inotify: fix lock ordering wrt do_page_fault's mmap_sem
authorNick Piggin <nickpiggin@yahoo.com.au>
Thu, 2 Oct 2008 21:50:12 +0000 (14:50 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 2 Oct 2008 22:53:13 +0000 (15:53 -0700)
Fix inotify lock order reversal with mmap_sem due to holding locks over
copy_to_user.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Reported-by: "Daniel J Blueman" <daniel.blueman@gmail.com>
Tested-by: "Daniel J Blueman" <daniel.blueman@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/inotify_user.c
include/asm-x86/uaccess_64.h

index 6024942..d85c7d9 100644 (file)
@@ -323,7 +323,7 @@ out:
 }
 
 /*
- * remove_kevent - cleans up and ultimately frees the given kevent
+ * remove_kevent - cleans up the given kevent
  *
  * Caller must hold dev->ev_mutex.
  */
@@ -334,7 +334,13 @@ static void remove_kevent(struct inotify_device *dev,
 
        dev->event_count--;
        dev->queue_size -= sizeof(struct inotify_event) + kevent->event.len;
+}
 
+/*
+ * free_kevent - frees the given kevent.
+ */
+static void free_kevent(struct inotify_kernel_event *kevent)
+{
        kfree(kevent->name);
        kmem_cache_free(event_cachep, kevent);
 }
@@ -350,6 +356,7 @@ static void inotify_dev_event_dequeue(struct inotify_device *dev)
                struct inotify_kernel_event *kevent;
                kevent = inotify_dev_get_event(dev);
                remove_kevent(dev, kevent);
+               free_kevent(kevent);
        }
 }
 
@@ -433,17 +440,15 @@ static ssize_t inotify_read(struct file *file, char __user *buf,
        dev = file->private_data;
 
        while (1) {
-               int events;
 
                prepare_to_wait(&dev->wq, &wait, TASK_INTERRUPTIBLE);
 
                mutex_lock(&dev->ev_mutex);
-               events = !list_empty(&dev->events);
-               mutex_unlock(&dev->ev_mutex);
-               if (events) {
+               if (!list_empty(&dev->events)) {
                        ret = 0;
                        break;
                }
+               mutex_unlock(&dev->ev_mutex);
 
                if (file->f_flags & O_NONBLOCK) {
                        ret = -EAGAIN;
@@ -462,7 +467,6 @@ static ssize_t inotify_read(struct file *file, char __user *buf,
        if (ret)
                return ret;
 
-       mutex_lock(&dev->ev_mutex);
        while (1) {
                struct inotify_kernel_event *kevent;
 
@@ -481,6 +485,13 @@ static ssize_t inotify_read(struct file *file, char __user *buf,
                        }
                        break;
                }
+               remove_kevent(dev, kevent);
+
+               /*
+                * Must perform the copy_to_user outside the mutex in order
+                * to avoid a lock order reversal with mmap_sem.
+                */
+               mutex_unlock(&dev->ev_mutex);
 
                if (copy_to_user(buf, &kevent->event, event_size)) {
                        ret = -EFAULT;
@@ -498,7 +509,9 @@ static ssize_t inotify_read(struct file *file, char __user *buf,
                        count -= kevent->event.len;
                }
 
-               remove_kevent(dev, kevent);
+               free_kevent(kevent);
+
+               mutex_lock(&dev->ev_mutex);
        }
        mutex_unlock(&dev->ev_mutex);
 
index 515d4dc..45806d6 100644 (file)
@@ -7,6 +7,7 @@
 #include <linux/compiler.h>
 #include <linux/errno.h>
 #include <linux/prefetch.h>
+#include <linux/lockdep.h>
 #include <asm/page.h>
 
 /*