[Ubi] Wrong major:minor

[Ubi] Wrong major:minor

[Gilles]

Hello

I used the following data to compile Ubi(fs) into a uClinux kernel:

========= device_table.txt
#UBIFS nodes
/dev/ubi_ctrl	c	644	0	0	10	63	0	0	-
/dev/ubi0	c	644	0	0	253	0	0	0	-
========= 

However, ubimkvol isn't happy with it:
========= 
root:~> ubimkvol /dev/ubi0 -s 20MiB -N ubifs0

libubi: error!: "/dev/ubi0" has major:minor 253:0, but this does not
correspond to any UBI device or volume

ubimkvol: error!: "/dev/ubi0" is not an UBI device node
========= 

Here's what the two device nodes look like in the running kernel:
========= 
root:~> ls -al /dev/ubi*
crw-r--r--    1 root     root      253,   0 May 27  2011 /dev/ubi0
crw-r--r--    1 root     root       10,  63 May 27  2011 /dev/ubi_ctrl
========= 

I'm no Linux expert, and was wondering how to find out what the
correct major:minor is for the Ubi device nodes? Are they defined
somewhere in the Linux source code?

Thank you.

Re: [Ubi] Wrong major:minor

[Gilles]

On Sun, 29 May 2011 12:39:33 +0200, Gilles <gilles.ganault@free.fr>
wrote:
>I'm no Linux expert, and was wondering how to find out what the
>correct major:minor is for the Ubi device nodes? Are they defined
>somewhere in the Linux source code?

I have a couple more newbie questions:

1. A tutorial mentionned using ubinfo, ubiformat, ubiattach, ubimkvol,
and finally mount: Why do we need all those commands to use NAND
memory from uClinux?

2. Am I correct in understanding that the stack looks like this?
_______________________
| Linux
|_______________________
|UBIFS
|_______________________
|UBI
|_______________________
|MTD
|_______________________
|Hardware NAND memory
|_______________________

Thank you.

Re: [Ubi] Wrong major:minor

[Mike Frysinger]

On Sun, May 29, 2011 at 06:39, Gilles wrote:
> I'm no Linux expert, and was wondering how to find out what the
> correct major:minor is for the Ubi device nodes?

they're most likely dynamic and there's no safe way you could create
the nodes ahead of time with a device table.  simply run mdev to take
care of /dev for you.

you can check /proc/devices to be sure ... anything over 250 is dynamic
-mike

Re: [Ubi] Wrong major:minor

[Gilles]

On Sun, 29 May 2011 11:54:36 -0400, Mike Frysinger
<vapier.adi@gmail.com> wrote:
>they're most likely dynamic and there's no safe way you could create
>the nodes ahead of time with a device table.  simply run mdev to take
>care of /dev for you.
>
>you can check /proc/devices to be sure ... anything over 250 is dynamic

Thanks for the infos.

I rebuilt uClinux with Busybox's mdev and added an rc.d script +
symlink. The dev nodes are created, but /dev/ubi0 is missing:
============
>  ubinfo /dev/ubi0 -a
libubi: error!: cannot get information about "/dev/ubi0"
        error 2 (No such file or directory)
ubinfo: error!: unrecognized device node "/dev/ubi0"
============

Here's what I did:

1. make config_menuconfig
    
    Miscellaneous Configuration
        Device Nodes (Dynamic)
    
    BusyBox
    [*] mdev
    [*]   Support /etc/mdev.conf
    [ ]     Support subdirs/symlinks (NEW)
    [ ]     Support command execution at device addition/removal (NEW)
    [ ]   Support loading of firmwares (NEW)

2. vi /usr/src/uClinux-dist-2010R1-RC5/vendors/Rowetel/IP04/mdev
    
    #!/bin/sh
    # Start up file for mdev
    
    case $1 in
            start) echo "Starting mdev...";
            #BAD echo /sbin/mdev > /proc/sys/kernel/hotplug;
            /sbin/mdev -s;
            ;; 
    
            stop) killall -9 mdev;;
            enable) rm -f /etc/rc.d/S20mdev; 
            ln -s /etc/init.d/mdev /etc/rc.d/S20mdev;;
            disable) rm -f /etc/rc.d/S20mdev;;
    
            *) cat <<EOF;;
    
