udev: Directory for custom device nodes.

[Harald Hoyer]

Due to
https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=144593
I am thinking of another directory to put device nodes in, which should be 
copied over to /dev on system start.

Any suggestions?

[Christof Damian]

On Mon, 10 Jan 2005, Harald Hoyer wrote:
>  Due to
>  https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=144593
>  I am thinking of another directory to put device nodes in, which should be 
>  copied over to /dev on system start.
>  
>  Any suggestions?
>   
Maybe there shouldn't be device files at all, but config files with
mknod parameters.

christof

-- 
Christof Damian         
christof(a)damian.net

[Harald Hoyer]

Christof Damian wrote:
>  On Mon, 10 Jan 2005, Harald Hoyer wrote:
>  
> >Due to
> >https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=144593
> >I am thinking of another directory to put device nodes in, which should be 
> >copied over to /dev on system start.
> >
> >Any suggestions?
> >
>  
>  
>  Maybe there shouldn't be device files at all, but config files with
>  mknod parameters. 
For that, there is /etc/makedev.d already. :)

[Felipe Alfaro Solana]

On 10 Jan 2005, at 11:53, Harald Hoyer wrote:

>  Due to
>  https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=144593
>  I am thinking of another directory to put device nodes in, which 
>  should be copied over to /dev on system start.
>
>  Any suggestions? 
"/udev/devices" or "/udev/custom" sounds great.

[Harald Hoyer]

Felipe Alfaro Solana wrote:
>  On 10 Jan 2005, at 11:53, Harald Hoyer wrote:
>  
> > Due to
> > https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=144593
> > I am thinking of another directory to put device nodes in, which 
> > should be copied over to /dev on system start.
> >
> > Any suggestions?
>  
>  
>  "/udev/devices" or "/udev/custom" sounds great.
>   
I really don't want to start a new toplevel directory...

[Bill Nottingham]

Harald Hoyer (harald(a)redhat.com) said: 
>  Due to
>  https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=144593
>  I am thinking of another directory to put device nodes in, which should 
>  be copied over to /dev on system start.
>  
>  Any suggestions? 
I'd think that a series of makedev invocations to run would be
safer.

Bill

[Harald Hoyer]

Bill Nottingham wrote:
>  Harald Hoyer (harald(a)redhat.com) said: 
>  
> >Due to
> >https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=144593
> >I am thinking of another directory to put device nodes in, which should 
> >be copied over to /dev on system start.
> >
> >Any suggestions?
>  
>  
>  I'd think that a series of makedev invocations to run would be
>  safer.
>  
>  Bill
>   
so how can we make this rpm friendly?
/etc/sysconfig/makedev.d/
	system.nodes
	custom1.nodes
	custom2.nodes

cat /etc/sysconfig/makedev.d/*.nodes | xargs MAKEDEV -x

doh, xargs is in /usr/bin ! :-/

hmm, btw is /etc/makedev.d '.rpmorig' '.rpmsave' save?

[Nils Philippsen]

On Thu, 2005-01-13 at 11:57 +0100, Harald Hoyer wrote:
>  Bill Nottingham wrote:
>  > Harald Hoyer (harald(a)redhat.com) said: 
>  > 
>  >>Due to
>  >>https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=144593
>  >>I am thinking of another directory to put device nodes in, which should 
>  >>be copied over to /dev on system start.
>  >>
>  >>Any suggestions?
>  > 
>  > 
>  > I'd think that a series of makedev invocations to run would be
>  > safer.
>  > 
>  > Bill
>  > 
>  
>  so how can we make this rpm friendly?
>  /etc/sysconfig/makedev.d/
>  	system.nodes
>  	custom1.nodes
>  	custom2.nodes
>  
>  cat /etc/sysconfig/makedev.d/*.nodes | xargs MAKEDEV -x
>  
>  doh, xargs is in /usr/bin ! :-/ 
how about this "xargs for the poor" (could be changed a bit and put
into /etc/init.d/functions):

--- 8< ---
FILES=/etc/sysconfig/makedev.d/*.nodes
MAXNR=100
COMMAND="MAKEDEV -x"

NR=$MAXNR
ARGS=""
cat $FILES | while read line; do
    if [ "$NR" -gt 0 ]; then
        ARGS="$ARGS $line"
        NR=$[$NR - 1]
    else
        $COMMAND $ARGS
        NR=$MAXNR
        ARGS="$line"
    fi
done
if [ -n "$ARGS" ]; then
    $COMMAND $ARGS
fi
--- >8 ---

