Re: Unmounting a busy RO-Filesystem

Re: Unmounting a busy RO-Filesystem

[Denis Vlasenko]

On 16 December 2002 23:06, Bourne wrote:
> I have 3 partitions. /dev/hda3 for '/' , /dev/hda1 for /boot and
> /dev/hda2 for swap.
>
> I boot & then i do a CTRL+ALT+SYSRQ+U.  '/' and '/boot' are now
> remounted ReadOnly.
>
> 1) cd '/boot'
> 2) umount /boot ----> This gives me an error "Device Busy"

How do you imagine unmounting a directory when you are in it? ;)

> 3) cd /
> 4) umount / -------> No error

This is special case: "umount /" == "mount -o remount,ro /"

> 5) echo $? -----> outputs '0' indicating success. !!!!!!!!
>
> When i do the above by skipping the Sysrq part, i get the usual
> expected errors.

Without SysRq,

# mount -o remount,ro /

fails 'coz you have files open for writing.
You might ask how kernel can do that ro remount with SysRq?
It cheats! ;)
--
vda

Re: Unmounting a busy RO-Filesystem

[Steven Dake]

Denis Vlasenko wrote:

>On 16 December 2002 23:06, Bourne wrote:
>  
>
>>I have 3 partitions. /dev/hda3 for '/' , /dev/hda1 for /boot and
>>/dev/hda2 for swap.
>>
>>I boot & then i do a CTRL+ALT+SYSRQ+U.  '/' and '/boot' are now
>>remounted ReadOnly.
>>
>>1) cd '/boot'
>>2) umount /boot ----> This gives me an error "Device Busy"
>>    
>>
>
>How do you imagine unmounting a directory when you are in it? ;)
>  
>
This is possible with a kernel patch called forced unmount.  It will 
blow away the mount point even if there are files open in it or not. 
 The best part is it properly closes the superblock so the filesystem 
doesn't have to be fsck'ed.

>  
>

Unmounting a busy RO-Filesystem

[Bourne]

I have 3 partitions. /dev/hda3 for '/' , /dev/hda1 for /boot and 
/dev/hda2 for swap.

I boot & then i do a CTRL+ALT+SYSRQ+U.  '/' and '/boot' are now 
remounted ReadOnly.

1) cd '/boot'
2) umount /boot ----> This gives me an error "Device Busy"
3) cd /
4) umount / -------> No error
5) echo $? -----> outputs '0' indicating success. !!!!!!!!

When i do the above by skipping the Sysrq part, i get the usual expected 
errors.

So is the above behaviour expected ? I tried this on 2.4.19 & 2.5.51. 
Same results.

I am subscribed to the list

TIA
Bourne