xfs: inodes are new until the dentry cache is set up

Al Viro noticed a generic set of issues to do with filehandle lookup
racing with dentry cache setup. They involve a filehandle lookup
occurring while an inode is being created and the filehandle lookup
racing with the dentry creation for the real file. This can lead to
multiple dentries for the one path being instantiated. There are a
host of other issues around this same set of paths.

The underlying cause is that file handle lookup only waits on inode
cache instantiation rather than full dentry cache instantiation. XFS
is mostly immune to the problems discovered due to it's own internal
inode cache, but there are a couple of corner cases where races can
happen.

We currently clear the XFS_INEW flag when the inode is fully set up
after insertion into the cache. Newly allocated inodes are inserted
locked and so aren't usable until the allocation transaction
commits. This, however, occurs before the dentry and security
information is fully initialised and hence the inode is unlocked and
available for lookups to find too early.

To solve the problem, only clear the XFS_INEW flag for newly created
inodes once the dentry is fully instantiated. This means lookups
will retry until the XFS_INEW flag is removed from the inode and
hence avoids the race conditions in questions.

THis also means that xfs_create(), xfs_create_tmpfile() and
xfs_symlink() need to finish the setup of the inode in their error
paths if we had allocated the inode but failed later in the creation
process. xfs_symlink(), in particular, needed a lot of help to make
it's error handling match that of xfs_create().

Signed-off-by: Dave Chinner <[e-mail]>
Reviewed-by: Brian Foster <[e-mail]>
Signed-off-by: Dave Chinner <[e-mail]>

fs/xfs/xfs_icache.c

···
 	*ipp = ip;
 
 	/*
-	 * If we have a real type for an on-disk inode, we can set ops(&unlock)
+	 * If we have a real type for an on-disk inode, we can setup the inode
 	 * now.	 If it's a new inode being created, xfs_ialloc will handle it.
 	 */
 	if (xfs_iflags_test(ip, XFS_INEW) && ip->i_d.di_mode != 0)
-		xfs_setup_inode(ip);
+		xfs_setup_existing_inode(ip);
 	return 0;
 
 out_error_or_again:

fs/xfs/xfs_inode.c

···
 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 	xfs_trans_log_inode(tp, ip, flags);
 
-	/* now that we have an i_mode we can setup inode ops and unlock */
+	/* now that we have an i_mode we can setup the inode structure */
 	xfs_setup_inode(ip);
 
 	*ipp = ip;
···
 	xfs_trans_cancel(tp, cancel_flags);
  out_release_inode:
 	/*
-	 * Wait until after the current transaction is aborted to
-	 * release the inode.  This prevents recursive transactions
-	 * and deadlocks from xfs_inactive.
+	 * Wait until after the current transaction is aborted to finish the
+	 * setup of the inode and release the inode.  This prevents recursive
+	 * transactions and deadlocks from xfs_inactive.
 	 */
-	if (ip)
+	if (ip) {
+		xfs_finish_inode_setup(ip);
 		IRELE(ip);
+	}
 
 	xfs_qm_dqrele(udqp);
 	xfs_qm_dqrele(gdqp);
···
 	xfs_trans_cancel(tp, cancel_flags);
  out_release_inode:
 	/*
-	 * Wait until after the current transaction is aborted to
-	 * release the inode.  This prevents recursive transactions
-	 * and deadlocks from xfs_inactive.
+	 * Wait until after the current transaction is aborted to finish the
+	 * setup of the inode and release the inode.  This prevents recursive
+	 * transactions and deadlocks from xfs_inactive.
 	 */
-	if (ip)
+	if (ip) {
+		xfs_finish_inode_setup(ip);
 		IRELE(ip);
+	}
 
 	xfs_qm_dqrele(udqp);
 	xfs_qm_dqrele(gdqp);

fs/xfs/xfs_inode.h

···
 int		xfs_iozero(struct xfs_inode *, loff_t, size_t);
 
 
+/* from xfs_iops.c */
+/*
+ * When setting up a newly allocated inode, we need to call
+ * xfs_finish_inode_setup() once the inode is fully instantiated at
+ * the VFS level to prevent the rest of the world seeing the inode
+ * before we've completed instantiation. Otherwise we can do it
+ * the moment the inode lookup is complete.
+ */
+extern void xfs_setup_inode(struct xfs_inode *ip);
+static inline void xfs_finish_inode_setup(struct xfs_inode *ip)
+{
+	xfs_iflags_clear(ip, XFS_INEW);
+	barrier();
+	unlock_new_inode(VFS_I(ip));
+}
+
+static inline void xfs_setup_existing_inode(struct xfs_inode *ip)
+{
+	xfs_setup_inode(ip);
+	xfs_finish_inode_setup(ip);
+}
+
 #define IHOLD(ip) \
 do { \
 	ASSERT(atomic_read(&VFS_I(ip)->i_count) > 0) ; \

fs/xfs/xfs_iops.c

···
 	else
 		d_instantiate(dentry, inode);
 
+	xfs_finish_inode_setup(ip);
+
  out_free_acl:
 	if (default_acl)
 		posix_acl_release(default_acl);
···
 	return error;
 
  out_cleanup_inode:
+	xfs_finish_inode_setup(ip);
 	if (!tmpfile)
 		xfs_cleanup_inode(dir, inode, dentry);
 	iput(inode);
···
 		goto out_cleanup_inode;
 
 	d_instantiate(dentry, inode);
+	xfs_finish_inode_setup(cip);
 	return 0;
 
  out_cleanup_inode:
+	xfs_finish_inode_setup(cip);
 	xfs_cleanup_inode(dir, inode, dentry);
 	iput(inode);
  out:
···
 }
 
 /*
- * Initialize the Linux inode, set up the operation vectors and
- * unlock the inode.
+ * Initialize the Linux inode and set up the operation vectors.
  *
- * When reading existing inodes from disk this is called directly
- * from xfs_iget, when creating a new inode it is called from
- * xfs_ialloc after setting up the inode.
- *
- * We are always called with an uninitialised linux inode here.
- * We need to initialise the necessary fields and take a reference
- * on it.
+ * When reading existing inodes from disk this is called directly from xfs_iget,
+ * when creating a new inode it is called from xfs_ialloc after setting up the
+ * inode. These callers have different criteria for clearing XFS_INEW, so leave
+ * it up to the caller to deal with unlocking the inode appropriately.
  */
 void
 xfs_setup_inode(
···
 		inode_has_no_xattr(inode);
 		cache_no_acl(inode);
 	}
-
-	xfs_iflags_clear(ip, XFS_INEW);
-	barrier();
-
-	unlock_new_inode(inode);
 }

fs/xfs/xfs_iops.h

···
 
 extern ssize_t xfs_vn_listxattr(struct dentry *, char *data, size_t size);
 
-extern void xfs_setup_inode(struct xfs_inode *);
-
 /*
  * Internal setattr interfaces.
  */

fs/xfs/xfs_qm.c

···
 	xfs_trans_t	*tp;
 	int		error;
 	int		committed;
+	bool		need_alloc = true;
 
 	*ip = NULL;
 	/*
···
 				return error;
 			mp->m_sb.sb_gquotino = NULLFSINO;
 			mp->m_sb.sb_pquotino = NULLFSINO;
+			need_alloc = false;
 		}
 	}
 
···
 		return error;
 	}
 
-	if (!*ip) {
+	if (need_alloc) {
 		error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0, 0, 1, ip,
 								&committed);
 		if (error) {
···
 	spin_unlock(&mp->m_sb_lock);
 	xfs_log_sb(tp);
 
-	if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) {
+	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
+	if (error) {
+		ASSERT(XFS_FORCED_SHUTDOWN(mp));
 		xfs_alert(mp, "%s failed (error %d)!", __func__, error);
-		return error;
 	}
-	return 0;
+	if (need_alloc)
+		xfs_finish_inode_setup(*ip);
+	return error;
 }
 
