Libguestfs - Jan 2018 - virtdf outputs on host differs from df in guest

hi guys,

   happy new year!!!

  I encountered  a problem when we used virt-df command to display my 
guest disks useage.

the virt-df command from pkg 
libguestfs-tools-c-1.36.3-6.el7_4.3.x86_64,version is  virt-df --version 
: virt-df 1.36.3rhel=7,release=6.el7_4.3,libvirt

the output in guest with the method referred in 
http://libguestfs.org/virt-df.1.html#STATVFS NUMBERS

is:

[root@localhost ~]# df
Filesystem              1K-blocks    Used Available Use% Mounted on
/dev/mapper/centos-root  23544596 4335172  19209424  19% /
devtmpfs                   931220       0    931220   0% /dev
tmpfs                      941860       0    941860   0% /dev/shm
tmpfs                      941860    8612    933248   1% /run
tmpfs                      941860       0    941860   0% /sys/fs/cgroup
/dev/sda1                  508588  133328    375260  27% /boot
tmpfs                      188376       0    188376   0% /run/user/0

using scripts:

python -c 'import os; s = os.statvfs ("/"); print s'
posix.statvfs_result(f_bsize=4096, f_frsize=4096, f_blocks=5886149, 
f_bfree=4802342, f_bavail=4802342, f_files=23556096, f_ffree=23435372, 
f_favail=23435372, f_flag=4096, f_namemax=255)

python -c 'import os; s = os.statvfs ("/boot"); print s'
posix.statvfs_result(f_bsize=4096, f_frsize=4096, f_blocks=127147, 
f_bfree=93815, f_bavail=93815, f_files=512000, f_ffree=511626, 
f_favail=511626, f_flag=4096, f_namemax=255)

and then on host using virt-df showing:

# sudo virt-df rpm-build-for-7.2

Filesystem                           1K-blocks       Used Available  Use%
rpm-build-for-7.2:/dev/sda1             508588     107728 400860   22%
rpm-build-for-7.2:/dev/centos/root    23544596    4298384 19246212   19%

using scripts:

# guestfish --ro -d rpm-build-for-7.2 -i statvfs /

bsize: 4096
frsize: 4096
blocks: 5886149
bfree: 4810534
bavail: 4810534
files: 23556096
ffree: 23435372
favail: 23435372
fsid: 64513
flag: 4097
namemax: 255

# sudo guestfish --ro -d rpm-build-for-7.2 -i statvfs /boot
bsize: 4096
frsize: 4096
blocks: 127147
bfree: 100215
bavail: 100215
files: 512000
ffree: 511626
favail: 511626
fsid: 2049
flag: 4097
namemax: 255

we can saw that in guest the statvfs get the value was the same with the 
"df" command output. but the bfree/bavail values are different between
using guestfish statvfs on guest and

using statvfs in guest. I want to know whether is this a bug? or Does 
same reason cause this appearance？

Thanks,

Chen

On Thu, Jan 04, 2018 at 12:58:40PM +0800, Chen Fan wrote:
[In guest]
> python -c 'import os; s = os.statvfs ("/"); print s'
> posix.statvfs_result(f_bsize=4096, f_frsize=4096, f_blocks=5886149,
> f_bfree=4802342, f_bavail=4802342, f_files=23556096,
> f_ffree=23435372, f_favail=23435372, f_flag=4096, f_namemax=255)
> 
> python -c 'import os; s = os.statvfs ("/boot"); print s'
> posix.statvfs_result(f_bsize=4096, f_frsize=4096, f_blocks=127147,
> f_bfree=93815, f_bavail=93815, f_files=512000, f_ffree=511626,
> f_favail=511626, f_flag=4096, f_namemax=255)
[From the host via libguestfs]
> # guestfish --ro -d rpm-build-for-7.2 -i statvfs /
> 
> bsize: 4096
> frsize: 4096
> blocks: 5886149
> bfree: 4810534
> bavail: 4810534
> files: 23556096
> ffree: 23435372
> favail: 23435372
> fsid: 64513
> flag: 4097
> namemax: 255
> 
> # sudo guestfish --ro -d rpm-build-for-7.2 -i statvfs /boot
> bsize: 4096
> frsize: 4096
> blocks: 127147
> bfree: 100215
> bavail: 100215
> files: 512000
> ffree: 511626
> favail: 511626
> fsid: 2049
> flag: 4097
> namemax: 255
Was the guest quiescent and fully sync()d before you ran the python
commands?

In any case all that libguestfs is doing is executing the
statvfs(2) system call on a copy of the disk:

 
https://github.com/libguestfs/libguestfs/blob/e6e23c4d77ee15779f833fd4fbe6c574940e85a4/daemon/statvfs.c#L52

So if there's any difference then it's something to do with the guest
kernel vs the host kernel calculations.

I wouldn't expect the numbers to be exactly the same, but libguestfs
doesn't alter the numbers, it just passes them through.

