[DLM] block scand during recovery [1/6]
authorDavid Teigland <teigland@redhat.com>
Fri, 18 May 2007 13:58:15 +0000 (08:58 -0500)
committerSteven Whitehouse <swhiteho@redhat.com>
Mon, 9 Jul 2007 07:22:31 +0000 (08:22 +0100)
Don't let dlm_scand run during recovery since it may try to do a resource
directory removal while the directory nodes are changing.

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

index d8d6e729f96b669b5a6ed16bfb92c776cfc4744c..09668ec2e27988a246af20d11fd05374f2969988 100644 (file)
@@ -194,17 +194,17 @@ void dlm_dump_rsb(struct dlm_rsb *r)
 
 /* Threads cannot use the lockspace while it's being recovered */
 
-static inline void lock_recovery(struct dlm_ls *ls)
+static inline void dlm_lock_recovery(struct dlm_ls *ls)
 {
        down_read(&ls->ls_in_recovery);
 }
 
-static inline void unlock_recovery(struct dlm_ls *ls)
+void dlm_unlock_recovery(struct dlm_ls *ls)
 {
        up_read(&ls->ls_in_recovery);
 }
 
-static inline int lock_recovery_try(struct dlm_ls *ls)
+int dlm_lock_recovery_try(struct dlm_ls *ls)
 {
        return down_read_trylock(&ls->ls_in_recovery);
 }
@@ -985,11 +985,10 @@ void dlm_scan_rsbs(struct dlm_ls *ls)
 {
        int i;
 
-       if (dlm_locking_stopped(ls))
-               return;
-
        for (i = 0; i < ls->ls_rsbtbl_size; i++) {
                shrink_bucket(ls, i);
+               if (dlm_locking_stopped(ls))
+                       break;
                cond_resched();
        }
 }
@@ -2274,7 +2273,7 @@ int dlm_lock(dlm_lockspace_t *lockspace,
        if (!ls)
                return -EINVAL;
 
-       lock_recovery(ls);
+       dlm_lock_recovery(ls);
 
        if (convert)
                error = find_lkb(ls, lksb->sb_lkid, &lkb);
@@ -2302,7 +2301,7 @@ int dlm_lock(dlm_lockspace_t *lockspace,
        if (error == -EAGAIN)
                error = 0;
  out:
-       unlock_recovery(ls);
+       dlm_unlock_recovery(ls);
        dlm_put_lockspace(ls);
        return error;
 }
@@ -2322,7 +2321,7 @@ int dlm_unlock(dlm_lockspace_t *lockspace,
        if (!ls)
                return -EINVAL;
 
-       lock_recovery(ls);
+       dlm_lock_recovery(ls);
 
        error = find_lkb(ls, lkid, &lkb);
        if (error)
@@ -2344,7 +2343,7 @@ int dlm_unlock(dlm_lockspace_t *lockspace,
  out_put:
        dlm_put_lkb(lkb);
  out:
-       unlock_recovery(ls);
+       dlm_unlock_recovery(ls);
        dlm_put_lockspace(ls);
        return error;
 }
@@ -3424,7 +3423,7 @@ int dlm_receive_message(struct dlm_header *hd, int nodeid, int recovery)
                        }
                }
 
-               if (lock_recovery_try(ls))
+               if (dlm_lock_recovery_try(ls))
                        break;
                schedule();
        }
@@ -3503,7 +3502,7 @@ int dlm_receive_message(struct dlm_header *hd, int nodeid, int recovery)
                log_error(ls, "unknown message type %d", ms->m_type);
        }
 
-       unlock_recovery(ls);
+       dlm_unlock_recovery(ls);
  out:
        dlm_put_lockspace(ls);
        dlm_astd_wake();
@@ -4040,7 +4039,7 @@ int dlm_user_request(struct dlm_ls *ls, struct dlm_user_args *ua,
        struct dlm_args args;
        int error;
 
-       lock_recovery(ls);
+       dlm_lock_recovery(ls);
 
        error = create_lkb(ls, &lkb);
        if (error) {
@@ -4094,7 +4093,7 @@ int dlm_user_request(struct dlm_ls *ls, struct dlm_user_args *ua,
        list_add_tail(&lkb->lkb_ownqueue, &ua->proc->locks);
        spin_unlock(&ua->proc->locks_spin);
  out:
-       unlock_recovery(ls);
+       dlm_unlock_recovery(ls);
        return error;
 }
 
