ocfs2: use the new DLM operation callbacks while requesting new lockspace
authorGoldwyn Rodrigues <rgoldwyn@suse.de>
Tue, 21 Jan 2014 23:48:32 +0000 (15:48 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 22 Jan 2014 00:19:42 +0000 (16:19 -0800)
Attempt to use the new DLM operations.  If it is not supported, use the
traditional ocfs2_controld.

To exchange ocfs2 versioning, we use the LVB of the version dlm lock.
It first attempts to take the lock in EX mode (non-blocking).  If
successful (which means it is the first mount), it writes the version
number and downconverts to PR lock.  If it is unsuccessful, it reads the
version from the lock.

If this becomes the standard (with o2cb as well), it could simplify
userspace tools to check if the filesystem is mounted on other nodes.

Dan: Since ocfs2_protocol_version are two u8 values, the additional
checks with LONG* don't make sense.

Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/ocfs2/stack_user.c

index d3867100aca8a5b02f0298d55168d624dd6e2d9e..673ab40eb81ba68bd33a4c09df21a132b8c12b37 100644 (file)
@@ -23,6 +23,7 @@
 #include <linux/mutex.h>
 #include <linux/slab.h>
 #include <linux/reboot.h>
+#include <linux/sched.h>
 #include <asm/uaccess.h>
 
 #include "stackglue.h"
@@ -122,6 +123,7 @@ struct ocfs2_live_connection {
        struct dlm_lksb                 oc_version_lksb;
        char                            oc_lvb[DLM_LVB_LEN];
        struct completion               oc_sync_wait;
+       wait_queue_head_t               oc_wait;
 };
 
 struct ocfs2_control_private {
@@ -218,7 +220,7 @@ static int ocfs2_live_connection_attach(struct ocfs2_cluster_connection *conn,
        mutex_lock(&ocfs2_control_lock);
        c->oc_conn = conn;
 
-       if (atomic_read(&ocfs2_control_opened))
+       if ((c->oc_type == NO_CONTROLD) || atomic_read(&ocfs2_control_opened))
                list_add(&c->oc_list, &ocfs2_live_connection_list);
        else {
                printk(KERN_ERR
@@ -897,6 +899,55 @@ static int version_unlock(struct ocfs2_cluster_connection *conn)
        return sync_unlock(conn, &lc->oc_version_lksb, VERSION_LOCK);
 }
 
+/* get_protocol_version()
+ *
+ * To exchange ocfs2 versioning, we use the LVB of the version dlm lock.
+ * The algorithm is:
+ * 1. Attempt to take the lock in EX mode (non-blocking).
+ * 2. If successful (which means it is the first mount), write the
+ *    version number and downconvert to PR lock.
+ * 3. If unsuccessful (returns -EAGAIN), read the version from the LVB after
+ *    taking the PR lock.
+ */
+
+static int get_protocol_version(struct ocfs2_cluster_connection *conn)
+{
+       int ret;
+       struct ocfs2_live_connection *lc = conn->cc_private;
+       struct ocfs2_protocol_version pv;
+
+       running_proto.pv_major =
+               ocfs2_user_plugin.sp_max_proto.pv_major;
+       running_proto.pv_minor =
+               ocfs2_user_plugin.sp_max_proto.pv_minor;
+
+       lc->oc_version_lksb.sb_lvbptr = lc->oc_lvb;
+       ret = version_lock(conn, DLM_LOCK_EX,
+                       DLM_LKF_VALBLK|DLM_LKF_NOQUEUE);
+       if (!ret) {
+               conn->cc_version.pv_major = running_proto.pv_major;
+               conn->cc_version.pv_minor = running_proto.pv_minor;
+               version_to_lvb(&running_proto, lc->oc_lvb);
+               version_lock(conn, DLM_LOCK_PR, DLM_LKF_CONVERT|DLM_LKF_VALBLK);
+       } else if (ret == -EAGAIN) {
+               ret = version_lock(conn, DLM_LOCK_PR, DLM_LKF_VALBLK);
+               if (ret)
+                       goto out;
+               lvb_to_version(lc->oc_lvb, &pv);
+
+               if ((pv.pv_major != running_proto.pv_major) ||
+                               (pv.pv_minor > running_proto.pv_minor)) {
+                       ret = -EINVAL;
+                       goto out;
+               }
+
+               conn->cc_version.pv_major = pv.pv_major;
+               conn->cc_version.pv_minor = pv.pv_minor;
+       }
+out:
+       return ret;
+}
+
 static void user_recover_prep(void *arg)
 {
 }
@@ -925,6 +976,7 @@ static void user_recover_done(void *arg, struct dlm_slot *slots,
                }
 
        lc->oc_our_slot = our_slot;
+       wake_up(&lc->oc_wait);
 }
 
 const struct dlm_lockspace_ops ocfs2_ls_ops = {
@@ -933,11 +985,21 @@ const struct dlm_lockspace_ops ocfs2_ls_ops = {
        .recover_done = user_recover_done,
 };
 
+static int user_cluster_disconnect(struct ocfs2_cluster_connection *conn)
+{
+       version_unlock(conn);
+       dlm_release_lockspace(conn->cc_lockspace, 2);
+       conn->cc_lockspace = NULL;
+       ocfs2_live_connection_drop(conn->cc_private);
+       conn->cc_private = NULL;
+       return 0;
+}
+
 static int user_cluster_connect(struct ocfs2_cluster_connection *conn)
 {
        dlm_lockspace_t *fsdlm;
        struct ocfs2_live_connection *lc;
-       int rc;
+       int rc, ops_rv;
 
        BUG_ON(conn == NULL);
 
@@ -947,11 +1009,44 @@ static int user_cluster_connect(struct ocfs2_cluster_connection *conn)
                goto out;
        }
 
-       lc->oc_type = WITH_CONTROLD;
+       init_waitqueue_head(&lc->oc_wait);
+       init_completion(&lc->oc_sync_wait);
+       atomic_set(&lc->oc_this_node, 0);
+       conn->cc_private = lc;
+       lc->oc_type = NO_CONTROLD;
+
+       rc = dlm_new_lockspace(conn->cc_name, conn->cc_cluster_name,
+                              DLM_LSFL_FS, DLM_LVB_LEN,
+                              &ocfs2_ls_ops, conn, &ops_rv, &fsdlm);
+       if (rc)
+               goto out;
+
+       if (ops_rv == -EOPNOTSUPP) {
+               lc->oc_type = WITH_CONTROLD;
+               printk(KERN_NOTICE "ocfs2: You seem to be using an older "
+                               "version of dlm_controld and/or ocfs2-tools."
+                               " Please consider upgrading.\n");
+       } else if (ops_rv) {
+               rc = ops_rv;
+               goto out;
+       }
+       conn->cc_lockspace = fsdlm;
+
        rc = ocfs2_live_connection_attach(conn, lc);
        if (rc)
                goto out;
 
+       if (lc->oc_type == NO_CONTROLD) {
+               rc = get_protocol_version(conn);
+               if (rc) {
+                       printk(KERN_ERR "ocfs2: Could not determine"
+                                       " locking version\n");
+                       user_cluster_disconnect(conn);
+                       goto out;
+               }
+               wait_event(lc->oc_wait, (atomic_read(&lc->oc_this_node) > 0));
+       }
+
        /*
         * running_proto must have been set before we allowed any mounts
         * to proceed.
@@ -959,40 +1054,20 @@ static int user_cluster_connect(struct ocfs2_cluster_connection *conn)
        if (fs_protocol_compare(&running_proto, &conn->cc_version)) {
                printk(KERN_ERR
                       "Unable to mount with fs locking protocol version "
-                      "%u.%u because the userspace control daemon has "
-                      "negotiated %u.%u\n",
+                      "%u.%u because negotiated protocol is %u.%u\n",
                       conn->cc_version.pv_major, conn->cc_version.pv_minor,
                       running_proto.pv_major, running_proto.pv_minor);
                rc = -EPROTO;
                ocfs2_live_connection_drop(lc);
                lc = NULL;
-               goto out;
-       }
-
-       rc = dlm_new_lockspace(conn->cc_name, NULL, DLM_LSFL_FS, DLM_LVB_LEN,
-                              NULL, NULL, NULL, &fsdlm);
-       if (rc) {
-               ocfs2_live_connection_drop(lc);
-               lc = NULL;
-               goto out;
        }
 
-       conn->cc_private = lc;
-       conn->cc_lockspace = fsdlm;
 out:
        if (rc && lc)
                kfree(lc);
        return rc;
 }
 
-static int user_cluster_disconnect(struct ocfs2_cluster_connection *conn)
-{
-       dlm_release_lockspace(conn->cc_lockspace, 2);
-       conn->cc_lockspace = NULL;
-       ocfs2_live_connection_drop(conn->cc_private);
-       conn->cc_private = NULL;
-       return 0;
-}
 
 static int user_cluster_this_node(struct ocfs2_cluster_connection *conn,
                                  unsigned int *this_node)
@@ -1002,8 +1077,11 @@ static int user_cluster_this_node(struct ocfs2_cluster_connection *conn,
 
        if (lc->oc_type == WITH_CONTROLD)
                rc = ocfs2_control_get_this_node();
+       else if (lc->oc_type == NO_CONTROLD)
+               rc = atomic_read(&lc->oc_this_node);
        else
                rc = -EINVAL;
+
        if (rc < 0)
                return rc;