misc: nct1008: avoid reading bogus temperature
[linux-3.10.git] / kernel / kmod.c
index 2b76dee..8241906 100644 (file)
 #include <linux/syscalls.h>
 #include <linux/unistd.h>
 #include <linux/kmod.h>
-#include <linux/smp_lock.h>
 #include <linux/slab.h>
-#include <linux/namespace.h>
 #include <linux/completion.h>
+#include <linux/cred.h>
 #include <linux/file.h>
+#include <linux/fdtable.h>
 #include <linux/workqueue.h>
 #include <linux/security.h>
 #include <linux/mount.h>
 #include <linux/kernel.h>
 #include <linux/init.h>
 #include <linux/resource.h>
+#include <linux/notifier.h>
+#include <linux/suspend.h>
+#include <linux/rwsem.h>
+#include <linux/ptrace.h>
+#include <linux/async.h>
 #include <asm/uaccess.h>
 
+#include <trace/events/module.h>
+
 extern int max_threads;
 
 static struct workqueue_struct *khelper_wq;
 
-#ifdef CONFIG_KMOD
+/*
+ * kmod_thread_locker is used for deadlock avoidance.  There is no explicit
+ * locking to protect this global - it is private to the singleton khelper
+ * thread and should only ever be modified by that thread.
+ */
+static const struct task_struct *kmod_thread_locker;
+
+#define CAP_BSET       (void *)1
+#define CAP_PI         (void *)2
+
+static kernel_cap_t usermodehelper_bset = CAP_FULL_SET;
+static kernel_cap_t usermodehelper_inheritable = CAP_FULL_SET;
+static DEFINE_SPINLOCK(umh_sysctl_lock);
+static DECLARE_RWSEM(umhelper_sem);
+
+#ifdef CONFIG_MODULES
 
 /*
        modprobe_path is set via /proc/sys.
 */
 char modprobe_path[KMOD_PATH_LEN] = "/sbin/modprobe";
 
+static void free_modprobe_argv(struct subprocess_info *info)
+{
+       kfree(info->argv[3]); /* check call_modprobe() */
+       kfree(info->argv);
+}
+
+static int call_modprobe(char *module_name, int wait)
+{
+       struct subprocess_info *info;
+       static char *envp[] = {
+               "HOME=/",
+               "TERM=linux",
+               "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
+               NULL
+       };
+
+       char **argv = kmalloc(sizeof(char *[5]), GFP_KERNEL);
+       if (!argv)
+               goto out;
+
+       module_name = kstrdup(module_name, GFP_KERNEL);
+       if (!module_name)
+               goto free_argv;
+
+       argv[0] = modprobe_path;
+       argv[1] = "-q";
+       argv[2] = "--";
+       argv[3] = module_name;  /* check free_modprobe_argv() */
+       argv[4] = NULL;
+
+       info = call_usermodehelper_setup(modprobe_path, argv, envp, GFP_KERNEL,
+                                        NULL, free_modprobe_argv, NULL);
+       if (!info)
+               goto free_module_name;
+
+       return call_usermodehelper_exec(info, wait | UMH_KILLABLE);
+
+free_module_name:
+       kfree(module_name);
+free_argv:
+       kfree(argv);
+out:
+       return -ENOMEM;
+}
+
 /**
- * request_module - try to load a kernel module
- * @fmt:     printf style format string for the name of the module
- * @varargs: arguements as specified in the format string
+ * __request_module - try to load a kernel module
+ * @wait: wait (or not) for the operation to complete
+ * @fmt: printf style format string for the name of the module
+ * @...: arguments as specified in the format string
  *
  * Load a module using the user mode module loader. The function returns
  * zero on success or a negative errno code on failure. Note that a
@@ -61,27 +129,34 @@ char modprobe_path[KMOD_PATH_LEN] = "/sbin/modprobe";
  * If module auto-loading support is disabled then this function
  * becomes a no-operation.
  */
-int request_module(const char *fmt, ...)
+int __request_module(bool wait, const char *fmt, ...)
 {
        va_list args;
        char module_name[MODULE_NAME_LEN];
        unsigned int max_modprobes;
        int ret;
-       char *argv[] = { modprobe_path, "-q", "--", module_name, NULL };
-       static char *envp[] = { "HOME=/",
-                               "TERM=linux",
-                               "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
-                               NULL };
        static atomic_t kmod_concurrent = ATOMIC_INIT(0);
 #define MAX_KMOD_CONCURRENT 50 /* Completely arbitrary value - KAO */
        static int kmod_loop_msg;
 
