capabilities: introduce per-process capability bounding set
[safe/jmp/linux-2.6] / security / commoncap.c
index 04c12f5..5aba826 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
@@ -7,7 +7,7 @@
  *
  */
 
-#include <linux/config.h>
+#include <linux/capability.h>
 #include <linux/module.h>
 #include <linux/init.h>
 #include <linux/kernel.h>
 #include <linux/mman.h>
 #include <linux/pagemap.h>
 #include <linux/swap.h>
-#include <linux/smp_lock.h>
 #include <linux/skbuff.h>
 #include <linux/netlink.h>
 #include <linux/ptrace.h>
 #include <linux/xattr.h>
 #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);
 
 int cap_netlink_send(struct sock *sk, struct sk_buff *skb)
 {
@@ -30,17 +36,21 @@ int cap_netlink_send(struct sock *sk, struct sk_buff *skb)
        return 0;
 }
 
-EXPORT_SYMBOL(cap_netlink_send);
-
-int cap_netlink_recv(struct sk_buff *skb)
+int cap_netlink_recv(struct sk_buff *skb, int cap)
 {
-       if (!cap_raised(NETLINK_CB(skb).eff_cap, CAP_NET_ADMIN))
+       if (!cap_raised(NETLINK_CB(skb).eff_cap, cap))
                return -EPERM;
        return 0;
 }
 
 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. */
@@ -59,8 +69,8 @@ int cap_settime(struct timespec *ts, struct timezone *tz)
 int cap_ptrace (struct task_struct *parent, struct task_struct *child)
 {
        /* Derived from arch/i386/kernel/ptrace.c:sys_ptrace. */
-       if (!cap_issubset (child->cap_permitted, current->cap_permitted) &&
-           !capable(CAP_SYS_PTRACE))
+       if (!cap_issubset(child->cap_permitted, parent->cap_permitted) &&
+           !__capable(parent, CAP_SYS_PTRACE))
                return -EPERM;
        return 0;
 }
@@ -69,20 +79,57 @@ 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);
+}
+
+#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; }
+
+#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;
        }
 
@@ -109,14 +156,153 @@ void cap_capset_set (struct task_struct *target, kernel_cap_t *effective,
        target->cap_permitted = *permitted;
 }
 
+static inline void bprm_clear_caps(struct linux_binprm *bprm)
+{
+       cap_clear(bprm->cap_inheritable);
+       cap_clear(bprm->cap_permitted);
+       bprm->cap_effective = false;
+}
+
+#ifdef CONFIG_SECURITY_FILE_CAPABILITIES
+
+int cap_inode_need_killpriv(struct dentry *dentry)
+{
+       struct inode *inode = dentry->d_inode;
+       int error;
+
+       if (!inode->i_op || !inode->i_op->getxattr)
+              return 0;
+
+       error = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, NULL, 0);
+       if (error <= 0)
+               return 0;
+       return 1;
+}
+
+int cap_inode_killpriv(struct dentry *dentry)
+{
+       struct inode *inode = dentry->d_inode;
+
+       if (!inode->i_op || !inode->i_op->removexattr)
+              return 0;
+
+       return inode->i_op->removexattr(dentry, XATTR_NAME_CAPS);
+}
+
+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 < sizeof(magic_etc))
+               return -EINVAL;
+
+       magic_etc = le32_to_cpu(caps->magic_etc);
+
+       switch ((magic_etc & VFS_CAP_REVISION_MASK)) {
+       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: */
+static int get_file_caps(struct linux_binprm *bprm)
+{
+       struct dentry *dentry;
+       int rc = 0;
+       struct vfs_cap_data vcaps;
+       struct inode *inode;
+
+       if (bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID) {
+               bprm_clear_caps(bprm);
+               return 0;
+       }
+
+       dentry = dget(bprm->file->f_dentry);
+       inode = dentry->d_inode;
+       if (!inode->i_op || !inode->i_op->getxattr)
+               goto out;
+
+       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;
+               goto out;
+       }
+       if (rc < 0)
+               goto out;
+
+       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);
+
+out:
+       dput(dentry);
+       if (rc)
+               bprm_clear_caps(bprm);
+
+       return rc;
+}
+
+#else
+int cap_inode_need_killpriv(struct dentry *dentry)
+{
+       return 0;
+}
+
+int cap_inode_killpriv(struct dentry *dentry)
+{
+       return 0;
+}
+
+static inline int get_file_caps(struct linux_binprm *bprm)
+{
+       bprm_clear_caps(bprm);
+       return 0;
+}
+#endif
+
 int cap_bprm_set_security (struct linux_binprm *bprm)
 {
-       /* Copied from fs/exec.c:prepare_binprm. */
+       int ret;
 
