Skip to contentSkip to conversation

allow disabling IMA at runtime

PATCH In review Kyle McMartin 6 participants 6 messages Aug 26, 2009 – Aug 27, 2009

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 allow disabling IMA at runtime

From: Kyle McMartin <[email protected]>

Due to a memory leak in IMA that we're currently debugging in Fedora rawhide, it would be nice to be able to disable that support at runtime. Currently it's only able to be built in, and there's no toggle to avoid initializing it.

Provide one, in order to enhance debuggability. If a user can reboot a machine and edit its command line, one can do a far sight worse things than disabling a security precaution.

---

Documentation/kernel-parameters.txt +5 -0
index 7936b80..0d1b1ed 100644
@@ -926,6 +926,11 @@ and is between 256 and 4096 characters. It is defined in the file
 	ihash_entries=	[KNL]
 			Set number of hash buckets for inode cache.
 
+	ima=		[IMA]
+			Format: { "0" | "1" }
+			0 -- disable IMA.
+			1 -- enable IMA. (default)
+
 	ima_audit=	[IMA]
 			Format: { "0" | "1" }
 			0 -- integrity auditing messages. (Default)
security/integrity/ima/ima_main.c +17 -0
index 101c512..cc7603e 100644
@@ -339,10 +339,27 @@ int ima_bprm_check(struct linux_binprm *bprm)
 	return 0;
 }
 
+static int ima_disabled = 0;
+static int __init ima_enabled(char *str)
+{
+	unsigned long enabled;
+
+	if (!strict_strtoul(str, 0, &enabled))
+		ima_disabled = enabled ? 0 : 1;
+
+	return 1;
+}
+__setup("ima=", ima_enabled);
+
 static int __init init_ima(void)
 {
 	int error;
 
+	if (ima_disabled) {
+		pr_info("IMA disabled at user request.\n");
+		return 0;
+	}
+
 	ima_iintcache_init();
 	error = ima_init();
 	ima_initialized = 1;
#1 lore raw

Kyle McMartin <[email protected]> wrote on 08/25/2009 10:10:05 PM:

Quoted 7 lines

From: Kyle McMartin <[email protected]>

Due to a memory leak in IMA that we're currently debugging in Fedora rawhide, it would be nice to be able to disable that support at runtime. Currently it's only able to be built in, and there's no toggle to avoid initializing it.

Quoted 5 lines

Provide one, in order to enhance debuggability. If a user can reboot a machine and edit its command line, one can do a far sight worse things than disabling a security precaution.

Have you tried using Eric's patch
"IMA: Minimal IMA policy and boot param for TCB IMA policy" (commit 5789ba3bd0a3cd20df5980ebf03358f2eb44fd67), which introduced the ima_tcb=1 command line option? It wasn't backported to 2.6.30.

Mimi

Review

Quoted 2 lines

---

(quoted diff) +0 -0

b/Documentation/kernel-parameters.txt

Documentation/kernel-parameters.txt and is between 256 and 4096 characters. It is

Quoted 4 lines

index 7936b80..0d1b1ed 100644

Documentation/kernel-parameters.txt +0 -0
@@ -926,6 +926,11 @@ and is between 256 and 4096 characters. It is 

defined in the file

Review

Quoted 12 lines
Documentation/kernel-parameters.txt +5 -0
    ihash_entries=   [KNL]
          Set number of hash buckets for inode cache.
 
+   ima=      [IMA]
+         Format: { "0" | "1" }
+         0 -- disable IMA.
+         1 -- enable IMA. (default)
+
    ima_audit=   [IMA]
          Format: { "0" | "1" }
          0 -- integrity auditing messages. (Default)
(quoted diff) +0 -0

b/security/integrity/ima/ima_main.c

Quoted 31 lines

index 101c512..cc7603e 100644

security/integrity/ima/ima_main.c +17 -0
@@ -339,10 +339,27 @@ int ima_bprm_check(struct linux_binprm *bprm)
    return 0;
 }
 
+static int ima_disabled = 0;
+static int __init ima_enabled(char *str)
+{
+   unsigned long enabled;
+
+   if (!strict_strtoul(str, 0, &enabled))
+      ima_disabled = enabled ? 0 : 1;
+
+   return 1;
+}
+__setup("ima=", ima_enabled);
+
 static int __init init_ima(void)
 {
    int error;
 
+   if (ima_disabled) {
+      pr_info("IMA disabled at user request.\n");
+      return 0;
+   }
+
    ima_iintcache_init();
    error = ima_init();
    ima_initialized = 1;
#2 lore raw
Quoted 13 lines · Mimi Zohar

Kyle McMartin <[email protected]> wrote on 08/25/2009 10:10:05 PM:

Quoted 11 lines

From: Kyle McMartin <[email protected]>

Due to a memory leak in IMA that we're currently debugging in Fedora rawhide, it would be nice to be able to disable that support at runtime. Currently it's only able to be built in, and there's no toggle to avoid initializing it.

Provide one, in order to enhance debuggability. If a user can reboot a machine and edit its command line, one can do a far sight worse things than disabling a security precaution.

Quoted 6 lines
Quoted 1 line

Have you tried using Eric's patch
"IMA: Minimal IMA policy and boot param for TCB IMA policy" (commit 5789ba3bd0a3cd20df5980ebf03358f2eb44fd67), which introduced the ima_tcb=1 command line option? It wasn't backported to 2.6.30.

