LSM: add static to security_ops variable
Subsystems (2)
PATCH 1/1 LSM: add static to security_ops variable
security_ops was declared as a global variable, so other drivers or kernel code can easily change its value, like:
extern struct security_operations *security_ops;security_ops = NULL;
then insmod this driver immediately, it will get an oops. Other evil drivers can aslo fake this variable as extern.
--- security/security.c | 25 ++++++++++++++++++++++++- security/selinux/hooks.c | 18 ++++++------------ 2 files changed, 30 insertions(+), 13 deletions(-)
security/security.c +1 -1
@@ -26,7 +26,12 @@ static __initdata char chosen_lsm[SECURITY_NAME_MAX + 1] =extern struct security_operations default_security_ops;extern void security_fixup_ops (struct security_operations *ops); -struct security_operations *security_ops;/* Initialized to NULL */ +static struct security_operations *security_ops;/* Initialized
to NULL */
+/*
+ * Minimal support for a secondary security module,
+ * just to allow the use of the capability module.
+ */
+static struct security_operations *secondary_ops;
static inline int verify(struct security_operations *ops) {
@@ -63,6 +68,24 @@ int __init security_init(void)
return 0;
}
+void reset_secondary_ops(void)
+{
+ secondary_ops = security_ops;
+ if (!secondary_ops)
+ panic("SELinux: No initial security operations\n");
+}
+
+void reset_security_ops(void)
+{
+ /* Reset security_ops to the secondary module, dummy or capability. */
+ security_ops = secondary_ops;
+}
/* Save user chosen LSM */
static int __init choose_lsm(char *str)
{
security/selinux/hooks.c +5 -12
@@ -92,7 +92,9 @@#define NUM_SEL_MNT_OPTS 5extern int selinux_nlmsg_lookup (u16 sclass,u16 nlmsg_type,u32 *perm); -extern struct security_operations *security_ops; +extern void reset_secondary_ops (void ); +extern void reset_security_ops (void );/* SECMARK reference count */ atomic_t selinux_secmark_refcount =ATOMIC_INIT (0 );
@@ -126,12 +128,6 @@ int selinux_enabled = 1;#endif -/* -* Minimal support for a secondary security module, -* just to allow the use of the capability module. -*/ -static struct security_operations *secondary_ops; -/* Lists of inode and superblock security structures initialized before the policy was loaded. */ static LIST_HEAD (superblock_security_head);
@@ -5672,9 +5668,8 @@ static __init int selinux_init(void) 0, SLAB_PANIC, NULL);avc_init (); - secondary_ops = security_ops; -if (!secondary_ops) -panic ("SELinux: No initial security operations\n" ); +reset_secondary_ops (); +if (register_security (&selinux_ops))panic ("SELinux: Unable to register with kernel.\n" );
@@ -5835,8 +5830,7 @@ int selinux_disable(void) selinux_disabled =1 ; selinux_enabled =0 ; -/* Reset security_ops to the secondary module, dummy or capability. */ - security_ops = secondary_ops; +reset_security_ops ();/* Try to destroy the avc node cache */ avc_disable ();
Signature · 1 line
1.6.6.1
Quoted 8 lines · wzt wzt
security_opswas declared as a global variable, so other drivers or kernel code can easily change its value, like:extern struct
security_operations*security_ops;security_ops= NULL;then insmod this driver immediately, it will get an oops. Other evil drivers can aslo fake this variable as extern.
Evil drivers can do lots of things, if you can load a kernel module on the system, all bets are off. Just making this a private variable does not mean much.
What external module are you trying to keep from using this variable?
thanks,
greg k-h
Quoted 8 lines · wzt wzt
security_opswas declared as a global variable, so other drivers or kernel code can easily change its value, like:extern struct
security_operations*security_ops;security_ops= NULL;then insmod this driver immediately, it will get an oops. Other evil drivers can aslo fake this variable as extern.
I'd support a patch along these lines (but with the changes below) for a different reason: at present, SELinux directly manipulates security_ops for the purpose of runtime disable support, whereas that ought to be handled by the security framework.
Quoted 17 lines
Signed-off-by wzt <[email protected]>--- security/security.c | 25 ++++++++++++++++++++++++- security/selinux/hooks.c | 18 ++++++------------ 2 files changed, 30 insertions(+), 13 deletions(-)security/security.c +1
-1@@ -26,7 +26,12 @@ static __initdata char chosen_lsm[SECURITY_NAME_MAX + 1] =extern struct security_operations default_security_ops;extern void security_fixup_ops (struct security_operations *ops); -struct security_operations *security_ops;/* Initialized to NULL */ +static struct security_operations *security_ops;/* Initialized
Quoted 5 lines
to NULL */
+/*
+ * Minimal support for a secondary security module,
+ * just to allow the use of the capability module.
