ksmbd: fix racy issue from smb2 close and logoff with multichannel
authorNamjae Jeon <linkinjeon@kernel.org>
Wed, 3 May 2023 05:03:40 +0000 (14:03 +0900)
committerSteve French <stfrench@microsoft.com>
Thu, 4 May 2023 04:03:02 +0000 (23:03 -0500)
When smb client send concurrent smb2 close and logoff request
with multichannel connection, It can cause racy issue. logoff request
free tcon and can cause UAF issues in smb2 close. When receiving logoff
request with multichannel, ksmbd should wait until all remaning requests
complete as well as ones in the current connection, and then make
session expired.

Cc: stable@vger.kernel.org
Reported-by: zdi-disclosures@trendmicro.com # ZDI-CAN-20796 ZDI-CAN-20595
Signed-off-by: Namjae Jeon <linkinjeon@kernel.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
fs/ksmbd/connection.c
fs/ksmbd/connection.h
fs/ksmbd/mgmt/tree_connect.c
fs/ksmbd/mgmt/user_session.c
fs/ksmbd/smb2pdu.c

index 1f5176d..4ed379f 100644 (file)
@@ -20,7 +20,7 @@ static DEFINE_MUTEX(init_lock);
 static struct ksmbd_conn_ops default_conn_ops;
 
 LIST_HEAD(conn_list);
-DEFINE_RWLOCK(conn_list_lock);
+DECLARE_RWSEM(conn_list_lock);
 
 /**
  * ksmbd_conn_free() - free resources of the connection instance
@@ -32,9 +32,9 @@ DEFINE_RWLOCK(conn_list_lock);
  */
 void ksmbd_conn_free(struct ksmbd_conn *conn)
 {
-       write_lock(&conn_list_lock);
+       down_write(&conn_list_lock);
        list_del(&conn->conns_list);
-       write_unlock(&conn_list_lock);
+       up_write(&conn_list_lock);
 
        xa_destroy(&conn->sessions);
        kvfree(conn->request_buf);
@@ -84,9 +84,9 @@ struct ksmbd_conn *ksmbd_conn_alloc(void)
        spin_lock_init(&conn->llist_lock);
        INIT_LIST_HEAD(&conn->lock_list);
 
-       write_lock(&conn_list_lock);
+       down_write(&conn_list_lock);
        list_add(&conn->conns_list, &conn_list);
-       write_unlock(&conn_list_lock);
+       up_write(&conn_list_lock);
        return conn;
 }
 
@@ -95,7 +95,7 @@ bool ksmbd_conn_lookup_dialect(struct ksmbd_conn *c)
        struct ksmbd_conn *t;
        bool ret = false;
 
-       read_lock(&conn_list_lock);
+       down_read(&conn_list_lock);
        list_for_each_entry(t, &conn_list, conns_list) {
                if (memcmp(t->ClientGUID, c->ClientGUID, SMB2_CLIENT_GUID_SIZE))
                        continue;
@@ -103,7 +103,7 @@ bool ksmbd_conn_lookup_dialect(struct ksmbd_conn *c)
                ret = true;
                break;
        }
-       read_unlock(&conn_list_lock);
+       up_read(&conn_list_lock);
        return ret;
 }
 
@@ -157,9 +157,37 @@ void ksmbd_conn_unlock(struct ksmbd_conn *conn)
        mutex_unlock(&conn->srv_mutex);
 }
 
-void ksmbd_conn_wait_idle(struct ksmbd_conn *conn)
+void ksmbd_all_conn_set_status(u64 sess_id, u32 status)
 {
+       struct ksmbd_conn *conn;
+
+       down_read(&conn_list_lock);
+       list_for_each_entry(conn, &conn_list, conns_list) {
+               if (conn->binding || xa_load(&conn->sessions, sess_id))
+                       WRITE_ONCE(conn->status, status);
+       }
+       up_read(&conn_list_lock);
+}
+
+void ksmbd_conn_wait_idle(struct ksmbd_conn *conn, u64 sess_id)
+{
+       struct ksmbd_conn *bind_conn;
+
        wait_event(conn->req_running_q, atomic_read(&conn->req_running) < 2);
+
+       down_read(&conn_list_lock);
+       list_for_each_entry(bind_conn, &conn_list, conns_list) {
+               if (bind_conn == conn)
+                       continue;
+
+               if ((bind_conn->binding || xa_load(&bind_conn->sessions, sess_id)) &&
+                   !ksmbd_conn_releasing(bind_conn) &&
+                   atomic_read(&bind_conn->req_running)) {
+                       wait_event(bind_conn->req_running_q,
+                               atomic_read(&bind_conn->req_running) == 0);
+               }
+       }
+       up_read(&conn_list_lock);
 }
 