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
Fedora Windows cross-compiler. Compile Windows programs, test, and
build Windows installers. Over 100 libraries supported.
http://fedoraproject.org/wiki/MinGW

hi Jones, Thanks for your reply.


At 2018-01-04 19:41:27, "Richard W.M. Jones" <rjones@redhat.com>
wrote:
>On Thu, Jan 04, 2018 at 12:58:40PM +0800, Chen Fan wrote:
>[In guest]
>> python -c 'import os; s = os.statvfs ("/"); print s'
>> posix.statvfs_result(f_bsize=4096, f_frsize=4096, f_blocks=5886149,
>> f_bfree=4802342, f_bavail=4802342, f_files=23556096,
>> f_ffree=23435372, f_favail=23435372, f_flag=4096, f_namemax=255)
>> 
>> python -c 'import os; s = os.statvfs ("/boot"); print
s'
>> posix.statvfs_result(f_bsize=4096, f_frsize=4096, f_blocks=127147,
>> f_bfree=93815, f_bavail=93815, f_files=512000, f_ffree=511626,
>> f_favail=511626, f_flag=4096, f_namemax=255)
>
>[From the host via libguestfs]
>> # guestfish --ro -d rpm-build-for-7.2 -i statvfs /
>> 
>> bsize: 4096
>> frsize: 4096
>> blocks: 5886149
>> bfree: 4810534
>> bavail: 4810534
>> files: 23556096
>> ffree: 23435372
>> favail: 23435372
>> fsid: 64513
>> flag: 4097
>> namemax: 255
>> 
>> # sudo guestfish --ro -d rpm-build-for-7.2 -i statvfs /boot
>> bsize: 4096
>> frsize: 4096
>> blocks: 127147
>> bfree: 100215
>> bavail: 100215
>> files: 512000
>> ffree: 511626
>> favail: 511626
>> fsid: 2049
>> flag: 4097
>> namemax: 255
>
>Was the guest quiescent and fully sync()d before you ran the python
>commands?

yes, almost the guest is quiescent, and after invoke sync(), the results have no
different.

>
>In any case all that libguestfs is doing is executing the
>statvfs(2) system call on a copy of the disk:
>
> 
https://github.com/libguestfs/libguestfs/blob/e6e23c4d77ee15779f833fd4fbe6c574940e85a4/daemon/statvfs.c#L52
>
>So if there's any difference then it's something to do with the
guest
>kernel vs the host kernel calculations.

I saw that when doing virt-df, virt-df created a  temporary VM with the host
kernel, so I
recreate a temporary QEMU VM with command line: 
/usr/libexec/qemu-kvm -cpu host -m 1024 -nographic -kernel
/var/tmp/.guestfs-0/appliance.d/kernel -initrd
/var/tmp/.guestfs-0/appliance.d/initrd -append "console=ttyS0 root=/dev/vda
guestfs_rescue=1" -device virtio-blk-pci,drive=disk0 -drive
file=/var/tmp/.guestfs-0/appliance.d/root,id=disk0,if=none -device virtio-serial
-chardev socket,id=charchannel0,path=/tmp/a,server,nowait -device
virtserialport,nr=1,chardev=charchannel0,id=channel0,name=org.libguestfs.channel.0
-device virtio-blk-pci,drive=disk1 -drive
file=/srv/imgs/centos-7.2.qcow2,if=none,id=disk1
then the console showing:
><rescue> mount /dev/vdb1 /sysroot/
[   72.552185] SGI XFS with ACLs, security attributes, no debug enabled
[   72.568377] XFS (vdb1): Mounting V4 Filesystem
[   73.379679] XFS (vdb1): Starting recovery (logdev: internal)
[   73.406398] XFS (vdb1): Ending recovery (logdev:
internal)
><rescue> statvfs
bash: statvfs: command not found
><rescue> stat -f /sysroot
  File: "/sysroot"
    ID: fd1100000000 Namelen: 255     Type: xfs
Block size: 4096       Fundamental block size: 4096
Blocks: Total: 127147     Free: 93815      Available: 93815
Inodes: Total: 512000     Free: 511626



here the mount /dev/vdb1 device is the /boot partition from my guest, we can saw
that
the free blocks are the same as python command, so I think the host kernel does
not affect the result.
if I am wrong, pls correct me :)


Thanks,
Chen





>
>I wouldn't expect the numbers to be exactly the same, but libguestfs
>doesn't alter the numbers, it just passes them through.
>
>Rich.
>
>-- 
>Richard Jones, Virtualization Group, Red Hat
http://people.redhat.com/~rjones
>Read my programming and virtualization blog: http://rwmj.wordpress.com
>Fedora Windows cross-compiler. Compile Windows programs, test, and
>build Windows installers. Over 100 libraries supported.
>http://fedoraproject.org/wiki/MinGW