I thought I looked pretty hard for this, but clearly not hard enough. Thank
you. :-)
I'll go comb through the differences and try to backport them.

On an unrelated note, I am trying to figure out the optimal fix for a long
standing issue of cache flushing on partitions in zfs-fuse. This fills the
syslog:
zfs-fuse: sending ioctl 2285 to a partition

As far as I can tell, this is because zfs-fuse tries to issue SCSI or ATA
(depending on what it detects of the underlying block device) commands to
flush the disk's write cache. Unfortunately, it does this regardless of
whether it is operating on a raw disk or on a partition. Since all the
implementations of late have been moving toward using GPT partitions by
default when told to use the raw disk, this causes problems when SCSI or
ATA commands are set to a partition rather than the disk.

I am not even sure what actually happens underneath - whether the kernel
silently discards the SCSI command or whether it redirects it to the disk.

Either way, I'm trying to figure out what the correct solution in this case
is. Here is one option:
https://github.com/gordan-bobic/zfs-fuse/commit/fe5acf32fa5f2e1a716d088a5bcd8ada411ef860

The potential issue with that is that if the kernel forwards the SCSI cache
flush command to the disk even when it is sent to a partition, this will
obviously make the cache safety situation worse (AFAICT fsync and BLKFLSBUF
only flush page cache to disk, not disk's hardware cache to permanent
media).

I tried to start the discussion here:
https://groups.google.com/forum/#!topic/zfs-fuse/XmR7Ehe5Fi4

In a nutshell, different implementations seem to handle this quite
differently, and I am not sure what direction to take zfs-fuse in.




On Sat, May 30, 2015 at 9:19 AM, Hajo Möller <[email protected]> wrote:

> Hi Gordan,
>
> Seth Heeren has a mirror of his original git repo on Github,
> https://github.com/sehe/zfs-fuse/commits/maint
>
> It seems to be a complete mirror, the commit log matches the one from
> git.zfs-fuse.net, according to archive.org:
>
> http://web.archive.org/web/20130329131725/http://gitweb.zfs-fuse.net/?p=official;a=summary
>
> --
> Regards,
> Hajo Möller
> _______________________________________________
> developer mailing list
> [email protected]
> http://lists.open-zfs.org/mailman/listinfo/developer
>
_______________________________________________
developer mailing list
[email protected]
http://lists.open-zfs.org/mailman/listinfo/developer

Reply via email to