    Syntax: /etc/init.d/mdev [command]
    Available commands:
            start   Start the service
            stop    Stop the service
            enable  Enable service autostart
            disable Disable service autostart
    EOF
    esac

3. vi /usr/src/uClinux-dist-2010R1-RC5/vendors/Rowetel/IP04/Makefile
    
    #Added
    $(ROMFSINST) /etc/init.d/mdev
    $(ROMFSINST) -s /etc/init.d/mdev /etc/rc.d/S20mdev

After rebooting, here are the relevant items in /dev:

crw-rw----    1 root     root       90,   0 Dec 31 16:00 mtd0
crw-rw----    1 root     root       90,   1 Dec 31 16:00 mtd0ro
crw-rw----    1 root     root       90,   2 Dec 31 16:00 mtd1
crw-rw----    1 root     root       90,   3 Dec 31 16:00 mtd1ro
crw-rw----    1 root     root       90,   4 Dec 31 16:00 mtd2
crw-rw----    1 root     root       90,   5 Dec 31 16:00 mtd2ro
brw-rw----    1 root     root       31,   0 Dec 31 16:00 mtdblock0
brw-rw----    1 root     root       31,   1 Dec 31 16:00 mtdblock1
brw-rw----    1 root     root       31,   2 Dec 31 16:00 mtdblock2
crw-rw----    1 root     root       10,  63 Dec 31 16:00 ubi_ctrl

>  cat /proc/mtd
dev:    size   erasesize  name
mtd0: 00800000 00001000 "ROMfs"
mtd1: 00800000 00020000 "linux kernel(nand)"
mtd2: 0f800000 00020000 "file system(nand)"

Any idea why mdev doesn't create /dev/ubi0?

Thanks.

Re: [Ubi] Wrong major:minor

[Gilles]

On Mon, 30 May 2011 11:53:44 +0200, Gilles <gilles.ganault@free.fr>
wrote:
>Any idea why mdev doesn't create /dev/ubi0?

Made some progress: To get /dev/ubi0, it's required to attach, and
_then_ rerun "mdev":
=================
> ubiattach /dev/ubi_ctrl -m 2

> mdev -s
mdev: /sys/class/block/mtdblock0/bdi: No such file or directory
mdev: /sys/class/block/mtdblock1/bdi: No such file or directory
mdev: /sys/class/block/mtdblock2/bdi: No such file or directory

> ls -al /dev/ubi*
crw-rw----    1 root     root      251,   0 Dec 31 16:14 /dev/ubi0
crw-rw----    1 root     root      251,   1 Dec 31 16:14 /dev/ubi0_0
crw-rw----    1 root     root       10,  63 Dec 31 16:00 /dev/ubi_ctrl

> ubimkvol /dev/ubi0 -m -N ubifs0
Set volume size to 253016064
mtd: Giving out device 3 to ubifs0
Volume ID 0, size 1961 LEBs (253016064 bytes, 241.3 MiB), LEB size
129024 bytes (126.0 KiB), dynamic, name "ubifs0", alignment 1

> cat /proc/mtd
dev:    size   erasesize  name
mtd0: 00800000 00001000 "ROMfs"
mtd1: 00800000 00020000 "linux kernel(nand)"
mtd2: 0f800000 00020000 "file system(nand)"
mtd3: 0f14b800 0001f800 "ubifs0"
=================

I have a couple of questions:

1. I tried populating the mtd2/ubifs0 partition with rootfs.ubifs from
the workstation, which failed:

> cd /var/tmp ; wget -c http://workstation/roots.ubifs
> ubiupdatevol /dev/ubi0_0 /var/tmp/rootfs.ubifs
> mount -t ubifs ubi0:ubifs0 /mnt
UBIFS error (pid 87): validate_sb: LEB size mismatch: 204800 in
superblock, 129024 real
UBIFS error (pid 87): validate_sb: bad superblock, error 1
mount: mounting ubi0:ubifs0 on /mnt failed: Invalid argument

So I reformated the partition, and tried with rootfs.ext2:

> ubidetach /dev/ubi_ctrl -m 2
> ubiformat /dev/mtd2 -s 512 -O 512
> ubiattach /dev/ubi_ctrl -m 2
> ubimkvol /dev/ubi0 -m -N ubifs0
> ubiupdatevol /dev/ubi0_0 /var/tmp/rootfs.ext2
> mount -t ubifs ubi0:ubifs0 /mnt
UBIFS error (pid 108): ubifs_read_node: bad node type (0 but expected
6)
UBIFS error (pid 108): ubifs_read_node: bad node at LEB 0:0
mount: mounting ubi0:ubifs0 on /mnt failed: Invalid argument

2. Since the appliance is a closed-box, users can't plug anything
anyway: In this case, besides taking more time to detect devices and
creating the relevant dev nodes, are there drawbacks about using mdev
+ dynamic dev nods over a static list of dev nodes?

