« Return to Thread: ZFS and fuseblk
Re: ZFS and fuseblk
by Miklos Szeredi
::
Rate this Message:
Hi Ricardo,
> Can you clarify how the new fuseblk filesystem type differs in
> functionality from the usual fuse filesystem type?
>
> The patch description says that the only difference is that the
> "dev_name" mount option is interpreted in fuseblk.
>
> The description even mentions how this is useful for ZFS, however:
>
> - ZFS pools (and by consequence, the ZFS filesystems inside them) are
> typically composed of multiple block devices.
OK, I was totally ignorant of this. I thought, that ZFS is similar to
other filsystems, that it uses a single block device.
If that is not true, then fuseblk features mostly don't apply to ZFS,
since the kernel itself doesn't have support for more than one device
per filesystem.
For a single-device filesystem the fuseblk type provides the following
features:
- locking the device on mount, unlocking on release
- sharing the filesystem for multiple mounts
- allowing swapfiles to bypass the filesystem in accessing the
underlying device
- allowing lilo to work on the filesystem
The last two features need that the data blocks on the filesystem
correspond exactly to data blocks on the device. I seem to remember
you saying that this is never true for ZFS.
In addition there's one feature, that isn't strictly related to block
devices, but was bundled with the fuseblk filesystem. This is the
ability to synchronously unmounting a filesystem: when the filesystem
is last unmounted (no lazy unmounts or bind mounts remain), then the
unmount will wait until the filesystem acknowledges this (e.g. flushes
buffers).
> - I'm not even sure where the dev_name mount option or the fuseblk
> filesystem type is supposed to go into (/etc/fstab?).
No, it doesn't need to go into /etc/fstab. Fuse mounting from
/etc/fstab is rather involved:
/bin/mount -> /sbin/mount.fuse -> /usr/bin/filesystem -> /usr/bin/fusermount
The blkdev and fsname=xxx options are inserted by the filesystem
before calling fusermount.
> I'm not sure how can I use that functionality, since the ZFS filesystems
> shouldn't even be added to /etc/fstab unless the administrator really
> wants it (ZFS has an optional "legacy" mode for this).
>
> So typically the filesystems are mounted by the zfs-fuse daemon itself,
> which, of course, uses fuse_mount() to mount them.
>
> What are your thoughts on this?
Yes, In light of what you said, most of the features that fuseblk
provides doesn't apply to ZFS.
Thanks,
Miklos
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
fuse-devel mailing list
fuse-devel@...
https://lists.sourceforge.net/lists/listinfo/fuse-devel
> Can you clarify how the new fuseblk filesystem type differs in
> functionality from the usual fuse filesystem type?
>
> The patch description says that the only difference is that the
> "dev_name" mount option is interpreted in fuseblk.
>
> The description even mentions how this is useful for ZFS, however:
>
> - ZFS pools (and by consequence, the ZFS filesystems inside them) are
> typically composed of multiple block devices.
OK, I was totally ignorant of this. I thought, that ZFS is similar to
other filsystems, that it uses a single block device.
If that is not true, then fuseblk features mostly don't apply to ZFS,
since the kernel itself doesn't have support for more than one device
per filesystem.
For a single-device filesystem the fuseblk type provides the following
features:
- locking the device on mount, unlocking on release
- sharing the filesystem for multiple mounts
- allowing swapfiles to bypass the filesystem in accessing the
underlying device
- allowing lilo to work on the filesystem
The last two features need that the data blocks on the filesystem
correspond exactly to data blocks on the device. I seem to remember
you saying that this is never true for ZFS.
In addition there's one feature, that isn't strictly related to block
devices, but was bundled with the fuseblk filesystem. This is the
ability to synchronously unmounting a filesystem: when the filesystem
is last unmounted (no lazy unmounts or bind mounts remain), then the
unmount will wait until the filesystem acknowledges this (e.g. flushes
buffers).
> - I'm not even sure where the dev_name mount option or the fuseblk
> filesystem type is supposed to go into (/etc/fstab?).
No, it doesn't need to go into /etc/fstab. Fuse mounting from
/etc/fstab is rather involved:
/bin/mount -> /sbin/mount.fuse -> /usr/bin/filesystem -> /usr/bin/fusermount
The blkdev and fsname=xxx options are inserted by the filesystem
before calling fusermount.
> I'm not sure how can I use that functionality, since the ZFS filesystems
> shouldn't even be added to /etc/fstab unless the administrator really
> wants it (ZFS has an optional "legacy" mode for this).
>
> So typically the filesystems are mounted by the zfs-fuse daemon itself,
> which, of course, uses fuse_mount() to mount them.
>
> What are your thoughts on this?
Yes, In light of what you said, most of the features that fuseblk
provides doesn't apply to ZFS.
Thanks,
Miklos
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
fuse-devel mailing list
fuse-devel@...
https://lists.sourceforge.net/lists/listinfo/fuse-devel
« Return to Thread: ZFS and fuseblk
| Free embeddable forum powered by Nabble | Forum Help |
