Skip to contentSkip to conversation

hotplug and "rw" on kernel command line

DISCUSSION Alexander E. Patrakov 3 participants 3 messages May 30, 2004 – Jun 1, 2004

lore
Share & tools
lore
#0 lore raw

This is a trouble report forwarded from a Linux From Scratch CVS HEAD user.

=====================================================
[Bug 842] New: udev bootscript can fail to run udevstart if hotplug is installed.

In the LFS-specific udev bootscript, a check is made to see if /dev/.udev.tdb exists, and if it does the remainder of the script is not run. This was added to handle the situation where an initrd or initramfs had already run udevstart and created the "extra" nodes, so the script would not generate errors.

However, now that hotplug is in the unstable book, there is a problem: if the root filesystem is mounted read-write when the kernel mounts it (the first mount) AND a hotplug event occurs before this script runs (small window), then udev will get run by the hotplug scripts and this file will get created. The result is that when the script runs it does not do what is needed.
=====================================================

I understand that this is mostly LFS-specific, and the report can be in fact treated as invalid since we don't officially support the "rw" flag on the kernel command line, but I have a question.

Is this normal that a hotplug event somehow managed to fire up in that small time window? The udev initscript is the second one that actually runs in LFS, and it just mounts ramfs on /dev, calls udevstart and creates extra nodes in /dev like /dev/fd symlink. The first initscript is called mountkernfs and it just mounts /proc, /sys and probably /proc/bus/usb. Both bootscripts are attached. You see that none of them does things that can lead to a hotplug event, so the source of that bogus event is a mystery for me.

Another question is: how are other distros supposed to deal with such unexpected hotplug events, when the root filesystem is mounted read-only and /usr is not mounted at all?

Signature · 1 line
Alexander E. Patrakov

#!/bin/sh
# Begin $rc_base/init.d/functions - Run Level Control Functions

# Based on functions script from LFS-3.1 and earlier. # Rewritten by Gerard Beekmans - [email protected]

# With code based on Matthias Benkmann's simpleinit-msb @ # http://winterdrache.de/linux/newboot/index.html

# Number of seconds between term signal and kill signal when stopping processes KILLDELAY=3

umask 022
export PATH="/bin:/usr/bin:/sbin:/usr/sbin"

# Width of the Screen
COLUMNS=$(stty size)
COLUMNS=${COLUMNS##* }
# When using remote connections, such as a serial port, stty size returns 0 if [ "$COLUMNS" = "0" ]; then COLUMNS=80; fi

# Measurements for positioning result messages
COL=$(($COLUMNS - 10))
WCOL=$(($COLUMNS - 30))

# Set Cursur Position Commands, used via echo -e
SET_COL="\\033[${COL}G"
SET_WCOL="\\033[${WCOL}G"
CURS_UP="\\033[A"

# Set color commands, used via echo -e
NORMAL="\\033[0;39m"
SUCCESS="\\033[1;32m"
WARNING="\\033[1;33m"
FAILURE="\\033[1;31m"

echo_ok()

{
        echo -e "$CURS_UP$SET_COL"["$SUCCESS""  OK  ""$NORMAL"]
}

echo_failure()

{
        echo -e "$CURS_UP$SET_COL"["$FAILURE"FAILED"$NORMAL"]
}

echo_warning()

{
        echo -e "$CURS_UP$SET_WCOL$@$SET_COL"["$WARNING" WARN  "$NORMAL"]
}

# $i is inherited by the rc script
print_error_msg()
{
echo -e -n $FAILURE
echo
echo "You should not be reading this error message. It means" echo "that an unforseen error took place in $i," echo "which exited with a return value of $error_value" echo
echo "If you're able to track this error down to a bug in one" echo "of the files provided by the LFS book, please be so kind" echo "to inform us at [email protected]" echo -e -n $NORMAL
echo
echo
echo "Press Enter to continue..."
read ENTER
}

# $i is inherited by the rc script
check_script_status()
{
if [ ! -f $i ]
then
echo "$i is not a valid symlink"
continue
fi

if [ ! -x $i ]
then
echo "$i is not executable, skipping" continue
fi
}

evaluate_retval()

{
        error_value=$?

        if [ $error_value = 0 ]
        then
                print_status success
        else
                print_status failure
		sleep 5
        fi

        return 0
        #return $error_value
}

print_status()

{
        if [ $# = 0 ]
        then
                echo "Usage: $0 {success|warning|failure}"
                return 1
        fi

        case "$1" in
                success)
                        echo_ok
                ;;
                warning)
                        case "$2" in
                                running)
                                        echo_warning "Already running"
                                ;;
                                not_running)
                                        echo_warning "Not running"
                                ;;
                        esac
                ;;
                failure)
                        echo_failure
                ;;
        esac
}