3. Since the root file system lives in mtd2, which is now formatted as
Ubifs, why does ubimkvol create a third mtd partition (mtd3)? Will
writing to mtd3 write actually data to mtd2 (for persistence)?

Thank you.

Re: [Ubi] Wrong major:minor

[Gilles]

On Mon, 30 May 2011 14:06:57 +0200, Gilles <gilles.ganault@free.fr>
wrote:
>> ubiupdatevol /dev/ubi0_0 /var/tmp/rootfs.ubifs
>> mount -t ubifs ubi0:ubifs0 /mnt
>UBIFS error (pid 87): validate_sb: LEB size mismatch: 204800 in
>superblock, 129024 real
>UBIFS error (pid 87): validate_sb: bad superblock, error 1
>mount: mounting ubi0:ubifs0 on /mnt failed: Invalid argument

After more googling, the issue was wrong numbers between the original
root image, the Ubi image, and the NAND partition.

But I'm a bit lost as to how to harmonize settings in mkfs.ubifs,
ubi.cf/ubinize, and ubiformat.

Things work OK until I run "ubimkvol":
==================
1. Turn rootfs.ubifs into ubi.img:

> vi ubi.cfg
[ubifs]
mode=ubi
image=./images/rootfs.ubifs
vol_id=0
#"vol_size field must be larger than UBIFS image,
#but smaller than total flash space"
;4.915.200 images/rootfs.ubifs
;
;mtd2: 0f800000 00020000 "file system(nand)"
;f800000 = 260.046.848
vol_size=200MiB
vol_type=dynamic
vol_name=rootfs
vol_flags=autoresize

> ubinize -o ubi.img -m 2048 -p 128KiB -s 512 ubi.cfg

> mv ubi.img /var/www/

2. Download ubi.img from appliance:

> wget -c http://workstation/ubi.img

3. Format NAND mtd2 with ubi.img

> ubiformat /dev/mtd2 -s 512 -f ubi.img

4. Attach mtd2 to Ubi:
           
> ubiattach /dev/ubi_ctrl -m 2

5. Ask mdev to create ubi0 dev node:

> mdev -s
> ls -al /dev/ubi*

6. Create Ubi volume:

> ubimkvol /dev/ubi0 -m -N ubifs0
ubimkvol: error!: UBI device does not have free logical eraseblocks
==================

Thank you.

Re: [Ubi] Wrong major:minor

[Mike Frysinger]

On Mon, May 30, 2011 at 05:53, Gilles wrote:
>            #BAD echo /sbin/mdev > /proc/sys/kernel/hotplug;

i dont know why you've disabled this.  if you want dynamic nodes after
boot, you need it.
-mike

Re: [Ubi] Wrong major:minor

[Artem Bityutskiy]

On Sun, 2011-05-29 at 11:54 -0400, Mike Frysinger wrote:
> On Sun, May 29, 2011 at 06:39, Gilles wrote:
> > I'm no Linux expert, and was wondering how to find out what the
> > correct major:minor is for the Ubi device nodes?
> 
> they're most likely dynamic and there's no safe way you could create
> the nodes ahead of time with a device table.  simply run mdev to take
> care of /dev for you.
> 
> you can check /proc/devices to be sure ... anything over 250 is dynamic
> -mike

Right, UBI allocates device numbers dynamically, so pre-creating device
nodes is not a good idea.

Major/minor for every UBI device / UBI volume can be found at:

/sys/class/ubi/ubiX/dev
/sys/class/ubi/ubiX_Y/dev

Or the major number can be indeed found at /proc/devices, and then the
minor numbers follow the following fomula:

UBI Volume major number = UBI device major number.
UBI Volume minor number = UBI volume ID + 1.

