Security: split proc ptrace checking into read vs. attach
[safe/jmp/linux-2.6] / security / commoncap.c
index 778cb0c..0b6537a 100644 (file)
@@ -1,4 +1,4 @@
-/* Common capabilities, needed by capability.o and root_plug.o 
+/* Common capabilities, needed by capability.o and root_plug.o
  *
  *     This program is free software; you can redistribute it and/or modify
  *     it under the terms of the GNU General Public License as published by
@@ -23,6 +23,9 @@
 #include <linux/xattr.h>
 #include <linux/hugetlb.h>
 #include <linux/mount.h>
+#include <linux/sched.h>
+#include <linux/prctl.h>
+#include <linux/securebits.h>
 
 int cap_netlink_send(struct sock *sk, struct sk_buff *skb)
 {
@@ -39,6 +42,12 @@ int cap_netlink_recv(struct sk_buff *skb, int cap)
 
 EXPORT_SYMBOL(cap_netlink_recv);
 
+/*
+ * NOTE WELL: cap_capable() cannot be used like the kernel's capable()
+ * function.  That is, it has the reverse semantics: cap_capable()
+ * returns 0 when a task has a capability, but the kernel's capable()
+ * returns 1 for this case.
+ */
 int cap_capable (struct task_struct *tsk, int cap)
 {
        /* Derived from include/linux/sched.h:capable. */
@@ -54,7 +63,8 @@ int cap_settime(struct timespec *ts, struct timezone *tz)
        return 0;
 }
 
-int cap_ptrace (struct task_struct *parent, struct task_struct *child)
+int cap_ptrace (struct task_struct *parent, struct task_struct *child,
+               unsigned int mode)
 {
        /* Derived from arch/i386/kernel/ptrace.c:sys_ptrace. */
        if (!cap_issubset(child->cap_permitted, parent->cap_permitted) &&
@@ -67,20 +77,63 @@ int cap_capget (struct task_struct *target, kernel_cap_t *effective,
                kernel_cap_t *inheritable, kernel_cap_t *permitted)
 {
        /* Derived from kernel/capability.c:sys_capget. */
-       *effective = cap_t (target->cap_effective);
-       *inheritable = cap_t (target->cap_inheritable);
-       *permitted = cap_t (target->cap_permitted);
+       *effective = target->cap_effective;
+       *inheritable = target->cap_inheritable;
+       *permitted = target->cap_permitted;
        return 0;
 }
 
+#ifdef CONFIG_SECURITY_FILE_CAPABILITIES
+
+static inline int cap_block_setpcap(struct task_struct *target)
+{
+       /*
+        * No support for remote process capability manipulation with
+        * filesystem capability support.
+        */
+       return (target != current);
+}
+
+static inline int cap_inh_is_capped(void)
+{
+       /*
+        * Return 1 if changes to the inheritable set are limited
+        * to the old permitted set. That is, if the current task
+        * does *not* possess the CAP_SETPCAP capability.
+        */
+       return (cap_capable(current, CAP_SETPCAP) != 0);
+}
+
+static inline int cap_limit_ptraced_target(void) { return 1; }
+
+#else /* ie., ndef CONFIG_SECURITY_FILE_CAPABILITIES */
+
+static inline int cap_block_setpcap(struct task_struct *t) { return 0; }
+static inline int cap_inh_is_capped(void) { return 1; }
+static inline int cap_limit_ptraced_target(void)
+{
+       return !capable(CAP_SETPCAP);
+}
+
+#endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
+
 int cap_capset_check (struct task_struct *target, kernel_cap_t *effective,
                      kernel_cap_t *inheritable, kernel_cap_t *permitted)
 {
-       /* Derived from kernel/capability.c:sys_capset. */
-       /* verify restrictions on target's new Inheritable set */
-       if (!cap_issubset (*inheritable,
-                          cap_combine (target->cap_inheritable,
-                                       current->cap_permitted))) {
+       if (cap_block_setpcap(target)) {
+               return -EPERM;
+       }
+       if (cap_inh_is_capped()
+           && !cap_issubset(*inheritable,
+                            cap_combine(target->cap_inheritable,
+                                        current->cap_permitted))) {
+               /* incapable of using this inheritable set */
+               return -EPERM;
+       }
+       if (!cap_issubset(*inheritable,
+                          cap_combine(target->cap_inheritable,
+                                      current->cap_bset))) {
+               /* no new pI capabilities outside bounding set */
                return -EPERM;
        }
 
@@ -140,28 +193,51 @@ int cap_inode_killpriv(struct dentry *dentry)
        return inode->i_op->removexattr(dentry, XATTR_NAME_CAPS);
 }
 
-static inline int cap_from_disk(__le32 *caps, struct linux_binprm *bprm,
-                               int size)
+static inline int cap_from_disk(struct vfs_cap_data *caps,
+                               struct linux_binprm *bprm, unsigned size)
 {
        __u32 magic_etc;
+       unsigned tocopy, i;
 