+       /*
+        * We don't allow synchronous module loading from async.  Module
+        * init may invoke async_synchronize_full() which will end up
+        * waiting for this task which already is waiting for the module
+        * loading to complete, leading to a deadlock.
+        */
+       WARN_ON_ONCE(wait && current_is_async());
+
        va_start(args, fmt);
        ret = vsnprintf(module_name, MODULE_NAME_LEN, fmt, args);
        va_end(args);
        if (ret >= MODULE_NAME_LEN)
                return -ENAMETOOLONG;
 
+       ret = security_kernel_module_request(module_name);
+       if (ret)
+               return ret;
+
        /* If modprobe needs a service that is in a module, we get a recursive
         * loop.  Limit the number of running kmod threads to max_threads/2 or
         * MAX_KMOD_CONCURRENT, whichever is the smaller.  A cleaner method
@@ -98,31 +173,25 @@ int request_module(const char *fmt, ...)
        atomic_inc(&kmod_concurrent);
        if (atomic_read(&kmod_concurrent) > max_modprobes) {
                /* We may be blaming an innocent here, but unlikely */
-               if (kmod_loop_msg++ < 5)
+               if (kmod_loop_msg < 5) {
                        printk(KERN_ERR
                               "request_module: runaway loop modprobe %s\n",
                               module_name);
+                       kmod_loop_msg++;
+               }
                atomic_dec(&kmod_concurrent);
                return -ENOMEM;
        }
 
-       ret = call_usermodehelper(modprobe_path, argv, envp, 1);
+       trace_module_request(module_name, wait, _RET_IP_);
+
+       ret = call_modprobe(module_name, wait ? UMH_WAIT_PROC : UMH_WAIT_EXEC);
+
        atomic_dec(&kmod_concurrent);
        return ret;
 }
-EXPORT_SYMBOL(request_module);
-#endif /* CONFIG_KMOD */
-
-struct subprocess_info {
-       struct completion *complete;
-       char *path;
-       char **argv;
-       char **envp;
-       struct key *ring;
-       int wait;
-       int retval;
-       struct file *stdin;
-};
+EXPORT_SYMBOL(__request_module);
+#endif /* CONFIG_MODULES */
 
 /*
  * This is the task which runs the usermode application
@@ -130,72 +199,98 @@ struct subprocess_info {
 static int ____call_usermodehelper(void *data)
 {
        struct subprocess_info *sub_info = data;
-       struct key *new_session, *old_session;
+       struct cred *new;
        int retval;
 
-       /* Unblock all signals and set the session keyring. */
-       new_session = key_get(sub_info->ring);
-       flush_signals(current);
        spin_lock_irq(&current->sighand->siglock);
-       old_session = __install_session_keyring(current, new_session);
        flush_signal_handlers(current, 1);
-       sigemptyset(&current->blocked);
-       recalc_sigpending();
        spin_unlock_irq(&current->sighand->siglock);
 
