Bug 1579050

Description of problem:

If a cifs mount becomes unavailable, requiring a reconnect, and an application has generated a request which is blocked, and the credentials are no longer valid (for example, password has changed), the cifs client will reattempt the session setup forever, using the now-invalid credentials.

Because there is no way to update the stored password, the mount will never be usable again, but the reconnect will be reattempted hundreds of times/second.

For a 'soft' mount, at least, the kernel should give up the reconnect attempt, and return an error to the application, 


Version-Release number of selected component (if applicable):

kernel-*



How reproducible:

easy

Steps to Reproduce:


set an 'idle' disconnect of 1 minute in /etc/smb.conf on samba server
	deadtime = 1

# useradd testuser
# echo -e 'test1\ntest1' | smbpasswd -L -s testuser


on the client:

# mount.cifs //server/testuser /mnt/test -o username=testuser,password=test1,soft

doio.c
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <string.h>

int main(int argc, char *argv[]) {
        char buf[] = "TESTING";
        int ret, fd;

        fd = open("testfile", O_RDWR|O_CREAT|O_TRUNC, 0660);
        while (42) {
                if ((ret = write(fd, buf, strlen(buf))) != strlen(buf))
                        break;
                sleep(1);
        }
        printf("write returned %d. error: %m\n", ret);
        return EXIT_FAILURE;
}

# cd /mnt/test
# gcc doio.c -o doio
# ./doio



back on the server, change the password:
# echo -e 'test2\ntest2' | smbpasswd -L -s testuser

firewall off the client for a minute or two:
# iptables -A INPUT -s client -j DROP ; sleep 120 ; iptables -F



Actual results:

the server will time out and disconnect the 'idle' client, and/or the client will restart the session setup

reconnection is impossible due to the changed password

processes hang, potentially unkillable
kernel attempts reconnect at high rate causing high network & cpu usage, syslog fills with NT_STATUS_LOGON_FAILURE messages


Expected results:

Customer expects 'soft' mount to return IO error to application, and not block indefinitely


Additional info: