vlocation.c

linux 内核源代码

	spin_unlock(&afs_vlocation_updates_lock);
}

/*
 * lookup volume location
 * - iterate through the VL servers in a cell until one of them admits knowing
 *   about the volume in question
 * - lookup in the local cache if not able to find on the VL server
 * - insert/update in the local cache if did get a VL response
 */
struct afs_vlocation *afs_vlocation_lookup(struct afs_cell *cell,
					   struct key *key,
					   const char *name,
					   size_t namesz)
{
	struct afs_vlocation *vl;
	int ret;

	_enter("{%s},{%x},%*.*s,%zu",
	       cell->name, key_serial(key),
	       (int) namesz, (int) namesz, name, namesz);

	if (namesz >= sizeof(vl->vldb.name)) {
		_leave(" = -ENAMETOOLONG");
		return ERR_PTR(-ENAMETOOLONG);
	}

	/* see if we have an in-memory copy first */
	down_write(&cell->vl_sem);
	spin_lock(&cell->vl_lock);
	list_for_each_entry(vl, &cell->vl_list, link) {
		if (vl->vldb.name[namesz] != '\0')
			continue;
		if (memcmp(vl->vldb.name, name, namesz) == 0)
			goto found_in_memory;
	}
	spin_unlock(&cell->vl_lock);

	/* not in the cell's in-memory lists - create a new record */
	vl = afs_vlocation_alloc(cell, name, namesz);
	if (!vl) {
		up_write(&cell->vl_sem);
		return ERR_PTR(-ENOMEM);
	}

	afs_get_cell(cell);

	list_add_tail(&vl->link, &cell->vl_list);
	vl->state = AFS_VL_CREATING;
	up_write(&cell->vl_sem);

fill_in_record:
	ret = afs_vlocation_fill_in_record(vl, key);
	if (ret < 0)
		goto error_abandon;
	spin_lock(&vl->lock);
	vl->state = AFS_VL_VALID;
	spin_unlock(&vl->lock);
	wake_up(&vl->waitq);

	/* schedule for regular updates */
	afs_vlocation_queue_for_updates(vl);
	goto success;

found_in_memory:
	/* found in memory */
	_debug("found in memory");
	atomic_inc(&vl->usage);
	spin_unlock(&cell->vl_lock);
	if (!list_empty(&vl->grave)) {
		spin_lock(&afs_vlocation_graveyard_lock);
		list_del_init(&vl->grave);
		spin_unlock(&afs_vlocation_graveyard_lock);
	}
	up_write(&cell->vl_sem);

	/* see if it was an abandoned record that we might try filling in */
	spin_lock(&vl->lock);
	while (vl->state != AFS_VL_VALID) {
		afs_vlocation_state_t state = vl->state;

		_debug("invalid [state %d]", state);

		if (state == AFS_VL_NEW || state == AFS_VL_NO_VOLUME) {
			vl->state = AFS_VL_CREATING;
			spin_unlock(&vl->lock);
			goto fill_in_record;
		}

		/* must now wait for creation or update by someone else to
		 * complete */
		_debug("wait");

		spin_unlock(&vl->lock);
		ret = wait_event_interruptible(vl->waitq,
					       vl->state == AFS_VL_NEW ||
					       vl->state == AFS_VL_VALID ||
					       vl->state == AFS_VL_NO_VOLUME);
		if (ret < 0)
			goto error;
		spin_lock(&vl->lock);
	}
	spin_unlock(&vl->lock);

success:
	_leave(" = %p",vl);
	return vl;

error_abandon:
	spin_lock(&vl->lock);
	vl->state = AFS_VL_NEW;
	spin_unlock(&vl->lock);
	wake_up(&vl->waitq);
error:
	ASSERT(vl != NULL);
	afs_put_vlocation(vl);
	_leave(" = %d", ret);
	return ERR_PTR(ret);
}

/*
 * finish using a volume location record
 */
void afs_put_vlocation(struct afs_vlocation *vl)
{
	if (!vl)
		return;

	_enter("%s", vl->vldb.name);

	ASSERTCMP(atomic_read(&vl->usage), >, 0);

	if (likely(!atomic_dec_and_test(&vl->usage))) {
		_leave("");
		return;
	}

	spin_lock(&afs_vlocation_graveyard_lock);
	if (atomic_read(&vl->usage) == 0) {
		_debug("buried");
		list_move_tail(&vl->grave, &afs_vlocation_graveyard);
		vl->time_of_death = get_seconds();
		schedule_delayed_work(&afs_vlocation_reap,
				      afs_vlocation_timeout * HZ);

		/* suspend updates on this record */
		if (!list_empty(&vl->update)) {
			spin_lock(&afs_vlocation_updates_lock);
			list_del_init(&vl->update);
			spin_unlock(&afs_vlocation_updates_lock);
		}
	}
	spin_unlock(&afs_vlocation_graveyard_lock);
	_leave(" [killed?]");
}

/*
 * destroy a dead volume location record
 */
static void afs_vlocation_destroy(struct afs_vlocation *vl)
{
	_enter("%p", vl);

#ifdef AFS_CACHING_SUPPORT
	cachefs_relinquish_cookie(vl->cache, 0);
#endif

	afs_put_cell(vl->cell);
	kfree(vl);
}

/*
 * reap dead volume location records
 */
static void afs_vlocation_reaper(struct work_struct *work)
{
	LIST_HEAD(corpses);
	struct afs_vlocation *vl;
	unsigned long delay, expiry;
	time_t now;

