Skip to contentSkip to conversation

KEYS: put keyring if install_session_keyring_to_cred() fails

PATCH In review Eric Biggers 2 participants 3 messages Apr 1, 2017 – Apr 3, 2017

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

PATCH 1/1 KEYS: put keyring if install_session_keyring_to_cred() fails

From: Eric Biggers <[email protected]>

In join_session_keyring(), if install_session_keyring_to_cred() were to fail, we would leak the keyring reference, just like in the bug fixed by commit 23567fd052a9 ("KEYS: Fix keyring ref leak in join_session_keyring()"). Fortunately this cannot happen currently, but we really should be more careful. Do this by adding and using a new error label at which the keyring reference is dropped.

---
 security/keys/process_keys.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
security/keys/process_keys.c +4 -3
index b6fdd22205b1..fc60f998c74d 100644
@@ -799,15 +799,14 @@ long join_session_keyring(const char *name)
 		ret = PTR_ERR(keyring);
 		goto error2;
 	} else if (keyring == new->session_keyring) {
-		key_put(keyring);
 		ret = 0;
-		goto error2;
+		goto error3;
 	}
 
 	/* we've got a keyring - now to install it */
 	ret = install_session_keyring_to_cred(new, keyring);
 	if (ret < 0)
-		goto error2;
+		goto error3;
 
 	commit_creds(new);
 	mutex_unlock(&key_session_mutex);
@@ -817,6 +816,8 @@ long join_session_keyring(const char *name)
 okay:
 	return ret;
 
+error3:
+	key_put(keyring);
 error2:
 	mutex_unlock(&key_session_mutex);
 error:
Signature · 1 line
2.12.1
maintainer
#1 lore raw

Review

Quoted 4 lines · Eric Biggers
(quoted diff) +2 -0
+error3:
+	key_put(keyring);
 error2:
 	mutex_unlock(&key_session_mutex);

I would prefer the put to be outside of the locked section. How about the attached instead?

David
---
commit 5f9e6fc4fdbbf9ada9bd03640b1c3ca50a3e5415

Author: Eric Biggers <[email protected]>
Date:   Sat Apr 1 11:54:16 2017 -0700

KEYS: put keyring if install_session_keyring_to_cred() fails

In join_session_keyring(), if install_session_keyring_to_cred() were to fail, we would leak the keyring reference, just like in the bug fixed by commit 23567fd052a9 ("KEYS: Fix keyring ref leak in join_session_keyring()"). Fortunately this cannot happen currently, but we really should be more careful. Do this by adding and using a new error label at which the keyring reference is dropped.

security/keys/process_keys.c +8 -6
index 44451af828c0..8e98e5d1b5fb 100644
@@ -793,21 +793,20 @@ long join_session_keyring(const char *name)
 			KEY_ALLOC_IN_QUOTA, NULL, NULL);
 		if (IS_ERR(keyring)) {
 			ret = PTR_ERR(keyring);
-			goto error2;
+			goto error_unlock_no_keyring;
 		}
 	} else if (IS_ERR(keyring)) {
 		ret = PTR_ERR(keyring);
-		goto error2;
+		goto error_unlock_no_keyring;
 	} else if (keyring == new->session_keyring) {
-		key_put(keyring);
 		ret = 0;
-		goto error2;
+		goto error_unlock;
 	}
 
 	/* we've got a keyring - now to install it */
 	ret = install_session_keyring_to_cred(new, keyring);
 	if (ret < 0)
-		goto error2;
+		goto error_unlock;
 
 	commit_creds(new);
 	mutex_unlock(&key_session_mutex);
@@ -817,8 +816,11 @@ long join_session_keyring(const char *name)
 okay:
 	return ret;
 
-error2:
+error_unlock_no_keyring:
+	keyring = NULL;
+error_unlock:
 	mutex_unlock(&key_session_mutex);
+	key_put(keyring);
 error:
 	abort_creds(new);
 	return ret;
#2 lore raw

Hi David,

Quoted 9 lines · David Howells

Review

Quoted 4 lines · Eric Biggers
(quoted diff) +2 -0
+error3:
+	key_put(keyring);
 error2:
 	mutex_unlock(&key_session_mutex);

I would prefer the put to be outside of the locked section. How about the attached instead?

[...]

Quoted 33 lines
security/keys/process_keys.c +4 -5
index 44451af828c0..8e98e5d1b5fb 100644
@@ -793,21 +793,20 @@ long join_session_keyring(const char *name)
 			KEY_ALLOC_IN_QUOTA, NULL, NULL);
 		if (IS_ERR(keyring)) {
 			ret = PTR_ERR(keyring);
-			goto error2;
+			goto error_unlock_no_keyring;
 		}
 	} else if (IS_ERR(keyring)) {
 		ret = PTR_ERR(keyring);
-		goto error2;
+		goto error_unlock_no_keyring;
 	} else if (keyring == new->session_keyring) {
-		key_put(keyring);
 		ret = 0;
-		goto error2;
+		goto error_unlock;
 	}
 
 	/* we've got a keyring - now to install it */
 	ret = install_session_keyring_to_cred(new, keyring);
 	if (ret < 0)
-		goto error2;
+		goto error_unlock;
 
 	commit_creds(new);
 	mutex_unlock(&key_session_mutex);
@@ -817,8 +816,11 @@ long join_session_keyring(const char *name)
 okay:
 	return ret;

security/keys/process_keys.c long join_session_keyring(const char *name) · line 826

Quoted 10 lines
security/keys/process_keys.c +4 -1
 
-error2:
+error_unlock_no_keyring:
+	keyring = NULL;
+error_unlock:
 	mutex_unlock(&key_session_mutex);
+	key_put(keyring);
 error:
 	abort_creds(new);
 	return ret;

I guess I'm okay with either patch, but I think jumping to the error labels in the way I proposed is more straightforward since it keeps things in the logical order. And I doubt there's much advantage in moving key_put() outside of the mutex, given that key_session_mutex is only used for joining session keyrings so shouldn't be contended, and key_put() hardly does any work anyway.

- Eric