Target/iser: Fix iscsit_accept_np and rdma_cm racy flow
authorSagi Grimberg <sagig@mellanox.com>
Tue, 29 Apr 2014 10:13:45 +0000 (13:13 +0300)
committerNicholas Bellinger <nab@linux-iscsi.org>
Fri, 16 May 2014 00:09:10 +0000 (17:09 -0700)
RDMA CM and iSCSI target flows are asynchronous and completely
uncorrelated. Relying on the fact that iscsi_accept_np will be called
after CM connection request event and will wait for it is a mistake.

When attempting to login to a few targets this flow is racy and
unpredictable, but for parallel login to dozens of targets will
race and hang every time.

The correct synchronizing mechanism in this case is pending on
a semaphore rather than a wait_for_event. We keep the pending
interruptible for iscsi_np cleanup stage.

(Squash patch to remove dead code into parent - nab)

Reported-by: Slava Shwartsman <valyushash@gmail.com>
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Cc: stable@vger.kernel.org # 3.10+
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
drivers/infiniband/ulp/isert/ib_isert.c
drivers/infiniband/ulp/isert/ib_isert.h

index fda3feb4500816d44823aa8f2a49a0c798320cb9..826eaf5741f64241727db72d73cfb47b38de2ecd 100644 (file)
@@ -28,6 +28,7 @@
 #include <target/target_core_base.h>
 #include <target/target_core_fabric.h>
 #include <target/iscsi/iscsi_transport.h>
+#include <linux/semaphore.h>
 
 #include "isert_proto.h"
 #include "ib_isert.h"
@@ -666,8 +667,8 @@ isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
        list_add_tail(&isert_conn->conn_accept_node, &isert_np->np_accept_list);
        mutex_unlock(&isert_np->np_accept_mutex);
 
-       pr_debug("isert_connect_request() waking up np_accept_wq: %p\n", np);
-       wake_up(&isert_np->np_accept_wq);
+       pr_debug("isert_connect_request() up np_sem np: %p\n", np);
+       up(&isert_np->np_sem);
        return 0;
 
 out_conn_dev:
@@ -2999,7 +3000,7 @@ isert_setup_np(struct iscsi_np *np,
                pr_err("Unable to allocate struct isert_np\n");
                return -ENOMEM;
        }
-       init_waitqueue_head(&isert_np->np_accept_wq);
+       sema_init(&isert_np->np_sem, 0);
        mutex_init(&isert_np->np_accept_mutex);
        INIT_LIST_HEAD(&isert_np->np_accept_list);
        init_completion(&isert_np->np_login_comp);
@@ -3047,18 +3048,6 @@ out:
        return ret;
 }
 
-static int
-isert_check_accept_queue(struct isert_np *isert_np)
-{
-       int empty;
-
-       mutex_lock(&isert_np->np_accept_mutex);
-       empty = list_empty(&isert_np->np_accept_list);
-       mutex_unlock(&isert_np->np_accept_mutex);
-
-       return empty;
-}
-
 static int
 isert_rdma_accept(struct isert_conn *isert_conn)
 {
@@ -3151,16 +3140,14 @@ isert_accept_np(struct iscsi_np *np, struct iscsi_conn *conn)
        int max_accept = 0, ret;
 
 accept_wait:
-       ret = wait_event_interruptible(isert_np->np_accept_wq,
-                       !isert_check_accept_queue(isert_np) ||
-                       np->np_thread_state == ISCSI_NP_THREAD_RESET);
+       ret = down_interruptible(&isert_np->np_sem);
        if (max_accept > 5)
                return -ENODEV;
 
        spin_lock_bh(&np->np_thread_lock);
        if (np->np_thread_state == ISCSI_NP_THREAD_RESET) {
                spin_unlock_bh(&np->np_thread_lock);
-               pr_err("ISCSI_NP_THREAD_RESET for isert_accept_np\n");
+               pr_debug("ISCSI_NP_THREAD_RESET for isert_accept_np\n");
                return -ENODEV;
        }
        spin_unlock_bh(&np->np_thread_lock);
index 4c072ae34c01a3021e57cb5378a3949b97d0876c..da6612e6800004b0984880d54ef57bcaab1b0be1 100644 (file)
@@ -182,7 +182,7 @@ struct isert_device {
 };
 
 struct isert_np {
-       wait_queue_head_t       np_accept_wq;
+       struct semaphore        np_sem;
        struct rdma_cm_id       *np_cm_id;
        struct mutex            np_accept_mutex;
        struct list_head        np_accept_list;