://pastebin.com/raw/dJx47uBp
> 
> 
> I'm only seeing these issues on UBIFS enabled volumes.
> Reverting ("ubifs: xattr: Don't operate on deleted inodes") fixes these
> weird issues.

Do you also have the same problem on Linus's latest tree, or 4.18, with
this change in the tree?

thanks,

greg k-h

Re: ubifs: regression since "ubifs: xattr: Don't operate on deleted inodes"

[Richard Weinberger]

Koen,

On Thu, Sep 13, 2018 at 12:09 PM gregkh@linuxfoundation.org
<gregkh@linuxfoundation.org> wrote:
>
> adding stable@ for stable kernel issues...
>
> On Thu, Sep 13, 2018 at 09:58:35AM +0200, Koen Vandeputte wrote:
> > Hi all,
> >
> > I'm currently in the process of updating the kernel version within OpenWrt.
> > (4.14.68 to 4.14.69)
> >
> > Testing shows some issues on devices using specifically UBIFS.
> > Altering a perfect valid writable file shows weird errors:
> >
> >
> > [ Node 2 | node-2 ] ls -l /root/custom/scripts/banner.sh
> > -rwxr-xr-x    1 root     root           283 Sep 11 09:52
> > /root/custom/scripts/banner.sh
> >
> > [ Node 2 | node-2 ] cat /root/custom/scripts/banner.sh
> > #!/bin/sh
> >
> > if [ ! -f /root/.banner_ok ]
> > then
> >     RELEASE=$(cat /root/build_date)
> >     VERSION=$(cat /root/version)
> >
> >     echo "Generating banner: $VERSION $RELEASE"
> >     sed s/VERSION/$VERSION/g /root/custom/banner > /etc/banner
> >     sed -i s/RELEASE/$RELEASE/g /etc/banner
> >
> >     touch /root/.banner_ok
> > fi
> >
> > [ Node 2 | node-2 ] echo "test" > /root/custom/scripts/banner.sh
> > -ash: can't create /root/custom/scripts/banner.sh: nonexistent directory
> >
> >
> >
> > I'm also noticing other apps fail because /etc doesn't exists yet after
> > UBIFS boot loading.
> > these 2 issues were not seen on 4.14.68.
> >
> > The bootlog doesn't show any error:
> > https://pastebin.com/raw/dJx47uBp
> >
> >
> > I'm only seeing these issues on UBIFS enabled volumes.
> > Reverting ("ubifs: xattr: Don't operate on deleted inodes") fixes these
> > weird issues.

Please see my answer to the other thread.
Always keep the patch author on CC and don't start multiple threads for the
same issue on two mailing lists.
Yes, I didn't answer for three days, I had no internet connection...

-- 
Thanks,
//richard

Re: ubifs: regression since "ubifs: xattr: Don't operate on deleted inodes"

[Koen Vandeputte]

On 15-09-18 09:15, Richard Weinberger wrote:
> Koen,
>
> On Thu, Sep 13, 2018 at 12:09 PM gregkh@linuxfoundation.org
> <gregkh@linuxfoundation.org> wrote:
>> adding stable@ for stable kernel issues...
>>
>> On Thu, Sep 13, 2018 at 09:58:35AM +0200, Koen Vandeputte wrote:
>>> Hi all,
>>>
>>> I'm currently in the process of updating the kernel version within OpenWrt.
>>> (4.14.68 to 4.14.69)
>>>
>>> Testing shows some issues on devices using specifically UBIFS.
>>> Altering a perfect valid writable file shows weird errors:
>>>
>>>
>>> [ Node 2 | node-2 ] ls -l /root/custom/scripts/banner.sh
>>> -rwxr-xr-x    1 root     root           283 Sep 11 09:52
>>> /root/custom/scripts/banner.sh
>>>
>>> [ Node 2 | node-2 ] cat /root/custom/scripts/banner.sh
>>> #!/bin/sh
>>>
>>> if [ ! -f /root/.banner_ok ]
>>> then
>>>      RELEASE=$(cat /root/build_date)
>>>      VERSION=$(cat /root/version)
>>>
>>>      echo "Generating banner: $VERSION $RELEASE"
>>>      sed s/VERSION/$VERSION/g /root/custom/banner > /etc/banner
>>>      sed -i s/RELEASE/$RELEASE/g /etc/banner
>>>
>>>      touch /root/.banner_ok
>>> fi
>>>
>>> [ Node 2 | node-2 ] echo "test" > /root/custom/scripts/banner.sh
>>> -ash: can't create /root/custom/scripts/banner.sh: nonexistent directory
>>>
>>>
>>>
>>> I'm also noticing other apps fail because /etc doesn't exists yet after
>>> UBIFS boot loading.
>>> these 2 issues were not seen on 4.14.68.
>>>
>>> The bootlog doesn't show any error:
>>> https://pastebin.com/raw/dJx47uBp
>>>
>>>
>>> I'm only seeing these issues on UBIFS enabled volumes.
>>> Reverting ("ubifs: xattr: Don't operate on deleted inodes") fixes these
>>> weird issues.
> Please see my answer to the other thread.
> Always keep the patch author on CC and don't start multiple threads for the
> same issue on two mailing lists.
> Yes, I didn't answer for three days, I had no internet connection...
>
Richard,

I indeed posted here too as we passed the 3 days marker, and I noticed 
4.14.70 RC2 got staged.
Also, a lot of developers tend to ignore questions unless the initial 
question got posted to the official mailinglist.

Let's continue on the OpenWrt list as you propose, were the initial 
question was raised.

Apologies for the (double) noise.


Thank you,

Koen

Re: ubifs: regression since "ubifs: xattr: Don't operate on deleted inodes"

[Greg KH]

On Sun, Sep 16, 2018 at 09:52:48PM +0200, Koen Vandeputte wrote:
> 
> 
> On 15-09-18 09:15, Richard Weinberger wrote:
> > Koen,
> > 
> > On Thu, Sep 13, 2018 at 12:09 PM gregkh@linuxfoundation.org
> > <gregkh@linuxfoundation.org> wrote:
> > > adding stable@ for stable kernel issues...
> > > 
> > > On Thu, Sep 13, 2018 at 09:58:35AM +0200, Koen Vandeputte wrote:
> > > > Hi all,
> > > > 
> > > > I'm currently in the process of updating the kernel version within OpenWrt.
> > > > (4.14.68 to 4.14.69)
> > > > 
> > > > Testing shows some issues on devices using specifically UBIFS.
> > > > Altering a perfect valid writable file shows weird errors:
> > > > 
> > > > 
> > > > [ Node 2 | node-2 ] ls -l /root/custom/scripts/banner.sh
> > > > -rwxr-xr-x    1 root     root           283 Sep 11 09:52
> > > > /root/custom/scripts/banner.sh
> > > > 
> > > > [ Node 2 | node-2 ] cat /root/custom/scripts/banner.sh
> > > > #!/bin/sh
> > > > 
> > > > if [ ! -f /root/.banner_ok ]
> > > > then
> > > >      RELEASE=$(cat /root/build_date)
> > > >      VERSION=$(cat /root/version)
> > > > 
> > > >      echo "Generating banner: $VERSION $RELEASE"
> > > >      sed s/VERSION/$VERSION/g /root/custom/banner > /etc/banner
> > > >      sed -i s/RELEASE/$RELEASE/g /etc/banner
> > > > 
> > > >      touch /root/.banner_ok
> > > > fi
> > > > 
> > > > [ Node 2 | node-2 ] echo "test" > /root/custom/scripts/banner.sh
> > > > -ash: can't create /root/custom/scripts/banner.sh: nonexistent directory
> > > > 
> > > > 
> > > > 
> > > > I'm also noticing other apps fail because /etc doesn't exists yet after
> > > > UBIFS boot loading.
> > > > these 2 issues were not seen on 4.14.68.
> > > > 
> > > > The bootlog doesn't show any error:
> > > > https://pastebin.com/raw/dJx47uBp
> > > > 
> > > > 
> > > > I'm only seeing these issues on UBIFS enabled volumes.
> > > > Reverting ("ubifs: xattr: Don't operate on deleted inodes") fixes these
> > > > weird issues.
> > Please see my answer to the other thread.
> > Always keep the patch author on CC and don't start multiple threads for the
> > same issue on two mailing lists.
> > Yes, I didn't answer for three days, I had no internet connection...
> > 
> Richard,
> 
> I indeed posted here too as we passed the 3 days marker, and I noticed
> 4.14.70 RC2 got staged.
> Also, a lot of developers tend to ignore questions unless the initial
> question got posted to the official mailinglist.
> 
> Let's continue on the OpenWrt list as you propose, were the initial question
> was raised.
> 
> Apologies for the (double) noise.

I will assume that if the resolution is that something needs to be done
to the stable releases and a patch added/reverted, someone will
eventually tell me?

thanks,

greg k-h

Re: ubifs: regression since "ubifs: xattr: Don't operate on deleted inodes"

[Richard Weinberger]

Am Montag, 17. September 2018, 11:38:04 CEST schrieb Greg KH:
> On Sun, Sep 16, 2018 at 09:52:48PM +0200, Koen Vandeputte wrote:
> > 
> > 
> > On 15-09-18 09:15, Richard Weinberger wrote:
> > > Koen,
> > > 
> > > On Thu, Sep 13, 2018 at 12:09 PM gregkh@linuxfoundation.org
> > > <gregkh@linuxfoundation.org> wrote:
> > > > adding stable@ for stable kernel issues...
> > > > 
> > > > On Thu, Sep 13, 2018 at 09:58:35AM +0200, Koen Vandeputte wrote:
> > > > > Hi all,
> > > > > 
> > > > > I'm currently in the process of updating the kernel version within OpenWrt.
> > > > > (4.14.68 to 4.14.69)
> > > > > 
> > > > > Testing shows some issues on devices using specifically UBIFS.
> > > > > Altering a perfect valid writable file shows weird errors:
> > > > > 
> > > > > 
> > > > > [ Node 2 | node-2 ] ls -l /root/custom/scripts/banner.sh
> > > > > -rwxr-xr-x    1 root     root           283 Sep 11 09:52
> > > > > /root/custom/scripts/banner.sh
> > > > > 
> > > > > [ Node 2 | node-2 ] cat /root/custom/scripts/banner.sh
> > > > > #!/bin/sh
> > > > > 
> > > > > if [ ! -f /root/.banner_ok ]
> > > > > then
> > > > >      RELEASE=$(cat /root/build_date)
> > > > >      VERSION=$(cat /root/version)
> > > > > 
> > > > >      echo "Generating banner: $VERSION $RELEASE"
> > > > >      sed s/VERSION/$VERSION/g /root/custom/banner > /etc/banner
> > > > >      sed -i s/RELEASE/$RELEASE/g /etc/banner
> > > > > 
> > > > >      touch /root/.banner_ok
> > > > > fi
> > > > > 
> > > > > [ Node 2 | node-2 ] echo "test" > /root/custom/scripts/banner.sh
> > > > > -ash: can't create /root/custom/scripts/banner.sh: nonexistent directory
> > > > > 
> > > > > 
> > > > > 
> > > > > I'm also noticing other apps fail because /etc doesn't exists yet after
> > > > > UBIFS boot loading.
> > > > > these 2 issues were not seen on 4.14.68.
> > > > > 
> > > > > The bootlog doesn't show any error:
> > > > > https://pastebin.com/raw/dJx47uBp
> > > > > 
> > > > > 
> > > > > I'm only seeing these issues on UBIFS enabled volumes.
> > > > > Reverting ("ubifs: xattr: Don't operate on deleted inodes") fixes these
> > > > > weird issues.
> > > Please see my answer to the other thread.
> > > Always keep the patch author on CC and don't start multiple threads for the
> > > same issue on two mailing lists.
> > > Yes, I didn't answer for three days, I had no internet connection...
> > > 
> > Richard,
> > 
> > I indeed posted here too as we passed the 3 days marker, and I noticed
> > 4.14.70 RC2 got staged.
> > Also, a lot of developers tend to ignore questions unless the initial
> > question got posted to the official mailinglist.
> > 
> > Let's continue on the OpenWrt list as you propose, were the initial question
> > was raised.
> > 
> > Apologies for the (double) noise.
> 
> I will assume that if the resolution is that something needs to be done
> to the stable releases and a patch added/reverted, someone will
> eventually tell me?

The revert was sent yesterday to the mailinglist.
I'll send Linus^Wyou a pull request soon and from here
it can go into stable.

Thanks,
//richard