zfs discuss - Dec 2007 - What does "dataset is busy" actually mean?

Hello,

I''m trying to track down a problem with taking zfs snapshots. On
occasion the zfs command will report:

cannot snapshot ''<dataset name>'': dataset is busy

The problem is, I don''t know what causes zfs to think the data set is
"busy". Anyone out there know what constitutes a busy dataset?

I did some testing and it seems that mounting, nfs sharing and writing
to the dataset don''t seem to affect its "busy" status. That
is, I can
take a snapshot of a dataset that is mounted, in use by nfs, and being
written continuously. This leaves me a little puzzled at the definition
of "busy". 

Any help would be greatly appreciated.

Thanks,

Reid Spencer
illumita.com
 
 
This message posted from opensolaris.org

I''ve hit the problem myself recently, and mounting the filesystem
cleared
something in the brains of ZFS and alowed me to snapshot.

http://www.mail-archive.com/zfs-discuss at opensolaris.org/msg00812.html

PS: I''ll use Google before asking some questions,  a''la (C)
Bart Simpson
That''s how I found your question ;)
 
 
This message posted from opensolaris.org

Yes, it seems that mounting it and unmounting it with the zfs command clears the
condition and allows the data set to be destroyed. Seems this is a bug in zfs,
or at least an annoyance. I verified with fuser that no processes were using the
file system.

Now, what I''d really like to know, is what causes a dataset to get into
this state?
 
 
This message posted from opensolaris.org

Hmm, actually, not.

I just ran into a dataset where the mount/unmount doesn''t clear the
condition. I still get "dataset is busy" when attempting to destroy
it.
 
 
This message posted from opensolaris.org

> what causes a dataset to get into this state?
while I''m not exactly sure, I do have the steps leading up to when
I saw it trying to create a snapshot. ie:

10 % zfs snapshot z/b80nd/var at 0107nd
cannot create snapshot ''z/b80nd/var at 0107nd'': dataset is
busy
13 % mount -F zfs z/b80nd/var /z/b80nd/var
mount: Mount point /z/b80nd/var does not exist.
14 % mount -F zfs z/b80nd/var /mnt
15 % zfs snapshot -r z/b80nd at 0107nd
16 % zfs list | grep 0107
root/0107nd                        455M   107G  6.03G  legacy
root/b80nd at 0107nd                 50.5M      -  6.02G  -
z/b80nd at 0107nd                        0      -   243M  -
z/b80nd/opt at 0107nd                    0      -  1.18G  -
z/b80nd/usr at 0107nd                    0      -  2.25G  -
z/b80nd/var at 0107nd                    0      -  56.3M  -

running 64bit opensol-20080107 on intel

to get there I was walking through this cookbook:

zfs snapshot root/b80nd at 0107nd
zfs clone root/b80nd at 0107nd root/0107nd
cat /etc/vfstab | sed s/^root/#root/ | sed s/^z/#z/  > /root/0107nd/ 
etc/vfstab
echo "root/0107nd - / zfs - no -" >> /root/0107nd/etc/vfstab
cat /root/0107nd/etc/vfstab
zfs snapshot -r z/b80nd at dump
rsync -a --del --verbose /usr/.zfs/snapshot/dump/ /root/0107nd/usr
rsync -a --del --verbose /opt/.zfs/snapshot/dump/ /root/0107nd/opt
rsync -a --del --verbose /var/.zfs/snapshot/dump/ /root/0107nd/var
zfs set mountpoint=legacy root/0107nd
zpool set bootfs=root/0107nd root
reboot

mkdir -p /z/tmp/bfu ; cd /z/tmp/bfu
wget http://dlc.sun.com/osol/on/downloads/20080107/SUNWonbld.i386.tar.bz2
bzip2 -d -c SUNWonbld.i386.tar.bz2 | tar -xvf -
pkgadd -d onbld
wget
http://dlc.sun.com/osol/on/downloads/20080107/on-bfu-nightly-osol-nd.i386.tar.bz2
bzip2 -d -c on-bfu-nightly-osol-nd.i386.tar.bz2 | tar -xvf -
setenv FASTFS /opt/onbld/bin/i386/fastfs
setenv BFULD /opt/onbld/bin/i386/bfuld
setenv GZIPBIN /usr/bin/gzip
/opt/onbld/bin/bfu /z/tmp/bfu/archives-nightly-osol-nd/i386
/opt/onbld/bin/acr
echo etc/zfs/zpool.cache >> /boot/solaris/filelist.ramdisk  ; echo bug  
in bfu
reboot

rm -rf /bfu* /.make* /.bfu*
zfs snapshot root/0107nd at dump
mount -F zfs z/b80nd/var /mnt  ; echo bug in zfs
zfs snapshot -r z/b80nd at 0107nd
zfs clone z/b80nd at 0107nd z/0107nd
zfs set compression=lzjb z/0107nd
zfs clone z/b80nd/usr at 0107nd z/0107nd/usr
zfs clone z/b80nd/var at 0107nd z/0107nd/var
zfs clone z/b80nd/opt at 0107nd z/0107nd/opt
rsync -a --del --verbose /.zfs/snapshot/dump/ /z/0107nd
zfs set mountpoint=legacy z/0107nd/usr
zfs set mountpoint=legacy z/0107nd/opt
zfs set mountpoint=legacy z/0107nd/var
echo "z/0107nd/usr - /usr zfs - yes -" >> /etc/vfstab
echo "z/0107nd/var - /var zfs - yes -" >> /etc/vfstab
echo "z/0107nd/opt - /opt zfs - yes -" >> /etc/vfstab
reboot

heh heh, booting from a clone of a clone... waisted space under
root/`uname -v`/usr for a few libs needed at boot, but having
/usr /var /opt on the compressed pool with two raidz vdevs boots
to login in 45secs rather than 52secs on the single vdev root pool.