 

fs/xfs/xfs_symlink.c

···
 	int			pathlen;
 	struct xfs_bmap_free	free_list;
 	xfs_fsblock_t		first_block;
-	bool			unlock_dp_on_error = false;
+	bool                    unlock_dp_on_error = false;
 	uint			cancel_flags;
 	int			committed;
 	xfs_fileoff_t		first_fsb;
···
 			XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
 			&udqp, &gdqp, &pdqp);
 	if (error)
-		goto std_return;
+		return error;
 
 	tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
 	cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
···
 	}
 	if (error) {
 		cancel_flags = 0;
-		goto error_return;
+		goto out_trans_cancel;
 	}
 
 	xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
···
 	 */
 	if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
 		error = -EPERM;
-		goto error_return;
+		goto out_trans_cancel;
 	}
 
 	/*
···
 	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
 						pdqp, resblks, 1, 0);
 	if (error)
-		goto error_return;
+		goto out_trans_cancel;
 
 	/*
 	 * Check for ability to enter directory entry, if no space reserved.
···
 	if (!resblks) {
 		error = xfs_dir_canenter(tp, dp, link_name);
 		if (error)
-			goto error_return;
+			goto out_trans_cancel;
 	}
 	/*
 	 * Initialize the bmap freelist prior to calling either
···
 	 */
 	error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT), 1, 0,
 			       prid, resblks > 0, &ip, NULL);
-	if (error) {
-		if (error == -ENOSPC)
-			goto error_return;
-		goto error1;
-	}
+	if (error)
+		goto out_trans_cancel;
 
 	/*
-	 * An error after we've joined dp to the transaction will result in the
-	 * transaction cancel unlocking dp so don't do it explicitly in the
+	 * Now we join the directory inode to the transaction.  We do not do it
+	 * earlier because xfs_dir_ialloc might commit the previous transaction
+	 * (and release all the locks).  An error from here on will result in
+	 * the transaction cancel unlocking dp so don't do it explicitly in the
 	 * error path.
 	 */
 	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
···
 				  XFS_BMAPI_METADATA, &first_block, resblks,
 				  mval, &nmaps, &free_list);
 		if (error)
-			goto error2;
+			goto out_bmap_cancel;
 
 		if (resblks)
 			resblks -= fs_blocks;
···
 					       BTOBB(byte_cnt), 0);
 			if (!bp) {
 				error = -ENOMEM;
-				goto error2;
+				goto out_bmap_cancel;
 			}
 			bp->b_ops = &xfs_symlink_buf_ops;
 
···
 	error = xfs_dir_createname(tp, dp, link_name, ip->i_ino,
 					&first_block, &free_list, resblks);
 	if (error)
-		goto error2;
+		goto out_bmap_cancel;
 	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
 	xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
 
···
 	}
 
 	error = xfs_bmap_finish(&tp, &free_list, &committed);
-	if (error) {
-		goto error2;
-	}
+	if (error)
+		goto out_bmap_cancel;
+
 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
+	if (error)
+		goto out_release_inode;
+
 	xfs_qm_dqrele(udqp);
 	xfs_qm_dqrele(gdqp);
 	xfs_qm_dqrele(pdqp);
···
 	*ipp = ip;
 	return 0;
 
- error2:
-	IRELE(ip);
- error1:
+out_bmap_cancel:
 	xfs_bmap_cancel(&free_list);
 	cancel_flags |= XFS_TRANS_ABORT;
- error_return:
+out_trans_cancel:
 	xfs_trans_cancel(tp, cancel_flags);
+out_release_inode:
+	/*
+	 * Wait until after the current transaction is aborted to finish the
+	 * setup of the inode and release the inode.  This prevents recursive
+	 * transactions and deadlocks from xfs_inactive.
+	 */
+	if (ip) {
+		xfs_finish_inode_setup(ip);
+		IRELE(ip);
+	}
+
 	xfs_qm_dqrele(udqp);
 	xfs_qm_dqrele(gdqp);
 	xfs_qm_dqrele(pdqp);
 
 	if (unlock_dp_on_error)
 		xfs_iunlock(dp, XFS_ILOCK_EXCL);
- std_return:
 	return error;
 }