[PATCH][RFC] V1 Remove SELinux dependencies from linux-audit via LSM

Fri Aug 3 16:18:41 UTC 2007

--- Stephen Smalley <sds at tycho.nsa.gov> wrote:

> On Thu, 2007-08-02 at 20:33 -0700, Casey Schaufler wrote:
> > From: Casey Schaufler <casey at schaufler-ca.com>
> > 
> > This patch removes SELinux specific code from the kernel auditing
> > system, replacing it with LSM hook invocations that perform the
> > functions appropriate to those behaviors.
> 
> (patch was whitespace damaged)

Bugger. I know what I did wrong. Sorry.

> > The LSM interface is extended to provide interfaces for a module
> > to add audit filters. Interfaces are added to get secids from
> > inodes and ipcs.
> > 
> > The audit code is revised to call these hooks instead of the SELinux
> > functions. This requires some structure definitions to change header
> > files.
> > 
> > The SELinux code is changed to export the old interfaces as LSM hooks
> > instead of doing so directly. The SELinux specific audit filter code
> > has been moved into the SELinux module.
> > 
> > Signed-off-by: Casey Schaufler <casey at schaufler-ca.com>
> > 
> > ---
> 
> Missing diffstat -p1 output.

 include/linux/audit.h          |   55 ++++++++++++++++++++++++---
 include/linux/security.h       |   83
+++++++++++++++++++++++++++++++++++++++++
 include/linux/selinux.h        |   15 -------
 kernel/audit.c                 |   22 ++++------
 kernel/audit.h                 |   41 --------------------
 kernel/auditfilter.c           |   56 +++++++--------------------
 kernel/auditsc.c               |   36 ++++++++---------
 security/selinux/hooks.c       |   30 ++++++++++++++
 security/selinux/ss/services.c |    7 ---
 9 files changed, 206 insertions(+), 139 deletions(-)

> > I hope for some suggestions regarding how appropriate the header file
> > changes I've made are. My confidence level in them is fairly low. Also,
> > in my haste to get some feedback the testing has been limited. There
> > are three projects (SELinux, audit, and LSM) whose code I've changed,
> > and I expect there may be differences of opinion about what's right,
> > what's wrong, and how to go about reconciling the differences. I see
> > this as the start.
> > 
> > Thank you.
> > 
> 
> > diff -uprN -X linux-2.6.22-base/Documentation/dontdiff
> > linux-2.6.22-base/include/linux/security.h
> > linux-2.6.22-audit/include/linux/security.h
> > --- linux-2.6.22-base/include/linux/security.h	2007-07-08
> 16:32:17.000000000
> > -0700
> > +++ linux-2.6.22-audit/include/linux/security.h	2007-08-01
> 20:14:18.000000000
> > -0700
> > @@ -35,6 +35,8 @@
> >  #include <net/flow.h>
> >  
> >  struct ctl_table;
> > +struct audit_krule;
> > +struct selinux_audit_rule;
> 
> selinux_audit_rule in LSM interface?

The structure needs a new name. Any objections to audit_rule_lsm?
I'd suggest security_audit_rule, but that doesn't say anything about
where to look to see how it gets used.

>  
> >  /*
> >   * These functions are in security/capability.c and are used
> > @@ -1328,6 +1330,16 @@ struct security_operations {
> >   	int (*setprocattr)(struct task_struct *p, char *name, void *value,
> size_t
> > size);
> >  	int (*secid_to_secctx)(u32 secid, char **secdata, u32 *seclen);
> >  	void (*release_secctx)(char *secdata, u32 seclen);
> > +	void (*inode_getsecid) (const struct inode *inode, u32 *secid);
> > +	void (*ipc_getsecid) (struct kern_ipc_perm *p, u32 *secid);
> > +	int (*audit_rule_supplied) (struct audit_krule *rule);
> > +	int (*audit_rule_match) (u32 sid, u32 field, u32 op,
> > +                          	 struct selinux_audit_rule *rule,
> 
> Ditto.
> 
> > +                      	         struct audit_context *actx);
> > +	int (*audit_rule_init) (u32 field, u32 op, char *rulestr,
> > +                                struct selinux_audit_rule **rule);
> > +	void (*audit_rule_free) (struct selinux_audit_rule *rule);
> > +
> >  
> >  #ifdef CONFIG_SECURITY_NETWORK
> >  	int (*unix_stream_connect) (struct socket * sock,
> 
> > diff -uprN -X linux-2.6.22-base/Documentation/dontdiff
> > linux-2.6.22-base/kernel/audit.c linux-2.6.22-audit/kernel/audit.c
> > --- linux-2.6.22-base/kernel/audit.c	2007-07-08 16:32:17.000000000 -0700
> > +++ linux-2.6.22-audit/kernel/audit.c	2007-08-01 12:16:36.000000000 -0700
> > @@ -252,7 +252,7 @@ static int audit_set_rate_limit(int limi
> >  	if (sid) {
> >  		char *ctx = NULL;
> >  		u32 len;
> > -		if ((rc = selinux_sid_to_string(sid, &ctx, &len)) == 0) {
> > +		if ((rc = security_secid_to_secctx(sid, &ctx, &len)) == 0) {
> >  			audit_log(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE,
> >  				"audit_rate_limit=%d old=%d by auid=%u"
> >  				" subj=%s res=%d",
> 
> If you do that, then you also need to use security_secctx_release() to
> free ctx rather than direct kfree.

