Re: null-ptr-deref in integrity_inode_free()

[Mimi Zohar <zohar@linux.ibm.com>]

On Wed, 2021-04-14 at 22:32 +0200, Christian Brauner wrote:

> On Wed, Apr 14, 2021, 22:21 Mimi Zohar <zohar@linux.ibm.com> wrote:
> > On Wed, 2021-04-14 at 20:19 +0200, Christian Brauner wrote:
> > > On Wed, Apr 14, 2021 at 01:46:46PM -0400, Mimi Zohar wrote:
> > > > Hi Christian,
> > > > 
> > > > On Wed, 2021-04-14 at 13:50 +0200, Christian Brauner wrote:
> > > > > Hey,
> > > > > 
> > > > > [Resending since the previous mail somehow hasn't made it onto the list.]
> > > > > 
> > > > > While working on a patch to port ecryptfs to use clone_private_mount() I
> > > > > hit the splat in [1] with v5.12-rc6 (Without any of my changes.). To
> > > > > reproduce this you can use the config in [5]. Then run the scripts [2]
> > > > > and [3] in two terminals. The kernel command line was [4]:
> > > > > 
> > > > > while true; do ./test.sh; done
> > > > > while true; do ./test2.sh; done
> > > > > 
> > > > > and wait for [1] to appear.
> > > > > 
> > > > > The two test scripts aren't specifically designed to trigger this issue.
> > > > > They were stress tests for my ecryptfs clone_private_mount() port. They
> > > > > just allow to trigger this issue.
> > > > > 
> > > > > From a very uninformed position it seemed that:
> > > > > 
> > > > > void integrity_inode_free(struct inode *inode)
> > > > > {
> > > > >         struct integrity_iint_cache *iint;
> > > > > 
> > > > >         if (!IS_IMA(inode))
> > > > >                 return;
> > > > 
> > > > Thank you for all the details.
> > > > 
> > > > A builtin IMA policy wasn't specified on the boot command line.  Unless
> > > > a custom IMA policy is loaded, it shouldn't get beyond this
> > > > point.   Before looking any farther, I would appreciate your confirming
> > > > that you've loaded a custom IMA policy.  To see if a policy has been
> > > > loaded, cat /sys/kernel/security/ima/policy.
> > > 
> > > Ah, interesting thank you. Here's the output:
> > > 
> > > sudo cat /sys/kernel/security/ima/policy
> > >   dont_measure fsmagic=0x9fa0
> > >   dont_measure fsmagic=0x62656572
> > >   dont_measure fsmagic=0x64626720
> > >   dont_measure fsmagic=0x1021994
> > >   dont_measure fsmagic=0x1cd1
> > >   dont_measure fsmagic=0x42494e4d
> > >   dont_measure fsmagic=0x73636673
> > >   dont_measure fsmagic=0xf97cff8c
> > >   dont_measure fsmagic=0x43415d53
> > >   dont_measure fsmagic=0x27e0eb
> > >   dont_measure fsmagic=0x63677270
> > >   dont_measure fsmagic=0x6e736673
> > >   dont_measure fsmagic=0xde5e81e4
> > >   measure func=MMAP_CHECK mask=MAY_EXEC
> > >   measure func=BPRM_CHECK mask=MAY_EXEC
> > >   measure func=FILE_CHECK mask=^MAY_READ euid=0
> > >   measure func=FILE_CHECK mask=^MAY_READ uid=0
> > >   measure func=MODULE_CHECK
> > >   measure func=FIRMWARE_CHECK
> > >   measure func=POLICY_CHECK
> > > 
> > > Hm, note that ecryptfs is not in this list so I guess ecryptfs would be
> > > measured?
> > 
> > That's the "tcb" policy, which can be specified as "policy=tcb" on the
> > boot command line.  The builtin policy rules are generic and would
> > normally be replaced with more specific policy rules based on LSM
> > labels.
> 
> Ah, thank you.
> 
> > Right, ecryptfs files are being measured.
> > 
> > There's somehow a race freeing the inode.  The fix is straight forward.
> > Did you want to fix it or should I?
> 
> Please do fix it. I've reach eod over here. :)
> You're simply going to handle the NULL case when freeing the iint?

Yes, but it would be nice to understand the race and whether this is
something or has been there from the beginning.

Mimi