Re: [PATCH v1.2 0/5] IMA: making i_readcount a first class inode citizen (reposting)

[David Safford <safford@watson.ibm.com>]

On Sun, 2010-11-21 at 16:33 -0500, Mimi Zohar wrote:
> On Sun, 2010-11-21 at 09:56 -0800, Linus Torvalds wrote:
> > On Sun, Nov 21, 2010 at 5:18 AM, Mimi Zohar <zohar@linux.vnet.ibm.com> wrote:
> > >
> > > IMA (and the proposed EVM/IMA-appraisal patches) detects file change
> > > based on i_version. When the file is closed, if the file has changed,
> > > IMA marks the file as needing to be re-measured. Of course this requires
> > > the filesystem to be mounted with iversion. Don't know if this helps.
> > 
> > If you only do this at close time, I see a _major_ security hole.
> > 
> > The attacker can just write to the file, and keep it open. Ta-daa,
> > everybody who reads it sees the new contents, but your IMA logic is
> > oblivious and thinks it doesn't need to be re-measured.
> > 
> >                             Linus
> 
> Not exactly.  While the file remains open for write, it doesn't make any
> sense to re-measure the file, as there is nothing preventing the file
> from continuing to change.  Any measurement would thus be meaningless.
> Only after the file closes, does it make sense to re-measure.  I did not
> mean to imply there isn't any indication of the problem in the
> measurement list, there obviously is.
> 
> Mimi
> 
To elaborate a bit on Mimi's response - in the case of a malicious
program keeping a file open for write to avoid measurement:
1. as she points out, the reason for i_writecount and i_readcount
   is to detect this "open_writer" problem and log it in both the 
   measurement list and in the audit log.
2. the attacker program itself must have been measured before it
   was executed. 

dave