Re: [PATCH v6 02/17] ima: Define ns_status for storing namespaced iint data

On Wed, 2021-12-15 at 21:37 -0500, Stefan Berger wrote:

On 12/15/21 16:12, Mimi Zohar wrote:

quoted

Hi Stefan,

On Fri, 2021-12-10 at 14:47 -0500, Stefan Berger wrote:

quoted

From: Mehmet Kayaalp <redacted>

This patch adds an rbtree to the IMA namespace structure that stores a
namespaced version of iint->flags in ns_status struct. Similar to the
integrity_iint_cache, both the iint ns_struct are looked up using the
inode pointer value. The lookup, allocate, and insertion code is also
similar, except ns_struct is not free'd when the inode is free'd.
Instead, the lookup verifies the i_ino and i_generation fields are also a
match.

Signed-off-by: Mehmet Kayaalp <redacted>
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>

Again, hopefully it isn't premature for generic comments:

- Function/inline comments would be appreciated, especially when the
code differs from the original code.  Example below.

quoted

diff --git a/security/integrity/ima/ima_init_ima_ns.c b/security/integrity/ima/ima_init_ima_ns.c
index f820686baf9f..08781a44f7bf 100644
--- a/security/integrity/ima/ima_init_ima_ns.c
+++ b/security/integrity/ima/ima_init_ima_ns.c

@@ -14,11 +14,18 @@
  #include <linux/user_namespace.h>
  #include <linux/ima.h>
  #include <linux/proc_ns.h>
+#include <linux/slab.h>
  
  #include "ima.h"
  
  int ima_init_namespace(struct ima_namespace *ns)
  {
+	ns->ns_status_tree = RB_ROOT;
+	rwlock_init(&ns->ns_status_lock);
+	ns->ns_status_cache = KMEM_CACHE(ns_status, SLAB_PANIC);
+	if (!ns->ns_status_cache)
+		return -ENOMEM;
+

For example, using KMEM_CACHE() is probably correct here, at least for
now, but it is different than the original code which uses
kmem_cache_alloc() with init_once().  Memory cleanup is done on free,

^  kmem_cache_create() with init_once.

quoted

before it is re-used.

KMEM_CACHE + kmem_cache_alloc/zalloc() are pretty common. What kind of 
comment would be helpful here?

The original reason for using kmem_cache_create() with init_once and
deferring memory cleanup to free was for performance.  Using
KMEM_CACHE() and kmem_cache_zalloc() instead could be for simplicity. 
The comment should note the change.

Mimi