-- 
Best Regards,
Artem Bityutskiy (Артём Битюцкий)

Re: [Ubi] Wrong major:minor

[Artem Bityutskiy]

On Sun, 2011-05-29 at 12:39 +0200, Gilles wrote:
> Hello
> 
> I used the following data to compile Ubi(fs) into a uClinux kernel:
> 
> ========= device_table.txt
> #UBIFS nodes
> /dev/ubi_ctrl	c	644	0	0	10	63	0	0	-
> /dev/ubi0	c	644	0	0	253	0	0	0	-
> ========= 

But I guess you may add an UBI module parameter which will ask UBI to
use major numbers you want - should be easy to do. E.g., if you add a
'major_base=200' parameter then UBI will use 200:0 for ubi0, 201:0 for
ubi1, 200:1 for ubi0_0, 200:2 for ubi0_1, etc.

Then you could do:

modprobe ubi major_base=200

or add

ubi.major_base=200

to you kernel parameters if you have UBI compiled-in.
 
-- 
Best Regards,
Artem Bityutskiy (Артём Битюцкий)

Re: [Ubi] Wrong major:minor

[Artem Bityutskiy]

On Wed, 2011-06-01 at 11:36 +0300, Artem Bityutskiy wrote:
> On Sun, 2011-05-29 at 12:39 +0200, Gilles wrote:
> > Hello
> > 
> > I used the following data to compile Ubi(fs) into a uClinux kernel:
> > 
> > ========= device_table.txt
> > #UBIFS nodes
> > /dev/ubi_ctrl	c	644	0	0	10	63	0	0	-
> > /dev/ubi0	c	644	0	0	253	0	0	0	-
> > ========= 
> 
> But I guess you may add an UBI module parameter which will ask UBI to
> use major numbers you want - should be easy to do. E.g., if you add a
> 'major_base=200' parameter then UBI will use 200:0 for ubi0, 201:0 for
> ubi1, 200:1 for ubi0_0, 200:2 for ubi0_1, etc.

Although I'm not sure - dynamic numbers were introduced because fixed
numbers are PITA, so adding such a parameter would be a step backward.
Not sure...

-- 
Best Regards,
Artem Bityutskiy (Артём Битюцкий)

Re: [Ubi] Wrong major:minor

[Mike Frysinger]

On Wed, Jun 1, 2011 at 04:37, Artem Bityutskiy wrote:
> On Wed, 2011-06-01 at 11:36 +0300, Artem Bityutskiy wrote:
>> On Sun, 2011-05-29 at 12:39 +0200, Gilles wrote:
>> > I used the following data to compile Ubi(fs) into a uClinux kernel:
>> >
>> > ========= device_table.txt
>> > #UBIFS nodes
>> > /dev/ubi_ctrl       c       644     0       0       10      63      0       0       -
>> > /dev/ubi0   c       644     0       0       253     0       0       0       -
>> > =========
>>
>> But I guess you may add an UBI module parameter which will ask UBI to
>> use major numbers you want - should be easy to do. E.g., if you add a
>> 'major_base=200' parameter then UBI will use 200:0 for ubi0, 201:0 for
>> ubi1, 200:1 for ubi0_0, 200:2 for ubi0_1, etc.
>
> Although I'm not sure - dynamic numbers were introduced because fixed
> numbers are PITA, so adding such a parameter would be a step backward.
> Not sure...

yeah, i dont think there should be.  at least nothing specific to ubi.
 perhaps a general kernel thing where you could match the registered
name to a static #, but then that wouldnt be ubi's problem.  that'd be
in the core logic and work for all devices.
-mike

Re: [Ubi] Wrong major:minor

[Artem Bityutskiy]

On Wed, 2011-06-01 at 06:11 -0400, Mike Frysinger wrote:
> On Wed, Jun 1, 2011 at 04:37, Artem Bityutskiy wrote:
> > On Wed, 2011-06-01 at 11:36 +0300, Artem Bityutskiy wrote:
> >> On Sun, 2011-05-29 at 12:39 +0200, Gilles wrote:
> >> > I used the following data to compile Ubi(fs) into a uClinux kernel:
> >> >
> >> > ========= device_table.txt
> >> > #UBIFS nodes
> >> > /dev/ubi_ctrl       c       644     0       0       10      63      0       0       -
> >> > /dev/ubi0   c       644     0       0       253     0       0       0       -
> >> > =========
> >>
> >> But I guess you may add an UBI module parameter which will ask UBI to
> >> use major numbers you want - should be easy to do. E.g., if you add a
> >> 'major_base=200' parameter then UBI will use 200:0 for ubi0, 201:0 for
> >> ubi1, 200:1 for ubi0_0, 200:2 for ubi0_1, etc.
> >
> > Although I'm not sure - dynamic numbers were introduced because fixed
> > numbers are PITA, so adding such a parameter would be a step backward.
> > Not sure...
> 
> yeah, i dont think there should be.  at least nothing specific to ubi.
>  perhaps a general kernel thing where you could match the registered
> name to a static #, but then that wouldnt be ubi's problem.  that'd be
> in the core logic and work for all devices.

