Remove dependency of rc.d/tmp on grep

[se]

Summary

The committed version of /etc/rc.d/tmp depends on "mountcritremote" due to its use of grep.

This dependency can easily be removed to allow rc.d/tmp to be executed before /usr/bin is mounted.

Use of a here document avoids complications due to the spawned sub-shell if "df /tmp | while ..." was used.
In that case, the sub-shell would return a success code which then needed to be tested.

If a version without here-document was found to be preferential, I'd suggest a shell function that returns the result of "df /tmp | grep ..." but which does use the while loop from this patch instead of grep.

Test Plan

Apply patch and verify that "mount_md" is invoked if there is no line matching "^/dev/md[0-9].* /tmp" in the output of "df /tmp".

[rgrimes]

This fixes at least one dependance on /usr, I see another later, mktemp which is /usr/bin/mktemp. I would of marked it above, but a full file context was not uploaded to this review.

[se]

Remove dependency on mktemp, too, since it also resides in /usr/bin, as noted by rgrimes.

[cy]

Otherwise it looks good. This has the advantage of:

1. Less dependency on /usr, as rgrimes mentioned on src-committers@.
2. Less fork()ing.

libexec/rc/rc.d/tmp

50

Another way to do this, might be df /tmp | while read line... To me it looks cleaner.

[se]

As written in the comment of the original submission, a version based on "df /tmp | while ..." can be used instead of the here-doc.
But after implementing both, I did consider the here-doc based version to be both easier to understand and more efficient and thus only proposed that version.

libexec/rc/rc.d/tmp

50

My initial attempt actually used "df /tmp | while ...", but I find the here doc version more straight forward.

Here is a version that uses "df /tmp | while ...":

mount_tmpmfs()
{
        df /tmp | while read line; do
                case $line in
                /dev/md[0-9]*\ /tmp)
                        return 1;;
                esac
        done
        if [ $? = 0 ]; then
            mount_md ${tmpsize} /tmp "${tmpmfs_flags}"
            chmod 01777 /tmp
        fi
}

The while loop is executed in a sub-shell and thus needs to be exited with "return 1" to pass a value to the if condition.

This is easy to understand if you know about the while loop being executed in a sub-shell, I think that the here-doc version does not require such knowledge.

The pipe-based version requires 2 forked sub-processes (the df command and the shell executing the while loop), but the here-doc version only forks the df process and I do prefer it for these reasons.

[glebius]

Stefan, thanks a lot of taking care of my fallout!

[glebius]

libexec/rc/rc.d/tmp

50

To me the here-doc looks more efficient than pipe.

[se]

I had tried to create the temporary directory name without dependency on /dev/random (and thus suggested to use "ps | sha256", or in my temporary version "ps lauxww | grep sha256").

But since mktemp uses random(4), the currently committed version did also depend on /dev/random being operational.
Therefore, the version I want to commit uses "dd if=/dev/random" to generate the test directory.
(If sufficient entropy was not available, "ps lauxww" might actually be preferential, but then I'd expect that there must have been reports of mktemp blocking ...)

[cy]

Agree with glebius.