Cache user_ns in struct cred
[linux-2.6.git] / kernel / capability.c
index 901e0fd..32a80e0 100644 (file)
@@ -7,21 +7,17 @@
  * 30 May 2002:        Cleanup, Robert M. Love <rml@tech9.net>
  */
 
+#include <linux/audit.h>
 #include <linux/capability.h>
 #include <linux/mm.h>
 #include <linux/module.h>
 #include <linux/security.h>
 #include <linux/syscalls.h>
 #include <linux/pid_namespace.h>
+#include <linux/user_namespace.h>
 #include <asm/uaccess.h>
 
 /*
- * This lock protects task->cap_* for all tasks including current.
- * Locking rule: acquire this prior to tasklist_lock.
- */
-static DEFINE_SPINLOCK(task_capability_lock);
-
-/*
  * Leveraged for setting/resetting capabilities
  */
 
@@ -33,6 +29,15 @@ EXPORT_SYMBOL(__cap_empty_set);
 EXPORT_SYMBOL(__cap_full_set);
 EXPORT_SYMBOL(__cap_init_eff_set);
 
+int file_caps_enabled = 1;
+
+static int __init file_caps_disable(char *str)
+{
+       file_caps_enabled = 0;
+       return 1;
+}
+__setup("no_file_caps", file_caps_disable);
+
 /*
  * More recent versions of libcap are available from:
  *
@@ -116,31 +121,34 @@ static int cap_validate_magic(cap_user_header_t header, unsigned *tocopy)
 }
 
 /*
- * For sys_getproccap() and sys_setproccap(), any of the three
- * capability set pointers may be NULL -- indicating that that set is
- * uninteresting and/or not to be changed.
+ * The only thing that can change the capabilities of the current
+ * process is the current process. As such, we can't be in this code
+ * at the same time as we are in the process of setting capabilities
+ * in this process. The net result is that we can limit our use of
+ * locks to when we are reading the caps of another process.
  */
-
-/*
- * Atomically modify the effective capabilities returning the original
- * value. No permission check is performed here - it is assumed that the
- * caller is permitted to set the desired effective capabilities.
- */
-kernel_cap_t cap_set_effective(const kernel_cap_t pE_new)
+static inline int cap_get_target_pid(pid_t pid, kernel_cap_t *pEp,
+                                    kernel_cap_t *pIp, kernel_cap_t *pPp)
 {
-       kernel_cap_t pE_old;
+       int ret;
 
-       spin_lock(&task_capability_lock);
+       if (pid && (pid != task_pid_vnr(current))) {
+               struct task_struct *target;
 
-       pE_old = current->cap_effective;
-       current->cap_effective = pE_new;
+               rcu_read_lock();
 
-       spin_unlock(&task_capability_lock);
+               target = find_task_by_vpid(pid);
+               if (!target)
+                       ret = -ESRCH;
+               else
+                       ret = security_capget(target, pEp, pIp, pPp);
 
-       return pE_old;
-}
+               rcu_read_unlock();
+       } else
+               ret = security_capget(current, pEp, pIp, pPp);
 
-EXPORT_SYMBOL(cap_set_effective);
+       return ret;
+}
 
 /**
  * sys_capget - get the capabilities of a given process.
@@ -151,17 +159,16 @@ EXPORT_SYMBOL(cap_set_effective);
  *
  * Returns 0 on success and < 0 on error.
  */
-asmlinkage long sys_capget(cap_user_header_t header, cap_user_data_t dataptr)
+SYSCALL_DEFINE2(capget, cap_user_header_t, header, cap_user_data_t, dataptr)
 {
        int ret = 0;
        pid_t pid;
-       struct task_struct *target;
        unsigned tocopy;
        kernel_cap_t pE, pI, pP;
 
        ret = cap_validate_magic(header, &tocopy);
-       if (ret != 0)
-               return ret;
+       if ((dataptr == NULL) || (ret != 0))
+               return ((dataptr == NULL) && (ret == -EINVAL)) ? 0 : ret;
 
        if (get_user(pid, &header->pid))
                return -EFAULT;
@@ -169,24 +176,7 @@ asmlinkage long sys_capget(cap_user_header_t header, cap_user_data_t dataptr)
        if (pid < 0)
                return -EINVAL;
 
-       spin_lock(&task_capability_lock);
-       read_lock(&tasklist_lock);
-
-       if (pid && pid != task_pid_vnr(current)) {
-               target = find_task_by_vpid(pid);
-               if (!target) {
-                       ret = -ESRCH;
-                       goto out;
-               }
-       } else
-               target = current;
-
-       ret = security_capget(target, &pE, &pI, &pP);
-
-out:
-       read_unlock(&tasklist_lock);
-       spin_unlock(&task_capability_lock);
-
+       ret = cap_get_target_pid(pid, &pE, &pI, &pP);
        if (!ret) {
                struct __user_cap_data_struct kdata[_KERNEL_CAPABILITY_U32S];
                unsigned i;
@@ -216,7 +206,6 @@ out:
                 * before modification is attempted and the application
                 * fails.
                 */
-
                if (copy_to_user(dataptr, kdata, tocopy
                                 * sizeof(struct __user_cap_data_struct))) {
                        return -EFAULT;
@@ -226,94 +215,30 @@ out:
        return ret;
 }
 
