vote.c

linux 内核源代码

	spin_unlock(&osb->net_response_lock);
}

static int ocfs2_broadcast_vote(struct ocfs2_super *osb,
				struct ocfs2_vote_msg *request,
				unsigned int response_id,
				int *response,
				struct ocfs2_net_response_cb *callback)
{
	int status, i, remote_err;
	struct ocfs2_net_wait_ctxt *w = NULL;
	int dequeued = 0;

	mlog_entry_void();

	w = ocfs2_new_net_wait_ctxt(response_id);
	if (!w) {
		status = -ENOMEM;
		mlog_errno(status);
		goto bail;
	}
	w->n_callback = callback;

	/* we're pretty much ready to go at this point, and this fills
	 * in n_response which we need anyway... */
	ocfs2_queue_net_wait_ctxt(osb, w);

	i = ocfs2_node_map_iterate(osb, &osb->mounted_map, 0);

	while (i != O2NM_INVALID_NODE_NUM) {
		if (i != osb->node_num) {
			mlog(0, "trying to send request to node %i\n", i);
			ocfs2_node_map_set_bit(osb, &w->n_node_map, i);

			remote_err = 0;
			status = o2net_send_message(OCFS2_MESSAGE_TYPE_VOTE,
						    osb->net_key,
						    request,
						    sizeof(*request),
						    i,
						    &remote_err);
			if (status == -ETIMEDOUT) {
				mlog(0, "remote node %d timed out!\n", i);
				status = -EAGAIN;
				goto bail;
			}
			if (remote_err < 0) {
				status = remote_err;
				mlog(0, "remote error %d on node %d!\n",
				     remote_err, i);
				mlog_errno(status);
				goto bail;
			}
			if (status < 0) {
				mlog_errno(status);
				goto bail;
			}
		}
		i++;
		i = ocfs2_node_map_iterate(osb, &osb->mounted_map, i);
		mlog(0, "next is %d, i am %d\n", i, osb->node_num);
	}
	mlog(0, "done sending, now waiting on responses...\n");

	wait_event(w->n_event, ocfs2_node_map_is_empty(osb, &w->n_node_map));

	ocfs2_dequeue_net_wait_ctxt(osb, w);
	dequeued = 1;

	*response = w->n_response;
	status = 0;
bail:
	if (w) {
		if (!dequeued)
			ocfs2_dequeue_net_wait_ctxt(osb, w);
		kfree(w);
	}

	mlog_exit(status);
	return status;
}

static struct ocfs2_vote_msg * ocfs2_new_vote_request(struct ocfs2_super *osb,
						      u64 blkno,
						      unsigned int generation,
						      enum ocfs2_vote_request type)
{
	struct ocfs2_vote_msg *request;
	struct ocfs2_msg_hdr *hdr;

	BUG_ON(!ocfs2_is_valid_vote_request(type));

	request = kzalloc(sizeof(*request), GFP_NOFS);
	if (!request) {
		mlog_errno(-ENOMEM);
	} else {
		hdr = &request->v_hdr;
		hdr->h_node_num = cpu_to_be32(osb->node_num);
		hdr->h_request = cpu_to_be32(type);
		hdr->h_blkno = cpu_to_be64(blkno);
		hdr->h_generation = cpu_to_be32(generation);
	}

	return request;
}

/* Complete the buildup of a new vote request and process the
 * broadcast return value. */
static int ocfs2_do_request_vote(struct ocfs2_super *osb,
				 struct ocfs2_vote_msg *request,
				 struct ocfs2_net_response_cb *callback)
{
	int status, response = -EBUSY;
	unsigned int response_id;
	struct ocfs2_msg_hdr *hdr;

	response_id = ocfs2_new_response_id(osb);

	hdr = &request->v_hdr;
	hdr->h_response_id = cpu_to_be32(response_id);

	status = ocfs2_broadcast_vote(osb, request, response_id, &response,
				      callback);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	status = response;
bail:

	return status;
}

int ocfs2_request_mount_vote(struct ocfs2_super *osb)
{
	int status;
	struct ocfs2_vote_msg *request = NULL;

	request = ocfs2_new_vote_request(osb, 0ULL, 0, OCFS2_VOTE_REQ_MOUNT);
	if (!request) {
		status = -ENOMEM;
		goto bail;
	}

	status = -EAGAIN;
	while (status == -EAGAIN) {
		if (!(osb->s_mount_opt & OCFS2_MOUNT_NOINTR) &&
		    signal_pending(current)) {
			status = -ERESTARTSYS;
			goto bail;
		}

		if (ocfs2_node_map_is_only(osb, &osb->mounted_map,
					   osb->node_num)) {
			status = 0;
			goto bail;
		}

		status = ocfs2_do_request_vote(osb, request, NULL);
	}

bail:
	kfree(request);
	return status;
}

int ocfs2_request_umount_vote(struct ocfs2_super *osb)
{
	int status;
	struct ocfs2_vote_msg *request = NULL;

	request = ocfs2_new_vote_request(osb, 0ULL, 0, OCFS2_VOTE_REQ_UMOUNT);
	if (!request) {
		status = -ENOMEM;
		goto bail;
	}

	status = -EAGAIN;
	while (status == -EAGAIN) {
		/* Do not check signals on this vote... We really want
		 * this one to go all the way through. */

		if (ocfs2_node_map_is_only(osb, &osb->mounted_map,
					   osb->node_num)) {
			status = 0;
			goto bail;
		}

