Append does not return error on full disks

Append does not return error on full disks

[Frederic Giasson]

Hi,

	I have observed that jffs2 nevers returns a 'disk full' error when
an application appends data to an existing file.  I have setup a test which
opens an existing file for appending, and fwrite() data to it until fwrite()
returns error.
My test never stopped!  So when I manually stoppped it, the file system was
corrupted.  I was unable to unmount it, and neither I could remount it after
reboot.  I have setup a second test which is a script instead of being a
program, using the append '>>' command.  Same results. 

	During my test, I did'nt open the file with OSYNC flag though, maybe
it could patch the problem.  I will try it.
Nevertheless, it shouldn't happen that a process be able to append data
indefinitely to a file without ever getting an error when the disk becomes
full.

	Does anyone have a clue about it?  Maybe there is a bug somewhere
down in the jffs2_reserve_space() function?

Frédéric Giasson

Re: Append does not return error on full disks

[tkeitel]

On Tue, May 28, 2002 at 08:30:03 -0400, Frederic Giasson wrote:
> Hi,
> 
> 	I have observed that jffs2 nevers returns a 'disk full' error when
> an application appends data to an existing file.  I have setup a test which
> opens an existing file for appending, and fwrite() data to it until fwrite()
> returns error.
> My test never stopped!  So when I manually stoppped it, the file system was
> corrupted.  I was unable to unmount it, and neither I could remount it after
> reboot.  I have setup a second test which is a script instead of being a
> program, using the append '>>' command.  Same results. 
> 
> 	During my test, I did'nt open the file with OSYNC flag though, maybe
> it could patch the problem.  I will try it.
> Nevertheless, it shouldn't happen that a process be able to append data
> indefinitely to a file without ever getting an error when the disk becomes
> full.
> 
> 	Does anyone have a clue about it?  Maybe there is a bug somewhere
> down in the jffs2_reserve_space() function?
> 

What version of JFFS2 did you use?

-- 
tino.keitel@innominate.com
dipl.-inf.                        Innominate Security Technologies AG
software engineer                                   networking people
tel: +49.30.6392-3308                      http://www.innominate.com/

Re: Append does not return error on full disks

[David Woodhouse]

fgiasson@mediatrix.com said:
> 	I have observed that jffs2 nevers returns a 'disk full' error when an
> application appends data to an existing file.  I have setup a test
> which opens an existing file for appending, and fwrite() data to it
> until fwrite() returns error. My test never stopped!  So when I
> manually stoppped it, the file system was corrupted.  I was unable to
> unmount it, and neither I could remount it after reboot.  I have setup
> a second test which is a script instead of being a program, using the
> append '>>' command.  Same results. 

Eep. Can you reproduce this with the current code from either the trunk or 
the jffs2-2_4-branch, and with CONFIG_JFFS2_FS_DEBUG=1, logging all the 
messages via a serial console?

> 	During my test, I did'nt open the file with OSYNC flag though, maybe
> it could patch the problem.  I will try it.

Doesn't matter. O_SYNC does nothing -- it's always synchronous anyway.


--
dwmw2

Re: Append does not return error on full disks

[Frederic Giasson]

>What version of JFFS2 did you use?

I use code from MTD CVS took on May22th.

Frederic Giasson