[linux-cifs-client] [PATCH] Cannot allocate memory

[Jeff Layton]

On Tue, 10 Mar 2009 18:31:36 +0300
Pavel Shilovsky <piastry at etersoft.ru> wrote:

> Hello!
> 
> This patch resolves problem "Cannot allocate memory" during loading 
> module when we do unload/load quickly. During unloading the 
> cifs_demultiplex_thread doesn't have enough time for clearing cache and 
> kmem_cache_destroy fails. Then during loading module it fails with 
> kmem_cache_create: duplicate cache cifs_request.
> 
> This script can show this bug:
> 
> rmmod cifs
> 
> for i in `seq 1 20`;
> do {
>         echo "trying to mount " $i
>         insmod ./cifs.ko
>         cifsmount [имя шары] [путь] -onoperm
>         umount [путь]
>         rmmod cifs
> };
> done
> 
> 
> 
> And it fails on 2nd iteration.
> 
> My patch fixes it.
> 
> diff --git a/fs/cifs/cifsglob.h b/fs/cifs/cifsglob.h
> index 1ae6314..18ba45a 100644
> --- a/fs/cifs/cifsglob.h
> +++ b/fs/cifs/cifsglob.h
> @@ -176,6 +176,7 @@ struct TCP_Server_Info {
>         struct mac_key mac_signing_key;
>         char ntlmv2_hash[16];
>         unsigned long lstrp; /* when we got last response from this 
> server */
> +       int running;
>  };
> 
>  /*
> diff --git a/fs/cifs/connect.c b/fs/cifs/connect.c
> index f254235..39daaf2 100644
> --- a/fs/cifs/connect.c
> +++ b/fs/cifs/connect.c
> @@ -347,6 +347,7 @@ cifs_demultiplex_thread(struct TCP_Server_Info *server)
>         bool isMultiRsp;
>         int reconnect;
> 
> +       server->running = 1;
>         current->flags |= PF_MEMALLOC;
>         cFYI(1, ("Demultiplex PID: %d", task_pid_nr(current)));
> 
> @@ -758,6 +759,8 @@ multi_t2_fnd:
> 
>         kfree(server->hostname);
>         task_to_wake = xchg(&server->tsk, NULL);
> +       server->running = 0;
> +       msleep(20);
>         kfree(server);
> 
>         length = atomic_dec_return(&tcpSesAllocCount);
> @@ -1408,6 +1411,9 @@ cifs_put_tcp_session(struct TCP_Server_Info *server)
>         task = xchg(&server->tsk, NULL);
>         if (task)
>                 force_sig(SIGKILL, task);
> +
> +       while(server->running == 1)
> +               msleep(10);
>  }
> 
>  static struct cifsSesInfo *
> 

On what kernel version are you reproducing this? Does it have this patch?

commit 0468a2cf914e79442b8309ce62e3f861599d8cab
Author: Jeff Layton <jlayton at redhat.com>
Date:   Mon Dec 1 07:09:35 2008 -0500

    cifs: take module reference when starting cifsd
    
    cifsd can outlive the last cifs mount. We need to hold a module
    reference until it exits to prevent someone from unplugging
    the module until we're ready.
    
    Signed-off-by: Jeff Layton <jlayton at redhat.com>
    Signed-off-by: Steve French <sfrench at us.ibm.com>

-- 
Jeff Layton <jlayton at redhat.com>