Re: [2.6.27-rc5] inotify_read's ev_mutex vs do_page_fault's mmap_sem...

[Nick Piggin <nickpiggin@yahoo.com.au>]

[-- Attachment #1: Type: text/plain, Size: 647 bytes --]

On Wednesday 10 September 2008 14:07, Nick Piggin wrote:
> On Wednesday 10 September 2008 07:03, Daniel J Blueman wrote:
> > I observed this locking violation [1] while gnome-panel was loading;
> > this was previously reported at
> > http://uwsg.iu.edu/hypermail/linux/kernel/0806.3/2881.html .
> >
> > Let me know for more information/config/testing. Thanks!
>
> Thanks for the report. I've attached a patch you could test. It compiles
> (and boots a UML here) but I don't think I've actually tested the inotify
> path at all, so it may explode on you.

Indeed it was wrong, I messed up the waitqueue handling. This one
survives longer for me...

[-- Attachment #2: inotify-lor-fix.patch --]
[-- Type: text/x-diff, Size: 2629 bytes --]

Fix inotify lock order reversal with mmap_sem due to holding locks over
copy_to_user.

Signed-off-by: Nick Piggin <npiggin@suse.de>
---
Index: linux-2.6/fs/inotify_user.c
===================================================================
--- linux-2.6.orig/fs/inotify_user.c
+++ linux-2.6/fs/inotify_user.c
@@ -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
 
 	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(st
 		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 
 	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 
 	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 
 			}
 			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 
 			count -= kevent->event.len;
 		}
 
-		remove_kevent(dev, kevent);
+		free_kevent(kevent);
+
+		mutex_lock(&dev->ev_mutex);
 	}
 	mutex_unlock(&dev->ev_mutex);
 
Index: linux-2.6/include/asm-x86/uaccess_64.h
===================================================================
--- linux-2.6.orig/include/asm-x86/uaccess_64.h
+++ linux-2.6/include/asm-x86/uaccess_64.h
@@ -7,6 +7,7 @@
 #include <linux/compiler.h>
 #include <linux/errno.h>
 #include <linux/prefetch.h>
+#include <linux/lockdep.h>
 #include <asm/page.h>
 
 /*