-       if (size != XATTR_CAPS_SZ)
+       if (size < sizeof(magic_etc))
                return -EINVAL;
 
-       magic_etc = le32_to_cpu(caps[0]);
+       magic_etc = le32_to_cpu(caps->magic_etc);
 
        switch ((magic_etc & VFS_CAP_REVISION_MASK)) {
-       case VFS_CAP_REVISION:
-               if (magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
-                       bprm->cap_effective = true;
-               else
-                       bprm->cap_effective = false;
-               bprm->cap_permitted = to_cap_t( le32_to_cpu(caps[1]) );
-               bprm->cap_inheritable = to_cap_t( le32_to_cpu(caps[2]) );
-               return 0;
+       case VFS_CAP_REVISION_1:
+               if (size != XATTR_CAPS_SZ_1)
+                       return -EINVAL;
+               tocopy = VFS_CAP_U32_1;
+               break;
+       case VFS_CAP_REVISION_2:
+               if (size != XATTR_CAPS_SZ_2)
+                       return -EINVAL;
+               tocopy = VFS_CAP_U32_2;
+               break;
        default:
                return -EINVAL;
        }
+
+       if (magic_etc & VFS_CAP_FLAGS_EFFECTIVE) {
+               bprm->cap_effective = true;
+       } else {
+               bprm->cap_effective = false;
+       }
+
+       for (i = 0; i < tocopy; ++i) {
+               bprm->cap_permitted.cap[i] =
+                       le32_to_cpu(caps->data[i].permitted);
+               bprm->cap_inheritable.cap[i] =
+                       le32_to_cpu(caps->data[i].inheritable);
+       }
+       while (i < VFS_CAP_U32) {
+               bprm->cap_permitted.cap[i] = 0;
+               bprm->cap_inheritable.cap[i] = 0;
+               i++;
+       }
+
+       return 0;
 }
 
 /* Locate any VFS capabilities: */
@@ -169,7 +245,7 @@ static int get_file_caps(struct linux_binprm *bprm)
 {
        struct dentry *dentry;
        int rc = 0;
-       __le32 v1caps[XATTR_CAPS_SZ];
+       struct vfs_cap_data vcaps;
        struct inode *inode;
 
        if (bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID) {
@@ -182,8 +258,8 @@ static int get_file_caps(struct linux_binprm *bprm)
        if (!inode->i_op || !inode->i_op->getxattr)
                goto out;
 
-       rc = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, &v1caps,
-                                                       XATTR_CAPS_SZ);
+       rc = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, &vcaps,
+                                  XATTR_CAPS_SZ);
        if (rc == -ENODATA || rc == -EOPNOTSUPP) {
                /* no data, that's ok */
                rc = 0;
@@ -192,10 +268,10 @@ static int get_file_caps(struct linux_binprm *bprm)
        if (rc < 0)
                goto out;
 
-       rc = cap_from_disk(v1caps, bprm, rc);
+       rc = cap_from_disk(&vcaps, bprm, rc);
        if (rc)
                printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n",
-                       __FUNCTION__, rc, bprm->filename);
+                       __func__, rc, bprm->filename);
 
 out:
        dput(dentry);
@@ -230,7 +306,7 @@ int cap_bprm_set_security (struct linux_binprm *bprm)
        ret = get_file_caps(bprm);
        if (ret)
                printk(KERN_NOTICE "%s: get_file_caps returned %d for %s\n",
-                       __FUNCTION__, ret, bprm->filename);
+                       __func__, ret, bprm->filename);
 
        /*  To support inheritance of root-permissions and suid-root
         *  executables under compatibility mode, we raise all three
@@ -257,10 +333,11 @@ void cap_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
        /* Derived from fs/exec.c:compute_creds. */
        kernel_cap_t new_permitted, working;
 
-       new_permitted = cap_intersect (bprm->cap_permitted, cap_bset);
-       working = cap_intersect (bprm->cap_inheritable,
+       new_permitted = cap_intersect(bprm->cap_permitted,
+                                current->cap_bset);
+       working = cap_intersect(bprm->cap_inheritable,
                                 current->cap_inheritable);
-       new_permitted = cap_combine (new_permitted, working);
+       new_permitted = cap_combine(new_permitted, working);
 
        if (bprm->e_uid != current->uid || bprm->e_gid != current->gid ||
            !cap_issubset (new_permitted, current->cap_permitted)) {
@@ -272,9 +349,10 @@ void cap_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
                                bprm->e_uid = current->uid;
                                bprm->e_gid = current->gid;
                        }
-                       if (!capable (CAP_SETPCAP)) {
-                               new_permitted = cap_intersect (new_permitted,
-                                                       current->cap_permitted);
+                       if (cap_limit_ptraced_target()) {
+                               new_permitted =
+                                       cap_intersect(new_permitted,
+                                                     current->cap_permitted);
                        }
                }
        }
@@ -285,15 +363,17 @@ void cap_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
        /* For init, we want to retain the capabilities set
         * in the init_task struct. Thus we skip the usual
         * capability rules */
-       if (!is_init(current)) {
+       if (!is_global_init(current)) {
                current->cap_permitted = new_permitted;
-               current->cap_effective = bprm->cap_effective ?
-                               new_permitted : 0;
+               if (bprm->cap_effective)
+                       current->cap_effective = new_permitted;
+               else
+                       cap_clear(current->cap_effective);
        }
 
        /* 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)
@@ -311,8 +391,8 @@ int cap_bprm_secureexec (struct linux_binprm *bprm)
                current->egid != current->gid);
 }
 
-int cap_inode_setxattr(struct dentry *dentry, char *name, void *value,
-                      size_t size, int flags)
+int cap_inode_setxattr(struct dentry *dentry, const char *name,
+                      const void *value, size_t size, int flags)
 {
        if (!strcmp(name, XATTR_NAME_CAPS)) {
                if (!capable(CAP_SETFCAP))
@@ -325,7 +405,7 @@ int cap_inode_setxattr(struct dentry *dentry, char *name, void *value,
        return 0;
 }
 
-int cap_inode_removexattr(struct dentry *dentry, char *name)
+int cap_inode_removexattr(struct dentry *dentry, const char *name)
 {
        if (!strcmp(name, XATTR_NAME_CAPS)) {
                if (!capable(CAP_SETFCAP))
@@ -373,7 +453,7 @@ static inline void cap_emulate_setxuid (int old_ruid, int old_euid,
 {
        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);
        }
@@ -410,13 +490,15 @@ int cap_task_post_setuid (uid_t old_ruid, uid_t old_euid, uid_t old_suid,
 
                        if (!issecure (SECURE_NO_SETUID_FIXUP)) {
                                if (old_fsuid == 0 && current->fsuid != 0) {
-                                       cap_t (current->cap_effective) &=
-                                           ~CAP_FS_MASK;
+                                       current->cap_effective =
+                                               cap_drop_fs_set(
+                                                   current->cap_effective);
                                }
                                if (old_fsuid != 0 && current->fsuid == 0) {
-                                       cap_t (current->cap_effective) |=
-                                           (cap_t (current->cap_permitted) &
-                                            CAP_FS_MASK);
+                                       current->cap_effective =
+                                               cap_raise_fs_set(
+                                                   current->cap_effective,
+                                                   current->cap_permitted);
                                }
                        }
                        break;
@@ -463,27 +545,23 @@ int cap_task_setnice (struct task_struct *p, int nice)
        return cap_safe_nice(p);
 }
 
-int cap_task_kill(struct task_struct *p, struct siginfo *info,
-                               int sig, u32 secid)
+/*
+ * called from kernel/sys.c for prctl(PR_CABSET_DROP)
+ * done without task_capability_lock() because it introduces
+ * no new races - i.e. only another task doing capget() on
+ * this task could get inconsistent info.  There can be no
+ * racing writer bc a task can only change its own caps.
+ */
+static long cap_prctl_drop(unsigned long cap)
 {
-       if (info != SEND_SIG_NOINFO && (is_si_special(info) || SI_FROMKERNEL(info)))
-               return 0;
-
-       if (secid)
-               /*
-                * Signal sent as a particular user.
-                * Capabilities are ignored.  May be wrong, but it's the
-                * only thing we can do at the moment.
-                * Used only by usb drivers?
-                */
-               return 0;
-       if (cap_issubset(p->cap_permitted, current->cap_permitted))
-               return 0;
-       if (capable(CAP_KILL))
-               return 0;
-
-       return -EPERM;
+       if (!capable(CAP_SETPCAP))
+               return -EPERM;
+       if (!cap_valid(cap))
+               return -EINVAL;
+       cap_lower(current->cap_bset, cap);
+       return 0;
 }
+
 #else
 int cap_task_setscheduler (struct task_struct *p, int policy,
                           struct sched_param *lp)
@@ -498,19 +576,101 @@ int cap_task_setnice (struct task_struct *p, int nice)
 {
        return 0;
 }
-int cap_task_kill(struct task_struct *p, struct siginfo *info,
-                               int sig, u32 secid)
+#endif
+
+int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
+                  unsigned long arg4, unsigned long arg5, long *rc_p)
 {
-       return 0;
+       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;
 }
-#endif
 
 void cap_task_reparent_to_init (struct task_struct *p)
 {
-       p->cap_effective = CAP_INIT_EFF_SET;
-       p->cap_inheritable = CAP_INIT_INH_SET;
-       p->cap_permitted = CAP_FULL_SET;
-       p->keep_capabilities = 0;
+       cap_set_init_eff(p->cap_effective);
+       cap_clear(p->cap_inheritable);
+       cap_set_full(p->cap_permitted);
+       p->securebits = SECUREBITS_DEFAULT;
        return;
 }