kfifo: add kfifo_skip, kfifo_from_user and kfifo_to_user
[safe/jmp/linux-2.6] / kernel / capability.c
index 39e8193..7f876e6 100644 (file)
@@ -7,6 +7,7 @@
  * 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/syscalls.h>
 #include <linux/pid_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);
+#include "cred-internals.h"
 
 /*
  * 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:
  *
@@ -53,28 +58,42 @@ static void warn_legacy_capability_use(void)
 }
 
 /*
- * 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.
+ * Version 2 capabilities worked fine, but the linux/capability.h file
+ * that accompanied their introduction encouraged their use without
+ * the necessary user-space source code changes. As such, we have
+ * created a version 3 with equivalent functionality to version 2, but
+ * with a header change to protect legacy source code from using
+ * version 2 when it wanted to use version 1. If your system has code
+ * that trips the following warning, it is using version 2 specific
+ * capabilities and may be doing so insecurely.
+ *
+ * The remedy is to either upgrade your version of libcap (to 2.10+,
+ * if the application is linked against it), or recompile your
+ * application with modern kernel headers and this warning will go
+ * away.
  */
 
-/**
- * sys_capget - get the capabilities of a given process.
- * @header: pointer to struct that contains capability version and
- *     target pid data
- * @dataptr: pointer to struct that contains the effective, permitted,
- *     and inheritable capabilities that are returned
- *
- * Returns 0 on success and < 0 on error.
+static void warn_deprecated_v2(void)
+{
+       static int warned;
+
+       if (!warned) {
+               char name[sizeof(current->comm)];
+
+               printk(KERN_INFO "warning: `%s' uses deprecated v2"
+                      " capabilities in a way that may be insecure.\n",
+                      get_task_comm(name, current));
+               warned = 1;
+       }
+}
+
+/*
+ * Version check. Return the number of u32s in each capability flag
+ * array, or a negative value on error.
  */
-asmlinkage long sys_capget(cap_user_header_t header, cap_user_data_t dataptr)
+static int cap_validate_magic(cap_user_header_t header, unsigned *tocopy)
 {
-       int ret = 0;
-       pid_t pid;
        __u32 version;
-       struct task_struct *target;
-       unsigned tocopy;
-       kernel_cap_t pE, pI, pP;
 
        if (get_user(version, &header->version))
                return -EFAULT;
@@ -82,43 +101,84 @@ asmlinkage long sys_capget(cap_user_header_t header, cap_user_data_t dataptr)
        switch (version) {
        case _LINUX_CAPABILITY_VERSION_1:
                warn_legacy_capability_use();
-               tocopy = _LINUX_CAPABILITY_U32S_1;
+               *tocopy = _LINUX_CAPABILITY_U32S_1;
                break;
        case _LINUX_CAPABILITY_VERSION_2:
-               tocopy = _LINUX_CAPABILITY_U32S_2;
+               warn_deprecated_v2();
+               /*
+                * fall through - v3 is otherwise equivalent to v2.
+                */
+       case _LINUX_CAPABILITY_VERSION_3:
+               *tocopy = _LINUX_CAPABILITY_U32S_3;
                break;
        default:
-               if (put_user(_LINUX_CAPABILITY_VERSION, &header->version))
+               if (put_user((u32)_KERNEL_CAPABILITY_VERSION, &header->version))
                        return -EFAULT;
                return -EINVAL;
        }
 
-       if (get_user(pid, &header->pid))
-               return -EFAULT;
+       return 0;
+}
 
-       if (pid < 0)
-               return -EINVAL;
+/*
+ * 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.
+ */
+static inline int cap_get_target_pid(pid_t pid, kernel_cap_t *pEp,
+                                    kernel_cap_t *pIp, kernel_cap_t *pPp)
+{
+       int ret;
+
+       if (pid && (pid != task_pid_vnr(current))) {
+               struct task_struct *target;
 
-       spin_lock(&task_capability_lock);
-       read_lock(&tasklist_lock);
+               read_lock(&tasklist_lock);
 
-       if (pid && pid != task_pid_vnr(current)) {
                target = find_task_by_vpid(pid);
-               if (!target) {
+               if (!target)
                        ret = -ESRCH;
-                       goto out;
-               }
+               else
+                       ret = security_capget(target, pEp, pIp, pPp);
+
+               read_unlock(&tasklist_lock);
        } else
-               target = current;
+               ret = security_capget(current, pEp, pIp, pPp);
 