Nils
-- 
     Nils Philippsen    /    Red Hat    /    nphilipp(a)redhat.com
"They that can give up essential liberty to obtain a little temporary
 safety deserve neither liberty nor safety."     -- B. Franklin, 1759
 PGP fingerprint:  C4A8 9474 5C4C ADE3 2B8F  656D 47D8 9B65 6951 3011

[Harald Hoyer]

Nils Philippsen wrote:
>  how about this "xargs for the poor" (could be changed a bit
> and put
>  into /etc/init.d/functions):
>  
>  --- 8< ---
>  FILES=/etc/sysconfig/makedev.d/*.nodes
>  MAXNR=100
>  COMMAND="MAKEDEV -x"
>  
>  NR=$MAXNR
>  ARGS=""
>  cat $FILES | while read line; do
>      if [ "$NR" -gt 0 ]; then
>          ARGS="$ARGS $line"
>          NR=$[$NR - 1]
>      else
>          $COMMAND $ARGS
>          NR=$MAXNR
>          ARGS="$line"
>      fi
>  done
>  if [ -n "$ARGS" ]; then
>      $COMMAND $ARGS
>  fi
>  --- >8 ---
>  
>  Nils 
what about this one? :)
for i in /etc/sysconfig/makedev.d/*.nodes; do
	MAKEDEV -x $(cat "$i")
done

[Harald Hoyer]

ok, this is what I came up with for FC4:

xargs_simple () {
         if [ "$1" = "-n" ]; then
                 shift
                 MAXNR="$1"
                 shift
         else
                 MAXNR=100
         fi
         NR=$MAXNR
         ARGS=""
         [ -z "$1" ] && set echo

         while read line; do
                 if [ $NR -gt 0 ]; then
                         ARGS="$ARGS $line"
                         NR=$[$NR - 1]
                 else
                         "$@" $ARGS
                         NR=$MAXNR
                         ARGS="$line"
                 fi
         done
         if [ -n "$ARGS" ]; then
                 "$@" $ARGS
         fi
}

make_extra_nodes () {
         ln -snf /proc/self/fd $udev_root/fd
         ln -snf /proc/self/fd/0 $udev_root/stdin
         ln -snf /proc/self/fd/1 $udev_root/stdout
         ln -snf /proc/self/fd/2 $udev_root/stderr
         ln -snf /proc/kcore $udev_root/core

         [ -d $udev_root/pts ] || mkdir -m 0755 $udev_root/pts
         [ -d $udev_root/shm ] || mkdir -m 0755 $udev_root/shm

         if [ -x $MAKEDEV ]; then
                 for i in /etc/sysconfig/makedev.d/*.nodes; do
                         cat "$i" | sed -e 's,#.*,,g' | \
                                 xargs_simple -n 100 $MAKEDEV -x
                 done
         fi
         for i in /etc/sysconfig/makedev.d/*.symlinks; do
                 cat "$i" | sed -e 's,#.*,,g' | xargs_simple -n 1 ln
-snv
         fi
}


This should MAKEDEV the nodes from /etc/sysconfig/makedev.d/*.nodes
# cat /etc/sysconfig/makedev.d/50-udev.nodes | \
>  sed -e 's,#.*,,g' | xargs_simple echo 
tty1 tty2 tty3
tty4 tty5 tty6 loop0 loop1 loop2 loop3 loop4 loop5 loop6 loop7 
lp0 lp1 lp2 lp3 parport0 parport1 parport2 parport3 net/tun ppp console null zero

The file can contain comments with "#". Also symlinks described in 
/etc/sysconfig/makedev.d/*.symlinks are created.

Packages can drop their own files in /etc/sysconfig/makedev.d and the 
definition of the nodes in /etc/makedev.d/.

Are the names fine or should we rename /etc/sysconfig/makedev.d in s.th. like 
/etc/sysconfig/init-makedev.d ??

