dumpon: accept "dump" type in fstab

[emaste]

PR: 175311

In contrast to the approach in the PR I went with a loop looking for dump first then swap, as dump should take precedence over swap, in e.g.

# Device                Mountpoint      FStype  Options         Dump    Pass#
/dev/nvd0p3             none    swap    sw              0       0
/dev/nvd0p4             none    dump    sw              0       0

[emaste]

or if we want to avoid reading /etc/fstab twice something like

dump_try=
swap_try=
while read dev mp type more ; do
        [ "${type}" = dump -o "${type}" = swap ] || continue
        case ${dev} in
        *.bde|*.eli)
                dumpon_warn_unencrypted
                dev=${dev%.*}
                ;;
        esac
        [ -c "${dev}" ] || continue
        eval ${type}_try=\"\${${type}_try} ${dev}\"
done </etc/fstab
for dev in ${dump_try} ${swap_try}; do
        dumpon_try "${dev}" 2>/dev/null && return 0
done

[des]

In D34518#782061, @emaste wrote:

or if we want to avoid reading /etc/fstab twice something like

I'm not sure it's worth the added complexity, especially in a zfs world where fstab is typically very short.

The commit message needs some work, it should mention that we prefer dump to swap if both are encountered.

libexec/rc/rc.d/dumpon

65

We should skip lines where ${dev} begins with #, otherwise commenting out the first dump or swap line will result in some strange console errors:

unable to specify #/dev/da0p2 as a dump device

66

The patch in the PR has it right here (except for a missing newline); case is faster than test for string comparisons. Although arguably we should fix that by giving sh a built-in [.

[jilles]

I agree that comment lines in /etc/fstab should be skipped.

libexec/rc/rc.d/dumpon

66

Our sh has had a built-in [ since 2001, but case is still somewhat faster because it avoids the overhead associated with simple commands and built-in utilities. I don't think this is sufficient reason to replace [ with case even where it looks unnatural, though.