From: Klemens Nanni <kn@openbsd.org>
Subject: Re: Earlier /tmp cleanup in /etc/rc
To: Andrew Hewus Fresh <afresh1@openbsd.org>, tech@openbsd.org
Date: Tue, 25 Aug 2026 20:32:49 +0000

24.08.2026 21:30, Andrew Hewus Fresh пишет:
> With the recent adjustments to clean up /tmp asynchronously, deraadt and
> others noticed an existing race that some of the relinking and early
> daemons using /tmp while it was being cleaned.
> 
> This patch solves those known problems and possibly some unknown (trying
> to ktrace the startup affects things) by moving /tmp mounting and
> cleanup before any async processes that would potentially access /tmp,
> making the cleanup safer.

Hoisting /tmp cleaning before most (any?) new uses seems sensible.

> 
> The interaction with find, rm, and flushing is a bit weird, so many of
> 30+ email in this solution were finding the correct incantation to make
> `find -exec rm | head` delete the expected number of entries, but ended
> up with a more reliable and easier to read solution.
> 
> I was also reminded that we are not on the install media and we have
> access to more reasonable tools and so can avoid the confusing `sh -c`
> and instead use xargs to make find and mv cooperate.

Reads solid to me, a few remarks inline.

> 
> 
> Index: rc
> ===================================================================
> RCS file: /cvs/src/etc/rc,v
> diff -u -p -r1.590 rc
> --- rc	17 Aug 2026 22:41:38 -0000	1.590
> +++ rc	24 Aug 2026 16:05:01 -0000
> @@ -502,6 +502,21 @@ sysctl_conf
>  mount -s /var >/dev/null 2>&1		# cannot be on NFS
>  mount -s /var/log >/dev/null 2>&1	# cannot be on NFS
>  mount -s /usr >/dev/null 2>&1		# if NFS, fstab must use IP address
> +mount -s /tmp >/dev/null 2>&1		# if NFS, fstab must use IP address
> +
> +# Prune /tmp before users can authenticate.  A few files are removed
> +# immediately, but most of the deletion happens asynchronously while
> +# daemons are being started
> +( cd /tmp && {
> +    echo clearing /tmp
> +    skip="! -path . ! -path ./lost+found ! -path ./quota.user 

'! -path .' is '-mindepth 1', if you prefer that to match -maxdepth 1.

> +	! -path ./quota.group ! -path ./vi.recover"
> +    find -d . $skip -print -exec rm -rf {} \; | head -50 >/dev/null
> +    tmpdir=$(mktemp -d .delXXXXXXXXXX) &&
> +	find . -maxdepth 1 $skip ! -name "$tmpdir" -print0 | \

\ is not needed here.

> +	    xargs -r0J % mv -f % "$tmpdir"
> +    rm -rf "$tmpdir" &

rm still runs on mktemp failure, which is fine as -f ensure zero-exit,
logically it reads like oversight, though, so I'd gate both commands:

	tmpdir=$(mktemp ...) && {
		find ... |
			xargs
		rm ... &
	}
> +} )
>  
>  reorder_libs 2>&1 |&
>  
> @@ -597,26 +612,6 @@ if [[ -f /etc/ptmp ]]; then
>  	    'password file may be incorrect -- /etc/ptmp exists'
>  fi
>  
> -echo clearing /tmp
> -
> -# Prune /tmp before users can authenticate.  A few files are removed
> -# immediately, but most of the deletion happens asyncronously while
> -# daemons are being started
> -clear_tmp_dir=$(cd /tmp && {
> -    rm -rf .{X11,ICE}-unix
> -
> -    set -- rm -rf
> -    TMPDIR=$(mktemp -d /tmp/.delXXXXXXXXXX) && export TMPDIR &&
> -	set -- sh -c 'mv -f -- "$@" "$TMPDIR"'
> -    find . -maxdepth 1 ! -name . ! -name lost+found ! -name quota.user \
> -	! -name quota.group ! -name vi.recover ! -name "$(basename "$TMPDIR")" \
> -	-exec "$@" -- {} \+
> -    rm -rfv "$TMPDIR" 2>&1 | head -20 >/dev/null
> -    echo "$TMPDIR"
> -} )
> -rm -rf "$clear_tmp_dir" &
> -clear_tmp_pid=$!
> -
>  # Create Unix sockets directories for X if needed and make sure they have
>  # correct permissions.
>  [[ -d /usr/X11R6/lib ]] && mkdir -m 1777 /tmp/.{X11,ICE}-unix
> @@ -666,8 +661,6 @@ start_daemon dhcp6leased rad hostapd lpd
>  start_daemon ftpproxy ftpproxy6 tftpd tftpproxy identd inetd rarpd bootparamd
>  start_daemon rbootd mopd vmd spamd spamlogd sndiod
>  echo '.'
> -
> -wait $clear_tmp_pid
>  
>  # If rc.firsttime exists, run it just once, and make sure it is deleted.
>  run_upgrade_script firsttime
>