-/*
- * cap_set_pg - set capabilities for all processes in a given process
- * group.  We call this holding task_capability_lock and tasklist_lock.
- */
-static inline int cap_set_pg(int pgrp_nr, kernel_cap_t *effective,
-                             kernel_cap_t *inheritable,
-                             kernel_cap_t *permitted)
-{
-       struct task_struct *g, *target;
-       int ret = -EPERM;
-       int found = 0;
-       struct pid *pgrp;
-
-       pgrp = find_vpid(pgrp_nr);
-       do_each_pid_task(pgrp, PIDTYPE_PGID, g) {
-               target = g;
-               while_each_thread(g, target) {
-                       if (!security_capset_check(target, effective,
-                                                       inheritable,
-                                                       permitted)) {
-                               security_capset_set(target, effective,
-                                                       inheritable,
-                                                       permitted);
-                               ret = 0;
-                       }
-                       found = 1;
-               }
-       } while_each_pid_task(pgrp, PIDTYPE_PGID, g);
-
-       if (!found)
-               ret = 0;
-       return ret;
-}
-
-/*
- * cap_set_all - set capabilities for all processes other than init
- * and self.  We call this holding task_capability_lock and tasklist_lock.
- */
-static inline int cap_set_all(kernel_cap_t *effective,
-                              kernel_cap_t *inheritable,
-                              kernel_cap_t *permitted)
-{
-     struct task_struct *g, *target;
-     int ret = -EPERM;
-     int found = 0;
-
-     do_each_thread(g, target) {
-             if (target == current || is_container_init(target->group_leader))
-                     continue;
-             found = 1;
-            if (security_capset_check(target, effective, inheritable,
-                                               permitted))
-                    continue;
-            ret = 0;
-            security_capset_set(target, effective, inheritable, permitted);
-     } while_each_thread(g, target);
-
-     if (!found)
-            ret = 0;
-     return ret;
-}
-
 /**
- * sys_capset - set capabilities for a process or a group of processes
+ * sys_capset - set capabilities for a process or (*) a group of processes
  * @header: pointer to struct that contains capability version and
  *     target pid data
  * @data: pointer to struct that contains the effective, permitted,
  *     and inheritable capabilities
  *
- * Set capabilities for a given process, all processes, or all
- * processes in a given process group.
+ * Set capabilities for the current process only.  The ability to any other
+ * process(es) has been deprecated and removed.
  *
  * The restrictions on setting capabilities are specified as:
  *
- * [pid is for the 'target' task.  'current' is the calling task.]
- *
- * I: any raised capabilities must be a subset of the (old current) permitted
- * P: any raised capabilities must be a subset of the (old current) permitted
- * E: must be set to a subset of (new target) permitted
+ * I: any raised capabilities must be a subset of the old permitted
+ * P: any raised capabilities must be a subset of the old permitted
+ * E: must be set to a subset of new permitted
  *
  * Returns 0 on success and < 0 on error.
  */
-asmlinkage long sys_capset(cap_user_header_t header, const cap_user_data_t data)
+SYSCALL_DEFINE2(capset, cap_user_header_t, header, const cap_user_data_t, data)
 {
        struct __user_cap_data_struct kdata[_KERNEL_CAPABILITY_U32S];
-       unsigned i, tocopy;
+       unsigned i, tocopy, copybytes;
        kernel_cap_t inheritable, permitted, effective;
-       struct task_struct *target;
+       struct cred *new;
        int ret;
        pid_t pid;
 
@@ -324,13 +249,16 @@ asmlinkage long sys_capset(cap_user_header_t header, const cap_user_data_t data)
        if (get_user(pid, &header->pid))
                return -EFAULT;
 
-       if (pid && pid != task_pid_vnr(current) && !capable(CAP_SETPCAP))
+       /* may only affect current now */
+       if (pid != 0 && pid != task_pid_vnr(current))
                return -EPERM;
 
-       if (copy_from_user(&kdata, data, tocopy
-                          * sizeof(struct __user_cap_data_struct))) {
+       copybytes = tocopy * sizeof(struct __user_cap_data_struct);
+       if (copybytes > sizeof(kdata))
+               return -EFAULT;
+
+       if (copy_from_user(&kdata, data, copybytes))
                return -EFAULT;
-       }
 
        for (i = 0; i < tocopy; i++) {
                effective.cap[i] = kdata[i].effective;
@@ -344,55 +272,142 @@ asmlinkage long sys_capset(cap_user_header_t header, const cap_user_data_t data)
                i++;
        }
 
-       spin_lock(&task_capability_lock);
-       read_lock(&tasklist_lock);
+       new = prepare_creds();
+       if (!new)
+               return -ENOMEM;
 
-       if (pid > 0 && pid != task_pid_vnr(current)) {
-               target = find_task_by_vpid(pid);
-               if (!target) {
-                       ret = -ESRCH;
-                       goto out;
-               }
-       } else
-               target = current;
-
-       ret = 0;
-
-       /* having verified that the proposed changes are legal,
-          we now put them into effect. */
-       if (pid < 0) {
-               if (pid == -1)  /* all procs other than current and init */
-                       ret = cap_set_all(&effective, &inheritable, &permitted);
-
-               else            /* all procs in process group */
-                       ret = cap_set_pg(-pid, &effective, &inheritable,
-                                        &permitted);
-       } else {
-               ret = security_capset_check(target, &effective, &inheritable,
-                                           &permitted);
-               if (!ret)
-                       security_capset_set(target, &effective, &inheritable,
-                                           &permitted);
-       }
+       ret = security_capset(new, current_cred(),
+                             &effective, &inheritable, &permitted);
+       if (ret < 0)
+               goto error;
 
-out:
-       read_unlock(&tasklist_lock);
-       spin_unlock(&task_capability_lock);
+       audit_log_capset(pid, new, current_cred());
 
+       return commit_creds(new);
+
+error:
+       abort_creds(new);
        return ret;
 }
 
-int __capable(struct task_struct *t, int cap)
+/**
+ * has_capability - Does a task have a capability in init_user_ns
+ * @t: The task in question
+ * @cap: The capability to be tested for
+ *
+ * Return true if the specified task has the given superior capability
+ * currently in effect to the initial user namespace, false if not.
+ *
+ * Note that this does not set PF_SUPERPRIV on the task.
+ */
+bool has_capability(struct task_struct *t, int cap)
 {
-       if (security_capable(t, cap) == 0) {
-               t->flags |= PF_SUPERPRIV;
-               return 1;
-       }
-       return 0;
+       int ret = security_real_capable(t, &init_user_ns, cap);
+
+       return (ret == 0);
 }
 
-int capable(int cap)
+/**
+ * has_capability - Does a task have a capability in a specific user ns
+ * @t: The task in question
+ * @ns: target user namespace
+ * @cap: The capability to be tested for
+ *
+ * Return true if the specified task has the given superior capability
+ * currently in effect to the specified user namespace, false if not.
+ *
+ * Note that this does not set PF_SUPERPRIV on the task.
+ */
+bool has_ns_capability(struct task_struct *t,
+                      struct user_namespace *ns, int cap)
+{
+       int ret = security_real_capable(t, ns, cap);
+
+       return (ret == 0);
+}
+
+/**
+ * has_capability_noaudit - Does a task have a capability (unaudited)
+ * @t: The task in question
+ * @cap: The capability to be tested for
+ *
+ * Return true if the specified task has the given superior capability
+ * currently in effect to init_user_ns, false if not.  Don't write an
+ * audit message for the check.
+ *
+ * Note that this does not set PF_SUPERPRIV on the task.
+ */
+bool has_capability_noaudit(struct task_struct *t, int cap)
+{
+       int ret = security_real_capable_noaudit(t, &init_user_ns, cap);
+
+       return (ret == 0);
+}
+
+/**
+ * capable - Determine if the current task has a superior capability in effect
+ * @cap: The capability to be tested for
+ *
+ * Return true if the current task has the given superior capability currently
+ * available for use, false if not.
+ *
+ * This sets PF_SUPERPRIV on the task if the capability is available on the
+ * assumption that it's about to be used.
+ */
+bool capable(int cap)
 {
-       return __capable(current, cap);
+       return ns_capable(&init_user_ns, cap);
 }
 EXPORT_SYMBOL(capable);
+
+/**
+ * ns_capable - Determine if the current task has a superior capability in effect
+ * @ns:  The usernamespace we want the capability in
+ * @cap: The capability to be tested for
+ *
+ * Return true if the current task has the given superior capability currently
+ * available for use, false if not.
+ *
+ * This sets PF_SUPERPRIV on the task if the capability is available on the
+ * assumption that it's about to be used.
+ */
+bool ns_capable(struct user_namespace *ns, int cap)
+{
+       if (unlikely(!cap_valid(cap))) {
+               printk(KERN_CRIT "capable() called with invalid cap=%u\n", cap);
+               BUG();
+       }
+
+       if (security_capable(ns, current_cred(), cap) == 0) {
+               current->flags |= PF_SUPERPRIV;
+               return true;
+       }
+       return false;
+}
+EXPORT_SYMBOL(ns_capable);
+
+/**
+ * task_ns_capable - Determine whether current task has a superior
+ * capability targeted at a specific task's user namespace.
+ * @t: The task whose user namespace is targeted.
+ * @cap: The capability in question.
+ *
+ *  Return true if it does, false otherwise.
+ */
+bool task_ns_capable(struct task_struct *t, int cap)
+{
+       return ns_capable(task_cred_xxx(t, user)->user_ns, cap);
+}
+EXPORT_SYMBOL(task_ns_capable);
+
+/**
+ * nsown_capable - Check superior capability to one's own user_ns
+ * @cap: The capability in question
+ *
+ * Return true if the current task has the given superior capability
+ * targeted at its own user namespace.
+ */
+bool nsown_capable(int cap)
+{
+       return ns_capable(current_user_ns(), cap);
+}