]> git.karo-electronics.de Git - linux-beck.git/commitdiff
audit: Fix sleep in atomic
authorJan Kara <jack@suse.cz>
Wed, 14 Dec 2016 13:40:05 +0000 (14:40 +0100)
committerPaul Moore <paul@paul-moore.com>
Tue, 3 Jan 2017 20:56:38 +0000 (15:56 -0500)
Audit tree code was happily adding new notification marks while holding
spinlocks. Since fsnotify_add_mark() acquires group->mark_mutex this can
lead to sleeping while holding a spinlock, deadlocks due to lock
inversion, and probably other fun. Fix the problem by acquiring
group->mark_mutex earlier.

CC: Paul Moore <paul@paul-moore.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Paul Moore <paul@paul-moore.com>
kernel/audit_tree.c

index b4b58400531f2bfdc9fdb2114c6e5ca029c59335..862969014cf6476b9feb4046eac79104fa5be316 100644 (file)
@@ -231,9 +231,11 @@ static void untag_chunk(struct node *p)
        if (size)
                new = alloc_chunk(size);
 
+       mutex_lock(&entry->group->mark_mutex);
        spin_lock(&entry->lock);
        if (chunk->dead || !entry->inode) {
                spin_unlock(&entry->lock);
+               mutex_unlock(&entry->group->mark_mutex);
                if (new)
                        free_chunk(new);
                goto out;
@@ -251,6 +253,7 @@ static void untag_chunk(struct node *p)
                list_del_rcu(&chunk->hash);
                spin_unlock(&hash_lock);
                spin_unlock(&entry->lock);
+               mutex_unlock(&entry->group->mark_mutex);
                fsnotify_destroy_mark(entry, audit_tree_group);
                goto out;
        }
@@ -258,8 +261,8 @@ static void untag_chunk(struct node *p)
        if (!new)
                goto Fallback;
 
-       if (fsnotify_add_mark(&new->mark,
-                             entry->group, entry->inode, NULL, 1)) {
+       if (fsnotify_add_mark_locked(&new->mark, entry->group, entry->inode,
+                                    NULL, 1)) {
                fsnotify_put_mark(&new->mark);
                goto Fallback;
        }
@@ -293,6 +296,7 @@ static void untag_chunk(struct node *p)
                owner->root = new;
        spin_unlock(&hash_lock);
        spin_unlock(&entry->lock);
+       mutex_unlock(&entry->group->mark_mutex);
        fsnotify_destroy_mark(entry, audit_tree_group);
        fsnotify_put_mark(&new->mark);  /* drop initial reference */
        goto out;
@@ -309,6 +313,7 @@ Fallback:
        put_tree(owner);
        spin_unlock(&hash_lock);
        spin_unlock(&entry->lock);
+       mutex_unlock(&entry->group->mark_mutex);
 out:
        fsnotify_put_mark(entry);
        spin_lock(&hash_lock);
@@ -386,18 +391,21 @@ static int tag_chunk(struct inode *inode, struct audit_tree *tree)
 
        chunk_entry = &chunk->mark;
 
+       mutex_lock(&old_entry->group->mark_mutex);
        spin_lock(&old_entry->lock);
        if (!old_entry->inode) {
                /* old_entry is being shot, lets just lie */
                spin_unlock(&old_entry->lock);
+               mutex_unlock(&old_entry->group->mark_mutex);
                fsnotify_put_mark(old_entry);
                free_chunk(chunk);
                return -ENOENT;
        }
 
-       if (fsnotify_add_mark(chunk_entry,
-                             old_entry->group, old_entry->inode, NULL, 1)) {
+       if (fsnotify_add_mark_locked(chunk_entry, old_entry->group,
+                                    old_entry->inode, NULL, 1)) {
                spin_unlock(&old_entry->lock);
+               mutex_unlock(&old_entry->group->mark_mutex);
                fsnotify_put_mark(chunk_entry);
                fsnotify_put_mark(old_entry);
                return -ENOSPC;
@@ -413,6 +421,7 @@ static int tag_chunk(struct inode *inode, struct audit_tree *tree)
                chunk->dead = 1;
                spin_unlock(&chunk_entry->lock);
                spin_unlock(&old_entry->lock);
+               mutex_unlock(&old_entry->group->mark_mutex);
 
                fsnotify_destroy_mark(chunk_entry, audit_tree_group);
 
@@ -445,6 +454,7 @@ static int tag_chunk(struct inode *inode, struct audit_tree *tree)
        spin_unlock(&hash_lock);
        spin_unlock(&chunk_entry->lock);
        spin_unlock(&old_entry->lock);
+       mutex_unlock(&old_entry->group->mark_mutex);
        fsnotify_destroy_mark(old_entry, audit_tree_group);
        fsnotify_put_mark(chunk_entry); /* drop initial reference */
        fsnotify_put_mark(old_entry); /* pair to fsnotify_find mark_entry */