mount(8) enhancement

[Reinoud Zandijk <reinoud%NetBSD.org@localhost>]

Dear folks,

there are file systems that although living in one partition/device support
multiple heads/labels to mount on that can't be expressed as a disk partition
scheme like disklabel or GPT since they all live in one partion.

A common way to mount such systems is to append the head-name to the device
like f.e.
	mount /dev/dk0:bart /mnt
	mount /dev/dk0%lisa /mnt
	mount /dev/dk0@maggy /mnt

Or whatever scheme the FS wants or is chosen. Hammer2 supports such a scheme
in DragonflyBSD:
	mount_hammer2 /dev/serno/QM00005.s1d@ROOT /mnt
	mount_hammer2 -o ro /dev/serno/QM00005.s1d@2018-05-30_163833 /mnt

Our current mount(8) and all other file systems uses pathadj() from
sbin/mount/pathadj.c that is really a wrapper around realpath(3).

Passing such a name trough realpath() results in the following when passing it
trough my FS in development :

	# mount -t sfs /dev/dk1/bart /mnt
	dev: '/dev/dk1/bart'
	dir: '/mnt'
	mount_sfs: Warning: realpath /dev/dk1/bart: Not a directory
	canondev: '/dev/dk1'
	canondir: '/mnt'

but also
	# mount -t sfs /dev/dk1@lisa /
	dev: '/dev/dk1@lisa'
	dir: '/'
	mount_sfs: Warning: realpath /dev/dk1@lisa: No such file or directory
	canondev: '/dev/dk1@lisa'
	canondir: '/'
	mount_sfs: Cannot mount /dev/dk1@lisa on /: Bad address

and
	# mount -t sfs /dev/dk1:lisa /
	dev: '/dev/dk1:lisa'
	dir: '/'
	mount_sfs: Warning: realpath /dev/dk1:lisa: No such file or directory
	canondev: '/dev/dk1:lisa'
	canondir: '/'
	mount_sfs: Cannot mount /dev/dk1:lisa on /: Bad address

Note that mount(8) itself considers any mention of '%@' in a filename to be
NFS and isn't that forgiving:
	# mount /dev/dk1/root /mnt
	mount: cannot open `/dev/dk1/root': Not a directory
	# mount /dev/dk1@root /mnt
	mount_nfs: can't get net id for host "root": Temporary failure in name resolution
	# mount /dev/dk1:root /mnt
	mount_nfs: can't get net id for host "/dev/dk1": Temporary failure in name resolution


My proposal is to adjust
	pathadj(const char *input, char *adjusted)
into
	pathadj(const char *input, char *adjusted, char *tail)

It first copies the tail from the input with the tail defined as anything
after one of [$%@:&]. This does make the assumption that names with such
characters are not used in /dev/. It then calls realpath() on the truncated
input giving the desired result so the filesystem type associated with the
device can be recovered like normal and the appropriate mount_$FS can be
called with the appended tail.A
Only if this fails, a fallback is provided for Sun style NFS if the result of
realpath() is not found or is not a device; its hostname:mountpoint or
user@hostname:mountpoint will fail to locate `user' or `hostname' and if they
do exist somehow they are more like not devices.

Any objections to this scheme before I try to implement it?

With regards,
Reinoud