@@ -4106,7 +4105,7 @@ int dlm_user_convert(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
        struct dlm_user_args *ua;
        int error;
 
-       lock_recovery(ls);
+       dlm_lock_recovery(ls);
 
        error = find_lkb(ls, lkid, &lkb);
        if (error)
@@ -4146,7 +4145,7 @@ int dlm_user_convert(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
  out_put:
        dlm_put_lkb(lkb);
  out:
-       unlock_recovery(ls);
+       dlm_unlock_recovery(ls);
        kfree(ua_tmp);
        return error;
 }
@@ -4159,7 +4158,7 @@ int dlm_user_unlock(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
        struct dlm_user_args *ua;
        int error;
 
-       lock_recovery(ls);
+       dlm_lock_recovery(ls);
 
        error = find_lkb(ls, lkid, &lkb);
        if (error)
@@ -4194,7 +4193,7 @@ int dlm_user_unlock(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
  out_put:
        dlm_put_lkb(lkb);
  out:
-       unlock_recovery(ls);
+       dlm_unlock_recovery(ls);
        kfree(ua_tmp);
        return error;
 }
@@ -4207,7 +4206,7 @@ int dlm_user_cancel(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
        struct dlm_user_args *ua;
        int error;
 
-       lock_recovery(ls);
+       dlm_lock_recovery(ls);
 
        error = find_lkb(ls, lkid, &lkb);
        if (error)
@@ -4231,7 +4230,7 @@ int dlm_user_cancel(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
  out_put:
        dlm_put_lkb(lkb);
  out:
-       unlock_recovery(ls);
+       dlm_unlock_recovery(ls);
        kfree(ua_tmp);
        return error;
 }
@@ -4314,7 +4313,7 @@ void dlm_clear_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
 {
        struct dlm_lkb *lkb, *safe;
 
-       lock_recovery(ls);
+       dlm_lock_recovery(ls);
 
        while (1) {
                lkb = del_proc_lock(ls, proc);
@@ -4347,7 +4346,7 @@ void dlm_clear_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
        }
 
        mutex_unlock(&ls->ls_clear_proc_locks);
-       unlock_recovery(ls);
+       dlm_unlock_recovery(ls);
 }
 
 static void purge_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
@@ -4429,12 +4428,12 @@ int dlm_user_purge(struct dlm_ls *ls, struct dlm_user_proc *proc,
        if (nodeid != dlm_our_nodeid()) {
                error = send_purge(ls, nodeid, pid);
        } else {
-               lock_recovery(ls);
+               dlm_lock_recovery(ls);
                if (pid == current->pid)
                        purge_proc_locks(ls, proc);
                else
                        do_purge(ls, nodeid, pid);
-               unlock_recovery(ls);
+               dlm_unlock_recovery(ls);
        }
        return error;
 }
index 64fc4ec406683b2b855a4e7b059f442c8c7e34c6..19403aa08739a72c74951f62d9eb37263d7f517d 100644 (file)
@@ -24,6 +24,8 @@ void dlm_put_rsb(struct dlm_rsb *r);
 void dlm_hold_rsb(struct dlm_rsb *r);
 int dlm_put_lkb(struct dlm_lkb *lkb);
 void dlm_scan_rsbs(struct dlm_ls *ls);
+int dlm_lock_recovery_try(struct dlm_ls *ls);
+void dlm_unlock_recovery(struct dlm_ls *ls);
 
 int dlm_purge_locks(struct dlm_ls *ls);
 void dlm_purge_mstcpy_locks(struct dlm_rsb *r);
index a677b2a5eed4dff8cc40be9c6f41c298565fa3bf..414a108df934870c63b50abc2be273bb332e917f 100644 (file)
@@ -234,8 +234,12 @@ static int dlm_scand(void *data)
        struct dlm_ls *ls;
 
        while (!kthread_should_stop()) {
-               list_for_each_entry(ls, &lslist, ls_list)
-                       dlm_scan_rsbs(ls);
+               list_for_each_entry(ls, &lslist, ls_list) {
+                       if (dlm_lock_recovery_try(ls)) {
+                               dlm_scan_rsbs(ls);
+                               dlm_unlock_recovery(ls);
+                       }
+               }
                schedule_timeout_interruptible(dlm_config.ci_scan_secs * HZ);
        }
        return 0;