Skip to contentSkip to conversation

hiddev: fix incorrect hiddev freeing

PATCH In review Johannes Weiner 1 participant 3 messages Mar 9, 2009

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 hiddev: fix incorrect hiddev freeing

When hiddev_open() fails for whatever reason, free the just allocated hiddev_list structure shared hiddev potentially still in use.

The hiddev is freed in device disconnect/last close of the device file and must not be freed while there are possibly existing references to it.

This is probably responsible for these

http://kerneloops.org/oops.php?number=221185
http://kerneloops.org/oops.php?number=220365

where a reader sleeps on the waitqueue, the device gets disconnected (exist -> 0) another user tries to open it, fails on the exist check and frees the hiddev from the table. The finish_wait() in the reader will then dereference the hiddev to get to the waitqueue and oopses.

This was introduced by commit 079034073faf974973baa0256b029451f6e768ad "HID: hiddev cleanup -- handle all error conditions properly".

---

drivers/hid/usbhid/hiddev.c +1 -1
index 4940e4d..00ea1ed 100644
@@ -306,7 +306,7 @@ static int hiddev_open(struct inode *inode, struct file *file)
 	return 0;
 bail:
 	file->private_data = NULL;
-	kfree(list->hiddev);
+	kfree(list);
 	return res;
 }
 
#1 lore raw
Quoted 2 lines · Johannes Weiner

When hiddev_open() fails for whatever reason, free the just allocated hiddev_list structure shared hiddev potentially still in use.

                       ^
                   instead of

Sorry, it's late (or early?) Need a resend?

Quoted 39 lines

The hiddev is freed in device disconnect/last close of the device file and must not be freed while there are possibly existing references to it.

This is probably responsible for these

http://kerneloops.org/oops.php?number=221185
http://kerneloops.org/oops.php?number=220365

where a reader sleeps on the waitqueue, the device gets disconnected (exist -> 0) another user tries to open it, fails on the exist check and frees the hiddev from the table. The finish_wait() in the reader will then dereference the hiddev to get to the waitqueue and oopses.

This was introduced by commit 079034073faf974973baa0256b029451f6e768ad "HID: hiddev cleanup -- handle all error conditions properly".

---

drivers/hid/usbhid/hiddev.c +1 -2
index 4940e4d..00ea1ed 100644
@@ -306,7 +306,7 @@ static int hiddev_open(struct inode *inode, struct file *file)
 	return 0;
 bail:
 	file->private_data = NULL;
-	kfree(list->hiddev);
+	kfree(list);
 	return res;
 }
 
--

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#2 lore raw
In reply to Johannes Weiner
Quoted 29 lines · Johannes Weiner

When hiddev_open() fails for whatever reason, free the just allocated hiddev_list structure shared hiddev potentially still in use.

The hiddev is freed in device disconnect/last close of the device file and must not be freed while there are possibly existing references to it.

This is probably responsible for these

http://kerneloops.org/oops.php?number=221185
http://kerneloops.org/oops.php?number=220365

where a reader sleeps on the waitqueue, the device gets disconnected (exist -> 0) another user tries to open it, fails on the exist check and frees the hiddev from the table. The finish_wait() in the reader will then dereference the hiddev to get to the waitqueue and oopses.

This was introduced by commit 079034073faf974973baa0256b029451f6e768ad "HID: hiddev cleanup -- handle all error conditions properly".

---

drivers/hid/usbhid/hiddev.c +0 -0
index 4940e4d..00ea1ed 100644

drivers/hid/usbhid/hiddev.c static int hiddev_open(struct inode *inode, struct file *file) · line 310

Quoted 6 lines
drivers/hid/usbhid/hiddev.c +1 -1
@@ -306,7 +306,7 @@ static int hiddev_open(struct inode *inode, struct file *file)
 	return 0;
 bail:
 	file->private_data = NULL;
-	kfree(list->hiddev);
+	kfree(list);
 	return res;

This isn't responsible for the above quoted oopsen but I think I found the real issue. Resend coming soon.

Dear stable team, I mixed up the commit date with the authoring date. The bugs were introduced after .28, so there is no need to backport anything unless my fixes fail to get into .29. So please ignore for now.

Hannes