-       key_put(old_session);
-
-       /* Install input pipe when needed */
-       if (sub_info->stdin) {
-               struct files_struct *f = current->files;
-               struct fdtable *fdt;
-               /* no races because files should be private here */
-               sys_close(0);
-               fd_install(0, sub_info->stdin);
-               spin_lock(&f->file_lock);
-               fdt = files_fdtable(f);
-               FD_SET(0, fdt->open_fds);
-               FD_CLR(0, fdt->close_on_exec);
-               spin_unlock(&f->file_lock);
-
-               /* and disallow core files too */
-               current->signal->rlim[RLIMIT_CORE] = (struct rlimit){0, 0};
+       /* We can run anywhere, unlike our parent keventd(). */
+       set_cpus_allowed_ptr(current, cpu_all_mask);
+
+       /*
+        * Our parent is keventd, which runs with elevated scheduling priority.
+        * Avoid propagating that into the userspace child.
+        */
+       set_user_nice(current, 0);
+
+       retval = -ENOMEM;
+       new = prepare_kernel_cred(current);
+       if (!new)
+               goto fail;
+
+       spin_lock(&umh_sysctl_lock);
+       new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
+       new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
+                                            new->cap_inheritable);
+       spin_unlock(&umh_sysctl_lock);
+
+       if (sub_info->init) {
+               retval = sub_info->init(sub_info, new);
+               if (retval) {
+                       abort_creds(new);
+                       goto fail;
+               }
        }
 
-       /* We can run anywhere, unlike our parent keventd(). */
-       set_cpus_allowed(current, CPU_MASK_ALL);
+       commit_creds(new);
 
-       retval = -EPERM;
-       if (current->fs->root)
-               retval = kernel_execve(sub_info->path,
-                               sub_info->argv, sub_info->envp);
+       retval = do_execve(sub_info->path,
+                          (const char __user *const __user *)sub_info->argv,
+                          (const char __user *const __user *)sub_info->envp);
+       if (!retval)
+               return 0;
 
        /* Exec failed? */
+fail:
        sub_info->retval = retval;
        do_exit(0);
 }
 
+static int call_helper(void *data)
+{
+       /* Worker thread started blocking khelper thread. */
+       kmod_thread_locker = current;
+       return ____call_usermodehelper(data);
+}
+
+static void call_usermodehelper_freeinfo(struct subprocess_info *info)
+{
+       if (info->cleanup)
+               (*info->cleanup)(info);
+       kfree(info);
+}
+
+static void umh_complete(struct subprocess_info *sub_info)
+{
+       struct completion *comp = xchg(&sub_info->complete, NULL);
+       /*
+        * See call_usermodehelper_exec(). If xchg() returns NULL
+        * we own sub_info, the UMH_KILLABLE caller has gone away.
+        */
+       if (comp)
+               complete(comp);
+       else
+               call_usermodehelper_freeinfo(sub_info);
+}
+
 /* Keventd can't block, but this (a child) can. */
 static int wait_for_helper(void *data)
 {
        struct subprocess_info *sub_info = data;
        pid_t pid;
-       struct k_sigaction sa;
 
-       /* Install a handler: if SIGCLD isn't handled sys_wait4 won't
-        * populate the status, but will return -ECHILD. */
-       sa.sa.sa_handler = SIG_IGN;
-       sa.sa.sa_flags = 0;
-       siginitset(&sa.sa.sa_mask, sigmask(SIGCHLD));
-       do_sigaction(SIGCHLD, &sa, NULL);
-       allow_signal(SIGCHLD);
+       /* If SIGCLD is ignored sys_wait4 won't populate the status. */
+       spin_lock_irq(&current->sighand->siglock);
+       current->sighand->action[SIGCHLD-1].sa.sa_handler = SIG_DFL;
+       spin_unlock_irq(&current->sighand->siglock);
 
        pid = kernel_thread(____call_usermodehelper, sub_info, SIGCHLD);
        if (pid < 0) {
                sub_info->retval = pid;
        } else {
-               int ret;
-
+               int ret = -ECHILD;
                /*
                 * Normally it is bogus to call wait4() from in-kernel because
                 * wait4() wants to write the exit code to a userspace address.
@@ -216,113 +311,420 @@ static int wait_for_helper(void *data)
                        sub_info->retval = ret;
        }
 
-       complete(sub_info->complete);
-       return 0;
+       umh_complete(sub_info);
+       do_exit(0);
 }
 
 /* This is run by khelper thread  */
-static void __call_usermodehelper(void *data)
+static void __call_usermodehelper(struct work_struct *work)
 {
-       struct subprocess_info *sub_info = data;
+       struct subprocess_info *sub_info =
+               container_of(work, struct subprocess_info, work);
+       int wait = sub_info->wait & ~UMH_KILLABLE;
        pid_t pid;
-       int wait = sub_info->wait;
 
        /* CLONE_VFORK: wait until the usermode helper has execve'd
         * successfully We need the data structures to stay around
         * until that is done.  */
-       if (wait)
+       if (wait == UMH_WAIT_PROC)
                pid = kernel_thread(wait_for_helper, sub_info,
                                    CLONE_FS | CLONE_FILES | SIGCHLD);
-       else
-               pid = kernel_thread(____call_usermodehelper, sub_info,
+       else {
+               pid = kernel_thread(call_helper, sub_info,
                                    CLONE_VFORK | SIGCHLD);
+               /* Worker thread stopped blocking khelper thread. */
+               kmod_thread_locker = NULL;
+       }
 
-       if (pid < 0) {
-               sub_info->retval = pid;
-               complete(sub_info->complete);
-       } else if (!wait)
-               complete(sub_info->complete);
+       switch (wait) {
+       case UMH_NO_WAIT:
+               call_usermodehelper_freeinfo(sub_info);
+               break;
+
+       case UMH_WAIT_PROC:
+               if (pid > 0)
+                       break;
+               /* FALLTHROUGH */
+       case UMH_WAIT_EXEC:
+               if (pid < 0)
+                       sub_info->retval = pid;
+               umh_complete(sub_info);
+       }
+}
+
+/*
+ * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
+ * (used for preventing user land processes from being created after the user
+ * land has been frozen during a system-wide hibernation or suspend operation).
+ * Should always be manipulated under umhelper_sem acquired for write.
+ */
+static enum umh_disable_depth usermodehelper_disabled = UMH_DISABLED;
+
+/* Number of helpers running */
+static atomic_t running_helpers = ATOMIC_INIT(0);
+
+/*
+ * Wait queue head used by usermodehelper_disable() to wait for all running
+ * helpers to finish.
+ */
+static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
+
+/*
+ * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
+ * to become 'false'.
+ */
+static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq);
+
+/*
+ * Time to wait for running_helpers to become zero before the setting of
+ * usermodehelper_disabled in usermodehelper_disable() fails
+ */
+#define RUNNING_HELPERS_TIMEOUT        (5 * HZ)
+
+int usermodehelper_read_trylock(void)
+{
+       DEFINE_WAIT(wait);
+       int ret = 0;
+
+       down_read(&umhelper_sem);
+       for (;;) {
+               prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
+                               TASK_INTERRUPTIBLE);
+               if (!usermodehelper_disabled)
+                       break;
+
+               if (usermodehelper_disabled == UMH_DISABLED)
+                       ret = -EAGAIN;
+
+               up_read(&umhelper_sem);
+
+               if (ret)
+                       break;
+
+               schedule();
+               try_to_freeze();
+
+               down_read(&umhelper_sem);
+       }
+       finish_wait(&usermodehelper_disabled_waitq, &wait);
+       return ret;
+}
+EXPORT_SYMBOL_GPL(usermodehelper_read_trylock);
+
+long usermodehelper_read_lock_wait(long timeout)
+{
+       DEFINE_WAIT(wait);
+
+       if (timeout < 0)
+               return -EINVAL;
+
+       down_read(&umhelper_sem);
+       for (;;) {
+               prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
+                               TASK_UNINTERRUPTIBLE);
+               if (!usermodehelper_disabled)
+                       break;
+
+               up_read(&umhelper_sem);
+
+               timeout = schedule_timeout(timeout);
+               if (!timeout)
+                       break;
+
+               down_read(&umhelper_sem);
+       }
+       finish_wait(&usermodehelper_disabled_waitq, &wait);
+       return timeout;
+}
+EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait);
+
+void usermodehelper_read_unlock(void)
+{
+       up_read(&umhelper_sem);
+}
+EXPORT_SYMBOL_GPL(usermodehelper_read_unlock);
+
+/**
+ * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
+ * @depth: New value to assign to usermodehelper_disabled.
+ *
+ * Change the value of usermodehelper_disabled (under umhelper_sem locked for
+ * writing) and wakeup tasks waiting for it to change.
+ */
+void __usermodehelper_set_disable_depth(enum umh_disable_depth depth)
+{
+       down_write(&umhelper_sem);
+       usermodehelper_disabled = depth;
+       wake_up(&usermodehelper_disabled_waitq);
+       up_write(&umhelper_sem);
+}
+
+/**
+ * __usermodehelper_disable - Prevent new helpers from being started.
+ * @depth: New value to assign to usermodehelper_disabled.
+ *
+ * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
+ */
+int __usermodehelper_disable(enum umh_disable_depth depth)
+{
+       long retval;
+
+       if (!depth)
+               return -EINVAL;
+
+       down_write(&umhelper_sem);
+       usermodehelper_disabled = depth;
+       up_write(&umhelper_sem);
+
+       /*
+        * From now on call_usermodehelper_exec() won't start any new
+        * helpers, so it is sufficient if running_helpers turns out to
+        * be zero at one point (it may be increased later, but that
+        * doesn't matter).
+        */
+       retval = wait_event_timeout(running_helpers_waitq,
+                                       atomic_read(&running_helpers) == 0,
+                                       RUNNING_HELPERS_TIMEOUT);
+       if (retval)
+               return 0;
+
+       __usermodehelper_set_disable_depth(UMH_ENABLED);
+       return -EAGAIN;
+}
+
+static void helper_lock(void)
+{
+       atomic_inc(&running_helpers);
+       smp_mb__after_atomic_inc();
+}
+
+static void helper_unlock(void)
+{
+       if (atomic_dec_and_test(&running_helpers))
+               wake_up(&running_helpers_waitq);
 }
 
