PROBLEM: GPT partition table + RAID: kernel could not load init while root filesystem is mounted

PROBLEM: GPT partition table + RAID: kernel could not load init while root filesystem is mounted

[manio]

Hello
I am trying to migrate from old root filesystem (linux raid1 on disks 
with msdos partition table) to new (4K sector) disks with GPT partition 
table (also raid1). I am using raid autodetection (i have superblock 
v0.90 on all raid devices) and i am passing /dev/mdX as a root device to 
the kernel. When i am passing root=/dev/md10 (old disks with msdos 
partition tables) booting process is fine (i am pasting relevant info):

[    6.695241] raid1: raid set md10 active with 2 out of 2 mirrors
[    6.695280] md10: detected capacity change from 0 to 6999572480
[    6.695332] md: ... autorun DONE.
[    6.695407]  md10: unknown partition table
[    6.696236] EXT3-fs (md10): error: couldn't mount because of 
unsupported optional features (240)
[    6.696508] EXT2-fs (md10): error: couldn't mount because of 
unsupported optional features (240)
[    6.743086] EXT4-fs (md10): mounted filesystem with ordered data mode
[    6.743185] VFS: Mounted root (ext4 filesystem) readonly on device 9:10.
[    6.743227] Freeing unused kernel memory: 424k freed
..... now comes INIT - and it boots correctly

BUT (!) the problem is when i am trying to pass root=/dev/md0 (raid1 on 
new disks with GPT partition tables). Then it seems like kernel cannot 
run /sbin/init from my ext4 root partition. The root filesystem seems to 
be mounted OK:

md: ... autorun DONE.
  md0: unknown partition table
EXT3-fs (md0): error: couldn't mount because of unsupported optional 
features (240)
EXT2-fs (md0): error: couldn't mount because of unsupported optional 
features (240)
EXT4-fs (md0): mounted filesystem with ordered data mode
VFS: Mounted root (ext4 filesystem) readonly on device 9:0.
Freeing unused kernel memory: 424k freed

..... now there is some USB init debug info and it stop :(
/sbin/init is not launched and it seems like the kernel hung :(
i was waiting several minutes then - and nothing happend.

If you would like to see how the boot process look like from the grub to 
the kernel hung (filmed with my digital camera) then download this:
http://manio.skyboo.net/kernel-gpt-init-problem.avi

please help me - any tips are welcome

best regards,
-- 
Mariusz Bialonczyk
jabber/e-mail: manio@skyboo.net
http://manio.skyboo.net

Re: PROBLEM: GPT partition table + RAID: kernel could not load init while root filesystem is mounted

[Mariusz Bialonczyk]

On 09/18/2010 02:52 AM, manio wrote:
> Hello
> I am trying to migrate from old root filesystem (linux raid1 on disks
> with msdos partition table) to new (4K sector) disks with GPT partition
> table (also raid1). I am using raid autodetection (i have superblock
> v0.90 on all raid devices) and i am passing /dev/mdX as a root device to
> the kernel. When i am passing root=/dev/md10 (old disks with msdos
> partition tables) booting process is fine (i am pasting relevant info):
>
> [ 6.695241] raid1: raid set md10 active with 2 out of 2 mirrors
> [ 6.695280] md10: detected capacity change from 0 to 6999572480
> [ 6.695332] md: ... autorun DONE.
> [ 6.695407] md10: unknown partition table
> [ 6.696236] EXT3-fs (md10): error: couldn't mount because of unsupported
> optional features (240)
> [ 6.696508] EXT2-fs (md10): error: couldn't mount because of unsupported
> optional features (240)
> [ 6.743086] EXT4-fs (md10): mounted filesystem with ordered data mode
> [ 6.743185] VFS: Mounted root (ext4 filesystem) readonly on device 9:10.
> [ 6.743227] Freeing unused kernel memory: 424k freed
> ..... now comes INIT - and it boots correctly
>
> BUT (!) the problem is when i am trying to pass root=/dev/md0 (raid1 on
> new disks with GPT partition tables). Then it seems like kernel cannot
> run /sbin/init from my ext4 root partition. The root filesystem seems to
> be mounted OK:
>
> md: ... autorun DONE.
> md0: unknown partition table
> EXT3-fs (md0): error: couldn't mount because of unsupported optional
> features (240)
> EXT2-fs (md0): error: couldn't mount because of unsupported optional
> features (240)
> EXT4-fs (md0): mounted filesystem with ordered data mode
> VFS: Mounted root (ext4 filesystem) readonly on device 9:0.
> Freeing unused kernel memory: 424k freed
>
> ..... now there is some USB init debug info and it stop :(
> /sbin/init is not launched and it seems like the kernel hung :(
> i was waiting several minutes then - and nothing happend.

after some investigation i am almost sure that the problem is with
kernel_execve() - i put some debug info and i can see that last call
from init/main.c is:
run_init_process("/sbin/init");
i was checking the boot partition - it is same correct ext4 filesystem
with same contents - the only difference between them is the partition
table - the problem is on GPT, while on msdos partition table it calls
INIT correctly after kernel_execve call. On GPT it hungs - maybe it is
not jumping into correct location?
Strange thing is - that the root filesystem in both cases is mounted,
so why it has trouble to run /sbin/init in GPT case?

Last time i forgot to mention that i am using kernel 2.6.36-rc4 but i
was also testing it on 2.6.34 with same results.
I am using x86_64 architecture.

regards,
-- 
Mariusz Bialonczyk
jabber/e-mail: manio@skyboo.net
http://manio.skyboo.net

Re: PROBLEM: GPT partition table + RAID: kernel could not load init while root filesystem is mounted

[Mariusz Bialonczyk]

On 09/19/2010 12:17 AM, Mariusz Bialonczyk wrote:
> after some investigation i am almost sure that the problem is with
> kernel_execve()

... and i was damn wrong!
I put a lot of debug printf's inside init and i must admit that INIT is
launched correctly! So i think it's not a kernel problem here because it
successfully starting init, sorry for bothering you, guys :(
I also bisected the place where init stops and it totally freezes when
calling: close(2). I was also testing upstart with same hung results
(upstart it also closing stderr).
Sorry for little off-topic question: Maybe you know why sysvinit and
upstart cannot close stderr when booting md from GPT partition table?
grub2 problem?

regards
-- 
Mariusz Bialonczyk
jabber/e-mail: manio@skyboo.net
http://manio.skyboo.net

[SOLVED] Re: PROBLEM: GPT partition table + RAID: kernel could not load init while root filesystem is mounted

[Mariusz Bialonczyk]

On 09/25/2010 02:03 AM, Mariusz Bialonczyk wrote:
> Sorry for little off-topic question: Maybe you know why sysvinit and
> upstart cannot close stderr when booting md from GPT partition table?
> grub2 problem?

everything is now clear
i was doing rsync with -x when copying data - so /dev nodes was not
copied because /dev was mounted as tmpfs - so as a result i have
empty /dev directory on /dev/md0 (boot partition)

regards,
-- 
Mariusz Bialonczyk
jabber/e-mail: manio@skyboo.net
http://manio.skyboo.net