Skip to contentSkip to conversation

compute_creds fixup in -mm

DISCUSSION Andy Lutomirski 4 participants 8 messages Apr 21, 2004 1 message not in the archive

Subsystems (5)
lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw
Quoted 9 lines · Chris Wright

[ I'm going through a bunch of security bugs/fixes, and I'm getting a little bleary eyed so perhaps I'm missing something obvious. ]

Andy, your patch appears to leave the core issue unfixed, unfortuantely. [...]
This is still vulnerable. I think the problem is that a ptrace detach never acquires the task_lock(), yet it will clear task->ptrace. So the above code is still racey.

Quoted 6 lines

ptrace_attach()
task_lock()
must_not_trace_exec() /*true, don't elevate uid*/
ptrace_detach()
must_not_trace_exec() /*false, give full caps*/ task_unlock()

Yeesh! You're right.

Quoted 5 lines

Code could be cleaned up to simply check must_not_trace_exec() only once (like below, untested). Otherwise, it'd be nice to task_lock() on __ptrace_unlink() before clearing task->ptrace, but this isn't proper lock netsing.

I think your patch is correct; I didn't do it that way because I didn't want to worry about whether it would change the semantics. I'm pretty sure now that it doesn't.

This doesn't fix selinux, though -- its apply_creds hook just blindly calls commoncap's. In fact, this breaks all attempts to get nested capability modules right. The problem is that, AFAICS, not only does ptrace_detach not take task_lock, _exit() doesn't either. So you get an equivalent race for the shared state check. I see two ways to fix that:

1. something checks for shared state _once_ and saves it either in the binprm or passes it as a parameter to apply_creds. apply_creds needs to be changed so that the task_lock is taken by the caller.

2. Don't nest apply_creds.

I'm favoring #1, as it doesn't break the whole programming model. Did I miss something?

This one also (hopefully) fixes selinux.

Patch against 2.6.5-mm5.

 fs/exec.c                |   36 +++++++++++++++++++++---------------
 include/linux/security.h |   19 +++++++++++++------
 security/commoncap.c     |   38 ++++++++++++--------------------------
 security/dummy.c         |   15 ++-------------
 security/selinux/hooks.c |   12 ++++--------
 5 files changed, 52 insertions(+), 68 deletions(-)
include/linux/security.h renamed from include/linux/security.h.ptlock +13 -6
@@ -44,7 +44,7 @@
 extern int cap_capset_check (struct task_struct *target, kernel_cap_t *effective, kernel_cap_t *inheritable, kernel_cap_t *permitted);
 extern void cap_capset_set (struct task_struct *target, kernel_cap_t *effective, kernel_cap_t *inheritable, kernel_cap_t *permitted);
 extern int cap_bprm_set_security (struct linux_binprm *bprm);
-extern void cap_bprm_apply_creds (struct linux_binprm *bprm);
+extern void cap_bprm_apply_creds (struct linux_binprm *bprm, int unsafe);
 extern int cap_bprm_secureexec(struct linux_binprm *bprm);
 extern int cap_inode_setxattr(struct dentry *dentry, char *name, void *value, size_t size, int flags);
 extern int cap_inode_removexattr(struct dentry *dentry, char *name);
@@ -86,6 +86,11 @@
 struct sched_param;
 struct swap_info_struct;
 
+/* brpm_apply_creds unsafe reasons */
+#define LSM_UNSAFE_SHARE	1
+#define LSM_UNSAFE_PTRACE	2
+#define LSM_UNSAFE_PTRACE_CAP	4
+
 #ifdef CONFIG_SECURITY
 
