Check out PR#bin/2977 that I submitted a while ago.  Basically, my
suggested patch modifies dump to only recognize ufs/ffs file system
entries in /etc/fstab.  IMHO, do recognize any others is ridiculous, 
since dump(8) couldn't dump them anyway.

On Tue, 14 Jan 1997, Brook Milligan wrote:

>    QUESTION:  Is there a reason for dump to construct its data structure
>    representing /etc/fstab in reverse order?
> 
>    Have I missed something obvious here?
> 
> I should have realized what I was missing (but a question still
> remains).  Of course, the last /etc/fstab entry is the one most
> recently mounted and therefore the one in use and to be backed up.
> Hence the reverse order linked list.  OK.
> 
> But, shouldn't union mounts be disregarded altogether since they are
> only namespace manipulations and not actual device mounts?  I.e.,
> shouldn't dump skip over union mounts in constructing its internal
> version of /etc/fstab so that it only matches "real" filesystem
> entries?
> 
> Now what am I missing?
> 
> One patch to skip union mounts is the following:
> 
> *** ./src/sbin/dump/optr.c.orig Sat May 18 10:16:17 1996
> --- ./src/sbin/dump/optr.c      Tue Jan 14 14:29:46 1997
> ***************
> *** 444,449 ****
> --- 444,451 ----
>                     strcmp(fs->fs_type, FSTAB_RO) &&
>                     strcmp(fs->fs_type, FSTAB_RQ))
>                         continue;
> +               if (strcmp(fs->fs_vfstype, "union") == 0)
> +                       continue;
>                 fs = allocfsent(fs);
>                 if ((pf = (struct pfstab *)malloc(sizeof (*pf))) == NULL)
>                         quit("%s\n", strerror(errno));
> 
> Thanks again for your help.
> 
> Cheers,
> Brook
>