[Féliciano Matias]

Le mardi 25 janvier 2005 à 12:45 +0100, Harald Hoyer a écrit :
>  ok, this is what I came up with for FC4:
>  
>  (snip)
>  
>  This should MAKEDEV the nodes from /etc/sysconfig/makedev.d/*.nodes
>  # cat /etc/sysconfig/makedev.d/50-udev.nodes | \
>   > sed -e 's,#.*,,g' | xargs_simple echo
>  tty1 tty2 tty3 tty4 tty5 tty6 loop0 loop1 loop2 loop3 loop4 loop5 loop6 loop7 
>  lp0 lp1 lp2 lp3 parport0 parport1 parport2 parport3 net/tun ppp console null zero
>  
>  The file can contain comments with "#". Also symlinks described in 
>  /etc/sysconfig/makedev.d/*.symlinks are created.
>  
>  Packages can drop their own files in /etc/sysconfig/makedev.d and the 
>  definition of the nodes in /etc/makedev.d/.
>   
Nice.

>  Are the names fine or should we rename /etc/sysconfig/makedev.d in
> s.th. like 
>  /etc/sysconfig/init-makedev.d ?? 
I prefer /etc/sysconfig/init-makedev.d.

[Rahul Sundaram]

On Tue, 25 Jan 2005 12:45:41 +0100, Harald Hoyer <harald(a)redhat.com&gt; wrote:
>  ok, this is what I came up with for FC4: 
I am not really fond of moving stuff like into the already crowded
/etc/sysconfig. can you please do something thats upstream compatible

-- 
Regards,
Rahul Sundaram

[Harald Hoyer]

Rahul Sundaram wrote:
>  On Tue, 25 Jan 2005 12:45:41 +0100, Harald Hoyer
> <harald(a)redhat.com&gt; wrote:
>  
> >ok, this is what I came up with for FC4:
>  
>  
>  I am not really fond of moving stuff like into the already crowded
>  /etc/sysconfig. can you please do something thats upstream compatible 
like what? upstream has no mechanism for manual device creation.

[Féliciano Matias]

Le mardi 25 janvier 2005 à 12:45 +0100, Harald Hoyer a écrit :
>  ok, this is what I came up with for FC4:
>  (...)
>   Also symlinks described in 
>  /etc/sysconfig/makedev.d/*.symlinks are created. 
Could it be possible to have symlinks in the same file ?

[Harald Hoyer]

Féliciano Matias wrote:
>  Le mardi 25 janvier 2005 à 12:45 +0100, Harald Hoyer a écrit :
>  
> >ok, this is what I came up with for FC4:
> >(...)
> > Also symlinks described in 
> >/etc/sysconfig/makedev.d/*.symlinks are created.
>  
>  
>  Could it be possible to have symlinks in the same file ? 
hmm, this would require parsing the file... but we could remove the symlinks, 
cause MAKEDEV understands symlinks like this:

# fgrep rfcomm0 /etc/makedev.d/redhat
l rfcomm0 ttyUB0
# MAKEDEV -x rfcomm0
# ll /dev/rfcomm0
lrwxrwxrwx  1 root root 6 25. Jan 13:33 /dev/rfcomm0 -> ttyUB0

[Féliciano Matias]

Le mardi 25 janvier 2005 à 12:45 +0100, Harald Hoyer a écrit :
>  Are the names fine or should we rename /etc/sysconfig/makedev.d in
> s.th. like 
>  /etc/sysconfig/init-makedev.d ?? 

/etc/sysconfig/makedev.d (or init-makedev.d) does not store things like
found in /etc/makedev.d/ .

proposition :
/etc/sysconfig/init-dev.d


Another point, sysinit does not provide a way to load a module at boot
time (other than using /etc/rc.d/rc.local).

For example, I use this driver (with mplayer) :
http://www.linuxops.net/~pw/mga_vid/

This module does not support auto-loading.
A /etc/sysconfig/init-dev.d/*.module would be appreciated.


Last, what about /etc/modprobe.conf.d by default ?

Modprobe support "include /etc/modprobe.conf.d" with /etc/modprobe.conf.

There is one piece missing "in my dream" :
 /etc/security/console.perms.d/mga_vid
where I can add "<v4l>=/dev/mga_vid".

Suppose I want to package the mga_vid driver, I'll use :
/etc/makedev.d/mga_vid      # /dev/mga_vid definition
/etc/sysconfig/init-dev.d/mga_vid.dev    # create /dev/mga_vid
/etc/sysconfig/init-dev.d/mga_vid.module   # load mga_vid
/etc/modprobe.conf.d/mga_vid.conf    # configure mga_vid
/etc/security/console.perms.d/mga_vid    # pam_console configuration

[Harald Hoyer]

Féliciano Matias wrote:
>  Le mardi 25 janvier 2005 à 12:45 +0100, Harald Hoyer a écrit :
>  
> >Are the names fine or should we rename /etc/sysconfig/makedev.d in s.th. like 
> >/etc/sysconfig/init-makedev.d ??
>  
>  
>  
>  /etc/sysconfig/makedev.d (or init-makedev.d) does not store things like
>  found in /etc/makedev.d/ .
>  
>  proposition :
>  /etc/sysconfig/init-dev.d
>  
>  
>  Another point, sysinit does not provide a way to load a module at boot
>  time (other than using /etc/rc.d/rc.local).
>  
>  For example, I use this driver (with mplayer) :
>  http://www.linuxops.net/~pw/mga_vid/
>  
>  This module does not support auto-loading.
>  A /etc/sysconfig/init-dev.d/*.module would be appreciated. 
Hmm, there is rc.modules mentioned in rc.sysinit, but I think that is deprecated.

>  
>  
>  Last, what about /etc/modprobe.conf.d by default ?
>  
>  Modprobe support "include /etc/modprobe.conf.d" with /etc/modprobe.conf.
>  
>  There is one piece missing "in my dream" :
>   /etc/security/console.perms.d/mga_vid
>  where I can add "<v4l>=/dev/mga_vid".
>  
>  Suppose I want to package the mga_vid driver, I'll use :
>  /etc/makedev.d/mga_vid      # /dev/mga_vid definition
>  /etc/sysconfig/init-dev.d/mga_vid.dev    # create /dev/mga_vid
>  /etc/sysconfig/init-dev.d/mga_vid.module   # load mga_vid
>  /etc/modprobe.conf.d/mga_vid.conf    # configure mga_vid
>  /etc/security/console.perms.d/mga_vid    # pam_console configuration
>   

Right... that would be cool.

[Féliciano Matias]

Le mardi 25 janvier 2005 à 14:47 +0100, Harald Hoyer a écrit :
>  
>  Hmm, there is rc.modules mentioned in rc.sysinit, but I think that is deprecated.
>  
/etc/rc.d/rc.sysinit (FC3) :
        # Load modules (for backward compatibility with VARs)
        if [ -f /etc/rc.modules ]; then
                /etc/rc.modules
        fi


$ ls /etc/rc.modules
ls: /etc/rc.modules: No such file or directory

It's not better than /etc/rc.d/rc.local .

[Dan Williams]

On Tue, 2005-01-25 at 14:38 +0100, Féliciano Matias wrote:
>  Suppose I want to package the mga_vid driver, I'll use :
>  /etc/makedev.d/mga_vid      # /dev/mga_vid definition
>  /etc/sysconfig/init-dev.d/mga_vid.dev    # create /dev/mga_vid
>  /etc/sysconfig/init-dev.d/mga_vid.module   # load mga_vid
>  /etc/modprobe.conf.d/mga_vid.conf    # configure mga_vid
>  /etc/security/console.perms.d/mga_vid    # pam_console configuration 
Does it seem utterly _pathetic_ to anyone else but me that there would
have to be 5 different places to add information about a particular
hardware device, just to get it to load and work?  Just a point of
observation.

Dan

[Féliciano Matias]

Le mardi 25 janvier 2005 à 09:49 -0500, Dan Williams a écrit :
>  On Tue, 2005-01-25 at 14:38 +0100, Féliciano Matias wrote:
>  > Suppose I want to package the mga_vid driver, I'll use :
>  > /etc/makedev.d/mga_vid      # /dev/mga_vid definition
>  > /etc/sysconfig/init-dev.d/mga_vid.dev    # create /dev/mga_vid
>  > /etc/sysconfig/init-dev.d/mga_vid.module   # load mga_vid
>  > /etc/modprobe.conf.d/mga_vid.conf    # configure mga_vid
>  > /etc/security/console.perms.d/mga_vid    # pam_console configuration
>  
>  Does it seem utterly _pathetic_ to anyone else but me that there would
>  have to be 5 different places to add information about a particular
>  hardware device, just to get it to load and work?  Just a point of
>  observation. 
Do you prefer hacks and "Ooops" in order to
configure /etc/modprobe.conf, /etc/rc.d/rc.local, /etc/security/console.perms ?

If the driver support udev, doesn't need configuration, do not need
pam_console, can be auto-loaded, ... all this stuff is useless.

But sometimes...
You know, the "real world".