 /**
@@ -112,6 +117,8 @@
  *	also perform other state changes on the process (e.g.  closing open
  *	file descriptors to which access is no longer granted if the attributes
  *	were changed).
+ *	bprm_apply_creds is called under task_lock.  @unsafe indicates various
+ *	reasons why it may be unsafe to change security state.
  *	@bprm contains the linux_binprm structure.
  * @bprm_set_security:
  *	Save security information in the bprm->security field, typically based
@@ -1026,7 +1033,7 @@
 
 	int (*bprm_alloc_security) (struct linux_binprm * bprm);
 	void (*bprm_free_security) (struct linux_binprm * bprm);
-	void (*bprm_apply_creds) (struct linux_binprm * bprm);
+	void (*bprm_apply_creds) (struct linux_binprm * bprm, int unsafe);
 	int (*bprm_set_security) (struct linux_binprm * bprm);
 	int (*bprm_check_security) (struct linux_binprm * bprm);
 	int (*bprm_secureexec) (struct linux_binprm * bprm);
@@ -1290,9 +1297,9 @@
 {
 	security_ops->bprm_free_security (bprm);
 }
-static inline void security_bprm_apply_creds (struct linux_binprm *bprm)
+static inline void security_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
 {
-	security_ops->bprm_apply_creds (bprm);
+	security_ops->bprm_apply_creds (bprm, unsafe);
 }
 static inline int security_bprm_set (struct linux_binprm *bprm)
 {
@@ -1962,9 +1969,9 @@
 static inline void security_bprm_free (struct linux_binprm *bprm)
 { }
 
-static inline void security_bprm_apply_creds (struct linux_binprm *bprm)
+static inline void security_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
 {
-	cap_bprm_apply_creds (bprm);
+	cap_bprm_apply_creds (bprm, unsafe);
 }
 
 static inline int security_bprm_set (struct linux_binprm *bprm)
fs/exec.c renamed from fs/exec.c.ptlock +21 -15
@@ -919,24 +919,30 @@
 
 EXPORT_SYMBOL(prepare_binprm);
 
-/*
- * This function is used to produce the new IDs and capabilities
- * from the old ones and the file's capabilities.
- *
- * The formula used for evolving capabilities is:
- *
- *       pI' = pI
- * (***) pP' = (fP & X) | (fI & pI)
- *       pE' = pP' & fE          [NB. fE is 0 or ~0]
- *
- * I=Inheritable, P=Permitted, E=Effective // p=process, f=file
- * ' indicates post-exec(), and X is the global 'cap_bset'.
- *
- */
+static inline int must_not_trace_exec (struct task_struct *p)
+{
+	int unsafe = 0;
+	if (p->ptrace & PT_PTRACED) {
+		if(p->ptrace & PT_PTRACE_CAP)
+			unsafe |= LSM_UNSAFE_PTRACE_CAP;
+		else
+			unsafe |= LSM_UNSAFE_PTRACE;
+	}
+	if (atomic_read(&p->fs->count) > 1 ||
+	    atomic_read(&p->files->count) > 1 ||
+	    atomic_read(&p->sighand->count) > 1)
+		unsafe |= LSM_UNSAFE_SHARE;
+
+	return unsafe;
+}
 
 void compute_creds(struct linux_binprm *bprm)
 {
-	security_bprm_apply_creds(bprm);
+	task_lock(current);
+
+	security_bprm_apply_creds(bprm, must_not_trace_exec(current));
+
+	task_unlock(current);
 }
 
 EXPORT_SYMBOL(compute_creds);
security/selinux/hooks.c renamed from security/selinux/hooks.c.ptlock +4 -8
@@ -1746,7 +1746,7 @@
 	spin_unlock(&files->file_lock);
 }
 
-static void selinux_bprm_apply_creds(struct linux_binprm *bprm)
+static void selinux_bprm_apply_creds(struct linux_binprm *bprm, int unsafe)
 {
 	struct task_security_struct *tsec, *psec;
 	struct bprm_security_struct *bsec;
@@ -1756,7 +1756,7 @@
 	struct rlimit *rlim, *initrlim;
 	int rc, i;
 
-	secondary_ops->bprm_apply_creds(bprm);
+	secondary_ops->bprm_apply_creds(bprm, unsafe);
 
 	tsec = current->security;
 
