*   Add any capability from current's capability bounding set
  *       to the current process' inheritable set
  *   Allow taking bits out of capability bounding set
+ *   Allow modification of the securebits for a process
  */
 
 #define CAP_SETPCAP          8
 int capable(int cap);
 int __capable(struct task_struct *t, int cap);
 
-extern long cap_prctl_drop(unsigned long cap);
-
 #endif /* __KERNEL__ */
 
 #endif /* !_LINUX_CAPABILITY_H */
 
 #include <linux/ipc.h>
 #include <linux/pid_namespace.h>
 #include <linux/user_namespace.h>
+#include <linux/securebits.h>
 #include <net/net_namespace.h>
 
 #define INIT_FDTABLE \
        .cap_inheritable = CAP_INIT_INH_SET,                            \
        .cap_permitted  = CAP_FULL_SET,                                 \
        .cap_bset       = CAP_INIT_BSET,                                \
-       .keep_capabilities = 0,                                         \
+       .securebits     = SECUREBITS_DEFAULT,                           \
        .user           = INIT_USER,                                    \
        .comm           = "swapper",                                    \
        .thread         = INIT_THREAD,                                  \
 
 # define PR_UNALIGN_NOPRINT    1       /* silently fix up unaligned user accesses */
 # define PR_UNALIGN_SIGBUS     2       /* generate SIGBUS on unaligned user access */
 
-/* Get/set whether or not to drop capabilities on setuid() away from uid 0 */
+/* Get/set whether or not to drop capabilities on setuid() away from
+ * uid 0 (as per security/commoncap.c) */
 #define PR_GET_KEEPCAPS   7
 #define PR_SET_KEEPCAPS   8
 
 #define PR_GET_SECCOMP 21
 #define PR_SET_SECCOMP 22
 
-/* Get/set the capability bounding set */
+/* Get/set the capability bounding set (as per security/commoncap.c) */
 #define PR_CAPBSET_READ 23
 #define PR_CAPBSET_DROP 24
 
 # define PR_TSC_ENABLE         1       /* allow the use of the timestamp counter */
 # define PR_TSC_SIGSEGV                2       /* throw a SIGSEGV instead of reading the TSC */
 
+/* Get/set securebits (as per security/commoncap.c) */
+#define PR_GET_SECUREBITS 27
+#define PR_SET_SECUREBITS 28
+
 #endif /* _LINUX_PRCTL_H */
 
 #include <linux/smp.h>
 #include <linux/sem.h>
 #include <linux/signal.h>
-#include <linux/securebits.h>
 #include <linux/fs_struct.h>
 #include <linux/compiler.h>
 #include <linux/completion.h>
        gid_t gid,egid,sgid,fsgid;
        struct group_info *group_info;
        kernel_cap_t   cap_effective, cap_inheritable, cap_permitted, cap_bset;
-       unsigned keep_capabilities:1;
+       unsigned securebits;
        struct user_struct *user;
 #ifdef CONFIG_KEYS
        struct key *request_key_auth;   /* assumed request_key authority */
 
 
 #define SECUREBITS_DEFAULT 0x00000000
 
-extern unsigned securebits;
-
 /* When set UID 0 has no special privileges. When unset, we support
    inheritance of root-permissions and suid-root executable under
    compatibility mode. We raise the effective and inheritable bitmasks
    *of the executable file* if the effective uid of the new process is
    0. If the real uid is 0, we raise the inheritable bitmask of the
    executable file. */
-#define SECURE_NOROOT            0
+#define SECURE_NOROOT                  0
+#define SECURE_NOROOT_LOCKED           1  /* make bit-0 immutable */
 
 /* When set, setuid to/from uid 0 does not trigger capability-"fixes"
    to be compatible with old programs relying on set*uid to loose
    privileges. When unset, setuid doesn't change privileges. */
-#define SECURE_NO_SETUID_FIXUP   2
+#define SECURE_NO_SETUID_FIXUP         2
+#define SECURE_NO_SETUID_FIXUP_LOCKED  3  /* make bit-2 immutable */
+
+/* When set, a process can retain its capabilities even after
+   transitioning to a non-root user (the set-uid fixup suppressed by
+   bit 2). Bit-4 is cleared when a process calls exec(); setting both
+   bit 4 and 5 will create a barrier through exec that no exec()'d
+   child can use this feature again. */
+#define SECURE_KEEP_CAPS               4
+#define SECURE_KEEP_CAPS_LOCKED                5  /* make bit-4 immutable */
 
 /* Each securesetting is implemented using two bits. One bit specify
    whether the setting is on or off. The other bit specify whether the
    setting is fixed or not. A setting which is fixed cannot be changed
    from user-level. */
+#define issecure_mask(X)       (1 << (X))
+#define issecure(X)            (issecure_mask(X) & current->securebits)
 
-#define issecure(X) ( (1 << (X+1)) & SECUREBITS_DEFAULT ?      \
-                     (1 << (X)) & SECUREBITS_DEFAULT :         \
-                     (1 << (X)) & securebits )
+#define SECURE_ALL_BITS                (issecure_mask(SECURE_NOROOT) | \
+                                issecure_mask(SECURE_NO_SETUID_FIXUP) | \
+                                issecure_mask(SECURE_KEEP_CAPS))
+#define SECURE_ALL_LOCKS       (SECURE_ALL_BITS << 1)
 
 #endif /* !_LINUX_SECUREBITS_H */
 
 #include <linux/xfrm.h>
 #include <net/flow.h>
 
-extern unsigned securebits;
-
 /* Maximum number of letters for an LSM name string */
 #define SECURITY_NAME_MAX      10
 
 extern int cap_inode_killpriv(struct dentry *dentry);
 extern int cap_task_post_setuid (uid_t old_ruid, uid_t old_euid, uid_t old_suid, int flags);
 extern void cap_task_reparent_to_init (struct task_struct *p);
+extern int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
+                         unsigned long arg4, unsigned long arg5, long *rc_p);
 extern int cap_task_setscheduler (struct task_struct *p, int policy, struct sched_param *lp);
 extern int cap_task_setioprio (struct task_struct *p, int ioprio);
 extern int cap_task_setnice (struct task_struct *p, int nice);
  *     @arg3 contains a argument.
  *     @arg4 contains a argument.
  *     @arg5 contains a argument.
- *     Return 0 if permission is granted.
+ *      @rc_p contains a pointer to communicate back the forced return code
+ *     Return 0 if permission is granted, and non-zero if the security module
+ *      has taken responsibility (setting *rc_p) for the prctl call.
  * @task_reparent_to_init:
  *     Set the security attributes in @p->security for a kernel thread that
  *     is being reparented to the init task.
        int (*task_wait) (struct task_struct * p);
        int (*task_prctl) (int option, unsigned long arg2,
                           unsigned long arg3, unsigned long arg4,
-                          unsigned long arg5);
+                          unsigned long arg5, long *rc_p);
        void (*task_reparent_to_init) (struct task_struct * p);
        void (*task_to_inode)(struct task_struct *p, struct inode *inode);
 
                        int sig, u32 secid);
 int security_task_wait(struct task_struct *p);
 int security_task_prctl(int option, unsigned long arg2, unsigned long arg3,
-                        unsigned long arg4, unsigned long arg5);
+                        unsigned long arg4, unsigned long arg5, long *rc_p);
 void security_task_reparent_to_init(struct task_struct *p);
 void security_task_to_inode(struct task_struct *p, struct inode *inode);
 int security_ipc_permission(struct kern_ipc_perm *ipcp, short flag);
 static inline int security_task_prctl (int option, unsigned long arg2,
                                       unsigned long arg3,
                                       unsigned long arg4,
-                                      unsigned long arg5)
+                                      unsigned long arg5, long *rc_p)
 {
-       return 0;
+       return cap_task_prctl(option, arg2, arg3, arg3, arg5, rc_p);
 }
 
 static inline void security_task_reparent_to_init (struct task_struct *p)
 
 asmlinkage long sys_prctl(int option, unsigned long arg2, unsigned long arg3,
                          unsigned long arg4, unsigned long arg5)
 {
-       long error;
+       long uninitialized_var(error);
 