 int ksmbd_conn_write(struct ksmbd_work *work)
@@ -360,10 +388,10 @@ int ksmbd_conn_handler_loop(void *p)
        }
 
 out:
+       ksmbd_conn_set_releasing(conn);
        /* Wait till all reference dropped to the Server object*/
        wait_event(conn->r_count_q, atomic_read(&conn->r_count) == 0);
 
-
        if (IS_ENABLED(CONFIG_UNICODE))
                utf8_unload(conn->um);
        unload_nls(conn->local_nls);
@@ -407,7 +435,7 @@ static void stop_sessions(void)
        struct ksmbd_transport *t;
 
 again:
-       read_lock(&conn_list_lock);
+       down_read(&conn_list_lock);
        list_for_each_entry(conn, &conn_list, conns_list) {
                struct task_struct *task;
 
@@ -418,12 +446,12 @@ again:
                                    task->comm, task_pid_nr(task));
                ksmbd_conn_set_exiting(conn);
                if (t->ops->shutdown) {
-                       read_unlock(&conn_list_lock);
+                       up_read(&conn_list_lock);
                        t->ops->shutdown(t);
-                       read_lock(&conn_list_lock);
+                       down_read(&conn_list_lock);
                }
        }
-       read_unlock(&conn_list_lock);
+       up_read(&conn_list_lock);
 
        if (!list_empty(&conn_list)) {
                schedule_timeout_interruptible(HZ / 10); /* 100ms */
index 98bb5f1..ad8dfaa 100644 (file)
@@ -26,7 +26,8 @@ enum {
        KSMBD_SESS_GOOD,
        KSMBD_SESS_EXITING,
        KSMBD_SESS_NEED_RECONNECT,
-       KSMBD_SESS_NEED_NEGOTIATE
+       KSMBD_SESS_NEED_NEGOTIATE,
+       KSMBD_SESS_RELEASING
 };
 
 struct ksmbd_stats {
@@ -140,10 +141,10 @@ struct ksmbd_transport {
 #define KSMBD_TCP_PEER_SOCKADDR(c)     ((struct sockaddr *)&((c)->peer_addr))
 
 extern struct list_head conn_list;
-extern rwlock_t conn_list_lock;
+extern struct rw_semaphore conn_list_lock;
 
 bool ksmbd_conn_alive(struct ksmbd_conn *conn);
-void ksmbd_conn_wait_idle(struct ksmbd_conn *conn);
+void ksmbd_conn_wait_idle(struct ksmbd_conn *conn, u64 sess_id);
 struct ksmbd_conn *ksmbd_conn_alloc(void);
 void ksmbd_conn_free(struct ksmbd_conn *conn);
 bool ksmbd_conn_lookup_dialect(struct ksmbd_conn *c);
@@ -191,6 +192,11 @@ static inline bool ksmbd_conn_exiting(struct ksmbd_conn *conn)
        return READ_ONCE(conn->status) == KSMBD_SESS_EXITING;
 }
 
+static inline bool ksmbd_conn_releasing(struct ksmbd_conn *conn)
+{
+       return READ_ONCE(conn->status) == KSMBD_SESS_RELEASING;
+}
+
 static inline void ksmbd_conn_set_new(struct ksmbd_conn *conn)
 {
        WRITE_ONCE(conn->status, KSMBD_SESS_NEW);
@@ -215,4 +221,11 @@ static inline void ksmbd_conn_set_exiting(struct ksmbd_conn *conn)
 {
        WRITE_ONCE(conn->status, KSMBD_SESS_EXITING);
 }
+
+static inline void ksmbd_conn_set_releasing(struct ksmbd_conn *conn)
+{
+       WRITE_ONCE(conn->status, KSMBD_SESS_RELEASING);
+}
+
+void ksmbd_all_conn_set_status(u64 sess_id, u32 status);
 #endif /* __CONNECTION_H__ */
index 8ce17b3..aa9c138 100644 (file)
@@ -129,6 +129,9 @@ int ksmbd_tree_conn_session_logoff(struct ksmbd_session *sess)
        struct ksmbd_tree_connect *tc;
        unsigned long id;
 
+       if (!sess)
+               return -EINVAL;
+
        xa_for_each(&sess->tree_conns, id, tc)
                ret |= ksmbd_tree_conn_disconnect(sess, tc);
        xa_destroy(&sess->tree_conns);
index 3840de7..8a5dcab 100644 (file)
@@ -144,10 +144,6 @@ void ksmbd_session_destroy(struct ksmbd_session *sess)
        if (!sess)
                return;
 
-       down_write(&sessions_table_lock);
-       hash_del(&sess->hlist);
-       up_write(&sessions_table_lock);
-
        if (sess->user)
                ksmbd_free_user(sess->user);
 
@@ -178,15 +174,18 @@ static void ksmbd_expire_session(struct ksmbd_conn *conn)
        unsigned long id;
        struct ksmbd_session *sess;
 
+       down_write(&sessions_table_lock);
        xa_for_each(&conn->sessions, id, sess) {
                if (sess->state != SMB2_SESSION_VALID ||
                    time_after(jiffies,
                               sess->last_active + SMB2_SESSION_TIMEOUT)) {
                        xa_erase(&conn->sessions, sess->id);
+                       hash_del(&sess->hlist);
                        ksmbd_session_destroy(sess);
                        continue;
                }
        }
+       up_write(&sessions_table_lock);
 }
 
 int ksmbd_session_register(struct ksmbd_conn *conn,
@@ -198,15 +197,16 @@ int ksmbd_session_register(struct ksmbd_conn *conn,
        return xa_err(xa_store(&conn->sessions, sess->id, sess, GFP_KERNEL));
 }
 
-static void ksmbd_chann_del(struct ksmbd_conn *conn, struct ksmbd_session *sess)
+static int ksmbd_chann_del(struct ksmbd_conn *conn, struct ksmbd_session *sess)
 {
        struct channel *chann;
 
        chann = xa_erase(&sess->ksmbd_chann_list, (long)conn);
        if (!chann)
-               return;
+               return -ENOENT;
 
        kfree(chann);
+       return 0;
 }
 
 void ksmbd_sessions_deregister(struct ksmbd_conn *conn)
@@ -214,13 +214,37 @@ void ksmbd_sessions_deregister(struct ksmbd_conn *conn)
        struct ksmbd_session *sess;
        unsigned long id;
 
+       down_write(&sessions_table_lock);
+       if (conn->binding) {
+               int bkt;
+               struct hlist_node *tmp;
+
+               hash_for_each_safe(sessions_table, bkt, tmp, sess, hlist) {
+                       if (!ksmbd_chann_del(conn, sess) &&
+                           xa_empty(&sess->ksmbd_chann_list)) {
+                               hash_del(&sess->hlist);
+                               ksmbd_session_destroy(sess);
+                       }
+               }
+       }
+
        xa_for_each(&conn->sessions, id, sess) {
+               unsigned long chann_id;
+               struct channel *chann;
+
+               xa_for_each(&sess->ksmbd_chann_list, chann_id, chann) {
+                       if (chann->conn != conn)
+                               ksmbd_conn_set_exiting(chann->conn);
+               }
+
                ksmbd_chann_del(conn, sess);
                if (xa_empty(&sess->ksmbd_chann_list)) {
                        xa_erase(&conn->sessions, sess->id);
+                       hash_del(&sess->hlist);
                        ksmbd_session_destroy(sess);
                }
        }
+       up_write(&sessions_table_lock);
 }
 
 struct ksmbd_session *ksmbd_session_lookup(struct ksmbd_conn *conn,
index 26db1dc..0ca1c5d 100644 (file)
@@ -2081,21 +2081,22 @@ int smb2_session_logoff(struct ksmbd_work *work)
        struct smb2_logoff_rsp *rsp = smb2_get_msg(work->response_buf);
        struct ksmbd_session *sess;
        struct smb2_logoff_req *req = smb2_get_msg(work->request_buf);
+       u64 sess_id = le64_to_cpu(req->hdr.SessionId);
 
        rsp->StructureSize = cpu_to_le16(4);
        inc_rfc1001_len(work->response_buf, 4);
 
        ksmbd_debug(SMB, "request\n");
 
-       ksmbd_conn_set_need_reconnect(conn);
+       ksmbd_all_conn_set_status(sess_id, KSMBD_SESS_NEED_RECONNECT);
        ksmbd_close_session_fds(work);
-       ksmbd_conn_wait_idle(conn);
+       ksmbd_conn_wait_idle(conn, sess_id);
 
        /*
         * Re-lookup session to validate if session is deleted
         * while waiting request complete
         */
-       sess = ksmbd_session_lookup(conn, le64_to_cpu(req->hdr.SessionId));
+       sess = ksmbd_session_lookup_all(conn, sess_id);
        if (ksmbd_tree_conn_session_logoff(sess)) {
                ksmbd_debug(SMB, "Invalid tid %d\n", req->hdr.Id.SyncId.TreeId);
                rsp->hdr.Status = STATUS_NETWORK_NAME_DELETED;
@@ -2108,7 +2109,7 @@ int smb2_session_logoff(struct ksmbd_work *work)
 
        ksmbd_free_user(sess->user);
        sess->user = NULL;
-       ksmbd_conn_set_need_negotiate(conn);
+       ksmbd_all_conn_set_status(sess_id, KSMBD_SESS_NEED_NEGOTIATE);
        return 0;
 }
 
@@ -6863,7 +6864,7 @@ int smb2_lock(struct ksmbd_work *work)
 
                nolock = 1;
                /* check locks in connection list */
-               read_lock(&conn_list_lock);
+               down_read(&conn_list_lock);
                list_for_each_entry(conn, &conn_list, conns_list) {
                        spin_lock(&conn->llist_lock);
                        list_for_each_entry_safe(cmp_lock, tmp2, &conn->lock_list, clist) {
@@ -6880,7 +6881,7 @@ int smb2_lock(struct ksmbd_work *work)
                                                list_del(&cmp_lock->flist);
                                                list_del(&cmp_lock->clist);
                                                spin_unlock(&conn->llist_lock);
-                                               read_unlock(&conn_list_lock);
+                                               up_read(&conn_list_lock);
 
                                                locks_free_lock(cmp_lock->fl);
                                                kfree(cmp_lock);
@@ -6902,7 +6903,7 @@ int smb2_lock(struct ksmbd_work *work)
                                    cmp_lock->start > smb_lock->start &&
                                    cmp_lock->start < smb_lock->end) {
                                        spin_unlock(&conn->llist_lock);
-                                       read_unlock(&conn_list_lock);
+                                       up_read(&conn_list_lock);
                                        pr_err("previous lock conflict with zero byte lock range\n");
                                        goto out;
                                }
@@ -6911,7 +6912,7 @@ int smb2_lock(struct ksmbd_work *work)
                                    smb_lock->start > cmp_lock->start &&
                                    smb_lock->start < cmp_lock->end) {
                                        spin_unlock(&conn->llist_lock);
-                                       read_unlock(&conn_list_lock);
+                                       up_read(&conn_list_lock);
                                        pr_err("current lock conflict with zero byte lock range\n");
                                        goto out;
                                }
@@ -6922,14 +6923,14 @@ int smb2_lock(struct ksmbd_work *work)
                                      cmp_lock->end >= smb_lock->end)) &&
                                    !cmp_lock->zero_len && !smb_lock->zero_len) {
                                        spin_unlock(&conn->llist_lock);
-                                       read_unlock(&conn_list_lock);
+                                       up_read(&conn_list_lock);
                                        pr_err("Not allow lock operation on exclusive lock range\n");
                                        goto out;
                                }
                        }
                        spin_unlock(&conn->llist_lock);
                }
-               read_unlock(&conn_list_lock);
+               up_read(&conn_list_lock);
 out_check_cl:
                if (smb_lock->fl->fl_type == F_UNLCK && nolock) {
                        pr_err("Try to unlock nolocked range\n");