>  On Tue, 2005-01-25 at 14:38 +0100, Féliciano Matias wrote:
>  > Suppose I want to package the mga_vid driver, I'll use : 
Now suppose /etc/modprobe.conf.d/, /etc/sysconfig/init-dev.d/, ... exist
and I created a rpm package for mga_vid.

Just install it and enjoy.

[Rahul Sundaram]

HI
>  
>  Now suppose /etc/modprobe.conf.d/, /etc/sysconfig/init-dev.d/, ... exist
>  and I created a rpm package for mga_vid. 
replace /etc/sysconfig/init-dev.d/ with /etc/udev/init-dev.d  and this
looks more good to me

-- 
Regards,
Rahul Sundaram

[Féliciano Matias]

Le mardi 25 janvier 2005 à 21:26 +0530, Rahul Sundaram a écrit :
>  HI
>  > 
>  > Now suppose /etc/modprobe.conf.d/, /etc/sysconfig/init-dev.d/, ... exist
>  > and I created a rpm package for mga_vid.
>  
>  replace /etc/sysconfig/init-dev.d/ with /etc/udev/init-dev.d  and this
>  looks more good to me
>   
/etc/udev/init-dev.d is /etc/udev/devices/ and it already exist.
But it "sucks".

/etc/sysconfig/init-dev.d/ will be used by start_udev (AFAIK) and
start_udev is called at boot time by /etc/rc.d/rc.sysinit. So for
me /etc/sysconfig/... is the right place (until there is something
upstream).

Note that udev (the daemon) does not use /sbin/start_udev
nor /etc/udev/devices/ . These files do not configure udev.

[Bill Nottingham]

Féliciano Matias (feliciano.matias(a)free.fr) said: 
>  There is one piece missing "in my dream" :
>   /etc/security/console.perms.d/mga_vid
>  where I can add "<v4l>=/dev/mga_vid". 
This will get done with HAL in the future, FWIW.

Bill

[Féliciano Matias]

Le mardi 25 janvier 2005 à 09:50 -0500, Bill Nottingham a écrit :
>  Féliciano Matias (feliciano.matias(a)free.fr) said: 
>  > There is one piece missing "in my dream" :
>  >  /etc/security/console.perms.d/mga_vid
>  > where I can add "<v4l>=/dev/mga_vid".
>  
>  This will get done with HAL in the future, FWIW.
>   
You mean HAL will replace or work with pam_console ?

Just an example. With mga_vid and a kernel with framebuffer enabled, I
can login in a console (not X11) and launch :
$ mplayer -vo mga dvd://

With X11 I use "mplayer -vo xmga ..."

[Ville Skyttä]

On Tue, 2005-01-25 at 09:50 -0500, Bill Nottingham wrote:
>  Féliciano Matias (feliciano.matias(a)free.fr) said: 
>  > There is one piece missing "in my dream" :
>  >  /etc/security/console.perms.d/mga_vid
>  > where I can add "<v4l>=/dev/mga_vid".
>  
>  This will get done with HAL in the future, FWIW. 
Excellent.  Any estimates to what "future" in this context expands to?

[Michael A. Peters]

On 01/10/2005 02:53:49 AM, Harald Hoyer wrote:
>  Due to
>  https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=144593 
That bug is very similar to one I filed that was closed because dmesg  
showed the grep segfault was due to the NVidia drivers.

I'm willing to bet that should the reporter disable the nvidia driver,  
his grep would work perfectly - even with the nvidia modules still in / 
etc/udev/devices.

I'll look for my bug number for you to refer to.

[Michael A. Peters]

On 01/11/2005 06:07:19 PM, Michael A. Peters wrote:
>  On 01/10/2005 02:53:49 AM, Harald Hoyer wrote:
> > Due to
> > https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=144593
>  
>  That bug is very similar to one I filed that was closed because dmesg  
>  showed the grep segfault was due to the NVidia drivers. 
I can't find the one I reported, I've tried bugzilla parameters that  
should imho show me closed bugs in grep, but it says zaro bugs found,  
so I'm not searching bugzilla right apparently - because I _know_ it  
was there (I filed it), I _know_ it was closed as duplicate of general  
nvidia bug, I _know_ that the issue was spinlock, and that dmesg showed  
it was nvidia's driver to blame. But I can't find it in bugzilla.

If you know how to search bugzilla better than me, it was reported  
against grep from my e-mail address < mpeters(a)mac.com >

[Michael A. Peters]

On 01/11/2005 06:07:19 PM, Michael A. Peters wrote:

>  
found it -

https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=141294