-       error = security_task_prctl(option, arg2, arg3, arg4, arg5);
-       if (error)
+       if (security_task_prctl(option, arg2, arg3, arg4, arg5, &error))
                return error;
 
        switch (option) {
                                error = -EINVAL;
                        break;
 
-               case PR_GET_KEEPCAPS:
-                       if (current->keep_capabilities)
-                               error = 1;
-                       break;
-               case PR_SET_KEEPCAPS:
-                       if (arg2 != 0 && arg2 != 1) {
-                               error = -EINVAL;
-                               break;
-                       }
-                       current->keep_capabilities = arg2;
-                       break;
                case PR_SET_NAME: {
                        struct task_struct *me = current;
                        unsigned char ncomm[sizeof(me->comm)];
                case PR_SET_SECCOMP:
                        error = prctl_set_seccomp(arg2);
                        break;
-
-               case PR_CAPBSET_READ:
-                       if (!cap_valid(arg2))
-                               return -EINVAL;
-                       return !!cap_raised(current->cap_bset, arg2);
-               case PR_CAPBSET_DROP:
-#ifdef CONFIG_SECURITY_FILE_CAPABILITIES
-                       return cap_prctl_drop(arg2);
-#else
-                       return -EINVAL;
-#endif
                case PR_GET_TSC:
                        error = GET_TSC_CTL(arg2);
                        break;
 
        .task_setioprio =               cap_task_setioprio,
        .task_setnice =                 cap_task_setnice,
        .task_post_setuid =             cap_task_post_setuid,
+       .task_prctl =                   cap_task_prctl,
        .task_reparent_to_init =        cap_task_reparent_to_init,
 
        .syslog =                       cap_syslog,
 
 #include <linux/hugetlb.h>
 #include <linux/mount.h>
 #include <linux/sched.h>
-
-/* Global security state */
-
-unsigned securebits = SECUREBITS_DEFAULT; /* systemwide security settings */
-EXPORT_SYMBOL(securebits);
+#include <linux/prctl.h>
+#include <linux/securebits.h>
 
 int cap_netlink_send(struct sock *sk, struct sk_buff *skb)
 {
 
        /* AUD: Audit candidate if current->cap_effective is set */
 
-       current->keep_capabilities = 0;
+       current->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
 }
 
 int cap_bprm_secureexec (struct linux_binprm *bprm)
 {
        if ((old_ruid == 0 || old_euid == 0 || old_suid == 0) &&
            (current->uid != 0 && current->euid != 0 && current->suid != 0) &&
-           !current->keep_capabilities) {
+           !issecure(SECURE_KEEP_CAPS)) {
                cap_clear (current->cap_permitted);
                cap_clear (current->cap_effective);
        }
  * this task could get inconsistent info.  There can be no
  * racing writer bc a task can only change its own caps.
  */
-long cap_prctl_drop(unsigned long cap)
+static long cap_prctl_drop(unsigned long cap)
 {
        if (!capable(CAP_SETPCAP))
                return -EPERM;
        cap_lower(current->cap_bset, cap);
        return 0;
 }
+
 #else
 int cap_task_setscheduler (struct task_struct *p, int policy,
                           struct sched_param *lp)
 }
 #endif
 