@@ -1767,9 +1767,7 @@
 	if (tsec->sid != sid) {
 		/* Check for shared state.  If not ok, leave SID
 		   unchanged and kill. */
-		if ((atomic_read(&current->fs->count) > 1 ||
-		     atomic_read(&current->files->count) > 1 ||
-		     atomic_read(&current->sighand->count) > 1)) {
+		if (unsafe & LSM_UNSAFE_SHARE) {
 			rc = avc_has_perm(tsec->sid, sid,
 					  SECCLASS_PROCESS, PROCESS__SHARE,
 					  NULL, NULL);
@@ -1782,14 +1780,13 @@
 		/* Check for ptracing, and update the task SID if ok.
 		   Otherwise, leave SID unchanged and kill. */
 		task_lock(current);
-		if (current->ptrace & PT_PTRACED) {
+		if (unsafe & (LSM_UNSAFE_PTRACE | LSM_UNSAFE_PTRACE_CAP)) {
 			psec = current->parent->security;
 			rc = avc_has_perm_noaudit(psec->sid, sid,
 					  SECCLASS_PROCESS, PROCESS__PTRACE,
 					  NULL, &avd);
 			if (!rc)
 				tsec->sid = sid;
-			task_unlock(current);
 			avc_audit(psec->sid, sid, SECCLASS_PROCESS,
 				  PROCESS__PTRACE, &avd, rc, NULL);
 			if (rc) {
@@ -1798,7 +1795,6 @@
 			}
 		} else {
 			tsec->sid = sid;
-			task_unlock(current);
 		}
 
 		/* Close files for which the new task SID is not authorized. */
security/commoncap.c renamed from security/commoncap.c.ptlock +12 -26
@@ -115,15 +115,7 @@
 	return 0;
 }
 
-static inline int must_not_trace_exec (struct task_struct *p)
-{
-	return ((p->ptrace & PT_PTRACED) && !(p->ptrace & PT_PTRACE_CAP))
-		|| atomic_read(&p->fs->count) > 1
-		|| atomic_read(&p->files->count) > 1
-		|| atomic_read(&p->sighand->count) > 1;
-}
-
-void cap_bprm_apply_creds (struct linux_binprm *bprm)
+void cap_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
 {
 	/* Derived from fs/exec.c:compute_creds. */
 	kernel_cap_t new_permitted, working;
@@ -133,30 +125,25 @@
 				 current->cap_inheritable);
 	new_permitted = cap_combine (new_permitted, working);
 
-	task_lock(current);
-
-	if (bprm->e_uid != current->uid || bprm->e_gid != current->gid) {
+	if (bprm->e_uid != current->uid || bprm->e_gid != current->gid ||
+	    !cap_issubset (new_permitted, current->cap_permitted)) {
 		current->mm->dumpable = 0;
 
-		if (must_not_trace_exec(current) && !capable(CAP_SETUID)) {
-			bprm->e_uid = current->uid;
-			bprm->e_gid = current->gid;
+		if (unsafe & ~LSM_UNSAFE_PTRACE_CAP) {
+			if (!capable(CAP_SETUID)) {
+				bprm->e_uid = current->uid;
+				bprm->e_gid = current->gid;
+			}
+			if (!capable (CAP_SETPCAP)) {
+				new_permitted = cap_intersect (new_permitted,
+							current->cap_permitted);
+			}
 		}
 	}
 
 	current->suid = current->euid = current->fsuid = bprm->e_uid;
 	current->sgid = current->egid = current->fsgid = bprm->e_gid;
 
-	if (!cap_issubset (new_permitted, current->cap_permitted)) {
-		current->mm->dumpable = 0;
-
-		if (must_not_trace_exec (current) && !capable (CAP_SETPCAP)) {
-			new_permitted = cap_intersect (new_permitted,
-						       current->
-						       cap_permitted);
-		}
-	}
-
 	/* For init, we want to retain the capabilities set
 	 * in the init_task struct. Thus we skip the usual
 	 * capability rules */
@@ -167,7 +154,6 @@
 	}
 
 	/* AUD: Audit candidate if current->cap_effective is set */
-	task_unlock(current);
 
 	current->keep_capabilities = 0;
 }
security/dummy.c renamed from security/dummy.c.ptlock +2 -13
@@ -171,21 +171,12 @@
 	return;
 }
 
-static inline int must_not_trace_exec (struct task_struct *p)
+static void dummy_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
 {
-	return ((p->ptrace & PT_PTRACED) && !(p->ptrace & PT_PTRACE_CAP))
-		|| atomic_read(&p->fs->count) > 1
-		|| atomic_read(&p->files->count) > 1
-		|| atomic_read(&p->sighand->count) > 1;
-}
-
-static void dummy_bprm_apply_creds (struct linux_binprm *bprm)
-{
-	task_lock(current);
 	if (bprm->e_uid != current->uid || bprm->e_gid != current->gid) {
 		current->mm->dumpable = 0;
 
-		if (must_not_trace_exec(current) && !capable(CAP_SETUID)) {
+		if (unsafe && !capable(CAP_SETUID)) {
 			bprm->e_uid = current->uid;
 			bprm->e_gid = current->gid;
 		}
@@ -193,8 +184,6 @@
 
 	current->suid = current->euid = current->fsuid = bprm->e_uid;
 	current->sgid = current->egid = current->fsgid = bprm->e_gid;
-
-	task_unlock(current);
 }
 
 static int dummy_bprm_set_security (struct linux_binprm *bprm)
#2 lore raw
Quoted 5 lines · Andy Lutomirski

This doesn't fix selinux, though -- its apply_creds hook just blindly calls commoncap's. In fact, this breaks all attempts to get nested capability modules right. The problem is that, AFAICS, not only does ptrace_detach not take task_lock, _exit() doesn't either. So you get an equivalent race for the shared state check. I see two ways to fix that:

I didn't see Chris' patch. I assume that the worst case is unexpected program failure due to lack of capability, right? The SELinux security transitions aren't tied to the uid/capability evolution anyway, and SELinux provides its own separation among differing security domains, including the kernel access controls and enabling glibc secure mode.

Side bar: Why does the uid/capability logic proceed under the old uid/capability set rather than aborting the exec? SELinux leaves the task SID unchanged and forces a SIGKILL, as we don't really want the program proceeding under the wrong permission set anyway.

Quoted 3 lines

1. something checks for shared state _once_ and saves it either in the binprm or passes it as a parameter to apply_creds. apply_creds needs to be changed so that the task_lock is taken by the caller.

Likely a good idea.

Quoted 3 lines

This one also (hopefully) fixes selinux.

Patch against 2.6.5-mm5.

Didn't apply for me. Also would help to rebase to 2.6.6-rc2-mm1.

security/selinux/hooks.c line 1782

Quoted 6 lines
security/selinux/hooks.c renamed from security/selinux/hooks.c.ptlock +0 -0
@@ -1782,14 +1780,13 @@
  		/* Check for ptracing, and update the task SID if ok.
  		   Otherwise, leave SID unchanged and kill. */
  		task_lock(current);

Above line needs to be deleted, right?

Signature · 2 lines
Stephen Smalley <[email protected]>
National Security Agency
#3 lore raw
Quoted 9 lines · * Stephen Smalley ([email protected]) wrote:
Quoted 5 lines · Andy Lutomirski

This doesn't fix selinux, though -- its apply_creds hook just blindly calls commoncap's. In fact, this breaks all attempts to get nested capability modules right. The problem is that, AFAICS, not only does ptrace_detach not take task_lock, _exit() doesn't either. So you get an equivalent race for the shared state check. I see two ways to fix that:

I didn't see Chris' patch. I assume that the worst case is unexpected program failure due to lack of capability, right? The SELinux security

The opposite. You'd get a program with non-root euid, but full capability set, and AT_SECURE set false. My patch is below.

security/commoncap.c renamed from security/commoncap.c~compute_creds-lock +12 -14
@@ -135,28 +135,26 @@
 
 	task_lock(current);
 
-	if (bprm->e_uid != current->uid || bprm->e_gid != current->gid) {
+	if (bprm->e_uid != current->uid || bprm->e_gid != current->gid ||
+	    !cap_issubset (new_permitted, current->cap_permitted)) {
 		current->mm->dumpable = 0;
 
-		if (must_not_trace_exec(current) && !capable(CAP_SETUID)) {
-			bprm->e_uid = current->uid;
-			bprm->e_gid = current->gid;
+		if (must_not_trace_exec(current)) {
+			if (!capable(CAP_SETUID)) {
+				bprm->e_uid = current->uid;
+				bprm->e_gid = current->gid;
+			}
+			if (!capable (CAP_SETPCAP)) {
+				new_permitted = cap_intersect (new_permitted,
+							current->cap_permitted);
+			}
+
 		}
 	}
 
 	current->suid = current->euid = current->fsuid = bprm->e_uid;
 	current->sgid = current->egid = current->fsgid = bprm->e_gid;
 
-	if (!cap_issubset (new_permitted, current->cap_permitted)) {
-		current->mm->dumpable = 0;
-
-		if (must_not_trace_exec (current) && !capable (CAP_SETPCAP)) {
-			new_permitted = cap_intersect (new_permitted,
-						       current->
-						       cap_permitted);
-		}
-	}
-
 	/* For init, we want to retain the capabilities set
 	 * in the init_task struct. Thus we skip the usual
 	 * capability rules */
#4 lore raw
Quoted 6 lines · Chris Wright
Quoted 2 lines · * Stephen Smalley ([email protected]) wrote:

I didn't see Chris' patch. I assume that the worst case is unexpected program failure due to lack of capability, right? The SELinux security

The opposite. You'd get a program with non-root euid, but full capability set, and AT_SECURE set false. My patch is below.

Sorry, I wasn't clear. I meant the worst case due to the share/ptrace state check being duplicated in SELinux and in commoncap, as opposed to being performed once as in Andy's patch.

Signature · 2 lines
Stephen Smalley <[email protected]>
National Security Agency
#5 lore raw
Quoted 11 lines · * Stephen Smalley ([email protected]) wrote:
Quoted 6 lines · Chris Wright
Quoted 2 lines · * Stephen Smalley ([email protected]) wrote:

I didn't see Chris' patch. I assume that the worst case is unexpected program failure due to lack of capability, right? The SELinux security

The opposite. You'd get a program with non-root euid, but full capability set, and AT_SECURE set false. My patch is below.

Sorry, I wasn't clear. I meant the worst case due to the share/ptrace state check being duplicated in SELinux and in commoncap, as opposed to being performed once as in Andy's patch.

Ah, indeed.

thanks,
-chris

Signature · 1 line
Linux Security Modules     http://lsm.immunix.org     http://lsm.bkbits.net
#6 lore raw
In reply to Stephen Smalley
Quoted 7 lines · Stephen Smalley
Quoted 5 lines · Chris Wright
Quoted 2 lines · * Stephen Smalley ([email protected]) wrote:

I didn't see Chris' patch. I assume that the worst case is unexpected program failure due to lack of capability, right? The SELinux security

Quoted 8 lines
Quoted 2 lines

The opposite. You'd get a program with non-root euid, but full capability set, and AT_SECURE set false. My patch is below.

Sorry, I wasn't clear. I meant the worst case due to the share/ptrace state check being duplicated in SELinux and in commoncap, as opposed to being performed once as in Andy's patch.

I was worried about sid changing but uid and caps staying the same if a ptrace_detach or _exit happens between the cap_bprm_apply_creds call and the rest of selinux_bprm_apply_creds. Remember the sendmail bug -- program failure due to lack of capabilities can cause privilege leaks (in this case selinux sid leaks).

If this isn't a concern, then Chris' patch should be fine. On the other hand, some other LSM may prefer my version. Here's my corrected patch (thanks, Chris):

include/linux/security.h renamed from include/linux/security.h.ptlock +13 -6
@@ -44,7 +44,7 @@
 extern int cap_capset_check (struct task_struct *target, kernel_cap_t *effective, kernel_cap_t *inheritable, kernel_cap_t *permitted);
 extern void cap_capset_set (struct task_struct *target, kernel_cap_t *effective, kernel_cap_t *inheritable, kernel_cap_t *permitted);
 extern int cap_bprm_set_security (struct linux_binprm *bprm);
-extern void cap_bprm_apply_creds (struct linux_binprm *bprm);
+extern void cap_bprm_apply_creds (struct linux_binprm *bprm, int unsafe);
 extern int cap_bprm_secureexec(struct linux_binprm *bprm);
 extern int cap_inode_setxattr(struct dentry *dentry, char *name, void *value, size_t size, int flags);
 extern int cap_inode_removexattr(struct dentry *dentry, char *name);
@@ -86,6 +86,11 @@
 struct sched_param;
 struct swap_info_struct;
 
+/* brpm_apply_creds unsafe reasons */
+#define LSM_UNSAFE_SHARE	1
+#define LSM_UNSAFE_PTRACE	2
+#define LSM_UNSAFE_PTRACE_CAP	4
+
 #ifdef CONFIG_SECURITY
 
 /**
@@ -112,6 +117,8 @@
  *	also perform other state changes on the process (e.g.  closing open
  *	file descriptors to which access is no longer granted if the attributes
  *	were changed).
+ *	bprm_apply_creds is called under task_lock.  @unsafe indicates various
+ *	reasons why it may be unsafe to change security state.
  *	@bprm contains the linux_binprm structure.
  * @bprm_set_security:
  *	Save security information in the bprm->security field, typically based
@@ -1026,7 +1033,7 @@
 
 	int (*bprm_alloc_security) (struct linux_binprm * bprm);
 	void (*bprm_free_security) (struct linux_binprm * bprm);
-	void (*bprm_apply_creds) (struct linux_binprm * bprm);
+	void (*bprm_apply_creds) (struct linux_binprm * bprm, int unsafe);
 	int (*bprm_set_security) (struct linux_binprm * bprm);
 	int (*bprm_check_security) (struct linux_binprm * bprm);
 	int (*bprm_secureexec) (struct linux_binprm * bprm);
@@ -1290,9 +1297,9 @@
 {
 	security_ops->bprm_free_security (bprm);
 }
-static inline void security_bprm_apply_creds (struct linux_binprm *bprm)
+static inline void security_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
 {
-	security_ops->bprm_apply_creds (bprm);
+	security_ops->bprm_apply_creds (bprm, unsafe);
 }
 static inline int security_bprm_set (struct linux_binprm *bprm)
 {
@@ -1962,9 +1969,9 @@
 static inline void security_bprm_free (struct linux_binprm *bprm)
 { }
 
-static inline void security_bprm_apply_creds (struct linux_binprm *bprm)
+static inline void security_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
 {
-	cap_bprm_apply_creds (bprm);
+	cap_bprm_apply_creds (bprm, unsafe);
 }
 
 static inline int security_bprm_set (struct linux_binprm *bprm)
fs/exec.c renamed from fs/exec.c.ptlock +19 -15
@@ -919,24 +919,28 @@
 
 EXPORT_SYMBOL(prepare_binprm);
 
-/*
- * This function is used to produce the new IDs and capabilities
- * from the old ones and the file's capabilities.
- *
- * The formula used for evolving capabilities is:
- *
- *       pI' = pI
- * (***) pP' = (fP & X) | (fI & pI)
- *       pE' = pP' & fE          [NB. fE is 0 or ~0]
- *
- * I=Inheritable, P=Permitted, E=Effective // p=process, f=file
- * ' indicates post-exec(), and X is the global 'cap_bset'.
- *
- */
+static inline int must_not_trace_exec (struct task_struct *p)
+{
+	int unsafe = 0;
+	if (p->ptrace & PT_PTRACED) {
+		if(p->ptrace & PT_PTRACE_CAP)
+			unsafe |= LSM_UNSAFE_PTRACE_CAP;
+		else
+			unsafe |= LSM_UNSAFE_PTRACE;
+	}
+	if (atomic_read(&p->fs->count) > 1 ||
+	    atomic_read(&p->files->count) > 1 ||
+	    atomic_read(&p->sighand->count) > 1)
+		unsafe |= LSM_UNSAFE_SHARE;
+
+	return unsafe;
+}
 
 void compute_creds(struct linux_binprm *bprm)
 {
-	security_bprm_apply_creds(bprm);
+	task_lock(current);
+	security_bprm_apply_creds(bprm, must_not_trace_exec(current));
+	task_unlock(current);
 }
 
 EXPORT_SYMBOL(compute_creds);
security/selinux/hooks.c renamed from security/selinux/hooks.c.ptlock +4 -9
@@ -1746,7 +1746,7 @@
 	spin_unlock(&files->file_lock);
 }
 
-static void selinux_bprm_apply_creds(struct linux_binprm *bprm)
+static void selinux_bprm_apply_creds(struct linux_binprm *bprm, int unsafe)
 {
 	struct task_security_struct *tsec, *psec;
 	struct bprm_security_struct *bsec;
@@ -1756,7 +1756,7 @@
 	struct rlimit *rlim, *initrlim;
 	int rc, i;
 
-	secondary_ops->bprm_apply_creds(bprm);
+	secondary_ops->bprm_apply_creds(bprm, unsafe);
 
 	tsec = current->security;
 
@@ -1767,9 +1767,7 @@
 	if (tsec->sid != sid) {
 		/* Check for shared state.  If not ok, leave SID
 		   unchanged and kill. */
-		if ((atomic_read(&current->fs->count) > 1 ||
-		     atomic_read(&current->files->count) > 1 ||
-		     atomic_read(&current->sighand->count) > 1)) {
+		if (unsafe & LSM_UNSAFE_SHARE) {
 			rc = avc_has_perm(tsec->sid, sid,
 					  SECCLASS_PROCESS, PROCESS__SHARE,
 					  NULL, NULL);
@@ -1781,15 +1779,13 @@
 
 		/* Check for ptracing, and update the task SID if ok.
 		   Otherwise, leave SID unchanged and kill. */
-		task_lock(current);
-		if (current->ptrace & PT_PTRACED) {
+		if (unsafe & (LSM_UNSAFE_PTRACE | LSM_UNSAFE_PTRACE_CAP)) {
 			psec = current->parent->security;
 			rc = avc_has_perm_noaudit(psec->sid, sid,
 					  SECCLASS_PROCESS, PROCESS__PTRACE,
 					  NULL, &avd);
 			if (!rc)
 				tsec->sid = sid;
-			task_unlock(current);
 			avc_audit(psec->sid, sid, SECCLASS_PROCESS,
 				  PROCESS__PTRACE, &avd, rc, NULL);
 			if (rc) {
@@ -1798,7 +1794,6 @@
 			}
 		} else {
 			tsec->sid = sid;
-			task_unlock(current);
 		}
 
 		/* Close files for which the new task SID is not authorized. */
security/commoncap.c renamed from security/commoncap.c.ptlock +12 -26
@@ -115,15 +115,7 @@
 	return 0;
 }
 
-static inline int must_not_trace_exec (struct task_struct *p)
-{
-	return ((p->ptrace & PT_PTRACED) && !(p->ptrace & PT_PTRACE_CAP))
-		|| atomic_read(&p->fs->count) > 1
-		|| atomic_read(&p->files->count) > 1
-		|| atomic_read(&p->sighand->count) > 1;
-}
-
-void cap_bprm_apply_creds (struct linux_binprm *bprm)
+void cap_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
 {
 	/* Derived from fs/exec.c:compute_creds. */
 	kernel_cap_t new_permitted, working;
@@ -133,30 +125,25 @@
 				 current->cap_inheritable);
 	new_permitted = cap_combine (new_permitted, working);
 
-	task_lock(current);
-
-	if (bprm->e_uid != current->uid || bprm->e_gid != current->gid) {
+	if (bprm->e_uid != current->uid || bprm->e_gid != current->gid ||
+	    !cap_issubset (new_permitted, current->cap_permitted)) {
 		current->mm->dumpable = 0;
 
-		if (must_not_trace_exec(current) && !capable(CAP_SETUID)) {
-			bprm->e_uid = current->uid;
-			bprm->e_gid = current->gid;
+		if (unsafe & ~LSM_UNSAFE_PTRACE_CAP) {
+			if (!capable(CAP_SETUID)) {
+				bprm->e_uid = current->uid;
+				bprm->e_gid = current->gid;
+			}
+			if (!capable (CAP_SETPCAP)) {
+				new_permitted = cap_intersect (new_permitted,
+							current->cap_permitted);
+			}
 		}
 	}
 
 	current->suid = current->euid = current->fsuid = bprm->e_uid;
 	current->sgid = current->egid = current->fsgid = bprm->e_gid;
 
-	if (!cap_issubset (new_permitted, current->cap_permitted)) {
-		current->mm->dumpable = 0;
-
-		if (must_not_trace_exec (current) && !capable (CAP_SETPCAP)) {
-			new_permitted = cap_intersect (new_permitted,
-						       current->
-						       cap_permitted);
-		}
-	}
-
 	/* For init, we want to retain the capabilities set
 	 * in the init_task struct. Thus we skip the usual
 	 * capability rules */
@@ -167,7 +154,6 @@
 	}
 
 	/* AUD: Audit candidate if current->cap_effective is set */
-	task_unlock(current);
 
 	current->keep_capabilities = 0;
 }
security/dummy.c renamed from security/dummy.c.ptlock +2 -13
@@ -171,21 +171,12 @@
 	return;
 }
 
-static inline int must_not_trace_exec (struct task_struct *p)
+static void dummy_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
 {
-	return ((p->ptrace & PT_PTRACED) && !(p->ptrace & PT_PTRACE_CAP))
-		|| atomic_read(&p->fs->count) > 1
-		|| atomic_read(&p->files->count) > 1
-		|| atomic_read(&p->sighand->count) > 1;
-}
-
-static void dummy_bprm_apply_creds (struct linux_binprm *bprm)
-{
-	task_lock(current);
 	if (bprm->e_uid != current->uid || bprm->e_gid != current->gid) {
 		current->mm->dumpable = 0;
 
-		if (must_not_trace_exec(current) && !capable(CAP_SETUID)) {
+		if (unsafe && !capable(CAP_SETUID)) {
 			bprm->e_uid = current->uid;
 			bprm->e_gid = current->gid;
 		}
@@ -193,8 +184,6 @@
 
 	current->suid = current->euid = current->fsuid = bprm->e_uid;
 	current->sgid = current->egid = current->fsgid = bprm->e_gid;
-
-	task_unlock(current);
 }
 
 static int dummy_bprm_set_security (struct linux_binprm *bprm)
#7 lore raw
Quoted 5 lines · Andy Lutomirski

I was worried about sid changing but uid and caps staying the same if a ptrace_detach or _exit happens between the cap_bprm_apply_creds call and the rest of selinux_bprm_apply_creds. Remember the sendmail bug -- program failure due to lack of capabilities can cause privilege leaks (in this case selinux sid leaks).

That particular issue shouldn't be a problem, as SELinux security transitions aren't controlled by Linux capabilities and SELinux specifically controls code execution (both entry into a domain and ability to execute anything else without changing domains). However, I do agree that it could yield an unexpected failure in the program that would be harmful, so I'm in favor of checking the state only once.

Signature · 2 lines
Stephen Smalley <[email protected]>
National Security Agency
#8 lore raw
In reply to Andy Lutomirski

fs/exec.c line 945

Quoted 8 lines · * Andy Lutomirski ([email protected]) wrote:
  void compute_creds(struct linux_binprm *bprm)
  {
fs/exec.c renamed from fs/exec.c.ptlock +5 -1
-	security_bprm_apply_creds(bprm);
+	task_lock(current);
+
+	security_bprm_apply_creds(bprm, must_not_trace_exec(current));
+
+	task_unlock(current);

unecessary extra lines.

security/selinux/hooks.c line 1782

Quoted 5 lines
security/selinux/hooks.c renamed from security/selinux/hooks.c.ptlock +1 -0
+static void selinux_bprm_apply_creds(struct linux_binprm *bprm, int unsafe)
@@ -1782,14 +1780,13 @@
  		/* Check for ptracing, and update the task SID if ok.
  		   Otherwise, leave SID unchanged and kill. */
  		task_lock(current);

oops ;-)

security/selinux/hooks.c line 1783

Quoted 2 lines
security/selinux/hooks.c renamed from security/selinux/hooks.c.ptlock +1 -1
-		if (current->ptrace & PT_PTRACED) {
+		if (unsafe & (LSM_UNSAFE_PTRACE | LSM_UNSAFE_PTRACE_CAP)) {

thanks,
-chris

Signature · 1 line
Linux Security Modules     http://lsm.immunix.org     http://lsm.bkbits.net