 /**
- * call_usermodehelper_keys - start a usermode application
- * @path: pathname for the application
- * @argv: null-terminated argument list
- * @envp: null-terminated environment list
- * @session_keyring: session keyring for process (NULL for an empty keyring)
+ * call_usermodehelper_setup - prepare to call a usermode helper
+ * @path: path to usermode executable
+ * @argv: arg vector for process
+ * @envp: environment for process
+ * @gfp_mask: gfp mask for memory allocation
+ * @cleanup: a cleanup function
+ * @init: an init function
+ * @data: arbitrary context sensitive data
+ *
+ * Returns either %NULL on allocation failure, or a subprocess_info
+ * structure.  This should be passed to call_usermodehelper_exec to
+ * exec the process and free the structure.
+ *
+ * The init function is used to customize the helper process prior to
+ * exec.  A non-zero return code causes the process to error out, exit,
+ * and return the failure to the calling process
+ *
+ * The cleanup function is just before ethe subprocess_info is about to
+ * be freed.  This can be used for freeing the argv and envp.  The
+ * Function must be runnable in either a process context or the
+ * context in which call_usermodehelper_exec is called.
+ */
+struct subprocess_info *call_usermodehelper_setup(char *path, char **argv,
+               char **envp, gfp_t gfp_mask,
+               int (*init)(struct subprocess_info *info, struct cred *new),
+               void (*cleanup)(struct subprocess_info *info),
+               void *data)
+{
+       struct subprocess_info *sub_info;
+       sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
+       if (!sub_info)
+               goto out;
+
+       INIT_WORK(&sub_info->work, __call_usermodehelper);
+       sub_info->path = path;
+       sub_info->argv = argv;
+       sub_info->envp = envp;
+
+       sub_info->cleanup = cleanup;
+       sub_info->init = init;
+       sub_info->data = data;
+  out:
+       return sub_info;
+}
+EXPORT_SYMBOL(call_usermodehelper_setup);
+
+/**
+ * call_usermodehelper_exec - start a usermode application
+ * @sub_info: information about the subprocessa
  * @wait: wait for the application to finish and return status.
+ *        when UMH_NO_WAIT don't wait at all, but you get no useful error back
+ *        when the program couldn't be exec'ed. This makes it safe to call
+ *        from interrupt context.
  *
  * Runs a user-space application.  The application is started
  * asynchronously if wait is not set, and runs as a child of keventd.
  * (ie. it runs with full root capabilities).
- *
- * Must be called from process context.  Returns a negative error code
- * if program was not execed successfully, or 0.
  */
-int call_usermodehelper_keys(char *path, char **argv, char **envp,
-                            struct key *session_keyring, int wait)
+int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
 {
        DECLARE_COMPLETION_ONSTACK(done);
-       struct subprocess_info sub_info = {
-               .complete       = &done,
-               .path           = path,
-               .argv           = argv,
-               .envp           = envp,
-               .ring           = session_keyring,
-               .wait           = wait,
-               .retval         = 0,
-       };
-       DECLARE_WORK(work, __call_usermodehelper, &sub_info);
+       int retval = 0;
 