+int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
+                  unsigned long arg4, unsigned long arg5, long *rc_p)
+{
+       long error = 0;
+
+       switch (option) {
+       case PR_CAPBSET_READ:
+               if (!cap_valid(arg2))
+                       error = -EINVAL;
+               else
+                       error = !!cap_raised(current->cap_bset, arg2);
+               break;
+#ifdef CONFIG_SECURITY_FILE_CAPABILITIES
+       case PR_CAPBSET_DROP:
+               error = cap_prctl_drop(arg2);
+               break;
+
+       /*
+        * The next four prctl's remain to assist with transitioning a
+        * system from legacy UID=0 based privilege (when filesystem
+        * capabilities are not in use) to a system using filesystem
+        * capabilities only - as the POSIX.1e draft intended.
+        *
+        * Note:
+        *
+        *  PR_SET_SECUREBITS =
+        *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
+        *    | issecure_mask(SECURE_NOROOT)
+        *    | issecure_mask(SECURE_NOROOT_LOCKED)
+        *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
+        *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
+        *
+        * will ensure that the current process and all of its
+        * children will be locked into a pure
+        * capability-based-privilege environment.
+        */
+       case PR_SET_SECUREBITS:
+               if ((((current->securebits & SECURE_ALL_LOCKS) >> 1)
+                    & (current->securebits ^ arg2))                  /*[1]*/
+                   || ((current->securebits & SECURE_ALL_LOCKS
+                        & ~arg2))                                    /*[2]*/
+                   || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS)) /*[3]*/
+                   || (cap_capable(current, CAP_SETPCAP) != 0)) {    /*[4]*/
+                       /*
+                        * [1] no changing of bits that are locked
+                        * [2] no unlocking of locks
+                        * [3] no setting of unsupported bits
+                        * [4] doing anything requires privilege (go read about
+                        *     the "sendmail capabilities bug")
+                        */
+                       error = -EPERM;  /* cannot change a locked bit */
+               } else {
+                       current->securebits = arg2;
+               }
+               break;
+       case PR_GET_SECUREBITS:
+               error = current->securebits;
+               break;
+
+#endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
+
+       case PR_GET_KEEPCAPS:
+               if (issecure(SECURE_KEEP_CAPS))
+                       error = 1;
+               break;
+       case PR_SET_KEEPCAPS:
+               if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
+                       error = -EINVAL;
+               else if (issecure(SECURE_KEEP_CAPS_LOCKED))
+                       error = -EPERM;
+               else if (arg2)
+                       current->securebits |= issecure_mask(SECURE_KEEP_CAPS);
+               else
+                       current->securebits &=
+                               ~issecure_mask(SECURE_KEEP_CAPS);
+               break;
+
+       default:
+               /* No functionality available - continue with default */
+               return 0;
+       }
+
+       /* Functionality provided */
+       *rc_p = error;
+       return 1;
+}
+
 void cap_task_reparent_to_init (struct task_struct *p)
 {
        cap_set_init_eff(p->cap_effective);
        cap_clear(p->cap_inheritable);
        cap_set_full(p->cap_permitted);
-       p->keep_capabilities = 0;
+       p->securebits = SECUREBITS_DEFAULT;
        return;
 }
 
 
 }
 
 static int dummy_task_prctl (int option, unsigned long arg2, unsigned long arg3,
-                            unsigned long arg4, unsigned long arg5)
+                            unsigned long arg4, unsigned long arg5, long *rc_p)
 {
        return 0;
 }
 
 }
 
 int security_task_prctl(int option, unsigned long arg2, unsigned long arg3,
-                        unsigned long arg4, unsigned long arg5)
+                        unsigned long arg4, unsigned long arg5, long *rc_p)
 {
-       return security_ops->task_prctl(option, arg2, arg3, arg4, arg5);
+       return security_ops->task_prctl(option, arg2, arg3, arg4, arg5, rc_p);
 }
 
 void security_task_reparent_to_init(struct task_struct *p)
 
                              unsigned long arg2,
                              unsigned long arg3,
                              unsigned long arg4,
-                             unsigned long arg5)
+                             unsigned long arg5,
+                             long *rc_p)
 {
        /* The current prctl operations do not appear to require
           any SELinux controls since they merely observe or modify
           the state of the current process. */
-       return 0;
+       return secondary_ops->task_prctl(option, arg2, arg3, arg4, arg5, rc_p);
 }
 
 static int selinux_task_wait(struct task_struct *p)