9p: add more conservative locking
Eric Van Hensbergen [Fri, 17 Oct 2008 17:45:40 +0000 (12:45 -0500)]
During the reorganization some of the multi-theaded locking assumptions were
accidently relaxed.  This patch moves us back towards a more conservative
locking strategy.

Signed-off-by: Eric Van Hensbergen <ericvh@gmail.com>

net/9p/trans_fd.c

index e8ebe2c..be65d82 100644 (file)
@@ -188,8 +188,16 @@ static void p9_conn_cancel(struct p9_conn *m, int err)
        LIST_HEAD(cancel_list);
 
        P9_DPRINTK(P9_DEBUG_ERROR, "mux %p err %d\n", m, err);
-       m->err = err;
+
        spin_lock_irqsave(&m->client->lock, flags);
+
+       if (m->err) {
+               spin_unlock_irqrestore(&m->client->lock, flags);
+               return;
+       }
+
+       m->err = err;
+
        list_for_each_entry_safe(req, rtmp, &m->req_list, req_list) {
                req->status = REQ_STATUS_ERROR;
                if (!req->t_err)
@@ -352,8 +360,9 @@ static void p9_read_work(struct work_struct *work)
        /* not an else because some packets (like clunk) have no payload */
        if ((m->req) && (m->rpos == m->rsize)) { /* packet is read in */
                P9_DPRINTK(P9_DEBUG_TRANS, "got new packet\n");
-
+               spin_lock(&m->client->lock);
                list_del(&m->req->req_list);
+               spin_unlock(&m->client->lock);
                p9_client_cb(m->client, m->req);
 
                m->rbuf = NULL;
@@ -651,9 +660,8 @@ static int p9_fd_request(struct p9_client *client, struct p9_req_t *req)
        if (m->err < 0)
                return m->err;
 
-       req->status = REQ_STATUS_UNSENT;
-
        spin_lock(&client->lock);
+       req->status = REQ_STATUS_UNSENT;
        list_add_tail(&req->req_list, &m->unsent_req_list);
        spin_unlock(&client->lock);
 
@@ -672,19 +680,21 @@ static int p9_fd_cancel(struct p9_client *client, struct p9_req_t *req)
 {
        struct p9_trans_fd *ts = client->trans;
        struct p9_conn *m = ts->conn;
+       int ret = 1;
 
        P9_DPRINTK(P9_DEBUG_TRANS, "mux %p req %p\n", m, req);
 
        spin_lock(&client->lock);
        list_del(&req->req_list);
-       spin_unlock(&client->lock);
 
        if (req->status == REQ_STATUS_UNSENT) {
                req->status = REQ_STATUS_FLSHD;
-               return 0;
+               ret = 0;
        }
 
-       return 1;
+       spin_unlock(&client->lock);
+
+       return ret;
 }
 
 /**