Skip to contentSkip to conversation

One more unlock missing in error case

PATCH Ulrich Drepper 3 participants 4 messages Feb 4, 2006 – Feb 5, 2006 VFS

lore
Share & tools
lore
#0 lore raw

This patch is needed to in addition to the other unlocking fix which is already applied. It should be obvious that the system will deadlock in case this isn't done.

(Andrew, I once again used a goto in the third error case because the common code is again larger.)

namei.c-new renamed from namei.c +5 -4
@@ -1096,6 +1096,7 @@
 
 		file = fget_light(dfd, &fput_needed);
 		if (!file) {
+			read_unlock(&current->fs->lock);
 			retval = -EBADF;
 			goto out_fail;
 		}
@@ -1104,15 +1105,15 @@
 
 		if (!S_ISDIR(dentry->d_inode->i_mode)) {
 			retval = -ENOTDIR;
+		unlock_fail:
 			fput_light(file, fput_needed);
+			read_unlock(&current->fs->lock);
 			goto out_fail;
 		}
 
 		retval = file_permission(file, MAY_EXEC);
-		if (retval) {
-			fput_light(file, fput_needed);
-			goto out_fail;
-		}
+		if (retval)
+			goto unlock_fail;
 
 		nd->mnt = mntget(file->f_vfsmnt);
 		nd->dentry = dget(dentry);
Signature · 1 line
➧ Ulrich Drepper ➧ Red Hat, Inc. ➧ 444 Castro St ➧ Mountain View, CA ❖
#1 lore raw
Quoted 4 lines · Ulrich Drepper

This patch is needed to in addition to the other unlocking fix which is
already applied. It should be obvious that the system will deadlock in case this isn't done.

hmm. How's about we undo that goto tangle while we're there?

fs/namei.c renamed from fs/namei.c~nameic-unlock-missing-in-error-case +16 -16
@@ -1070,6 +1070,8 @@ static int fastcall do_path_lookup(int d
 				unsigned int flags, struct nameidata *nd)
 {
 	int retval = 0;
+	int fput_needed;
+	struct file *file;
 
 	nd->last_type = LAST_ROOT; /* if there are only slashes... */
 	nd->flags = flags;
@@ -1091,29 +1093,22 @@ static int fastcall do_path_lookup(int d
 		nd->mnt = mntget(current->fs->pwdmnt);
 		nd->dentry = dget(current->fs->pwd);
 	} else {
-		struct file *file;
-		int fput_needed;
 		struct dentry *dentry;
 
 		file = fget_light(dfd, &fput_needed);
-		if (!file) {
-			retval = -EBADF;
-			goto out_fail;
-		}
+		retval = -EBADF;
+		if (!file)
+			goto unlock_fail;
 
 		dentry = file->f_dentry;
 
-		if (!S_ISDIR(dentry->d_inode->i_mode)) {
-			retval = -ENOTDIR;
-			fput_light(file, fput_needed);
-			goto out_fail;
-		}
+		retval = -ENOTDIR;
+		if (!S_ISDIR(dentry->d_inode->i_mode))
+			goto fput_unlock_fail;
 
 		retval = file_permission(file, MAY_EXEC);
-		if (retval) {
-			fput_light(file, fput_needed);
-			goto out_fail;
-		}
+		if (retval)
+			goto fput_unlock_fail;
 
 		nd->mnt = mntget(file->f_vfsmnt);
 		nd->dentry = dget(dentry);
@@ -1127,7 +1122,12 @@ out:
 	if (unlikely(current->audit_context
 		     && nd && nd->dentry && nd->dentry->d_inode))
 		audit_inode(name, nd->dentry->d_inode, flags);
-out_fail:
+	return retval;
+
+fput_unlock_fail:
+	fput_light(file, fput_needed);
+unlock_fail:
+	read_unlock(&current->fs->lock);
 	return retval;
 }
 

_

#2 lore raw
Quoted 1 line · Andrew Morton

hmm. How's about we undo that goto tangle while we're there?

Fine with me.

Signature · 1 line
➧ Ulrich Drepper ➧ Red Hat, Inc. ➧ 444 Castro St ➧ Mountain View, CA ❖
#3 lore raw
In reply to Andrew Morton
Quoted 8 lines · Andrew Morton
Quoted 4 lines · Ulrich Drepper

This patch is needed to in addition to the other unlocking fix which is
already applied. It should be obvious that the system will deadlock in case this isn't done.

hmm. How's about we undo that goto tangle while we're there?
...

The fput_unlock_fail label change looks good, but IMHO moving of retval settings out of the if's makes the code less readable.

cu
Adrian

Signature · 4 lines
       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed