md/raid4: permit raid0 takeover
[safe/jmp/linux-2.6] / drivers / md / dm-mpath.c
index c70604a..826bce7 100644 (file)
@@ -64,10 +64,12 @@ struct multipath {
        spinlock_t lock;
 
        const char *hw_handler_name;
+       char *hw_handler_params;
        unsigned nr_priority_groups;
        struct list_head priority_groups;
        unsigned pg_init_required;      /* pg_init needs calling? */
        unsigned pg_init_in_progress;   /* Only one pg_init allowed at once */
+       wait_queue_head_t pg_init_wait; /* Wait for pg_init completion */
 
        unsigned nr_valid_paths;        /* Total number of usable paths */
        struct pgpath *current_pgpath;
@@ -92,6 +94,8 @@ struct multipath {
         * can resubmit bios on error.
         */
        mempool_t *mpio_pool;
+
+       struct mutex work_mutex;
 };
 
 /*
@@ -197,6 +201,8 @@ static struct multipath *alloc_multipath(struct dm_target *ti)
                m->queue_io = 1;
                INIT_WORK(&m->process_queued_ios, process_queued_ios);
                INIT_WORK(&m->trigger_event, trigger_event);
+               init_waitqueue_head(&m->pg_init_wait);
+               mutex_init(&m->work_mutex);
                m->mpio_pool = mempool_create_slab_pool(MIN_IOS, _mpio_cache);
                if (!m->mpio_pool) {
                        kfree(m);
@@ -219,6 +225,7 @@ static void free_multipath(struct multipath *m)
        }
 
        kfree(m->hw_handler_name);
+       kfree(m->hw_handler_params);
        mempool_destroy(m->mpio_pool);
        kfree(m);
 }
@@ -228,6 +235,21 @@ static void free_multipath(struct multipath *m)
  * Path selection
  *-----------------------------------------------*/
 
+static void __pg_init_all_paths(struct multipath *m)
+{
+       struct pgpath *pgpath;
+
+       m->pg_init_count++;
+       m->pg_init_required = 0;
+       list_for_each_entry(pgpath, &m->current_pg->pgpaths, list) {
+               /* Skip failed paths */
+               if (!pgpath->is_active)
+                       continue;
+               if (queue_work(kmpath_handlerd, &pgpath->activate_path))
+                       m->pg_init_in_progress++;
+       }
+}
+
 static void __switch_pg(struct multipath *m, struct pgpath *pgpath)
 {
        m->current_pg = pgpath->pg;
@@ -432,7 +454,7 @@ static void process_queued_ios(struct work_struct *work)
 {
        struct multipath *m =
                container_of(work, struct multipath, process_queued_ios);
-       struct pgpath *pgpath = NULL, *tmp;
+       struct pgpath *pgpath = NULL;
        unsigned must_queue = 1;
        unsigned long flags;
 
@@ -450,14 +472,9 @@ static void process_queued_ios(struct work_struct *work)
            (!pgpath && !m->queue_if_no_path))
                must_queue = 0;
 
-       if (m->pg_init_required && !m->pg_init_in_progress && pgpath) {
-               m->pg_init_count++;
-               m->pg_init_required = 0;
-               list_for_each_entry(tmp, &pgpath->pg->pgpaths, list) {
-                       if (queue_work(kmpath_handlerd, &tmp->activate_path))
-                               m->pg_init_in_progress++;
-               }
-       }
+       if (m->pg_init_required && !m->pg_init_in_progress && pgpath)
+               __pg_init_all_paths(m);
+
 out:
        spin_unlock_irqrestore(&m->lock, flags);
        if (!must_queue)
@@ -590,8 +607,8 @@ static struct pgpath *parse_path(struct arg_set *as, struct path_selector *ps,
        if (!p)
                return ERR_PTR(-ENOMEM);
 
-       r = dm_get_device(ti, shift(as), ti->begin, ti->len,
-                         dm_table_get_mode(ti->table), &p->path.dev);
+       r = dm_get_device(ti, shift(as), dm_table_get_mode(ti->table),
+                         &p->path.dev);
        if (r) {
                ti->error = "error getting device";
                goto bad;
@@ -615,6 +632,17 @@ static struct pgpath *parse_path(struct arg_set *as, struct path_selector *ps,
                        dm_put_device(ti, p->path.dev);
                        goto bad;
                }
+
+               if (m->hw_handler_params) {
+                       r = scsi_dh_set_params(q, m->hw_handler_params);
+                       if (r < 0) {
+                               ti->error = "unable to set hardware "
+                                                       "handler parameters";
+                               scsi_dh_detach(q);
+                               dm_put_device(ti, p->path.dev);
+                               goto bad;
+                       }
+               }
        }
 
        r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error);
@@ -705,6 +733,7 @@ static struct priority_group *parse_priority_group(struct arg_set *as,
 static int parse_hw_handler(struct arg_set *as, struct multipath *m)
 {
        unsigned hw_argc;
+       int ret;
        struct dm_target *ti = m->ti;
 
        static struct param _params[] = {
@@ -726,17 +755,33 @@ static int parse_hw_handler(struct arg_set *as, struct multipath *m)
        request_module("scsi_dh_%s", m->hw_handler_name);
        if (scsi_dh_handler_exist(m->hw_handler_name) == 0) {
                ti->error = "unknown hardware handler type";
-               kfree(m->hw_handler_name);
-               m->hw_handler_name = NULL;
-               return -EINVAL;
+               ret = -EINVAL;
+               goto fail;
        }
 
-       if (hw_argc > 1)
-               DMWARN("Ignoring user-specified arguments for "
-                      "hardware handler \"%s\"", m->hw_handler_name);
+       if (hw_argc > 1) {
+               char *p;
+               int i, j, len = 4;
+
+               for (i = 0; i <= hw_argc - 2; i++)
+                       len += strlen(as->argv[i]) + 1;
+               p = m->hw_handler_params = kzalloc(len, GFP_KERNEL);
+               if (!p) {
+                       ti->error = "memory allocation failed";
+                       ret = -ENOMEM;
+                       goto fail;
+               }
+               j = sprintf(p, "%d", hw_argc - 1);
+               for (i = 0, p+=j+1; i <= hw_argc - 2; i++, p+=j+1)
+                       j = sprintf(p, "%s", as->argv[i]);
+       }
        consume(as, hw_argc - 1);
 
        return 0;
+fail:
+       kfree(m->hw_handler_name);
+       m->hw_handler_name = NULL;
+       return ret;
 }
 
 static int parse_features(struct arg_set *as, struct multipath *m)
@@ -855,13 +900,43 @@ static int multipath_ctr(struct dm_target *ti, unsigned int argc,
        return r;
 }
 
-static void multipath_dtr(struct dm_target *ti)
+static void multipath_wait_for_pg_init_completion(struct multipath *m)
 {
-       struct multipath *m = (struct multipath *) ti->private;
+       DECLARE_WAITQUEUE(wait, current);
+       unsigned long flags;
+
+       add_wait_queue(&m->pg_init_wait, &wait);
+
+       while (1) {
+               set_current_state(TASK_UNINTERRUPTIBLE);
+
+               spin_lock_irqsave(&m->lock, flags);
+               if (!m->pg_init_in_progress) {
+                       spin_unlock_irqrestore(&m->lock, flags);
+                       break;
+               }
+               spin_unlock_irqrestore(&m->lock, flags);
+
+               io_schedule();
+       }
+       set_current_state(TASK_RUNNING);
+
+       remove_wait_queue(&m->pg_init_wait, &wait);
+}
 
+static void flush_multipath_work(struct multipath *m)
+{
        flush_workqueue(kmpath_handlerd);
+       multipath_wait_for_pg_init_completion(m);
        flush_workqueue(kmultipathd);
        flush_scheduled_work();
+}
+
+static void multipath_dtr(struct dm_target *ti)
+{
+       struct multipath *m = ti->private;
+
+       flush_multipath_work(m);
        free_multipath(m);
 }
 
@@ -1086,9 +1161,9 @@ static int pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath)
        return limit_reached;
 }
 
-static void pg_init_done(struct dm_path *path, int errors)
+static void pg_init_done(void *data, int errors)
 {
-       struct pgpath *pgpath = path_to_pgpath(path);
+       struct pgpath *pgpath = data;
        struct priority_group *pg = pgpath->pg;
        struct multipath *m = pg->m;
        unsigned long flags;
@@ -1102,8 +1177,8 @@ static void pg_init_done(struct dm_path *path, int errors)
                        errors = 0;
                        break;
                }
-               DMERR("Cannot failover device because scsi_dh_%s was not "
-                     "loaded.", m->hw_handler_name);
+               DMERR("Could not failover the device: Handler scsi_dh_%s "
+                     "Error %d.", m->hw_handler_name, errors);
                /*
                 * Fail path for now, so we do not ping pong
                 */
@@ -1140,25 +1215,34 @@ static void pg_init_done(struct dm_path *path, int errors)
                        m->current_pgpath = NULL;
                        m->current_pg = NULL;
                }
-       } else if (!m->pg_init_required) {
-               m->queue_io = 0;
+       } else if (!m->pg_init_required)
                pg->bypassed = 0;
-       }
 
-       m->pg_init_in_progress--;
-       if (!m->pg_init_in_progress)
-               queue_work(kmultipathd, &m->process_queued_ios);
+       if (--m->pg_init_in_progress)
+               /* Activations of other paths are still on going */
+               goto out;
+
+       if (!m->pg_init_required)
+               m->queue_io = 0;
+
+       queue_work(kmultipathd, &m->process_queued_ios);
+
+       /*
+        * Wake up any thread waiting to suspend.
+        */
+       wake_up(&m->pg_init_wait);
+
+out:
        spin_unlock_irqrestore(&m->lock, flags);
 }
 
 static void activate_path(struct work_struct *work)
 {
-       int ret;
        struct pgpath *pgpath =
                container_of(work, struct pgpath, activate_path);
 
-       ret = scsi_dh_activate(bdev_get_queue(pgpath->path.dev->bdev));
-       pg_init_done(&pgpath->path, ret);
+       scsi_dh_activate(bdev_get_queue(pgpath->path.dev->bdev),
+                               pg_init_done, pgpath);
 }
 
 /*
@@ -1231,6 +1315,15 @@ static void multipath_presuspend(struct dm_target *ti)
        queue_if_no_path(m, 0, 1);
 }
 
+static void multipath_postsuspend(struct dm_target *ti)
+{
+       struct multipath *m = ti->private;
+
+       mutex_lock(&m->work_mutex);
+       flush_multipath_work(m);
+       mutex_unlock(&m->work_mutex);
+}
+
 /*
  * Restore the queue_if_no_path setting.
  */
@@ -1367,51 +1460,65 @@ static int multipath_status(struct dm_target *ti, status_type_t type,
 
 static int multipath_message(struct dm_target *ti, unsigned argc, char **argv)
 {
-       int r;
+       int r = -EINVAL;
        struct dm_dev *dev;
        struct multipath *m = (struct multipath *) ti->private;
        action_fn action;
 
+       mutex_lock(&m->work_mutex);
+
+       if (dm_suspended(ti)) {
+               r = -EBUSY;
+               goto out;
+       }
+
        if (argc == 1) {
-               if (!strnicmp(argv[0], MESG_STR("queue_if_no_path")))
-                       return queue_if_no_path(m, 1, 0);
-               else if (!strnicmp(argv[0], MESG_STR("fail_if_no_path")))
-                       return queue_if_no_path(m, 0, 0);
+               if (!strnicmp(argv[0], MESG_STR("queue_if_no_path"))) {
+                       r = queue_if_no_path(m, 1, 0);
+                       goto out;
+               } else if (!strnicmp(argv[0], MESG_STR("fail_if_no_path"))) {
+                       r = queue_if_no_path(m, 0, 0);
+                       goto out;
+               }
        }
 
-       if (argc != 2)
-               goto error;
+       if (argc != 2) {
+               DMWARN("Unrecognised multipath message received.");
+               goto out;
+       }
 
-       if (!strnicmp(argv[0], MESG_STR("disable_group")))
-               return bypass_pg_num(m, argv[1], 1);
-       else if (!strnicmp(argv[0], MESG_STR("enable_group")))
-               return bypass_pg_num(m, argv[1], 0);
-       else if (!strnicmp(argv[0], MESG_STR("switch_group")))
-               return switch_pg_num(m, argv[1]);
-       else if (!strnicmp(argv[0], MESG_STR("reinstate_path")))
+       if (!strnicmp(argv[0], MESG_STR("disable_group"))) {
+               r = bypass_pg_num(m, argv[1], 1);
+               goto out;
+       } else if (!strnicmp(argv[0], MESG_STR("enable_group"))) {
+               r = bypass_pg_num(m, argv[1], 0);
+               goto out;
+       } else if (!strnicmp(argv[0], MESG_STR("switch_group"))) {
+               r = switch_pg_num(m, argv[1]);
+               goto out;
+       } else if (!strnicmp(argv[0], MESG_STR("reinstate_path")))
                action = reinstate_path;
        else if (!strnicmp(argv[0], MESG_STR("fail_path")))
                action = fail_path;
-       else
-               goto error;
+       else {
+               DMWARN("Unrecognised multipath message received.");
+               goto out;
+       }
 
-       r = dm_get_device(ti, argv[1], ti->begin, ti->len,
-                         dm_table_get_mode(ti->table), &dev);
+       r = dm_get_device(ti, argv[1], dm_table_get_mode(ti->table), &dev);
        if (r) {
                DMWARN("message: error getting device %s",
                       argv[1]);
-               return -EINVAL;
+               goto out;
        }
 
        r = action_dev(m, dev, action);
 
        dm_put_device(ti, dev);
 
+out:
+       mutex_unlock(&m->work_mutex);
        return r;
-
-error:
-       DMWARN("Unrecognised multipath message received.");
-       return -EINVAL;
 }
 
 static int multipath_ioctl(struct dm_target *ti, unsigned int cmd,
@@ -1453,7 +1560,7 @@ static int multipath_iterate_devices(struct dm_target *ti,
 
        list_for_each_entry(pg, &m->priority_groups, list) {
                list_for_each_entry(p, &pg->pgpaths, list) {
-                       ret = fn(ti, p->path.dev, ti->begin, data);
+                       ret = fn(ti, p->path.dev, ti->begin, ti->len, data);
                        if (ret)
                                goto out;
                }
@@ -1537,13 +1644,14 @@ out:
  *---------------------------------------------------------------*/
 static struct target_type multipath_target = {
        .name = "multipath",
-       .version = {1, 1, 0},
+       .version = {1, 1, 1},
        .module = THIS_MODULE,
        .ctr = multipath_ctr,
        .dtr = multipath_dtr,
        .map_rq = multipath_map,
        .rq_end_io = multipath_end_io,
        .presuspend = multipath_presuspend,
+       .postsuspend = multipath_postsuspend,
        .resume = multipath_resume,
        .status = multipath_status,
        .message = multipath_message,