-       ret = security_capget(target, &pE, &pI, &pP);
+       return ret;
+}
 
-out:
-       read_unlock(&tasklist_lock);
-       spin_unlock(&task_capability_lock);
+/**
+ * sys_capget - get the capabilities of a given process.
+ * @header: pointer to struct that contains capability version and
+ *     target pid data
+ * @dataptr: pointer to struct that contains the effective, permitted,
+ *     and inheritable capabilities that are returned
+ *
+ * Returns 0 on success and < 0 on error.
+ */
+SYSCALL_DEFINE2(capget, cap_user_header_t, header, cap_user_data_t, dataptr)
+{
+       int ret = 0;
+       pid_t pid;
+       unsigned tocopy;
+       kernel_cap_t pE, pI, pP;
+
+       ret = cap_validate_magic(header, &tocopy);
+       if ((dataptr == NULL) || (ret != 0))
+               return ((dataptr == NULL) && (ret == -EINVAL)) ? 0 : ret;
 
+       if (get_user(pid, &header->pid))
+               return -EFAULT;
+
+       if (pid < 0)
+               return -EINVAL;
+
+       ret = cap_get_target_pid(pid, &pE, &pI, &pP);
        if (!ret) {
-               struct __user_cap_data_struct kdata[_LINUX_CAPABILITY_U32S];
+               struct __user_cap_data_struct kdata[_KERNEL_CAPABILITY_U32S];
                unsigned i;
 
                for (i = 0; i < tocopy; i++) {
@@ -128,7 +188,7 @@ out:
                }
 
                /*
-                * Note, in the case, tocopy < _LINUX_CAPABILITY_U32S,
+                * Note, in the case, tocopy < _KERNEL_CAPABILITY_U32S,
                 * we silently drop the upper capabilities here. This
                 * has the effect of making older libcap
                 * implementations implicitly drop upper capability
@@ -146,7 +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;
@@ -156,187 +215,102 @@ 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[_LINUX_CAPABILITY_U32S];
-       unsigned i, tocopy;
+       struct __user_cap_data_struct kdata[_KERNEL_CAPABILITY_U32S];
+       unsigned i, tocopy, copybytes;
        kernel_cap_t inheritable, permitted, effective;
-       __u32 version;
-       struct task_struct *target;
+       struct cred *new;
        int ret;
        pid_t pid;
 
-       if (get_user(version, &header->version))
-               return -EFAULT;
-
-       switch (version) {
-       case _LINUX_CAPABILITY_VERSION_1:
-               warn_legacy_capability_use();
-               tocopy = _LINUX_CAPABILITY_U32S_1;
-               break;
-       case _LINUX_CAPABILITY_VERSION_2:
-               tocopy = _LINUX_CAPABILITY_U32S_2;
-               break;
-       default:
-               if (put_user(_LINUX_CAPABILITY_VERSION, &header->version))
-                       return -EFAULT;
-               return -EINVAL;
-       }
+       ret = cap_validate_magic(header, &tocopy);
+       if (ret != 0)
+               return ret;
 
        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;
                permitted.cap[i] = kdata[i].permitted;
                inheritable.cap[i] = kdata[i].inheritable;
        }
-       while (i < _LINUX_CAPABILITY_U32S) {
+       while (i < _KERNEL_CAPABILITY_U32S) {
                effective.cap[i] = 0;
                permitted.cap[i] = 0;
                inheritable.cap[i] = 0;
                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)
+/**
+ * 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.
+ */
+int capable(int cap)
 {
-       if (security_capable(t, cap) == 0) {
-               t->flags |= PF_SUPERPRIV;
+       if (unlikely(!cap_valid(cap))) {
+               printk(KERN_CRIT "capable() called with invalid cap=%u\n", cap);
+               BUG();
+       }
+
+       if (security_capable(cap) == 0) {
+               current->flags |= PF_SUPERPRIV;
                return 1;
        }
        return 0;
 }
-
-int capable(int cap)
-{
-       return __capable(current, cap);
-}
 EXPORT_SYMBOL(capable);