Skip to contentSkip to conversation

inotify 0.18

ANNOUNCE Al Viro 3 participants 4 messages Jan 25, 2005 – Jan 27, 2005 1 message not in the archive Char and misc drivers

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): 1106682112.23615.3.camel@vertex

Al Viro

author
#1 lore raw

+static struct inode * find_inode(const char __user *dirname) +{
+ struct inode *inode;
+ struct nameidata nd;
+ int error;
+
+ error = __user_walk(dirname, LOOKUP_FOLLOW, &nd);
+ if (error)
+ return ERR_PTR(error);
+
+ inode = nd.dentry->d_inode;
+
+ /* you can only watch an inode if you have read permissions on it */
+ error = permission(inode, MAY_READ, NULL);
+ if (error) {
+ inode = ERR_PTR(error);
+ goto release_and_out;
+ }
+
+ spin_lock(&inode_lock);
+ __iget(inode);
+ spin_unlock(&inode_lock);
+release_and_out:
+ path_release(&nd);
+ return inode;
+}

Yawn... OK, so what happens if we get umount in the middle of your find_inode(), so that path_release() in there drops the last remaining reference to vfsmount (and superblock)?

#2 lore raw

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Quoted 27 lines · Al Viro
(quoted diff) +26 -0
+static struct inode * find_inode(const char __user *dirname)
+{
+	struct inode *inode;
+	struct nameidata nd;
+	int error;
+
+	error = __user_walk(dirname, LOOKUP_FOLLOW, &nd);
+	if (error)
+		return ERR_PTR(error);
+
+	inode = nd.dentry->d_inode;
+
+	/* you can only watch an inode if you have read permissions on it */
+	error = permission(inode, MAY_READ, NULL);
+	if (error) {
+		inode = ERR_PTR(error);
+		goto release_and_out;
+	}
+
+	spin_lock(&inode_lock);
+	__iget(inode);
+	spin_unlock(&inode_lock);
+release_and_out:
+	path_release(&nd);
+	return inode;
+}
Quoted 3 lines

Yawn... OK, so what happens if we get umount in the middle of your find_inode(), so that path_release() in there drops the last remaining reference to vfsmount (and superblock)?

How about inotify hold on to the nameidata until it is sure to have updated the watches? That way, the inode will be seen by inotify_super_block_umount when called by generic_shutdown_super.

There remains a small race though, in that the inotify ioctl may return success for an added watch after the umount, but this would have to be resolved by synchronizing in userspace anyway.

Incremental patch attached.

- --
Mike Waychison
Sun Microsystems, Inc.
1 (650) 352-5299 voice
1 (416) 202-8336 voice

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
NOTICE: The opinions expressed in this email are held by me, and may not represent the views of Sun Microsystems, Inc.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.5 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFB9rpKdQs4kOxk3/MRAl0RAJ9vssocI3P93AnfO+zwyhf/BX1V9wCffy9N r6t7TV0ZxpT3bT5pcYfpCHk=
=RmBk
-----END PGP SIGNATURE-----

Hold on to the nameidata of the watched file until we have successfully added/updated the watch on the inode.

---

 inotify.c |   48 ++++++++++++++++++++++++++----------------------
 1 files changed, 26 insertions(+), 22 deletions(-)
drivers/char/inotify.c +26 -22
Index: linux-2.6.10/drivers/char/inotify.c
===================================================================
@@ -181,33 +181,24 @@ static inline void put_inode_data(struct
 }
 
 /*
- * find_inode - resolve a user-given path to a specific inode and iget() it
+ * find_inode - resolve a user-given path to a specific inode and return a nd
  */
-static struct inode * find_inode(const char __user *dirname)
+static int find_inode(const char __user *dirname, struct nameidata *nd)
 {
-	struct inode *inode;
-	struct nameidata nd;
 	int error;
 
-	error = __user_walk(dirname, LOOKUP_FOLLOW, &nd);
+	error = __user_walk(dirname, LOOKUP_FOLLOW, nd);
 	if (error)
-		return ERR_PTR(error);
-
-	inode = nd.dentry->d_inode;
+		return error;
 
 	/* you can only watch an inode if you have read permissions on it */
-	error = permission(inode, MAY_READ, NULL);
+	error = permission(nd->dentry->d_inode, MAY_READ, NULL);
 	if (error) {
-		inode = ERR_PTR(error);
 		goto release_and_out;
 	}
 
-	spin_lock(&inode_lock);
-	__iget(inode);
-	spin_unlock(&inode_lock);
 release_and_out:
-	path_release(&nd);
-	return inode;
+	return error;
 }
 
 struct inotify_kernel_event * kernel_event(s32 wd, u32 mask, u32 cookie,
@@ -911,11 +902,15 @@ static int inotify_add_watch(struct inot
 {
 	struct inode *inode;
 	struct inotify_watch *watch;
+	struct nameidata nd;
 	int ret;
 
-	inode = find_inode((const char __user*) request->dirname);
-	if (IS_ERR(inode))
-		return PTR_ERR(inode);
+	ret = find_inode((const char __user*) request->dirname, &nd);
+	if (ret)
+		return ret;
+
+	/* Held in place by references in nd */
+	inode = nd.dentry->d_inode;
 
 	spin_lock(&dev->lock);
 
@@ -930,7 +925,7 @@ static int inotify_add_watch(struct inot
 		owatch->mask = request->mask;
 		inode_update_watch_mask(inode);
 		spin_unlock(&dev->lock);
-		iput(inode);
+		path_release(&nd);
 
 		return owatch->wd;
 	}
@@ -939,7 +934,7 @@ static int inotify_add_watch(struct inot
 
 	watch = create_watch(dev, request->mask, inode);
 	if (!watch) {
-		iput(inode);
+		path_release(&nd);
 		return -ENOSPC;
 	}
 
@@ -949,7 +944,7 @@ static int inotify_add_watch(struct inot
 	if (inotify_dev_add_watch(dev, watch)) {
 		delete_watch(dev, watch);
 		spin_unlock(&dev->lock);
-		iput(inode);
+		path_release(&nd);
 		return -EINVAL;
 	}
 
@@ -958,12 +953,21 @@ static int inotify_add_watch(struct inot
 		list_del_init(&watch->d_list);
 		delete_watch(dev, watch);
 		spin_unlock(&dev->lock);
-		iput(inode);
+		path_release(&nd);
 		return ret;
 	}
 
 	spin_unlock(&dev->lock);
 
+	/*
+	 * demote the references in nd to a reference to inode held
+	 * by the watch.
+	 */
+	spin_lock(&inode_lock);
+	__iget(inode);
+	spin_unlock(&inode_lock);
+	path_release(&nd);
+
 	return watch->wd;
 }
 
#3 lore raw
Quoted 3 lines · Mike Waychison

How about inotify hold on to the nameidata until it is sure to have updated the watches? That way, the inode will be seen by inotify_super_block_umount when called by generic_shutdown_super.

I was chatting with John as soon as Al pointed out the problem and suggested we look at something like this. I am glad you were able to crank it out so readily. Thank you!

Barring a more elegant solution, I think just holding nameidata down until the watches are added is the way to go. We will test this and then merge it into the next patch.

Thanks,

Robert Love