Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Message-ID: <1593117123.27152.385.camel@linux.ibm.com>
Subject: Re: [PATCH 02/12] ima: Create a function to free a rule entry
From:   Mimi Zohar <zohar@linux.ibm.com>
To:     Tyler Hicks <tyhicks@linux.microsoft.com>
Cc:     Dmitry Kasatkin <dmitry.kasatkin@gmail.com>,
        James Morris <jmorris@namei.org>,
        "Serge E . Hallyn" <serge@hallyn.com>,
        Lakshmi Ramasubramanian <nramas@linux.microsoft.com>,
        Prakhar Srivastava <prsriva02@gmail.com>,
        linux-kernel@vger.kernel.org, linux-integrity@vger.kernel.org,
        linux-security-module@vger.kernel.org
Date:   Thu, 25 Jun 2020 16:32:03 -0400
In-Reply-To: <20200625195647.GB4694@sequoia>
References: <20200623003236.830149-1-tyhicks@linux.microsoft.com>
         <20200623003236.830149-3-tyhicks@linux.microsoft.com>
         <1593113613.27152.345.camel@linux.ibm.com> <20200625195647.GB4694@sequoia>
Content-Type: text/plain; charset="UTF-8"
Mime-Version: 1.0
Content-Transfer-Encoding: 8bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, 2020-06-25 at 14:56 -0500, Tyler Hicks wrote:
> On 2020-06-25 15:33:33, Mimi Zohar wrote:
> > On Mon, 2020-06-22 at 19:32 -0500, Tyler Hicks wrote:
> > > There are several possible pieces of allocated memory in a rule entry.
> > > Create a function that can free all allocated memory for a given rule
> > > entry.
> > > 
> > > This patch introduces no functional changes but sets the groundwork for
> > > some memory leak fixes.
> > > 
> > > Signed-off-by: Tyler Hicks <tyhicks@linux.microsoft.com>
> > 
> > Having a function to release all memory associated with a policy rule
> > in general is a good idea.  However, in the case of the shallow copy,
> > we're not removing any IMA rules, just updating the LSM info.
> > 
> > There is an opportunity to transition from the builtin policy rules to
> > a custom IMA policy.  Afterwards IMA policy rules may only be
> > appended.
> > 
> > An IMA custom policy based on LSM info may be loaded prior to the LSM
> > policy.  These LSM based rules are inactive until the corresponding
> > LSM rule is loaded.  In some environments, LSM policies are loaded and
> > removed frequently.  The IMA rules themselves are not removed, just
> > the LSM info is updated to reflect the current LSM info.
> > 
> > > ---
> > >  security/integrity/ima/ima_policy.c | 33 +++++++++++++++++++++++++++--
> > >  1 file changed, 31 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/security/integrity/ima/ima_policy.c b/security/integrity/ima/ima_policy.c
> > > index 236a731492d1..1320333201c6 100644
> > > --- a/security/integrity/ima/ima_policy.c
> > > +++ b/security/integrity/ima/ima_policy.c
> > > @@ -261,6 +261,27 @@ static void ima_lsm_free_rule(struct ima_rule_entry *entry)
> > >  		security_filter_rule_free(entry->lsm[i].rule);
> > >  		kfree(entry->lsm[i].args_p);
> > >  	}
> > > +}
> > > +
> > > +static void ima_free_rule(struct ima_rule_entry *entry)
> > > +{
> > > +	if (!entry)
> > > +		return;
> > > +
> > > +	/*
> > > +	 * entry->template->fields may be allocated in ima_parse_rule() but that
> > > +	 * reference is owned by the corresponding ima_template_desc element in
> > > +	 * the defined_templates list and cannot be freed here
> > > +	 */
> > > +
> > > +	/*
> > > +	 * When freeing newly added ima_rule_entry members, consider if you
> > > +	 * need to disown any references after the shallow copy in
> > > +	 * ima_lsm_copy_rule()
> > > +	 */
> > > +	kfree(entry->fsname);
> > > +	kfree(entry->keyrings);
> > > +	ima_lsm_free_rule(entry);
> > >  	kfree(entry);
> > >  }
> > >  
> > > @@ -298,10 +319,18 @@ static struct ima_rule_entry *ima_lsm_copy_rule(struct ima_rule_entry *entry)
> > >  			pr_warn("rule for LSM \'%s\' is undefined\n",
> > >  				(char *)entry->lsm[i].args_p);
> > >  	}
> > > +
> > > +	/* Disown all references that were shallow copied */
> > > +	entry->fsname = NULL;
> > > +	entry->keyrings = NULL;
> > > +	entry->template = NULL;
> > >  	return nentry;
> > >  
> > >  out_err:
> > > -	ima_lsm_free_rule(nentry);
> > > +	nentry->fsname = NULL;
> > > +	nentry->keyrings = NULL;
> > > +	nentry->template = NULL;
> > > +	ima_free_rule(nentry);
> > 
> > >  	return NULL;
> > >  }
> > >  
> > > @@ -315,7 +344,7 @@ static int ima_lsm_update_rule(struct ima_rule_entry *entry)
> > >  
> > >  	list_replace_rcu(&entry->list, &nentry->list);
> > >  	synchronize_rcu();
> > > -	ima_lsm_free_rule(entry);
> > > +	ima_free_rule(entry);
> > 
> > This should only update the LSM info, nothing else.
> 
> That's effectively what's happening since the fsname, keyrings, and
> template pointers are being set to NULL, before exiting
> ima_lsm_copy_rule(), in the ima_rule_entry that's going to be freed.

Ah, that clarified the reason for setting fsname, keyrings, ... to
null before calling ima_free_rule.

> 
> This patch is only introducing the function which can free all memory
> associated with a rule and is starting to use it in place that a rule
> entry is freed.
> 
> Would you rather me introduce ima_free_rule() for the upcoming memory
> leak fixes in the series but not make use of it in
> ima_lsm_update_rule()?

You could add a comment explaining the NULLs, but it might be clearer
to keep the direct call to ima_lsm_free_rule().

Mimi