# Returns all of the pid #'s for $1 process
getpids()

{
        base=${1##*/}
        local lpids=""
        local pid

        pidlist=""
        lpids=$(pidof $base)
        for pid in $lpids
        do
                if [ $pid -ne $$ ] && [ $pid -ne $PPID ]
                then
                        pidlist="$pidlist $pid"
                fi
        done
}

# Starts a program if it is currently not running
loadproc()

{
        if [ $# = 0 ]
        then
                echo "Usage: loadproc {program}"
                exit 1
        fi

	if [ -n "$PIDFILE" -a -r "$PIDFILE" ]; then
		pidlist=`cat $PIDFILE`
	else
		getpids $1
	fi

        if [ -z "$pidlist" ]
        then
                "$@"
                evaluate_retval
        else
                print_status warning running
        fi
}

# Stops a process if it is running
killproc()

{
        if [ $# = 0 ]
        then
                echo "Usage: killproc {program} [signal]"
                exit 1
        fi

        if [ -z "$2" ]; then
                signal=TERM
                fallback=KILL
        else
                signal=${2##-}
                signal=${signal##SIG}
                fallback=""
        fi

	if [ -n "$PIDFILE" -a -r "$PIDFILE" ]; then
		pidlist=`cat $PIDFILE`
	else
		getpids $1
	fi

        if [ -n "$pidlist" ]; then
                local i=0

                for pid in $pidlist
                do
                        kill -$signal $pid 2>/dev/null

                        while [ $i -lt $KILLDELAY ]; do
                                kill -0 $pid 2>/dev/null || break
                                sleep 1
                                i=$(($i+1))
                        done

                        if [ -n "$fallback" ]; then
				kill -$fallback $pid 2>/dev/null
			fi
                done

		getpids $1

                base=${1##*/}

                if [ -n "$pidlist" ]; then
			failure=1
		else
			failure=0
			if [ ! -z $PIDFILE ]; then
				rm -f $PIDFILE
			else
				rm -f /var/run/$base.pid
			fi
		fi

                (exit $failure)
                evaluate_retval
        else
                print_status warning not_running
        fi
}

reloadproc()

{
        if [ $# = 0 ]
        then
                echo "Usage: reloadproc {program} [signal]"
                exit 1
        fi

        if [ -z "$2" ]; then
                signal=HUP
        else
                signal=${2##-}
                signal=${signal##SIG}

        fi

        getpids $1

        if [ -n "$pidlist" ]
        then
                failure=0

                for pid in $pidlist
                do
                        kill -$signal $pid || failure=1
                done

                (exit $failure)
                evaluate_retval
        else
                print_status warning not_running
        fi
}

statusproc()
{
if [ $# = 0 ]
then
echo "Usage: statusproc {program}" exit 1
fi

base=${1##*/}
getpids $base

if [ -n "$pidlist" ]
then
echo "$base is running with Process ID(s) $pidlist"
else
if [ -s /var/run/$base.pid ]
then
echo "$1 is not running but /var/run/$base.pid exists" return 1
else
echo "$1 is not running"
fi
fi
}

# End $rc_base/init.d/functions

#!/bin/sh
# Begin $rc_base/init.d/mountkernfs

# Rewritten by Gerard Beekmans - [email protected]

. /etc/sysconfig/rc
. $rc_functions

case "$1" in
start)
if [ ! -e /proc/mounts ]; then
echo "Mounting proc file system..." mount -n /proc evaluate_retval
fi

# This should be mounted on a 2.6 kernel. if grep -q '[[:space:]]sysfs' /proc/filesystems; then
if [ -d /sys -a ! -d /sys/block ]; then
echo "Mounting sysfs file system..." mount -n /sys evaluate_retval
fi
fi

if grep -q '[[:space:]]usbfs' /proc/filesystems; then
echo "Mounting USB filesystem" mount -n /proc/bus/usb evaluate_retval
fi
;;
*)
echo "Usage $0 {start}"
exit 1
;;
esac

# End $rc_base/init.d/mountkernfs

#!/bin/sh
# Begin $rc_base/init.d/udev - Udev cold-plugging script

# Written by Zack Winkles - [email protected]

. /etc/sysconfig/rc
. $rc_functions

# Assure that sysfs is mounted and that udev is present. [ -d /sys/block -a -x /sbin/udev ] || exit 0

# Create some things that sysfs does not, and should not export for us. Feel

# free to add devices to this list.
make_extra_nodes() {
	ln -s /proc/self/fd /dev/fd
	ln -s /proc/self/fd/0 /dev/stdin
	ln -s /proc/self/fd/1 /dev/stdout
	ln -s /proc/self/fd/2 /dev/stderr
	ln -s /proc/kcore /dev/core
	mkdir /dev/pts
	mkdir /dev/shm
}

case "$1" in
start)
# When the hotplug package isn't installed, make sure that udev # still gets hotplug events.
if [ ! -x /sbin/hotplug ]; then
echo /sbin/udev > /proc/sys/kernel/hotplug
fi

# Don't attempt to populate the /dev directory when something # else has already set it up.
[ -f /dev/.udev.tdb ] && exit 0

echo "Populating /dev with device nodes..."

