nfs4: fix potential race with rapid nfs_callback_up/down cycle
authorJeff Layton <jlayton@redhat.com>
Wed, 11 Jun 2008 14:03:11 +0000 (10:03 -0400)
committerTrond Myklebust <Trond.Myklebust@netapp.com>
Wed, 9 Jul 2008 16:09:32 +0000 (12:09 -0400)
If the nfsv4 callback thread is rapidly brought up and down, it's
possible that nfs_callback_svc might never get a chance to run. If
this happens, the cleanup at thread exit might never occur, throwing
the refcounting off and nfs_callback_info in an incorrect state.

Move the clean functions into nfs_callback_down. Also change the
nfs_callback_info struct to track the svc_rqst rather than svc_serv
since we need to know that to call svc_exit_thread.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
fs/nfs/callback.c

index 9e713d2d5d7476b1ac44a33edce36bd20b038f6b..f447f4b4476cc37b26144ffdfcef83cd80eab53e 100644 (file)
@@ -27,7 +27,7 @@
 
 struct nfs_callback_data {
        unsigned int users;
-       struct svc_serv *serv;
+       struct svc_rqst *rqst;
        struct task_struct *task;
 };
 
@@ -91,18 +91,15 @@ nfs_callback_svc(void *vrqstp)
                svc_process(rqstp);
        }
        unlock_kernel();
-       nfs_callback_info.task = NULL;
-       svc_exit_thread(rqstp);
        return 0;
 }
 
 /*
- * Bring up the server process if it is not already up.
+ * Bring up the callback thread if it is not already up.
  */
 int nfs_callback_up(void)
 {
        struct svc_serv *serv = NULL;
-       struct svc_rqst *rqstp;
        int ret = 0;
 
        mutex_lock(&nfs_callback_mutex);
@@ -120,22 +117,23 @@ int nfs_callback_up(void)
        nfs_callback_tcpport = ret;
        dprintk("Callback port = 0x%x\n", nfs_callback_tcpport);
 
-       rqstp = svc_prepare_thread(serv, &serv->sv_pools[0]);
-       if (IS_ERR(rqstp)) {
-               ret = PTR_ERR(rqstp);
+       nfs_callback_info.rqst = svc_prepare_thread(serv, &serv->sv_pools[0]);
+       if (IS_ERR(nfs_callback_info.rqst)) {
+               ret = PTR_ERR(nfs_callback_info.rqst);
+               nfs_callback_info.rqst = NULL;
                goto out_err;
        }
 
        svc_sock_update_bufs(serv);
-       nfs_callback_info.serv = serv;
 
-       nfs_callback_info.task = kthread_run(nfs_callback_svc, rqstp,
+       nfs_callback_info.task = kthread_run(nfs_callback_svc,
+                                            nfs_callback_info.rqst,
                                             "nfsv4-svc");
        if (IS_ERR(nfs_callback_info.task)) {
                ret = PTR_ERR(nfs_callback_info.task);
-               nfs_callback_info.serv = NULL;
+               svc_exit_thread(nfs_callback_info.rqst);
+               nfs_callback_info.rqst = NULL;
                nfs_callback_info.task = NULL;
-               svc_exit_thread(rqstp);
                goto out_err;
        }
 out:
@@ -157,14 +155,18 @@ out_err:
 }
 
 /*
- * Kill the server process if it is not already down.
+ * Kill the callback thread if it's no longer being used.
  */
 void nfs_callback_down(void)
 {
        mutex_lock(&nfs_callback_mutex);
        nfs_callback_info.users--;
-       if (nfs_callback_info.users == 0 && nfs_callback_info.task != NULL)
+       if (nfs_callback_info.users == 0 && nfs_callback_info.task != NULL) {
                kthread_stop(nfs_callback_info.task);
+               svc_exit_thread(nfs_callback_info.rqst);
+               nfs_callback_info.rqst = NULL;
+               nfs_callback_info.task = NULL;
+       }
        mutex_unlock(&nfs_callback_mutex);
 }