Libguestfs - Nov 2011 - virt-df: ext2/3/4 statvfs(2) output changed between two recent Linux kernels (3.2.0)

[Jim, as coreutils developer, this might interest you]

In libguestfs we run some quite detailed checks of the output of the
statvfs system call for our tool "virt-df".  I noticed that the output
of virt-df changed slightly.

Narrowing it down, the difference only happens between:

  Fedora kernel 3.2.0-0.rc0.git4.1.fc17 (3rd Nov)
  Fedora kernel 3.2.0-0.rc1.git2.1.fc17 (11th Nov)

With 3.2.0-0.rc0.git4.1.fc17, and all earlier versions, the output of
statvfs(2) on a freshly created 20M ext2 filesystem is:

  bsize: 1024
  frsize: 1024
  blocks: 19827     <-- NB
  bfree: 19655
  bavail: 18631
  files: 5136
  ffree: 5125
  favail: 5125
  fsid: 1411286354856931562
  flag: 4096
  namemax: 255

With 3.2.0-0.rc1.git2.1.fc17, it changed to:

  bsize: 1024
  frsize: 1024
  blocks: 19939     <-- NB
  bfree: 19655
  bavail: 18631
  files: 5136
  ffree: 5125
  favail: 5125
  fsid: -9088586685126948527
  flag: 4096
  namemax: 255

I am able to demonstrate the difference by keeping every other package
identical, and just booting the two different kernels.  As far as I
know, everything else including e2fsprogs is completely identical.

Reproducer:
  export FEBOOTSTRAP_KERNEL=/boot/vmlinuz-<version>
  guestfish sparse /tmp/test.img 20M : run : \
    mkfs ext2 /dev/sda : mount /dev/sda / : statvfs /

I tracked this down to the way the filesystem overhead is calculated.
It changed slightly in the commit below.  The description seems to be
wrong; it seems to affect any filesystem going through the ext4 driver
(this driver is used for ext2 & ext3 in recent Fedora), and doesn't
appear to have anything to do with 'bigalloc'.

  commit f975d6bcc7a698a10cc755115e27d3612dcfe322
  Author: Theodore Ts'o <tytso at mit.edu>
  Date:   Fri Sep 9 19:00:51 2011 -0400

      ext4: teach ext4_statfs() to deal with clusters if bigalloc is enabled

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
New in Fedora 11: Fedora Windows cross-compiler. Compile Windows
programs, test, and build Windows installers. Over 70 libraries supprt'd
http://fedoraproject.org/wiki/MinGW http://www.annexia.org/fedora_mingw

Richard W.M. Jones wrote:
> [Jim, as coreutils developer, this might interest you]
>
> In libguestfs we run some quite detailed checks of the output of the
> statvfs system call for our tool "virt-df".  I noticed that the
output
> of virt-df changed slightly.
>
> Narrowing it down, the difference only happens between:
>
>   Fedora kernel 3.2.0-0.rc0.git4.1.fc17 (3rd Nov)
>   Fedora kernel 3.2.0-0.rc1.git2.1.fc17 (11th Nov)
>
> With 3.2.0-0.rc0.git4.1.fc17, and all earlier versions, the output of
> statvfs(2) on a freshly created 20M ext2 filesystem is:
>
>   bsize: 1024
>   frsize: 1024
>   blocks: 19827     <-- NB
>   bfree: 19655
>   bavail: 18631
>   files: 5136
>   ffree: 5125
>   favail: 5125
>   fsid: 1411286354856931562
>   flag: 4096
>   namemax: 255
>
> With 3.2.0-0.rc1.git2.1.fc17, it changed to:
>
>   bsize: 1024
>   frsize: 1024
>   blocks: 19939     <-- NB
Good catch!
>   bfree: 19655
>   bavail: 18631
>   files: 5136
>   ffree: 5125
>   favail: 5125
>   fsid: -9088586685126948527
>   flag: 4096
>   namemax: 255
>
> I am able to demonstrate the difference by keeping every other package
> identical, and just booting the two different kernels.  As far as I
> know, everything else including e2fsprogs is completely identical.
>
> Reproducer:
>   export FEBOOTSTRAP_KERNEL=/boot/vmlinuz-<version>
>   guestfish sparse /tmp/test.img 20M : run : \
>     mkfs ext2 /dev/sda : mount /dev/sda / : statvfs /
>
> I tracked this down to the way the filesystem overhead is calculated.
> It changed slightly in the commit below.  The description seems to be
> wrong; it seems to affect any filesystem going through the ext4 driver
> (this driver is used for ext2 & ext3 in recent Fedora), and doesn't
> appear to have anything to do with 'bigalloc'.
>
>   commit f975d6bcc7a698a10cc755115e27d3612dcfe322
>   Author: Theodore Ts'o <tytso at mit.edu>
>   Date:   Fri Sep 9 19:00:51 2011 -0400
>
>       ext4: teach ext4_statfs() to deal with clusters if bigalloc is
enabled
Nice detective work.

Even if it's only a mere 0.5% difference, it seems like
something that should be announced more clearly -- that is,
assuming it was indeed intended.  If bigalloc-related changes
should not induce such changes in an ext2 file system, then
maybe it's a bug.