Yes, I agree. Gilles, please, disregard my suggestion.

-- 
Best Regards,
Artem Bityutskiy (Артём Битюцкий)

Re: [Ubi] Wrong major:minor

[Gilles]

On Wed, 01 Jun 2011 11:37:35 +0300, Artem Bityutskiy
<dedekind1@gmail.com> wrote:
>Although I'm not sure - dynamic numbers were introduced because fixed
>numbers are PITA, so adding such a parameter would be a step backward.

No problem. Mdev seems to work fine, so I'll stop using static dev
nodes. Thanks for the infos.

Re: [Ubi] Wrong major:minor

[Gilles]

On Mon, 30 May 2011 12:29:56 -0400, Mike Frysinger
<vapier.adi@gmail.com> wrote:

>On Mon, May 30, 2011 at 05:53, Gilles wrote:
>>            #BAD echo /sbin/mdev > /proc/sys/kernel/hotplug;
>
>i dont know why you've disabled this.  if you want dynamic nodes after
>boot, you need it.

Because...
1. of that error:
====
Starting mdev...
sh: can't open '/proc/sys/kernel/hotplug': No such file or directory
====

2. and because I don't actually need hotplugging, since the appliance
is a closed-box, so users can't actually (un)plug devices anyway.

I just needed mdev so that Linux would create the right dev nodes for
Ubi.

Thank you.

Re: [Ubi] Wrong major:minor

[Peter Korsgaard]

>>>>> "Mike" == Mike Frysinger <vapier.adi@gmail.com> writes:

Hi,

 >> Although I'm not sure - dynamic numbers were introduced because
 >> fixed numbers are PITA, so adding such a parameter would be a step
 >> backward.  Not sure...

 Mike> yeah, i dont think there should be.  at least nothing specific to
 Mike> ubi.  perhaps a general kernel thing where you could match the
 Mike> registered name to a static #, but then that wouldnt be ubi's
 Mike> problem.  that'd be in the core logic and work for all devices.

That already exists - CONFIG_DEVTMPFS.

-- 
Bye, Peter Korsgaard

Re: [Ubi] Wrong major:minor

[Gilles]

On Fri, 24 Jun 2011 13:46:29 +0200, Peter Korsgaard
<jacmet@sunsite.dk> wrote:
>>>>> "Mike" == Mike Frysinger <vapier.adi@gmail.com> writes:
> Mike> yeah, i dont think there should be.  at least nothing specific to
> Mike> ubi.  perhaps a general kernel thing where you could match the
> Mike> registered name to a static #, but then that wouldnt be ubi's
> Mike> problem.  that'd be in the core logic and work for all devices.
>
>That already exists - CONFIG_DEVTMPFS.

Thanks for the tip. Turns out mdev is fast enough to create the dev
nodes at run-time, but it's good to know there's an alternative.

Re: [Ubi] Wrong major:minor

[Mike Frysinger]

On Fri, Jun 24, 2011 at 07:46, Peter Korsgaard wrote:
>>>>>> "Mike" == Mike Frysinger writes:
>  >> Although I'm not sure - dynamic numbers were introduced because
>  >> fixed numbers are PITA, so adding such a parameter would be a step
>  >> backward.  Not sure...
> >
> > yeah, i dont think there should be.  at least nothing specific to
> > ubi.  perhaps a general kernel thing where you could match the
> > registered name to a static #, but then that wouldnt be ubi's
> > problem.  that'd be in the core logic and work for all devices.
>
> That already exists - CONFIG_DEVTMPFS.

while useful, that isnt what we were talking about.  that lets the
kernel populate /dev, that doesnt hardcode major/minor #s.
-mike