Indeed. Done.

> Also, the sid passed to this (and other functions) came from
> NETLINK_CB(skb).sid, which was set by netlink_sendmsg() with a direct
> call to selinux_get_task_sid.  So if you convert these calls to using
> LSM hooks w/o changing netlink to do likewise, you could end up with a
> mismatch (sid provided by selinux, defaulting to 0 when selinux
> disabled; secid_to_secctx hook provided by your module).

Yup. You are correct. Fixed.

> > diff -uprN -X linux-2.6.22-base/Documentation/dontdiff
> > linux-2.6.22-base/kernel/auditfilter.c
> linux-2.6.22-audit/kernel/auditfilter.c
> > --- linux-2.6.22-base/kernel/auditfilter.c	2007-07-08 16:32:17.000000000
> -0700
> > +++ linux-2.6.22-audit/kernel/auditfilter.c	2007-08-01 20:22:36.000000000
> -0700
> > @@ -29,6 +29,7 @@
> >  #include <linux/sched.h>
> >  #include <linux/inotify.h>
> >  #include <linux/selinux.h>
> > +#include <linux/security.h>
> 
> Shouldn't you be removing #include <linux/selinux.h> from all of the
> audit code?  That will help you catch all lingering references to
> selinux-specific references as well.

Yes.

> > @@ -1210,7 +1211,8 @@ static inline int audit_add_rule(struct 
> >  	struct audit_entry *e;
> >  	struct audit_field *inode_f = entry->rule.inode_f;
> >  	struct audit_watch *watch = entry->rule.watch;
> > -	struct nameidata *ndp, *ndw;
> > +	struct nameidata *ndp = NULL;
> > +	struct nameidata *ndw = NULL;
> 
> Not related to your changes, and seems to already be present in git (but
> without splitting into two lines).

Warnings are bad enough when they're in remote bits of code, when
they show up in code I'm whacking on they tend to disappear. I'll
put them back together to reduce merge issues.

> > diff -uprN -X linux-2.6.22-base/Documentation/dontdiff
> > linux-2.6.22-base/security/selinux/hooks.c
> > linux-2.6.22-audit/security/selinux/hooks.c
> > --- linux-2.6.22-base/security/selinux/hooks.c	2007-07-08
> 16:32:17.000000000
> > -0700
> > +++ linux-2.6.22-audit/security/selinux/hooks.c	2007-08-01
> 21:42:32.000000000
> > -0700
> > @@ -4838,6 +4862,12 @@ static struct security_operations selinu
> >  
> >  	.secid_to_secctx =		selinux_secid_to_secctx,
> >  	.release_secctx =		selinux_release_secctx,
> > +	.inode_getsecid =		selinux_get_inode_sid,
> > +	.ipc_getsecid = 		selinux_get_ipc_sid,
> > +	.audit_rule_supplied = 		selinux_audit_rule_supplied,
> > +	.audit_rule_match = 		selinux_audit_rule_match,
> > +	.audit_rule_init = 		selinux_audit_rule_init,
> > +	.audit_rule_free = 		selinux_audit_rule_free,
> 
> So I'd then expect you to also remove the function prototypes from
> include/linux/selinux.h unless there is some other caller, and move the
> functions from selinux/exports.c to selinux/hooks.c and make them
> static.  And drop the selinux_enabled check from them, as the hook
> function will only be registered if SELinux is enabled.

If that's the way you'd like the SELinux code arranged, that's what
I'll do.

> -- 
> Stephen Smalley
> National Security Agency

Thank you for the helpful comments.

Casey Schaufler
casey at schaufler-ca.com