From: Jeff Dike <jdike@ka...> - 2001-08-13 18:46:57
> Also, I don't think passing device number information between
> userspace and UML like this is correct.
No, it's definitely wrong.
> The device information does
> not mean the same thing between the two, possibly allowing for a
> collision, and I am afraid that returning the same device+inode number
> for two different files can potentially screw things up badly.
Yeah. Although it hadn't occurred to me that the device number would be
visible from stat. I was more worried about two host filesystems being
available under a single hostfs mount (and looking like a single device)
and having something encounter files on the host fs's that share an inode
I think the visibility of device numbers improves the situation somewhat,
although I think it introduces a discrepancy between stat and statfs about
whether a hostfs mount is a single mount or multiple mounts.
From: Jeff Dike <jdike@ka...> - 2001-08-13 18:47:00
> hostfs_user.c seems to assumes stat64 is the kernel structures, which
> is not neccesarily the case. In my case userspace struct stat64 is the
> glibc version, which is somewhat different from the kernel version and
> not dependent on the kernel headers.
Yeah, I just changed that. It now greps for __st_ino instead of making
assumptions about the kernel version implying things about the headers.