# Mount a temporary file system over /dev, so that any devices # made or removed during this boot don't affect the next one. # The reason we don't write to mtab is because we don't ever # want /dev to be unavailable (such as by `umount -a'). mount -n -t ramfs ramfs /dev

# Populate /dev with all the devices that are already available, # and save it's status so we can report failures. udevstart || failed=1

# Now, create some required files/directories/devices that sysfs # doesn't export for us.
make_extra_nodes

# When reporting the status, base it on the success or failure # of the `udevstart' command, since that's the most important. (exit $failed)
evaluate_retval
;;
*)
echo "Usage $0 {start}"
exit 1
;;
esac

# End $rc_base/init.d/udev

#1 lore raw
Quoted 3 lines · Alexander E. Patrakov

/proc/bus/usb. Both bootscripts are attached. You see that none of them does things that can lead to a hotplug event, so the source of that bogus event is a mystery for me.

Speaking as the original bug reporter, I'll jump in here: I am not suggesting that the unexpected hotplug event is bogus at all. Hotplug events are not only caused by the actions taken by the bootscripts, they can be caused by almost anything at all.

Consider the possibility that the user has a USB storage device that takes some time to respond to USB enumeration... the kernel could easily have finished its own initialization mounted the root filesystem read-write before this device is finally found and accepted by the USB storage driver. At the instant the driver binds to the device, one or more hotplug events will be generated. If the system's boot scripts have not yet had a chance to finish their work, the hotplug event will occur at an inopportune time and things can get weird.

The only solution I can see for this (well, proper solution that works in all cases I can think of) is to _not_ have the kernel default to calling /sbin/hotplug, but rather have the system bootscripts have to forcibly set /proc/sys/kernel/hotplug to that value. Alternatively, the /sbin/hotplug script from the hotplug package could be renamed after installation and then appropriate script could set /proc/sys/kernel/hotplug to /sbin/hotplug.real.

Either of these changes will get around this problem, because the boot scripts can handle the appropriate cold-plugging when ready, and they can very closely control when userspace will actually start to receive hotplug events.

-------------------------------------------------------
This SF.Net email is sponsored by: Oracle 10g
Get certified on the hottest thing ever to hit the market... Oracle 10g. Take an Oracle 10g class now, and we'll give you the exam FREE.
http://ads.osdn.com/?ad_id149&alloc_id�66&op=click
_______________________________________________
Linux-hotplug-devel mailing list http://linux-hotplug.sourceforge.net
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

#2 lore raw
In reply to Alexander E. Patrakov

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Quoted 4 lines

The only solution I can see for this (well, proper solution that works in all cases I can think of) is to _not_ have the kernel default to calling /sbin/hotplug, but rather have the system bootscripts have to forcibly set /proc/sys/kernel/hotplug to that value.

I absolutely agree. The initscripts should have a way to tell the kernel that the system is ready to handle hotplug events. We DO have such a way already, that's the /proc filesystem, exactlier, /proc/sys/kernel/hotplug. It were nice to have it disabled by default. No problem to add "echo /sbin/hotplug" somewhere in the initscripts.

If you are afraid of backward compatibility, one could add one more configuration option to the kernel, like that:
[x] /proc/sys/kernel/hotplug defaults to /sbin/hotplug [DEPRECATED] I suggest to make it deprecated in order to avoid any possible misreadings. IMHO, the user always has the right to misconfigure his system :)

Quoted 4 lines

Alternatively, the
/sbin/hotplug script from the hotplug package could be renamed after installation and then appropriate script could set /proc/sys/kernel/hotplug to /sbin/hotplug.real.

Hmmm... not a very nice kludge... in fact, this heavily raises the chances for nonintentional misconfiguration. However, this kludge may work for now for the users who do not want to wait for next kernel release.

IMHO, for this hack, /sbin/hotplug should be a script that calls real /sbin/hotplug.real if the system is ready in a distribution-specific manner (i.e., reads /proc/sys/kernel/hotplug and checks if it is set to /sbin/hotplug.real). It is necessary because some scripts may rely on the existance of /sbin/hotplug. Again, don't do that if you can patch the kernel instead.

Regards,
Alex
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (GNU/Linux)

iD8DBQFAvEWOaA+H7vY+1kURAoXdAJ4yCOn63hf8Hcx/rPYkONhYmfrYwQCeNhC5 hfwvZ3rkbflFhQnu0KM+Qtc=M+mD
-----END PGP SIGNATURE-----

-------------------------------------------------------
This SF.Net email is sponsored by: Oracle 10g
Get certified on the hottest thing ever to hit the market... Oracle 10g. Take an Oracle 10g class now, and we'll give you the exam FREE.
http://ads.osdn.com/?ad_id149&alloc_id�66&op�k
_______________________________________________
Linux-hotplug-devel mailing list http://linux-hotplug.sourceforge.net
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel