Bind mounts record fstype of none

Bind mounts record fstype of none

[Phillip Susi]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

I came across an old bug filed in debian noting that mount records the
fstype for bind mounts as 'none' and this causes df to suppress
showing it.  This seems like an error to me.  The type should be what
the actual filesystem type is, and if df wants to suppress showing
bind mounts, it should do so based on the bind flag.

Thoughts?

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.17 (MingW32)
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/

iQEcBAEBAgAGBQJScSKuAAoJEJrBOlT6nu75tJwIAKk1IQyvsQIpPUj9ugSrHVDp
CdsGB5D8VNznUJ+a+AReNp57AGi/UprdCXsWl33yOuz3zjQKNlYaGEMCsN/pSJnq
nn+8pJ5GSggPekUh2FO82sLGaEh4y+rMPlN6rwCyOERkP1V+MaYI9Z2KzCrpnpfI
LaDm17q1IzpDtJry55Vg+yd0vuJZ3qpfvLUmD4rj/h7N5gt/lXWeSzcUyIIQvHaA
IR7Nch1YaA612Tqz0Zf+dews1ERwERNw2+s8BBlsvTW4FzbR5gfMdn3Xzv8WFnTY
yr54S7Krrxn7jbBFCF44GQddxptIJ7jjI2GqYIq1D8L9DoTwSIGEk552/y2negc=
=k4dY
-----END PGP SIGNATURE-----

Re: Bind mounts record fstype of none

[Bernhard Voelker]

> On October 30, 2013 at 4:15 PM Phillip Susi <psusi@ubuntu.com> wrote:
> I came across an old bug filed in debian noting that mount records the
> fstype for bind mounts as 'none' and this causes df to suppress
> showing it.  This seems like an error to me.  The type should be what
> the actual filesystem type is, and if df wants to suppress showing
> bind mounts, it should do so based on the bind flag.
>
> Thoughts?

yes, df(1) is from coreutils. ;-)

However, latest df(1) will skip such bind mounts per default - but
now because of some 'bind' information (which is not available
anyway), but just because it skips "duplicates" based on the device
numbers. For the same device number, it will output that with the
shortest mount point name.
You have to use -a to see all.

Have a nice day,
Berny

Re: Bind mounts record fstype of none

[Phillip Susi]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 10/30/2013 11:39 AM, Bernhard Voelker wrote:
> yes, df(1) is from coreutils. ;-)
> 
> However, latest df(1) will skip such bind mounts per default - but 
> now because of some 'bind' information (which is not available 
> anyway), but just because it skips "duplicates" based on the
> device numbers. For the same device number, it will output that
> with the shortest mount point name. You have to use -a to see all.

So is mount wrong for recording the filesystem type as 'none' or not?


-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.17 (MingW32)
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/

iQEcBAEBAgAGBQJScSzwAAoJEJrBOlT6nu7587EH/j35wWxXuNqljBAn9B4fimyd
UcTuchJuPoeWg6fmB6XGJB+Zuv/sari8ur/ihVuV+SiC9+MvBYrkoy+xlTW+vmMc
B4usMZRvZQK9QW74nvqibDgp2S1mG2iEbpO58oJK2E6ILQoFLNPY2Awuz2VR7LY9
sLxseN4eSzRdPFRiFfzFIaI6s6PO4fvcJceNukO2LIvbhKEhufduTPEJrPKMDIYK
OAgmsDm5PUDCITDyyO9YYlZdVm/RxJmRvssdN+WycfsXzYgad+/BAdc0WENtw81R
JHg7MciHqNfNBJ5bmlqwnnynud37gWe7dYRLNoMgXDcAMsIwjBL1NsrEpKcmqmk=
=05sY
-----END PGP SIGNATURE-----

Re: Bind mounts record fstype of none

[Bernhard Voelker]

> On October 30, 2013 at 4:59 PM Phillip Susi <psusi@ubuntu.com> wrote:
> > On 10/30/2013 11:39 AM, Bernhard Voelker wrote:
> > yes, df(1) is from coreutils. ;-)
> >
> > However, latest df(1) will skip such bind mounts per default - but
> > now because of some 'bind' information (which is not available

typo: s/now/not/, of course.

> > anyway), but just because it skips "duplicates" based on the
> > device numbers. For the same device number, it will output that
> > with the shortest mount point name. You have to use -a to see all.
>
> So is mount wrong for recording the filesystem type as 'none' or not?

Where do you see 'none'?

$ df /home
Filesystem     1K-blocks    Used Available Use% Mounted on
/dev/sdb        10321208 1233056   8563864  13% /home

$ mkdir /tmp/home

$ mount -o bind /home /tmp/home

$ df -a | grep home
/dev/sdb        10321208 1233056   8563864  13% /home
/dev/sdb        10321208 1233056   8563864  13% /tmp/home

$ mount | grep home
/dev/sdb on /home type ext4 (rw,relatime,data=ordered)
/dev/sdb on /tmp/home type ext4 (rw,relatime,data=ordered)

$ grep home /proc/mounts /proc/self/mountinfo /proc/self/mounts
/proc/mounts:/dev/sdb /home ext4 rw,relatime,data=ordered 0 0
/proc/mounts:/dev/sdb /tmp/home ext4 rw,relatime,data=ordered 0 0
/proc/self/mountinfo:40 21 8:16 / /home rw,relatime shared:24 - ext4 /dev/sdb
rw,data=ordered
/proc/self/mountinfo:109 21 8:16 / /tmp/home rw,relatime shared:24 - ext4
/dev/sdb rw,data=ordered
/proc/self/mounts:/dev/sdb /home ext4 rw,relatime,data=ordered 0 0
/proc/self/mounts:/dev/sdb /tmp/home ext4 rw,relatime,data=ordered 0 0

[That was on kernel 3.7.10 of openSUSE-12.3,
mount/df built from {util-linux,coreutils}.git]

Have a nice day,
Berny

Re: Bind mounts record fstype of none

[Helmut Hullen]

Hallo, Bernhard,

Du meintest am 30.10.13:

>> So is mount wrong for recording the filesystem type as 'none' or
>> not?

> Where do you see 'none'?

        df -a -T

Maybe that problem is related to the option "-type"; "bind" is no  
allowed "type".

Viele Gruesse!
Helmut

Re: Bind mounts record fstype of none

[Dave Reisner]

On Wed, Oct 30, 2013 at 05:31:00PM +0100, Helmut Hullen wrote:
> Hallo, Bernhard,
> 
> Du meintest am 30.10.13:
> 
> >> So is mount wrong for recording the filesystem type as 'none' or
> >> not?
> 
> > Where do you see 'none'?
> 
>         df -a -T
> 
> Maybe that problem is related to the option "-type"; "bind" is no  
> allowed "type".

The fstype of "none" only shows up on older systems where mtab is a
regular file and not a symlink to /proc/self/mounts.

Cheers,
Dave

Re: Bind mounts record fstype of none

[Helmut Hullen]

Hallo, Dave,

Du meintest am 30.10.13:

[...]

>>> Where do you see 'none'?

>>         df -a -T
>>
>> Maybe that problem is related to the option "-type"; "bind" is no
>> allowed "type".

> The fstype of "none" only shows up on older systems where mtab is a
> regular file and not a symlink to /proc/self/mounts.

Yes - that's the difference on my system ("mtab" is a regular file).

Viele Gruesse!
Helmut

Re: Bind mounts record fstype of none

[Bernhard Voelker]

> On October 30, 2013 at 5:31 PM Helmut Hullen <Hullen@t-online.de> wrote:
>> Du meintest am 30.10.13:
>
> >> So is mount wrong for recording the filesystem type as 'none' or
> >> not?
>
> > Where do you see 'none'?
>
>         df -a -T
>
> Maybe that problem is related to the option "-type"; "bind" is no 
> allowed "type".