-       /* We don't have VFS support for capabilities yet */
-       cap_clear (bprm->cap_inheritable);
-       cap_clear (bprm->cap_permitted);
-       cap_clear (bprm->cap_effective);
+       ret = get_file_caps(bprm);
+       if (ret)
+               printk(KERN_NOTICE "%s: get_file_caps returned %d for %s\n",
+                       __FUNCTION__, ret, bprm->filename);
 
        /*  To support inheritance of root-permissions and suid-root
         *  executables under compatibility mode, we raise all three
@@ -132,9 +318,10 @@ int cap_bprm_set_security (struct linux_binprm *bprm)
                        cap_set_full (bprm->cap_permitted);
                }
                if (bprm->e_uid == 0)
-                       cap_set_full (bprm->cap_effective);
+                       bprm->cap_effective = true;
        }
-       return 0;
+
+       return ret;
 }
 
 void cap_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
@@ -142,14 +329,16 @@ 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)) {
-               current->mm->dumpable = suid_dumpable;
+               set_dumpable(current->mm, suid_dumpable);
+               current->pdeath_signal = 0;
 
                if (unsafe & ~LSM_UNSAFE_PTRACE_CAP) {
                        if (!capable(CAP_SETUID)) {
@@ -169,10 +358,12 @@ 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 (current->pid != 1) {
+       if (!is_global_init(current)) {
                current->cap_permitted = new_permitted;
-               current->cap_effective =
-                   cap_intersect (new_permitted, bprm->cap_effective);
+               if (bprm->cap_effective)
+                       current->cap_effective = new_permitted;
+               else
+                       cap_clear(current->cap_effective);
        }
 
        /* AUD: Audit candidate if current->cap_effective is set */
@@ -182,11 +373,15 @@ void cap_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
 
 int cap_bprm_secureexec (struct linux_binprm *bprm)
 {
-       /* If/when this module is enhanced to incorporate capability
-          bits on files, the test below should be extended to also perform a 
-          test between the old and new capability sets.  For now,
-          it simply preserves the legacy decision algorithm used by
-          the old userland. */
+       if (current->uid != 0) {
+               if (bprm->cap_effective)
+                       return 1;
+               if (!cap_isclear(bprm->cap_permitted))
+                       return 1;
+               if (!cap_isclear(bprm->cap_inheritable))
+                       return 1;
+       }
+
        return (current->euid != current->uid ||
                current->egid != current->gid);
 }
@@ -194,7 +389,11 @@ int cap_bprm_secureexec (struct linux_binprm *bprm)
 int cap_inode_setxattr(struct dentry *dentry, char *name, void *value,
                       size_t size, int flags)
 {
-       if (!strncmp(name, XATTR_SECURITY_PREFIX,
+       if (!strcmp(name, XATTR_NAME_CAPS)) {
+               if (!capable(CAP_SETFCAP))
+                       return -EPERM;
+               return 0;
+       } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
                     sizeof(XATTR_SECURITY_PREFIX) - 1)  &&
            !capable(CAP_SYS_ADMIN))
                return -EPERM;
@@ -203,7 +402,11 @@ int cap_inode_setxattr(struct dentry *dentry, char *name, void *value,
 
 int cap_inode_removexattr(struct dentry *dentry, char *name)
 {
-       if (!strncmp(name, XATTR_SECURITY_PREFIX,
+       if (!strcmp(name, XATTR_NAME_CAPS)) {
+               if (!capable(CAP_SETFCAP))
+                       return -EPERM;
+               return 0;
+       } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
                     sizeof(XATTR_SECURITY_PREFIX) - 1)  &&
            !capable(CAP_SYS_ADMIN))
                return -EPERM;
@@ -282,13 +485,15 @@ 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;
@@ -300,11 +505,118 @@ int cap_task_post_setuid (uid_t old_ruid, uid_t old_euid, uid_t old_suid,
        return 0;
 }
 
+#ifdef CONFIG_SECURITY_FILE_CAPABILITIES
+/*
+ * Rationale: code calling task_setscheduler, task_setioprio, and
+ * task_setnice, assumes that
+ *   . if capable(cap_sys_nice), then those actions should be allowed
+ *   . if not capable(cap_sys_nice), but acting on your own processes,
+ *     then those actions should be allowed
+ * This is insufficient now since you can call code without suid, but
+ * yet with increased caps.
+ * So we check for increased caps on the target process.
+ */
+static inline int cap_safe_nice(struct task_struct *p)
+{
+       if (!cap_issubset(p->cap_permitted, current->cap_permitted) &&
+           !__capable(current, CAP_SYS_NICE))
+               return -EPERM;
+       return 0;
+}
+
+int cap_task_setscheduler (struct task_struct *p, int policy,
+                          struct sched_param *lp)
+{
+       return cap_safe_nice(p);
+}
+
+int cap_task_setioprio (struct task_struct *p, int ioprio)
+{
+       return cap_safe_nice(p);
+}
+
+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)
+{
+       if (info != SEND_SIG_NOINFO && (is_si_special(info) || SI_FROMKERNEL(info)))
+               return 0;
+
+       /*
+        * Running a setuid root program raises your capabilities.
+        * Killing your own setuid root processes was previously
+        * allowed.
+        * We must preserve legacy signal behavior in this case.
+        */
+       if (p->euid == 0 && p->uid == current->uid)
+               return 0;
+
+       /* sigcont is permitted within same session */
+       if (sig == SIGCONT && (task_session_nr(current) == task_session_nr(p)))
+               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;
+}
+
+/*
+ * 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.
+ */
+long cap_prctl_drop(unsigned long cap)
+{
+       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)
+{
+       return 0;
+}
+int cap_task_setioprio (struct task_struct *p, int ioprio)
+{
+       return 0;
+}
+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)
+{
+       return 0;
+}
+#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;
+       cap_set_init_eff(p->cap_effective);
+       cap_clear(p->cap_inheritable);
+       cap_set_full(p->cap_permitted);
        p->keep_capabilities = 0;
        return;
 }
@@ -316,30 +628,12 @@ int cap_syslog (int type)
        return 0;
 }
 
-int cap_vm_enough_memory(long pages)
+int cap_vm_enough_memory(struct mm_struct *mm, long pages)
 {
        int cap_sys_admin = 0;
 
        if (cap_capable(current, CAP_SYS_ADMIN) == 0)
                cap_sys_admin = 1;
-       return __vm_enough_memory(pages, cap_sys_admin);
-}
-
-EXPORT_SYMBOL(cap_capable);
-EXPORT_SYMBOL(cap_settime);
-EXPORT_SYMBOL(cap_ptrace);
-EXPORT_SYMBOL(cap_capget);
-EXPORT_SYMBOL(cap_capset_check);
-EXPORT_SYMBOL(cap_capset_set);
-EXPORT_SYMBOL(cap_bprm_set_security);
-EXPORT_SYMBOL(cap_bprm_apply_creds);
-EXPORT_SYMBOL(cap_bprm_secureexec);
-EXPORT_SYMBOL(cap_inode_setxattr);
-EXPORT_SYMBOL(cap_inode_removexattr);
-EXPORT_SYMBOL(cap_task_post_setuid);
-EXPORT_SYMBOL(cap_task_reparent_to_init);
-EXPORT_SYMBOL(cap_syslog);
-EXPORT_SYMBOL(cap_vm_enough_memory);
-
-MODULE_DESCRIPTION("Standard Linux Common Capabilities Security Module");
-MODULE_LICENSE("GPL");
+       return __vm_enough_memory(mm, pages, cap_sys_admin);
+}
+