If function security_inode_alloc() returns a nonzero value due to an
error (e.g. fail to allocate memory), then some of the fields, including
'i_private', will not be initialized.

After that, if the fs-specfic free_inode function is called in
i_callback(), the nonzero value of 'i_private' field can be interpreted
as initialized. As a result, this can cause dereferencing of random
value pointer (e.g. nilfs2).

In earlier versions, a similar situation could occur with the 'u' union
in 'inode' structure.

Found by Linux Verification Center (linuxtesting.org) with syzkaller.

Signed-off-by: Rustam Subkhankulov <subkhankulov@ispras.ru>
Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
---
 fs/inode.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/fs/inode.c b/fs/inode.c
index ba1de23c13c1..a2892d85993d 100644
--- a/fs/inode.c
+++ b/fs/inode.c
@@ -192,8 +192,6 @@ int inode_init_always(struct super_block *sb, struct inode *inode)
 	inode->i_wb_frn_history = 0;
 #endif
 
-	if (security_inode_alloc(inode))
-		goto out;
 	spin_lock_init(&inode->i_lock);
 	lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
 
@@ -228,6 +226,10 @@ int inode_init_always(struct super_block *sb, struct inode *inode)
 	inode->i_fsnotify_mask = 0;
 #endif
 	inode->i_flctx = NULL;
+
+	if (security_inode_alloc(inode))
+		goto out;
+
 	this_cpu_inc(nr_inodes);
 
 	return 0;
-- 
2.34.1

On Mon, Aug 29, 2022 at 10:25:21PM +0300, Rustam Subkhankulov wrote:
> If function security_inode_alloc() returns a nonzero value due to an
> error (e.g. fail to allocate memory), then some of the fields, including
> 'i_private', will not be initialized.
> 
> After that, if the fs-specfic free_inode function is called in
> i_callback(), the nonzero value of 'i_private' field can be interpreted
> as initialized. As a result, this can cause dereferencing of random
> value pointer (e.g. nilfs2).
> 
> In earlier versions, a similar situation could occur with the 'u' union
> in 'inode' structure.

See vfs.git#work.inode (included into #for-next); I agree that your
commit message looks better, but...