-       if (!khelper_wq)
-               return -EBUSY;
+       helper_lock();
+       if (!sub_info->path) {
+               retval = -EINVAL;
+               goto out;
+       }
 
-       if (path[0] == '\0')
-               return 0;
+       if (sub_info->path[0] == '\0')
+               goto out;
+
+       if (!khelper_wq || usermodehelper_disabled) {
+               retval = -EBUSY;
+               goto out;
+       }
+       /*
+        * Worker thread must not wait for khelper thread at below
+        * wait_for_completion() if the thread was created with CLONE_VFORK
+        * flag, for khelper thread is already waiting for the thread at
+        * wait_for_completion() in do_fork().
+        */
+       if (wait != UMH_NO_WAIT && current == kmod_thread_locker) {
+               retval = -EBUSY;
+               goto out;
+       }
+
+       sub_info->complete = &done;
+       sub_info->wait = wait;
+
+       queue_work(khelper_wq, &sub_info->work);
+       if (wait == UMH_NO_WAIT)        /* task has freed sub_info */
+               goto unlock;
+
+       if (wait & UMH_KILLABLE) {
+               retval = wait_for_completion_killable(&done);
+               if (!retval)
+                       goto wait_done;
+
+               /* umh_complete() will see NULL and free sub_info */
+               if (xchg(&sub_info->complete, NULL))
+                       goto unlock;
+               /* fallthrough, umh_complete() was already called */
+       }
 
-       queue_work(khelper_wq, &work);
        wait_for_completion(&done);
-       return sub_info.retval;
+wait_done:
+       retval = sub_info->retval;
+out:
+       call_usermodehelper_freeinfo(sub_info);
+unlock:
+       helper_unlock();
+       return retval;
 }
-EXPORT_SYMBOL(call_usermodehelper_keys);
+EXPORT_SYMBOL(call_usermodehelper_exec);
 