	_enter("");

	now = get_seconds();
	spin_lock(&afs_vlocation_graveyard_lock);

	while (!list_empty(&afs_vlocation_graveyard)) {
		vl = list_entry(afs_vlocation_graveyard.next,
				struct afs_vlocation, grave);

		_debug("check %p", vl);

		/* the queue is ordered most dead first */
		expiry = vl->time_of_death + afs_vlocation_timeout;
		if (expiry > now) {
			delay = (expiry - now) * HZ;
			_debug("delay %lu", delay);
			if (!schedule_delayed_work(&afs_vlocation_reap,
						   delay)) {
				cancel_delayed_work(&afs_vlocation_reap);
				schedule_delayed_work(&afs_vlocation_reap,
						      delay);
			}
			break;
		}

		spin_lock(&vl->cell->vl_lock);
		if (atomic_read(&vl->usage) > 0) {
			_debug("no reap");
			list_del_init(&vl->grave);
		} else {
			_debug("reap");
			list_move_tail(&vl->grave, &corpses);
			list_del_init(&vl->link);
		}
		spin_unlock(&vl->cell->vl_lock);
	}

	spin_unlock(&afs_vlocation_graveyard_lock);

	/* now reap the corpses we've extracted */
	while (!list_empty(&corpses)) {
		vl = list_entry(corpses.next, struct afs_vlocation, grave);
		list_del(&vl->grave);
		afs_vlocation_destroy(vl);
	}

	_leave("");
}

/*
 * initialise the VL update process
 */
int __init afs_vlocation_update_init(void)
{
	afs_vlocation_update_worker =
		create_singlethread_workqueue("kafs_vlupdated");
	return afs_vlocation_update_worker ? 0 : -ENOMEM;
}

/*
 * discard all the volume location records for rmmod
 */
void afs_vlocation_purge(void)
{
	afs_vlocation_timeout = 0;

	spin_lock(&afs_vlocation_updates_lock);
	list_del_init(&afs_vlocation_updates);
	spin_unlock(&afs_vlocation_updates_lock);
	cancel_delayed_work(&afs_vlocation_update);
	queue_delayed_work(afs_vlocation_update_worker,
			   &afs_vlocation_update, 0);
	destroy_workqueue(afs_vlocation_update_worker);

	cancel_delayed_work(&afs_vlocation_reap);
	schedule_delayed_work(&afs_vlocation_reap, 0);
}

/*
 * update a volume location
 */
static void afs_vlocation_updater(struct work_struct *work)
{
	struct afs_cache_vlocation vldb;
	struct afs_vlocation *vl, *xvl;
	time_t now;
	long timeout;
	int ret;

	_enter("");

	now = get_seconds();

	/* find a record to update */
	spin_lock(&afs_vlocation_updates_lock);
	for (;;) {
		if (list_empty(&afs_vlocation_updates)) {
			spin_unlock(&afs_vlocation_updates_lock);
			_leave(" [nothing]");
			return;
		}

		vl = list_entry(afs_vlocation_updates.next,
				struct afs_vlocation, update);
		if (atomic_read(&vl->usage) > 0)
			break;
		list_del_init(&vl->update);
	}

	timeout = vl->update_at - now;
	if (timeout > 0) {
		queue_delayed_work(afs_vlocation_update_worker,
				   &afs_vlocation_update, timeout * HZ);
		spin_unlock(&afs_vlocation_updates_lock);
		_leave(" [nothing]");
		return;
	}

	list_del_init(&vl->update);
	atomic_inc(&vl->usage);
	spin_unlock(&afs_vlocation_updates_lock);

	/* we can now perform the update */
	_debug("update %s", vl->vldb.name);
	vl->state = AFS_VL_UPDATING;
	vl->upd_rej_cnt = 0;
	vl->upd_busy_cnt = 0;

	ret = afs_vlocation_update_record(vl, NULL, &vldb);
	spin_lock(&vl->lock);
	switch (ret) {
	case 0:
		afs_vlocation_apply_update(vl, &vldb);
		vl->state = AFS_VL_VALID;
		break;
	case -ENOMEDIUM:
		vl->state = AFS_VL_VOLUME_DELETED;
		break;
	default:
		vl->state = AFS_VL_UNCERTAIN;
		break;
	}
	spin_unlock(&vl->lock);
	wake_up(&vl->waitq);

	/* and then reschedule */
	_debug("reschedule");
	vl->update_at = get_seconds() + afs_vlocation_update_timeout;

	spin_lock(&afs_vlocation_updates_lock);

	if (!list_empty(&afs_vlocation_updates)) {
		/* next update in 10 minutes, but wait at least 1 second more
		 * than the newest record already queued so that we don't spam
		 * the VL server suddenly with lots of requests
		 */
		xvl = list_entry(afs_vlocation_updates.prev,
				 struct afs_vlocation, update);
		if (vl->update_at <= xvl->update_at)
			vl->update_at = xvl->update_at + 1;
		xvl = list_entry(afs_vlocation_updates.next,
				 struct afs_vlocation, update);
		timeout = xvl->update_at - now;
		if (timeout < 0)
			timeout = 0;
	} else {
		timeout = afs_vlocation_update_timeout;
	}

	ASSERT(list_empty(&vl->update));

	list_add_tail(&vl->update, &afs_vlocation_updates);

	_debug("timeout %ld", timeout);
	queue_delayed_work(afs_vlocation_update_worker,
			   &afs_vlocation_update, timeout * HZ);
	spin_unlock(&afs_vlocation_updates_lock);
	afs_put_vlocation(vl);
}