Hey Mimi, I was going to get in touch with you today, I don't really think this patch is necessary. Kyle hacked it together because it was a quick and dirty 'fix' for a memory leak that he didn't want to hunt down and he knows I won't let him compile IMA out *smile*. Intended to try to track it down this morning, but I'm getting swamped already, maybe you can try to figure out what's going on before I get a chance to come back to it this afternoon?

nfs_inode_cache       34     34   1824   17    8 : tunables    0    0    0 : slabdata      2      2      0
fuse_inode            22     22   1472   22    8 : tunables    0    0    0 : slabdata      1      1      0
rpc_inode_cache       40     40   1600   20    8 : tunables    0    0    0 : slabdata      2      2      0
btrfs_inode_cache  10622  10668   2328   14    8 : tunables    0    0    0 : slabdata    762    762      0
iint_cache        369714 369720    312   26    2 : tunables    0    0    0 : slabdata  14220  14220      0
mqueue_inode_cache     19     19   1664   19    8 : tunables    0    0    0 : slabdata      1      1      0
isofs_inode_cache      0      0   1288   25    8 : tunables    0    0    0 : slabdata      0      0      0
hugetlbfs_inode_cache     24     24   1312   24    8 : tunables    0    0    0 : slabdata      1      1      0
ext4_inode_cache       0      0   1864   17    8 : tunables    0    0    0 : slabdata      0      0      0
ext3_inode_cache      19     19   1656   19    8 : tunables    0    0    0 : slabdata      1      1      0
inotify_inode_mark_entry    253    255    240   17    1 : tunables    0    0    0 : slabdata     15     15      0
shmem_inode_cache   2740   3003   1560   21    8 : tunables    0    0    0 : slabdata    143    143      0
sock_inode_cache     902    920   1408   23    8 : tunables    0    0    0 : slabdata     40     40      0
proc_inode_cache    3060   3075   1288   25    8 : tunables    0    0    0 : slabdata    123    123      0
inode_cache         9943  10192   1240   26    8 : tunables    0    0    0 : slabdata    392    392      0
selinux_inode_security  27237  27838    264   31    2 : tunables    0    0    0 : slabdata    898    898      0

So the iint_cache is a LOT larger than all of the inode caches put together. This is a 2.6.31-0.167.rc6.git6.fc12.x86_64 kernel without any kernel options.

-Eric

-Eric

#3 lore raw
In reply to Kyle McMartin
Quoted 11 lines · On Tue, 25 Aug 2009 22:10:05 -0400 Kyle McMartin <[email protected]> wrote:

From: Kyle McMartin <[email protected]>

Due to a memory leak in IMA that we're currently debugging in Fedora rawhide, it would be nice to be able to disable that support at runtime. Currently it's only able to be built in, and there's no toggle to avoid initializing it.

Provide one, in order to enhance debuggability. If a user can reboot a machine and edit its command line, one can do a far sight worse things than disabling a security precaution.

nits:

Quoted 25 lines

---

Documentation/kernel-parameters.txt +5 -0
index 7936b80..0d1b1ed 100644
@@ -926,6 +926,11 @@ and is between 256 and 4096 characters. It is defined in the file
 	ihash_entries=	[KNL]
 			Set number of hash buckets for inode cache.
 
+	ima=		[IMA]
+			Format: { "0" | "1" }
+			0 -- disable IMA.
+			1 -- enable IMA. (default)
+
 	ima_audit=	[IMA]
 			Format: { "0" | "1" }
 			0 -- integrity auditing messages. (Default)
security/integrity/ima/ima_main.c +0 -0
index 101c512..cc7603e 100644
@@ -339,10 +339,27 @@ int ima_bprm_check(struct linux_binprm *bprm)
 	return 0;
 }

security/integrity/ima/ima_main.c int ima_bprm_check(struct linux_binprm *bprm) · line 350

Quoted 10 lines
security/integrity/ima/ima_main.c +9 -0
 
+static int ima_disabled = 0;
+static int __init ima_enabled(char *str)
+{
+	unsigned long enabled;
+
+	if (!strict_strtoul(str, 0, &enabled))
+		ima_disabled = enabled ? 0 : 1;
+	return 1;
+}

- documentation says "0 or 1" but the implementation says "0 or non-zero".

- implementation returns `1' whether or not the argument was successfully parsed.

What happens if a __setup() functions returns non-1?

<spends a while fossicking through prehistory and RustyCode>

From my reading of kernel/params.c:parse_args(), every __setup() function which returns 1' should result in printk("%s: %s' invalid for parameter `%s'), so I'm all confused and giving up.

security/integrity/ima/ima_main.c int ima_bprm_check(struct linux_binprm *bprm) · line 352

Quoted 1 line
security/integrity/ima/ima_main.c +1 -0
+__setup("ima=", ima_enabled);

Are we supposed to use core_param() nowadays?

Quoted 12 lines
security/integrity/ima/ima_main.c +5 -0
 static int __init init_ima(void)
 {
 	int error;
 
+	if (ima_disabled) {
+		pr_info("IMA disabled at user request.\n");
+		return 0;
+	}
+
 	ima_iintcache_init();
 	error = ima_init();
 	ima_initialized = 1;
#5 lore raw
In reply to Andrew Morton
Quoted 3 lines · Andrew Morton

From my reading of kernel/params.c:parse_args(), every __setup() function which returns 1' should result in printk("%s: %s' invalid for parameter `%s'), so I'm all confused and giving up.

Nope, it's an "obsolete_param", so non-zero means success.

Quoted 3 lines
Quoted 1 line

+__setup("ima=", ima_enabled);

Are we supposed to use core_param() nowadays?

Yeah, if this had backwards compatibility requirements. But for new, non-core params like this:

1) make sure the module is called "ima" (even if not a module: make the object ima.o)

2) use module_param(), and make your parameters "disable" and "audit" (ima_disable and ima_audit if must be nonstatic and use module_param_named). You don't need to write any parse functions at all!

3) Tell users to use ima.disable and ima.audit.

Thanks,
Rusty.