From: Namjae Jeon <linkinjeon@kernel.org>
Date: Sat, 27 Jun 2026 23:56:09 +0000 (+0900)
Subject: ksmbd: fix app-instance durable supersede session UAF
X-Git-Tag: v7.2-rc2~9^2
X-Git-Url: http://git.ipfire.org/?a=commitdiff_plain;h=f363a0fb134a3eb9e47368b1edbd251fd76be84b;p=thirdparty%2Fkernel%2Flinux.git

ksmbd: fix app-instance durable supersede session UAF

ksmbd_close_fd_app_instance_id() looks up a prior durable handle by
AppInstanceId and closes it through opinfo->sess->file_table. This is
unsafe after the original session has been torn down. session_fd_check()
preserves reconnectable durable handles in the global table and clears
opinfo->conn/fp->conn, but opinfo->sess can still point to the freed
ksmbd_session.

Use opinfo->conn as the orphan sentinel, but make the check reliable by
serializing it with session_fd_check(). That path clears opinfo->conn
under fp->f_ci->m_lock, so hold the same lock while testing opinfo->conn
and while dereferencing opinfo->sess->file_table. Also avoid closing
through the session file table if the volatile id has already been
unpublished by session teardown.

Durable reconnect must keep the two fields consistent. Rebinding only
opinfo->conn leaves opinfo->sess pointing at the old freed session, so
a later app-instance supersede can pass the conn check and write-lock the
freed session's file table. Clear opinfo->sess when preserving a durable
handle during session teardown, and set it to the reconnecting session
when opinfo->conn is rebound in ksmbd_reopen_durable_fd().

Fixes: 16c30649709d ("ksmbd: handle durable v2 app instance id")
Reported-by: Gil Portnoy <dddhkts1@gmail.com>
Co-developed-by: Gil Portnoy <dddhkts1@gmail.com>
Signed-off-by: Gil Portnoy <dddhkts1@gmail.com>
Signed-off-by: Namjae Jeon <linkinjeon@kernel.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
---

diff --git a/fs/smb/server/vfs_cache.c b/fs/smb/server/vfs_cache.c
index 73d28942dc0a..d95c405eab11 100644
--- a/fs/smb/server/vfs_cache.c
+++ b/fs/smb/server/vfs_cache.c
@@ -853,19 +853,24 @@ int ksmbd_close_fd_app_instance_id(char *app_instance_id)
 		return 0;
 
 	opinfo = opinfo_get(fp);
-	if (!opinfo || !opinfo->sess)
+	if (!opinfo)
+		goto out;
+
+	down_read(&fp->f_ci->m_lock);
+	if (!opinfo->conn) {
+		up_read(&fp->f_ci->m_lock);
 		goto out;
+	}
 
 	ft = &opinfo->sess->file_table;
 	write_lock(&ft->lock);
-	if (fp->f_state == FP_INITED) {
-		if (has_file_id(fp->volatile_id)) {
-			idr_remove(ft->idr, fp->volatile_id);
-			fp->volatile_id = KSMBD_NO_FID;
-		}
+	if (fp->f_state == FP_INITED && has_file_id(fp->volatile_id)) {
+		idr_remove(ft->idr, fp->volatile_id);
+		fp->volatile_id = KSMBD_NO_FID;
 		n_to_drop = ksmbd_mark_fp_closed(fp);
 	}
 	write_unlock(&ft->lock);
+	up_read(&fp->f_ci->m_lock);
 	opinfo_put(opinfo);
 	opinfo = NULL;
 
@@ -1562,6 +1567,7 @@ static bool session_fd_check(struct ksmbd_tree_connect *tcon,
 			continue;
 		ksmbd_conn_put(op->conn);
 		op->conn = NULL;
+		op->sess = NULL;
 	}
 	up_write(&ci->m_lock);
 
@@ -1717,6 +1723,7 @@ int ksmbd_reopen_durable_fd(struct ksmbd_work *work, struct ksmbd_file *fp)
 		if (op->conn)
 			continue;
 		op->conn = ksmbd_conn_get(fp->conn);
+		op->sess = work->sess;
 	}
 	up_write(&ci->m_lock);