]> pilppa.com Git - linux-2.6-omap-h63xx.git/commitdiff
[DLM] wait for config check during join [6/6]
authorDavid Teigland <teigland@redhat.com>
Fri, 18 May 2007 14:03:35 +0000 (09:03 -0500)
committerSteven Whitehouse <swhiteho@redhat.com>
Mon, 9 Jul 2007 07:22:42 +0000 (08:22 +0100)
Joining the lockspace should wait for the initial round of inter-node
config checks to complete before returning.  This way, if there's a
configuration mismatch between the joining node and the existing nodes,
the join can fail and return an error to the application.

Signed-off-by: David Teigland <teigland@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
fs/dlm/dlm_internal.h
fs/dlm/lockspace.c
fs/dlm/member.c
fs/dlm/rcom.c

index a8d6e993697cdb01290da356ce6f1917d69917fe..03ba6c4fd5c2a0bc7603e45d4add4fbd2defbeb2 100644 (file)
@@ -472,6 +472,8 @@ struct dlm_ls {
 
        wait_queue_head_t       ls_uevent_wait; /* user part of join/leave */
        int                     ls_uevent_result;
+       struct completion       ls_members_done;
+       int                     ls_members_result;
 
        struct miscdevice       ls_device;
 
index a3a50e67e4dd4855e993612f55d541916faae06d..c8f0c15ac166460ee07e0476b4a6d9b8c5a70833 100644 (file)
@@ -197,13 +197,24 @@ static int do_uevent(struct dlm_ls *ls, int in)
        else
                kobject_uevent(&ls->ls_kobj, KOBJ_OFFLINE);
 
+       log_debug(ls, "%s the lockspace group...", in ? "joining" : "leaving");
+
+       /* dlm_controld will see the uevent, do the necessary group management
+          and then write to sysfs to wake us */
+
        error = wait_event_interruptible(ls->ls_uevent_wait,
                        test_and_clear_bit(LSFL_UEVENT_WAIT, &ls->ls_flags));
+
+       log_debug(ls, "group event done %d %d", error, ls->ls_uevent_result);
+
        if (error)
                goto out;
 
        error = ls->ls_uevent_result;
  out:
+       if (error)
+               log_error(ls, "group %s failed %d %d", in ? "join" : "leave",
+                         error, ls->ls_uevent_result);
        return error;
 }
 
@@ -490,6 +501,8 @@ static int new_lockspace(char *name, int namelen, void **lockspace,
 
        init_waitqueue_head(&ls->ls_uevent_wait);
        ls->ls_uevent_result = 0;
+       init_completion(&ls->ls_members_done);
+       ls->ls_members_result = -1;
 
        ls->ls_recoverd_task = NULL;
        mutex_init(&ls->ls_recoverd_active);
@@ -540,10 +553,21 @@ static int new_lockspace(char *name, int namelen, void **lockspace,
        /* let kobject handle freeing of ls if there's an error */
        do_unreg = 1;
 
+       /* This uevent triggers dlm_controld in userspace to add us to the
+          group of nodes that are members of this lockspace (managed by the
+          cluster infrastructure.)  Once it's done that, it tells us who the
+          current lockspace members are (via configfs) and then tells the
+          lockspace to start running (via sysfs) in dlm_ls_start(). */
+
        error = do_uevent(ls, 1);
        if (error)
                goto out_stop;
 
+       wait_for_completion(&ls->ls_members_done);
+       error = ls->ls_members_result;
+       if (error)
+               goto out_members;
+
        dlm_create_debug_file(ls);
 
        log_debug(ls, "join complete");
@@ -551,6 +575,10 @@ static int new_lockspace(char *name, int namelen, void **lockspace,
        *lockspace = ls;
        return 0;
 
+ out_members:
+       do_uevent(ls, 0);
+       dlm_clear_members(ls);
+       kfree(ls->ls_node_array);
  out_stop:
        dlm_recoverd_stop(ls);
  out_delist:
@@ -588,6 +616,8 @@ int dlm_new_lockspace(char *name, int namelen, void **lockspace,
        error = new_lockspace(name, namelen, lockspace, flags, lvblen);
        if (!error)
                ls_count++;
+       else if (!ls_count)
+               threads_stop();
  out:
        mutex_unlock(&ls_lock);
        return error;
index f08faec3d854a70a07a4a82e4efcabf9cbc57f20..073599dced2ab4b2562f8f7df44db66aeaa079cc 100644 (file)
@@ -233,6 +233,12 @@ int dlm_recover_members(struct dlm_ls *ls, struct dlm_recover *rv, int *neg_out)
        *neg_out = neg;
 
        error = ping_members(ls);
+       if (!error || error == -EPROTO) {
+               /* new_lockspace() may be waiting to know if the config
+                  is good or bad */
+               ls->ls_members_result = error;
+               complete(&ls->ls_members_done);
+       }
        if (error)
                goto out;
 
index 6bfbd61538094f72cb1001bc782cc246faf55552..f71c23542f0fe70c105154534ccb21e5cd71acc2 100644 (file)
@@ -90,7 +90,7 @@ static int check_config(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
                log_error(ls, "version mismatch: %x nodeid %d: %x",
                          DLM_HEADER_MAJOR | DLM_HEADER_MINOR, nodeid,
                          rc->rc_header.h_version);
-               return -EINVAL;
+               return -EPROTO;
        }
 
        if (rf->rf_lvblen != ls->ls_lvblen ||
@@ -98,7 +98,7 @@ static int check_config(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
                log_error(ls, "config mismatch: %d,%x nodeid %d: %d,%x",
                          ls->ls_lvblen, ls->ls_exflags,
                          nodeid, rf->rf_lvblen, rf->rf_lsflags);
-               return -EINVAL;
+               return -EPROTO;
        }
        return 0;
 }