I don't get that.
Is it possible you hit the "mtab is evil" (TM Kzak) issue?
  http://karelzak.blogspot.de/2011_04_01_archive.html

Have a nice day,
Berny

Re: Bind mounts record fstype of none

[Phillip Susi]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 10/30/2013 12:23 PM, Bernhard Voelker wrote:
> Where do you see 'none'?
> 
> $ df /home Filesystem     1K-blocks    Used Available Use% Mounted
> on /dev/sdb        10321208 1233056   8563864  13% /home
> 
> $ mkdir /tmp/home
> 
> $ mount -o bind /home /tmp/home
> 
> $ df -a | grep home /dev/sdb        10321208 1233056   8563864  13%
> /home /dev/sdb        10321208 1233056   8563864  13% /tmp/home
> 
> $ mount | grep home /dev/sdb on /home type ext4
> (rw,relatime,data=ordered) /dev/sdb on /tmp/home type ext4
> (rw,relatime,data=ordered)

I see none in the output of mount and df.  Just rebuilt from git head
to make sure and still says none.


-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.17 (MingW32)
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/

iQEcBAEBAgAGBQJScVmZAAoJEJrBOlT6nu75lywH/i+iIcuIC4M2aQ7XD+uPnFmx
e4uW3qs94QdJi0ita9OjCgPKS7jWzCgk5+PiUbj8sWtRvol8WTdbl7yh7tcfjuFz
tu2TKg84Q9O+7xHVN6CbwyCmq8VCJQTLYB5fs/+OKByDYxzAiANJnh0P7h8Abu++
4HEAssrfSNFp3mP2urdZl0XLbEXt5lBdN9q1oNVYhyg6Ts/pTtcipg/+keM49uH0
+2QiJrCdKy9zhcIendFx/FCqvoiKXS/Lm4Hetn9qZwU5s2Oqcdf5BVTK3G+dQ7kv
pkY1K2dU18lN4osxjNOVcq/DYFGdSY8jHg3+jbLdn3xmxgKENoOKkvpd9Lea9ao=
=ta/U
-----END PGP SIGNATURE-----

Re: Bind mounts record fstype of none

[Helmut Hullen]

Hallo, Phillip,

Du meintest am 30.10.13:

> I came across an old bug filed in debian noting that mount records
> the fstype for bind mounts as 'none' and this causes df to suppress
> showing it.  This seems like an error to me.  The type should be what
> the actual filesystem type is, and if df wants to suppress showing
> bind mounts, it should do so based on the bind flag.

For seeing "bind" I'd prefer

        mount

or

        mount | grep -w bind

Viele Gruesse!
Helmut

Re: Bind mounts record fstype of none

[Karel Zak]

On Wed, Oct 30, 2013 at 11:15:59AM -0400, Phillip Susi wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> I came across an old bug filed in debian noting that mount records the
> fstype for bind mounts as 'none' and this causes df to suppress

 It was explicitly requested by Al Viro to use "none" for some 
 VFS operations years ago (c4966ccb16868fa748009a826340fac9d1b1ce39,
 0fae284a7a13d4d2dba7a908e0662a6d9c46f877).

 I think it's correct, because FS type does not make any sense there
 and it's waste of time to try to determine FS type for MS_BINS or
 MS_PROPAGATION operations.

> showing it.  This seems like an error to me.  The type should be what
> the actual filesystem type is, and if df wants to suppress showing
> bind mounts, it should do so based on the bind flag.

 Seems like you're victim of the mtab file. My Fedora 19:

    # mount --bind /home /mnt/test
    # mount | grep /mnt/test
    /dev/sda5 on /mnt/test type ext4 (rw,relatime,data=ordered)


   Karel
 
-- 
 Karel Zak  <kzak@redhat.com>
 http://karelzak.blogspot.com

Re: Bind mounts record fstype of none

[Phillip Susi]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 10/31/2013 6:57 AM, Karel Zak wrote:
> On Wed, Oct 30, 2013 at 11:15:59AM -0400, Phillip Susi wrote:
>> -----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
>> 
>> I came across an old bug filed in debian noting that mount
>> records the fstype for bind mounts as 'none' and this causes df
>> to suppress
> 
> It was explicitly requested by Al Viro to use "none" for some VFS
> operations years ago (c4966ccb16868fa748009a826340fac9d1b1ce39, 
> 0fae284a7a13d4d2dba7a908e0662a6d9c46f877).
> 
> I think it's correct, because FS type does not make any sense
> there and it's waste of time to try to determine FS type for
> MS_BINS or MS_PROPAGATION operations.

How does it not make sense?

>> showing it.  This seems like an error to me.  The type should be
>> what the actual filesystem type is, and if df wants to suppress
>> showing bind mounts, it should do so based on the bind flag.
> 
> Seems like you're victim of the mtab file. My Fedora 19:
> 
> # mount --bind /home /mnt/test # mount | grep /mnt/test /dev/sda5
> on /mnt/test type ext4 (rw,relatime,data=ordered)

Since the kernel reports the correct type, then I'd say mount should
behave the same when using mtab, and record it right instead of
forcing it to none.


-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.17 (MingW32)
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/

iQEcBAEBAgAGBQJScmawAAoJEJrBOlT6nu75B6cIAMHuRn7YaD6ZO2oS0e68RlyO
FGBNVkiUezG/cw8GBxihiSejlZP99cR0VMeD7ImLjLpuKbJ/tOdqHUGIFneDZu6I
hBz7luqVTnyOiimEaX8AXMWy4LOZzXiFd/etWTG9QpePsj+e+NRGj826cXmEpYoX
uIuWze+IRZc7NLzb+TmTtJcv/2a6yFZHxIdTt9SYzsPWWJ23m+t9GWtTbMt4CnXZ
04BxJpryA04KoZhJBeliHu87xQ5cQBOfZkBavx2rpoGoiQRez9XVorif3hP7SBwT
KD370LQr0GWbTQo1YEuIHXGaAcqEW/yQcPWKmekJXkg+1dhgE5c8Rvly37jeKRs=
=+osw
-----END PGP SIGNATURE-----

Re: Bind mounts record fstype of none

[Karel Zak]

On Thu, Oct 31, 2013 at 10:18:24AM -0400, Phillip Susi wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> On 10/31/2013 6:57 AM, Karel Zak wrote:
> > On Wed, Oct 30, 2013 at 11:15:59AM -0400, Phillip Susi wrote:
> >> -----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
> >> 
> >> I came across an old bug filed in debian noting that mount
> >> records the fstype for bind mounts as 'none' and this causes df
> >> to suppress
> > 
> > It was explicitly requested by Al Viro to use "none" for some VFS
> > operations years ago (c4966ccb16868fa748009a826340fac9d1b1ce39, 
> > 0fae284a7a13d4d2dba7a908e0662a6d9c46f877).
> > 
> > I think it's correct, because FS type does not make any sense
> > there and it's waste of time to try to determine FS type for
> > MS_BINS or MS_PROPAGATION operations.
> 
> How does it not make sense?

 It is not FS operation, FS type is completely irrelevant here, kernel
 does not use this argument of the mount(2) syscall.

> >> showing it.  This seems like an error to me.  The type should be
> >> what the actual filesystem type is, and if df wants to suppress
> >> showing bind mounts, it should do so based on the bind flag.
> > 
> > Seems like you're victim of the mtab file. My Fedora 19:
> > 
> > # mount --bind /home /mnt/test # mount | grep /mnt/test /dev/sda5
> > on /mnt/test type ext4 (rw,relatime,data=ordered)
> 
> Since the kernel reports the correct type, then I'd say mount should
> behave the same when using mtab, and record it right instead of
> forcing it to none.

 There is *many disadvantages* when you use mtab, not sure if it make
 sense to fix the issues especially when mtab is currently unused by 
 (almost) all distributions.

 Maybe we can keep the fstype and not replace it with "none" for mtab,
 but is it so important when we use "none" since 2009 and mtab is
 almost dead?

    Karel

-- 
 Karel Zak  <kzak@redhat.com>
 http://karelzak.blogspot.com

Re: Bind mounts record fstype of none

[Phillip Susi]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 10/31/2013 11:10 AM, Karel Zak wrote:
> There is *many disadvantages* when you use mtab, not sure if it
> make sense to fix the issues especially when mtab is currently
> unused by (almost) all distributions.
> 
> Maybe we can keep the fstype and not replace it with "none" for
> mtab, but is it so important when we use "none" since 2009 and mtab
> is almost dead?

I know there were a number of issues with this transition, have they
all been worked out at this point?  Is all you need to do to make the
switch is replace /etc/mtab with a symlink to /proc/mounts?  Maybe
I'll try doing that.


-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.17 (MingW32)
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/

iQEcBAEBAgAGBQJScnc5AAoJEJrBOlT6nu75HBEH/i7nKpCaq7TkgArB0IgnmSqM
fHkp1jMIltOxkfVDj8ma6hErYUJ+2g4abgKF8nrP39NWiFldrCBqTjOAr2kOl9wn
jO40x9ur6SJQpVbSrmtHOam05J+P09N2SRXb/LAVIsUzTyIvw3kBFaGpl0ElJz8Q
Jf3yTWr5gPyR14oAdOKdzkW70rktob2lRGM4DetChQd8H0LtfiaODtaBSbDyJMQ7
P8ErR42Isosf2+RBW+c/EC3biE82QsMcJjMP4eJ+A3uG8ZMlOTtDpthvMjQPY6+0
xCBz9Qd9yjvfgGTxnuV+AyX9Xa+mQNdS4/MtvyCwTF3TTKekhkYMK2CMSJ9gSdI=
=OpPE
-----END PGP SIGNATURE-----

Re: Bind mounts record fstype of none

[Karel Zak]

On Thu, Oct 31, 2013 at 11:28:57AM -0400, Phillip Susi wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> On 10/31/2013 11:10 AM, Karel Zak wrote:
> > There is *many disadvantages* when you use mtab, not sure if it
> > make sense to fix the issues especially when mtab is currently
> > unused by (almost) all distributions.
> > 
> > Maybe we can keep the fstype and not replace it with "none" for
> > mtab, but is it so important when we use "none" since 2009 and mtab
> > is almost dead?
> 
> I know there were a number of issues with this transition, have they
> all been worked out at this point?  Is all you need to do to make the
> switch is replace /etc/mtab with a symlink to /proc/mounts?  Maybe
> I'll try doing that.

 Yes, the symlink and libmount based mount(8). We had some issues with
 df(1) but I guess all is resolved by recent coreutils.

    Karel

-- 
 Karel Zak  <kzak@redhat.com>
 http://karelzak.blogspot.com

Re: Bind mounts record fstype of none

[Karel Zak]

On Thu, Oct 31, 2013 at 11:28:57AM -0400, Phillip Susi wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> On 10/31/2013 11:10 AM, Karel Zak wrote:
> > There is *many disadvantages* when you use mtab, not sure if it
> > make sense to fix the issues especially when mtab is currently
> > unused by (almost) all distributions.
> > 
> > Maybe we can keep the fstype and not replace it with "none" for
> > mtab, but is it so important when we use "none" since 2009 and mtab
> > is almost dead?
> 
> I know there were a number of issues with this transition, have they
> all been worked out at this point?  Is all you need to do to make the
> switch is replace /etc/mtab with a symlink to /proc/mounts?  Maybe
> I'll try doing that.

 ... and it's also good to link mount.nfs from nfs-utils with libmount
 as NFS maintains some mount options in userspace (in this case we use
 private library /run/mount/utab file).

    Karel

-- 
 Karel Zak  <kzak@redhat.com>
 http://karelzak.blogspot.com

Re: Bind mounts record fstype of none

[Karel Zak]

On Thu, Oct 31, 2013 at 10:18:24AM -0400, Phillip Susi wrote:
> >> I came across an old bug filed in debian

 I forgot... Thanks! We had very good relationship with Debian 
 (thanks to LaMont Jones), but in last 5 years it seems that 
 nobody from Debian camp cares. It's strange as Debian plays
 important role in Linux world.

    Karel

-- 
 Karel Zak  <kzak@redhat.com>
 http://karelzak.blogspot.com