-int call_usermodehelper_pipe(char *path, char **argv, char **envp,
-                            struct file **filp)
+/**
+ * call_usermodehelper() - prepare and start a usermode application
+ * @path: path to usermode executable
+ * @argv: arg vector for process
+ * @envp: environment for process
+ * @wait: wait for the application to finish and return status.
+ *        when UMH_NO_WAIT don't wait at all, but you get no useful error back
+ *        when the program couldn't be exec'ed. This makes it safe to call
+ *        from interrupt context.
+ *
+ * This function is the equivalent to use call_usermodehelper_setup() and
+ * call_usermodehelper_exec().
+ */
+int call_usermodehelper(char *path, char **argv, char **envp, int wait)
 {
-       DECLARE_COMPLETION(done);
-       struct subprocess_info sub_info = {
-               .complete       = &done,
-               .path           = path,
-               .argv           = argv,
-               .envp           = envp,
-               .retval         = 0,
-       };
-       struct file *f;
-       DECLARE_WORK(work, __call_usermodehelper, &sub_info);
+       struct subprocess_info *info;
+       gfp_t gfp_mask = (wait == UMH_NO_WAIT) ? GFP_ATOMIC : GFP_KERNEL;
 
-       if (!khelper_wq)
-               return -EBUSY;
+       info = call_usermodehelper_setup(path, argv, envp, gfp_mask,
+                                        NULL, NULL, NULL);
+       if (info == NULL)
+               return -ENOMEM;
 
-       if (path[0] == '\0')
-               return 0;
+       return call_usermodehelper_exec(info, wait);
+}
+EXPORT_SYMBOL(call_usermodehelper);
+
+static int proc_cap_handler(struct ctl_table *table, int write,
+                        void __user *buffer, size_t *lenp, loff_t *ppos)
+{
+       struct ctl_table t;
+       unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
+       kernel_cap_t new_cap;
+       int err, i;
+
+       if (write && (!capable(CAP_SETPCAP) ||
+                     !capable(CAP_SYS_MODULE)))
+               return -EPERM;
+
+       /*
+        * convert from the global kernel_cap_t to the ulong array to print to
+        * userspace if this is a read.
+        */
+       spin_lock(&umh_sysctl_lock);
+       for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)  {
+               if (table->data == CAP_BSET)
+                       cap_array[i] = usermodehelper_bset.cap[i];
+               else if (table->data == CAP_PI)
+                       cap_array[i] = usermodehelper_inheritable.cap[i];
+               else
+                       BUG();
+       }
+       spin_unlock(&umh_sysctl_lock);
 
-       f = create_write_pipe();
-       if (IS_ERR(f))
-               return PTR_ERR(f);
-       *filp = f;
+       t = *table;
+       t.data = &cap_array;
 
-       f = create_read_pipe(f);
-       if (IS_ERR(f)) {
-               free_write_pipe(*filp);
-               return PTR_ERR(f);
+       /*
+        * actually read or write and array of ulongs from userspace.  Remember
+        * these are least significant 32 bits first
+        */
+       err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
+       if (err < 0)
+               return err;
+
+       /*
+        * convert from the sysctl array of ulongs to the kernel_cap_t
+        * internal representation
+        */
+       for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
+               new_cap.cap[i] = cap_array[i];
+
+       /*
+        * Drop everything not in the new_cap (but don't add things)
+        */
+       spin_lock(&umh_sysctl_lock);
+       if (write) {
+               if (table->data == CAP_BSET)
+                       usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
+               if (table->data == CAP_PI)
+                       usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
        }
-       sub_info.stdin = f;
+       spin_unlock(&umh_sysctl_lock);
 
-       queue_work(khelper_wq, &work);
-       wait_for_completion(&done);
-       return sub_info.retval;
+       return 0;
 }
-EXPORT_SYMBOL(call_usermodehelper_pipe);
+
+struct ctl_table usermodehelper_table[] = {
+       {
+               .procname       = "bset",
+               .data           = CAP_BSET,
+               .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
+               .mode           = 0600,
+               .proc_handler   = proc_cap_handler,
+       },
+       {
+               .procname       = "inheritable",
+               .data           = CAP_PI,
+               .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
+               .mode           = 0600,
+               .proc_handler   = proc_cap_handler,
+       },
+       { }
+};
 
 void __init usermodehelper_init(void)
 {