From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mx2.suse.de ([195.135.220.15]:35162 "EHLO mx1.suse.de" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1751515AbeGJKCe (ORCPT ); Tue, 10 Jul 2018 06:02:34 -0400 From: Jan Kara To: linux-audit@redhat.com Cc: Paul Moore , , Al Viro , Richard Guy Briggs , Amir Goldstein , Jan Kara Subject: [PATCH 03/10] audit: Fix possible tagging failures Date: Tue, 10 Jul 2018 12:02:09 +0200 Message-Id: <20180710100217.12866-4-jack@suse.cz> In-Reply-To: <20180710100217.12866-1-jack@suse.cz> References: <20180710100217.12866-1-jack@suse.cz> Sender: linux-fsdevel-owner@vger.kernel.org List-ID: Audit tree code is replacing marks attached to inodes in non-atomic way. Thus fsnotify_find_mark() in tag_chunk() may find a mark that belongs to a chunk that is no longer valid one and will soon be destroyed. Tags added to such chunk will be simply lost. Fix the problem by making sure old mark is marked as going away (through fsnotify_detach_mark()) before dropping mark_mutex and thus in an atomic way wrt tag_chunk(). Note that this does not fix the problem completely as if tag_chunk() finds a mark that is going away, it fails with -ENOENT. But at least the failure is not silent and currently there's no way to search for another fsnotify mark attached to the inode. We'll fix this problem in later patch. Signed-off-by: Jan Kara --- kernel/audit_tree.c | 17 ++++++++++------- 1 file changed, 10 insertions(+), 7 deletions(-) diff --git a/kernel/audit_tree.c b/kernel/audit_tree.c index de8d344d91b1..7f9df8c66227 100644 --- a/kernel/audit_tree.c +++ b/kernel/audit_tree.c @@ -278,8 +278,9 @@ static void untag_chunk(struct node *p) list_del_init(&p->list); list_del_rcu(&chunk->hash); spin_unlock(&hash_lock); + fsnotify_detach_mark(entry); mutex_unlock(&entry->group->mark_mutex); - fsnotify_destroy_mark(entry, audit_tree_group); + fsnotify_free_mark(entry); goto out; } @@ -320,8 +321,9 @@ static void untag_chunk(struct node *p) list_for_each_entry(owner, &new->trees, same_root) owner->root = new; spin_unlock(&hash_lock); + fsnotify_detach_mark(entry); mutex_unlock(&entry->group->mark_mutex); - fsnotify_destroy_mark(entry, audit_tree_group); + fsnotify_free_mark(entry); fsnotify_put_mark(&new->mark); /* drop initial reference */ goto out; @@ -364,8 +366,9 @@ static int create_chunk(struct inode *inode, struct audit_tree *tree) if (tree->goner) { spin_unlock(&hash_lock); chunk->dead = 1; + fsnotify_detach_mark(entry); mutex_unlock(&audit_tree_group->mark_mutex); - fsnotify_destroy_mark(entry, audit_tree_group); + fsnotify_free_mark(entry); fsnotify_put_mark(entry); return 0; } @@ -446,10 +449,9 @@ static int tag_chunk(struct inode *inode, struct audit_tree *tree) if (tree->goner) { spin_unlock(&hash_lock); chunk->dead = 1; + fsnotify_detach_mark(chunk_entry); mutex_unlock(&audit_tree_group->mark_mutex); - - fsnotify_destroy_mark(chunk_entry, audit_tree_group); - + fsnotify_free_mark(chunk_entry); fsnotify_put_mark(chunk_entry); fsnotify_put_mark(old_entry); return 0; @@ -477,8 +479,9 @@ static int tag_chunk(struct inode *inode, struct audit_tree *tree) list_add(&tree->same_root, &chunk->trees); } spin_unlock(&hash_lock); + fsnotify_detach_mark(old_entry); mutex_unlock(&audit_tree_group->mark_mutex); - fsnotify_destroy_mark(old_entry, audit_tree_group); + fsnotify_free_mark(old_entry); fsnotify_put_mark(chunk_entry); /* drop initial reference */ fsnotify_put_mark(old_entry); /* pair to fsnotify_find mark_entry */ return 0; -- 2.16.4