		status = ocfs2_do_request_vote(osb, request, NULL);
	}

bail:
	kfree(request);
	return status;
}

/* TODO: This should eventually be a hash table! */
static struct ocfs2_net_wait_ctxt * __ocfs2_find_net_wait_ctxt(struct ocfs2_super *osb,
							       u32 response_id)
{
	struct list_head *p;
	struct ocfs2_net_wait_ctxt *w = NULL;

	list_for_each(p, &osb->net_response_list) {
		w = list_entry(p, struct ocfs2_net_wait_ctxt, n_list);
		if (response_id == w->n_response_id)
			break;
		w = NULL;
	}

	return w;
}

/* Translate response codes into local node errno values */
static inline int ocfs2_translate_response(int response)
{
	int ret;

	switch (response) {
	case OCFS2_RESPONSE_OK:
		ret = 0;
		break;

	case OCFS2_RESPONSE_BUSY:
		ret = -EBUSY;
		break;

	default:
		ret = -EINVAL;
	}

	return ret;
}

static int ocfs2_handle_response_message(struct o2net_msg *msg,
					 u32 len,
					 void *data, void **ret_data)
{
	unsigned int response_id, node_num;
	int response_status;
	struct ocfs2_super *osb = data;
	struct ocfs2_response_msg *resp;
	struct ocfs2_net_wait_ctxt * w;
	struct ocfs2_net_response_cb *resp_cb;

	resp = (struct ocfs2_response_msg *) msg->buf;

	response_id = be32_to_cpu(resp->r_hdr.h_response_id);
	node_num = be32_to_cpu(resp->r_hdr.h_node_num);
	response_status = 
		ocfs2_translate_response(be32_to_cpu(resp->r_response));

	mlog(0, "received response message:\n");
	mlog(0, "h_response_id = %u\n", response_id);
	mlog(0, "h_request = %u\n", be32_to_cpu(resp->r_hdr.h_request));
	mlog(0, "h_blkno = %llu\n",
	     (unsigned long long)be64_to_cpu(resp->r_hdr.h_blkno));
	mlog(0, "h_generation = %u\n", be32_to_cpu(resp->r_hdr.h_generation));
	mlog(0, "h_node_num = %u\n", node_num);
	mlog(0, "r_response = %d\n", response_status);

	spin_lock(&osb->net_response_lock);
	w = __ocfs2_find_net_wait_ctxt(osb, response_id);
	if (!w) {
		mlog(0, "request not found!\n");
		goto bail;
	}
	resp_cb = w->n_callback;

	if (response_status && (!w->n_response)) {
		/* we only really need one negative response so don't
		 * set it twice. */
		w->n_response = response_status;
	}

	if (resp_cb) {
		spin_unlock(&osb->net_response_lock);

		resp_cb->rc_cb(resp_cb->rc_priv, resp);

		spin_lock(&osb->net_response_lock);
	}

	__ocfs2_mark_node_responded(osb, w, node_num);
bail:
	spin_unlock(&osb->net_response_lock);

	return 0;
}

static int ocfs2_handle_vote_message(struct o2net_msg *msg,
				     u32 len,
				     void *data, void **ret_data)
{
	int status;
	struct ocfs2_super *osb = data;
	struct ocfs2_vote_work *work;

	work = kmalloc(sizeof(struct ocfs2_vote_work), GFP_NOFS);
	if (!work) {
		status = -ENOMEM;
		mlog_errno(status);
		goto bail;
	}

	INIT_LIST_HEAD(&work->w_list);
	memcpy(&work->w_msg, msg->buf, sizeof(struct ocfs2_vote_msg));

	mlog(0, "scheduling vote request:\n");
	mlog(0, "h_response_id = %u\n",
	     be32_to_cpu(work->w_msg.v_hdr.h_response_id));
	mlog(0, "h_request = %u\n", be32_to_cpu(work->w_msg.v_hdr.h_request));
	mlog(0, "h_blkno = %llu\n",
	     (unsigned long long)be64_to_cpu(work->w_msg.v_hdr.h_blkno));
	mlog(0, "h_generation = %u\n",
	     be32_to_cpu(work->w_msg.v_hdr.h_generation));
	mlog(0, "h_node_num = %u\n",
	     be32_to_cpu(work->w_msg.v_hdr.h_node_num));

	spin_lock(&osb->vote_task_lock);
	list_add_tail(&work->w_list, &osb->vote_list);
	osb->vote_count++;
	spin_unlock(&osb->vote_task_lock);

	ocfs2_kick_vote_thread(osb);

	status = 0;
bail:
	return status;
}

void ocfs2_unregister_net_handlers(struct ocfs2_super *osb)
{
	if (!osb->net_key)
		return;

	o2net_unregister_handler_list(&osb->osb_net_handlers);

	if (!list_empty(&osb->net_response_list))
		mlog(ML_ERROR, "net response list not empty!\n");

	osb->net_key = 0;
}

int ocfs2_register_net_handlers(struct ocfs2_super *osb)
{
	int status = 0;

	if (ocfs2_mount_local(osb))
		return 0;

	status = o2net_register_handler(OCFS2_MESSAGE_TYPE_RESPONSE,
					osb->net_key,
					sizeof(struct ocfs2_response_msg),
					ocfs2_handle_response_message,
					osb, NULL, &osb->osb_net_handlers);
	if (status) {
		mlog_errno(status);
		goto bail;
	}

	status = o2net_register_handler(OCFS2_MESSAGE_TYPE_VOTE,
					osb->net_key,
					sizeof(struct ocfs2_vote_msg),
					ocfs2_handle_vote_message,
					osb, NULL, &osb->osb_net_handlers);
	if (status) {
		mlog_errno(status);
		goto bail;
	}
bail:
	if (status < 0)
		ocfs2_unregister_net_handlers(osb);

	return status;
}