cifs: Fix a race condition with cifs_echo_request (f2caf901) · Commits · 戴 / test

There is a race condition with how we send (or supress and don't send) smb echos that will cause the client to incorrectly think the server is unresponsive and thus needs to be reconnected. Summary of the race condition: 1) Daisy chaining scheduling creates a gap. 2) If traffic comes unfortunate shortly after the last echo, the planned echo is suppressed. 3) Due to the gap, the next echo transmission is delayed until after the timeout, which is set hard to twice the echo interval. This is fixed by changing the timeouts from 2 to three times the echo interval. Detailed description of the bug: https://lutz.donnerhacke.de/eng/Blog/Groundhog-Day-with-SMB-remount Signed-off-by:

Ronnie Sahlberg <lsahlber@redhat.com> Reviewed-by:

Pavel Shilovsky <pshilov@microsoft.com> Signed-off-by:

Steve French <stfrench@microsoft.com>

fs/cifs/connect.c

+4 −4

Original line number	Diff line number	Diff line
		@@ -706,10 +706,10 @@ static bool
		server_unresponsive(struct TCP_Server_Info *server)
		{
		/*
		* We need to wait 2 echo intervals to make sure we handle such
		* We need to wait 3 echo intervals to make sure we handle such
		* situations right:
		* 1s client sends a normal SMB request
		* 2s client gets a response
		* 3s client gets a response
		* 30s echo workqueue job pops, and decides we got a response recently
		* and don't need to send another
		* ...
		@@ -718,9 +718,9 @@ server_unresponsive(struct TCP_Server_Info *server)
		*/
		if ((server->tcpStatus == CifsGood \|\|
		server->tcpStatus == CifsNeedNegotiate) &&
		time_after(jiffies, server->lstrp + 2 * server->echo_interval)) {
		time_after(jiffies, server->lstrp + 3 * server->echo_interval)) {
		cifs_dbg(VFS, "Server %s has not responded in %lu seconds. Reconnecting...\n",
		server->hostname, (2 * server->echo_interval) / HZ);
		server->hostname, (3 * server->echo_interval) / HZ);
		cifs_reconnect(server);
		wake_up(&server->response_q);
		return true;

Admin message