+ */
The comment is no longer accurate - secondary_ops was originally used by SELinux to call the "secondary" security module (capability or dummy), but that was replaced by direct calls to capability and the only remaining use is to save and restore the original security ops pointer value if SELinux is disabled by early userspace based
on /etc/selinux/config. Further, if we support this directly in the security framework, then we can just use &default_security_ops for this purpose since that is now available. So I don't believe we need secondary_ops at all.
Quoted 1 line
+static struct
security_operations*secondary_ops;
We don't need the above variable at all.
security/security.c int __init security_init(void) · line 76
Quoted 12 lines
static inline int
verify(struct security_operations *ops){security/security.c +6
-0@@ -63,6 +68,24 @@ int __init security_init(void)return 0 ; } +void reset_secondary_ops (void ) +{ + secondary_ops = security_ops; +if (!secondary_ops) +panic ("SELinux: No initial security operations\n" ); +}
We don't need the above function at all.
security/security.c int __init security_init(void) · line 80
Quoted 4 lines
security/security.c +4
-0+ +void reset_security_ops (void ) +{ +/* Reset security_ops to the secondary module, dummy or capability. */
The dummy module was removed so this can only be capability.
security/security.c int __init security_init(void) · line 81
Quoted 1 line
security/security.c +1
-0+ security_ops = secondary_ops;
This can just be:
security_ops = &default_security_ops;
Quoted 10 lines
security/security.c +1
-0+}/* Save user chosen LSM */ static int __init choose_lsm (char *str) {security/selinux/hooks.c +0
-0
security/selinux/hooks.c line 96
Quoted 6 lines
security/selinux/hooks.c +2
-1@@ -92,7 +92,9 @@#define NUM_SEL_MNT_OPTS 5extern int selinux_nlmsg_lookup (u16 sclass,u16 nlmsg_type,u32 *perm); -extern struct security_operations *security_ops; +extern void reset_secondary_ops (void ); +extern void reset_security_ops (void );
The extern declaration for reset_security_ops() would properly go in include/linux/security.h for general use by security modules.
Quoted 16 lines
security/selinux/hooks.c +0
-6/* SECMARK reference count */ atomic_t selinux_secmark_refcount =ATOMIC_INIT (0 );@@ -126,12 +128,6 @@ int selinux_enabled = 1;#endif -/* -* Minimal support for a secondary security module, -* just to allow the use of the capability module. -*/ -static struct security_operations *secondary_ops; -/* Lists of inode and superblock security structures initialized before the policy was loaded. */ static LIST_HEAD (superblock_security_head);
security/selinux/hooks.c static __init int selinux_init(void) · line 5672
Quoted 8 lines
security/selinux/hooks.c +2
-3@@ -5672,9 +5668,8 @@ static __init int selinux_init(void) 0, SLAB_PANIC, NULL);avc_init (); - secondary_ops = security_ops; -if (!secondary_ops) -panic ("SELinux: No initial security operations\n" ); +reset_secondary_ops (); +
We don't need to save this value as it is available via &default_security_ops and there is now only one possible value (dummy module killed).
security/selinux/hooks.c int selinux_disable(void) · line 5833
Quoted 10 lines
security/selinux/hooks.c +1
-2if (register_security (&selinux_ops))panic ("SELinux: Unable to register with kernel.\n" );@@ -5835,8 +5830,7 @@ int selinux_disable(void) selinux_disabled =1 ; selinux_enabled =0 ; -/* Reset security_ops to the secondary module, dummy or capability. */ - security_ops = secondary_ops; +reset_security_ops ();
So this is the only change needed here.
Quoted 3 lines
security/selinux/hooks.c +0
-0/* Try to destroy the avc node cache */ avc_disable ();
Signature · 2 lines
Stephen Smalley National Security Agency
I rewrite the patch, thx.
--- include/linux/security.h | 2 ++ security/security.c | 7 ++++++- security/selinux/hooks.c | 14 ++------------ 3 files changed, 10 insertions(+), 13 deletions(-)
include/linux/security.h +2 -0
@@ -95,6 +95,8 @@ struct seq_file;extern int cap_netlink_send (struct sock *sk,struct sk_buff *skb);extern int cap_netlink_recv (struct sk_buff *skb,int cap); +extern void reset_security_ops (void ); +#ifdef CONFIG_MMU extern unsigned long mmap_min_addr;extern unsigned long dac_mmap_min_addr;
security/security.c +1 -1
@@ -26,7 +26,7 @@ static __initdata char chosen_lsm[SECURITY_NAME_MAX + 1] =extern struct security_operations default_security_ops;extern void security_fixup_ops (struct security_operations *ops); -struct security_operations *security_ops;/* Initialized to NULL */ +static struct security_operations *security_ops;/* Initialized
to NULL */
static inline int verify(struct security_operations *ops) {
@@ -63,6 +63,11 @@ int __init security_init(void)
return 0;
}
+void reset_security_ops(void)
+{
+ security_ops = &default_security_ops;
+}
+
/* Save user chosen LSM */
static int __init choose_lsm(char *str)
{
security/selinux/hooks.c +2 -12
@@ -93,6 +93,7 @@extern int selinux_nlmsg_lookup (u16 sclass,u16 nlmsg_type,u32 *perm);extern struct security_operations *security_ops; +extern struct security_operations default_security_ops;/* SECMARK reference count */ atomic_t selinux_secmark_refcount =ATOMIC_INIT (0 );
@@ -125,13 +126,6 @@ __setup("selinux=", selinux_enabled_setup);int selinux_enabled =1 ;#endif - -/* -* Minimal support for a secondary security module, -* just to allow the use of the capability module. -*/ -static struct security_operations *secondary_ops; -/* Lists of inode and superblock security structures initialized before the policy was loaded. */ static LIST_HEAD (superblock_security_head);
@@ -5672,9 +5666,6 @@ static __init int selinux_init(void) 0, SLAB_PANIC, NULL);avc_init (); - secondary_ops = security_ops; -if (!secondary_ops) -panic ("SELinux: No initial security operations\n" );if (register_security (&selinux_ops))panic ("SELinux: Unable to register with kernel.\n" );
@@ -5835,8 +5826,7 @@ int selinux_disable(void) selinux_disabled =1 ; selinux_enabled =0 ; -/* Reset security_ops to the secondary module, dummy or capability. */ - security_ops = secondary_ops; +reset_security_ops ();/* Try to destroy the avc node cache */ avc_disable ();
Signature · 1 line
1.6.5.3
Quoted 149 lines · Stephen Smalley
Quoted 8 lines · wzt wzt
security_opswas declared as a global variable, so other drivers or kernel code can easily change its value, like:extern struct
security_operations*security_ops;security_ops= NULL;then insmod this driver immediately, it will get an oops. Other evil drivers can aslo fake this variable as extern.
I'd support a patch along these lines (but with the changes below) for a different reason: at present, SELinux directly manipulates
security_opsfor the purpose of runtime disable support, whereas that ought to be handled by the security framework.Quoted 17 lines
Signed-off-by wzt <[email protected]>---
security/security.c | 25 ++++++++++++++++++++++++-security/selinux/hooks.c| 18 ++++++------------2 files changed, 30 insertions(+), 13 deletions(-)security/security.c +1
-1@@ -26,7 +26,12 @@ static __initdata char chosen_lsm[SECURITY_NAME_MAX + 1] = extern struct security_operations default_security_ops; extern void security_fixup_ops(struct security_operations *ops); -struct security_operations *security_ops; /* Initialized to NULL */ +static struct security_operations *security_ops; /* InitializedQuoted 5 lines
to NULL */
+/*
+ * Minimal support for a secondary security module,
+ * just to allow the use of the capability module.
+ */The comment is no longer accurate -
secondary_opswas originally used by SELinux to call the "secondary" security module (capability or dummy), but that was replaced by direct calls to capability and the only remaining use is to save and restore the original security ops pointer value if SELinux is disabled by early userspace based
on/etc/selinux/config. Further, if we support this directly in the security framework, then we can just use&default_security_opsfor this purpose since that is now available. So I don't believe we needsecondary_opsat all.Quoted 1 line
+static struct
security_operations*secondary_ops;We don't need the above variable at all.
security/security.c int __init security_init(void) · line 76
Quoted 12 lines
static inline int
verify(struct security_operations *ops){security/security.c +6
-0@@ -63,6 +68,24 @@ int __init security_init(void) return 0; } +void reset_secondary_ops (void ) +{ + secondary_ops = security_ops; + if (!secondary_ops) + panic("SELinux: No initial security operations\n"); +}We don't need the above function at all.
security/security.c int __init security_init(void) · line 80
Quoted 4 lines
security/security.c +4
-0+ +void reset_security_ops (void ) +{ + /* Reset security_ops to the secondary module, dummy or capability. */The dummy module was removed so this can only be capability.
security/security.c int __init security_init(void) · line 81
Quoted 1 line
security/security.c +1
-0+ security_ops = secondary_ops;This can just be:
security_ops =&default_security_ops;Quoted 10 lines
security/security.c +1
-0+} /* Save user chosen LSM */ static int __init choose_lsm(char *str){
security/selinux/hooks.c +0
-0security/selinux/hooks.c line 96
Quoted 6 lines
security/selinux/hooks.c +2
-1@@ -92,7 +92,9 @@ #define NUM_SEL_MNT_OPTS 5 extern int selinux_nlmsg_lookup(u16 sclass, u16 nlmsg_type, u32 *perm); -extern struct security_operations *security_ops; +extern void reset_secondary_ops (void ); +extern void reset_security_ops (void );The extern declaration for
reset_security_ops()would properly go ininclude/linux/security.hfor general use by security modules.Quoted 16 lines
security/selinux/hooks.c +0
-6/* SECMARK reference count */ atomic_t selinux_secmark_refcount = ATOMIC_INIT(0);@@ -126,12 +128,6 @@ int selinux_enabled = 1; #endif -/* -* Minimal support for a secondary security module, -* just to allow the use of the capability module. -*/ -static struct security_operations *secondary_ops; - /* Lists of inode and superblock security structures initializedbefore the policy was loaded. */
staticLIST_HEAD(superblock_security_head);security/selinux/hooks.c static __init int selinux_init(void) · line 5672
Quoted 8 lines
security/selinux/hooks.c +2
-3@@ -5672,9 +5668,8 @@ static __init int selinux_init(void) 0, SLAB_PANIC, NULL); avc_init(); - secondary_ops = security_ops; - if (!secondary_ops) - panic("SELinux: No initial security operations\n"); + reset_secondary_ops(); +We don't need to save this value as it is available via
&default_security_opsand there is now only one possible value (dummy module killed).security/selinux/hooks.c int selinux_disable(void) · line 5833
Quoted 10 lines
security/selinux/hooks.c +1
-2if (register_security(&selinux_ops)) panic("SELinux: Unable to register with kernel.\n");@@ -5835,8 +5830,7 @@ int selinux_disable(void) selinux_disabled = 1; selinux_enabled = 0; - /* Reset security_ops to the secondary module, dummy or capability. */ - security_ops = secondary_ops; + reset_security_ops();So this is the only change needed here.
security/selinux/hooks.c int selinux_disable(void) · line 5835
Quoted 3 lines
security/selinux/hooks.c +0
-0/* Try to destroy the avc node cache */ avc_disable();--
Stephen Smalley
National Security Agency
Quoted 1 line · wzt wzt
I rewrite the patch, thx.
Patch description needs to be descriptive, e.g.:
"Enhance the security framework to support resetting the active security module. This eliminates the need for direct use of the security_ops variable outside of security.c, so make security_ops static."
Subject line could be more descriptive too, and likely just use
security: as the prefix.
You also need a Signed-off-by line, with a real name.
Also, see the comments below.
Quoted 17 lines
--- include/linux/security.h | 2 ++ security/security.c | 7 ++++++- security/selinux/hooks.c | 14 ++------------ 3 files changed, 10 insertions(+), 13 deletions(-)security/security.c +1
-1@@ -26,7 +26,7 @@ static __initdata char chosen_lsm[SECURITY_NAME_MAX + 1] =extern struct security_operations default_security_ops;extern void security_fixup_ops (struct security_operations *ops); -struct security_operations *security_ops;/* Initialized to NULL */ +static struct security_operations *security_ops;/* Initialized
Quoted 1 line
to NULL */
You can drop the /* Initialized to NULL */ comment since it is now static.
security/selinux/hooks.c line 96
Quoted 9 lines
security/selinux/hooks.c +1
-0@@ -93,6 +93,7 @@extern int selinux_nlmsg_lookup (u16 sclass,u16 nlmsg_type,u32 *perm);extern struct security_operations *security_ops; +extern struct security_operations default_security_ops;
We don't need this extern declaration.
The next obvious cleanup would be to make default_security_ops static. That will require some code reorganization though - it is presently defined in capability.c and manipulated by security.c.
Signature · 2 lines
Stephen Smalley National Security Agency
Quoted 2 lines · wzt wzt
security_opswas declared as a global variable, so other drivers or kernel code can easily change its value, like:
With your patch they still can.
Quoted 5 lines
extern struct
security_operations*security_ops;security_ops= NULL;then insmod this driver immediately, it will get an oops. Other evil drivers can aslo fake this variable as extern.
Maybe, but The attackers will use a complicated way to find the security_ops address, it's a barrier to attackers. LSM is security framework, we don't want the attackers can easily to break it. Just like the sys_call_table variable in kernel 2.4.x(global and writeable), evil drivers can extern the variable, then replace the Sys_X functions.
Quoted 12 lines · Alexey Dobriyan
Quoted 2 lines · wzt wzt
security_opswas declared as a global variable, so other drivers or kernel code can easily change its value, like:With your patch they still can.
Quoted 5 lines
extern struct
security_operations*security_ops;security_ops= NULL;then insmod this driver immediately, it will get an oops. Other evil drivers can aslo fake this variable as extern.
Quoted 2 lines · wzt wzt
Maybe, but The attackers will use a complicated way to find the
security_opsaddress, it's a barrier to attackers.
It's not a barrier, it's garbage. Once you know the adress security_ops ended up at, you simply write to it.
Quoted 2 lines
LSM is security framework, we don't want the attackers can easily to break it.
LSM doesn't protect kernel from kernel.
Quoted 3 lines
Just like the
sys_call_tablevariable in kernel 2.4.x(global and writeable), evil drivers can extern the variable, then replace theSys_Xfunctions.
Not that easily, but they still can.
Quoted 2 lines
It's not a barrier, it's garbage. Once you know the adress
security_opsended up at, you simply write to it.
How to find the security_ops address if the variable is static? Would you please make an example?
Quoted 1 line
Not that easily, but they still can.
That's why i suggest to make the variable to static, if you had wrote a rootkit, you will find that in kernel 2.4.x, there are many many rootkits, but in kernel 2.6.x, rootkit became fewer. Not all the kernel driver writers can master this method to find the variable's address.
The patch also delete the secondary_ops variable.
Quoted 18 lines · Alexey Dobriyan
Quoted 2 lines · wzt wzt
Maybe, but The attackers will use a complicated way to find the
security_opsaddress, it's a barrier to attackers.It's not a barrier, it's garbage. Once you know the adress
security_opsended up at, you simply write to it.Quoted 2 lines
LSM is security framework, we don't want the attackers can easily to break it.
LSM doesn't protect kernel from kernel.
Quoted 3 lines
Just like the
sys_call_tablevariable in kernel 2.4.x(global and writeable), evil drivers can extern the variable, then replace theSys_Xfunctions.Not that easily, but they still can.
Quoted 5 lines · wzt wzt
Quoted 2 lines
It's not a barrier, it's garbage. Once you know the adress
security_opsended up at, you simply write to it.How to find the
security_opsaddress if the variable is static? Would you please make an example?
See /proc/kallsyms .
Quoted 6 lines
Quoted 1 line
Not that easily, but they still can.
That's why i suggest to make the variable to static, if you had wrote a rootkit, you will find that in kernel 2.4.x, there are many many rootkits, but in kernel 2.6.x, rootkit became fewer. Not all the kernel driver writers can master this method to find the variable's address.
Please.
Quoted 1 line
The patch also delete the
secondary_opsvariable.
security_ops is not static, so you can find the address with kallsysm, but you can try secondary_ops:
static struct security_operations *secondary_ops = NULL;
cat /proc/kallsysm|grep secondary_ops
Quoted 20 lines · Alexey Dobriyan
Quoted 5 lines · wzt wzt
Quoted 2 lines
It's not a barrier, it's garbage. Once you know the adress
security_opsended up at, you simply write to it.How to find the
security_opsaddress if the variable is static? Would you please make an example?See
/proc/kallsyms.Quoted 6 lines
Quoted 1 line
Not that easily, but they still can.
That's why i suggest to make the variable to static, if you had wrote a rootkit, you will find that in kernel 2.4.x, there are many many rootkits, but in kernel 2.6.x, rootkit became fewer. Not all the kernel driver writers can master this method to find the variable's address.
Please.
Quoted 1 line
The patch also delete the
secondary_opsvariable.
Quoted 5 lines · wzt wzt
Quoted 2 lines
It's not a barrier, it's garbage. Once you know the adress
security_opsended up at, you simply write to it.How to find the
security_opsaddress if the variable is static? Would you please make an example?
- Find the accessor function (which is surely non-static).
- Find the address where writes the parameter.
Other must decide, if it's more secure (as in "obscure") with the accessor funtion or not.
Bernd
Signature · 2 lines
Bernd Petrovitsch Email : [email protected] LUGA : http://www.luga.at