From: Andrew Hewus Fresh <andrew@afresh1.com>
Subject: Re: Use dmesg(8) in fw_update(8) instead of dmesg.boot
To: tech@openbsd.org, Theo de Raadt <deraadt@openbsd.org>
Cc: Klemens Nanni <kn@openbsd.org>
Date: Thu, 25 Dec 2025 13:03:51 -0800

On Thu, Dec 25, 2025 at 12:55:11PM -0700, Theo de Raadt wrote:
> Klemens Nanni <kn@openbsd.org> wrote:
> 
> > 25.12.2025 19:07, Theo de Raadt пишет:
> > > Hmm.
> > > 
> > > I think we chose dmesg.boot for a reason before, but I'm not sure I
> > > remember the reason.
> > > 
> > > Maybe we thought $(dmesg) could have rolled over with noisy debug messages,
> > > and thus hidden names of devices we want to identify, and therefore
> > > dmesg.boot was more reliable?
> > > 
> > > Here's a thought:  Should we look at both $(dmesg) and dmesg.boot ??
> > 
> > I certainly have machines where dmesg gets flooded with log spam,
> > so any $(dmesg) with a few hours of uptime won't have a single device.
> > 
> > If fw_update stops looking at dmesg.boot, there won't be any chance
> > for it to update existing firmware, right?
> 
> Look at my suggestion.

Or what I said on this:


On Thu, Dec 25, 2025 at 08:01:28AM -0800, Andrew Hewus Fresh wrote:
> The previous implementation looked at both,  I assume in case anything
> had "fallen off" the dmesg filling up.  I think it is OK not to make
> that more intrusive change because we would have checked at boot and we
> also look for updates to already installed firmware.  The only failure
> point I see is if dmesg had filled up past dmesg.boot, new firmware came
> to be that didn't previously exist, and someone wanted to install it
> without rebooting.
> 
> https://github.com/openbsd/src/blob/master/usr.sbin/pkg_add/OpenBSD/FwUpdate.pm#L158-L175

In any case, here is a patch to scan both.  I haven't tried my alpha to
see how much it slows things down, but folks are showing up to visit so
I won't have too much time for revisions today.

Index: fw_update.8
===================================================================
RCS file: /cvs/src/usr.sbin/fw_update/fw_update.8,v
diff -u -p -r1.8 fw_update.8
--- fw_update.8	22 Mar 2025 19:51:29 -0000	1.8
+++ fw_update.8	25 Dec 2025 20:56:42 -0000
@@ -59,11 +59,11 @@ If used in conjunction with
 .Fl a ,
 delete firmware for all drivers.
 .It Fl D Ar path
-Use the
-.Xr dmesg 8
-output from
+Use the content of
 .Ar path
-rather than
+rather than output from 
+.Xr dmesg 8
+and
 .Pa /var/run/dmesg.boot
 to determine which firmware are needed.
 .It Fl F
Index: fw_update.sh
===================================================================
RCS file: /cvs/src/usr.sbin/fw_update/fw_update.sh,v
diff -u -p -r1.65 fw_update.sh
--- fw_update.sh	12 May 2025 23:48:12 -0000	1.65
+++ fw_update.sh	25 Dec 2025 20:56:42 -0000
@@ -239,12 +239,25 @@ verify_existing() {
 }
 
 devices_in_dmesg() {
+	if [ "${DMESG:-}" ]; then
+		_devices_in_dmesg "$DMESG"
+		return
+	fi
+
+	dmesg > "$FD_DIR/dmesg"
+	
+	_devices_in_dmesg /var/run/dmesg.boot
+	_devices_in_dmesg "$FD_DIR/dmesg"
+}
+
+_devices_in_dmesg() {
+	local _dmesg=$1
 	local IFS
 	local _d _m _dmesgtail _last='' _nl='
 '
 
 	# The dmesg can contain multiple boots, only look in the last one
-	_dmesgtail="$( echo ; sed -n 'H;/^OpenBSD/h;${g;p;}' "$DMESG" )"
+	_dmesgtail="$( echo ; sed -n 'H;/^OpenBSD/h;${g;p;}' "$_dmesg" )"
 
 	grep -v '^[[:space:]]*#' "$FWPATTERNS" |
 	    while read -r _d _m; do
@@ -489,7 +502,7 @@ set_fw_paths() {
 	if [ ! "$_version" ]; then
 		_version=$(sed -nE \
 		    '/^OpenBSD ([0-9]+\.[0-9][^ ]*) .*/{s//\1/;h;};${g;p;}' \
-		    "$DMESG")
+		    "${DMESG:-/var/run/dmesg.boot")
 	
 		# If VNAME was set in the environment instead of the DMESG,
 		# looking in the DMESG for "current" is wrong.
@@ -514,7 +527,6 @@ usage() {
 
 ALL=false
 LIST=false
-DMESG=/var/run/dmesg.boot
 
 while getopts :adD:Flnp:v name
 do
@@ -561,7 +573,7 @@ if [ "${FWURL:-}" ] && ! "$INSTALL" ; th
 	usage
 fi
 
-if [ ! -s "$DMESG" ]; then
+if [ "${DMESG:-}" ] && [ ! -s "$DMESG" ]; then
 	warn "${0##*/}: $DMESG: No such file or directory"
 	exit 1
 fi