RDMA/ucma: Put a lock around every call to the rdma_cm layer

The rdma_cm must be used single threaded.

This appears to be a bug in the design, as it does have lots of locking
that seems like it should allow concurrency. However, when it is all said
and done every single place that uses the cma_exch() scheme is broken, and
all the unlocked reads from the ucma of the cm_id data are wrong too.

syzkaller has been finding endless bugs related to this.

Fixing this in any elegant way is some enormous amount of work. Take a
very big hammer and put a mutex around everything to do with the
ucma_context at the top of every syscall.

Fixes: 75216638572f ("RDMA/cma: Export rdma cm interface to userspace")
Link: https://lore.kernel.org/r/[e-mail]
Reported-by: [e-mail]
Reported-by: [e-mail]
Reported-by: [e-mail]
Reported-by: [e-mail]
Reported-by: [e-mail]
Reported-by: [e-mail]
Reported-by: [e-mail]
Reported-by: [e-mail]
Reported-by: [e-mail]
Reported-by: [e-mail]
Reported-by: [e-mail]
Signed-off-by: Jason Gunthorpe <[e-mail]>

drivers/infiniband/core/ucma.c

···
 
 	struct ucma_file	*file;
 	struct rdma_cm_id	*cm_id;
+	struct mutex		mutex;
 	u64			uid;
 
 	struct list_head	list;
···
 	init_completion(&ctx->comp);
 	INIT_LIST_HEAD(&ctx->mc_list);
 	ctx->file = file;
+	mutex_init(&ctx->mutex);
 
 	if (xa_alloc(&ctx_table, &ctx->id, ctx, xa_limit_32b, GFP_KERNEL))
 		goto error;
···
 	}
 
 	events_reported = ctx->events_reported;
+	mutex_destroy(&ctx->mutex);
 	kfree(ctx);
 	return events_reported;
 }
···
 	if (IS_ERR(ctx))
 		return PTR_ERR(ctx);
 
+	mutex_lock(&ctx->mutex);
 	ret = rdma_bind_addr(ctx->cm_id, (struct sockaddr *) &cmd.addr);
+	mutex_unlock(&ctx->mutex);
+
 	ucma_put_ctx(ctx);
 	return ret;
 }
···
 	if (IS_ERR(ctx))
 		return PTR_ERR(ctx);
 
+	mutex_lock(&ctx->mutex);
 	ret = rdma_bind_addr(ctx->cm_id, (struct sockaddr *) &cmd.addr);
+	mutex_unlock(&ctx->mutex);
 	ucma_put_ctx(ctx);
 	return ret;
 }
···
 	if (IS_ERR(ctx))
 		return PTR_ERR(ctx);
 
+	mutex_lock(&ctx->mutex);
 	ret = rdma_resolve_addr(ctx->cm_id, (struct sockaddr *) &cmd.src_addr,
 				(struct sockaddr *) &cmd.dst_addr, cmd.timeout_ms);
+	mutex_unlock(&ctx->mutex);
 	ucma_put_ctx(ctx);
 	return ret;
 }
···
 	if (IS_ERR(ctx))
 		return PTR_ERR(ctx);
 
+	mutex_lock(&ctx->mutex);
 	ret = rdma_resolve_addr(ctx->cm_id, (struct sockaddr *) &cmd.src_addr,
 				(struct sockaddr *) &cmd.dst_addr, cmd.timeout_ms);
+	mutex_unlock(&ctx->mutex);
 	ucma_put_ctx(ctx);
 	return ret;
 }
···
 	if (IS_ERR(ctx))
 		return PTR_ERR(ctx);
 
+	mutex_lock(&ctx->mutex);
 	ret = rdma_resolve_route(ctx->cm_id, cmd.timeout_ms);
+	mutex_unlock(&ctx->mutex);
 	ucma_put_ctx(ctx);
 	return ret;
 }
···
 	if (IS_ERR(ctx))
 		return PTR_ERR(ctx);
 
+	mutex_lock(&ctx->mutex);
 	memset(&resp, 0, sizeof resp);
 	addr = (struct sockaddr *) &ctx->cm_id->route.addr.src_addr;
 	memcpy(&resp.src_addr, addr, addr->sa_family == AF_INET ?
···
 		ucma_copy_iw_route(&resp, &ctx->cm_id->route);
 
 out:
+	mutex_unlock(&ctx->mutex);
 	if (copy_to_user(u64_to_user_ptr(cmd.response),
 			 &resp, sizeof(resp)))
 		ret = -EFAULT;
···
 	if (IS_ERR(ctx))
 		return PTR_ERR(ctx);
 
+	mutex_lock(&ctx->mutex);
 	switch (cmd.option) {
 	case RDMA_USER_CM_QUERY_ADDR:
 		ret = ucma_query_addr(ctx, response, out_len);
···
 		ret = -ENOSYS;
 		break;
 	}
+	mutex_unlock(&ctx->mutex);
 
 	ucma_put_ctx(ctx);
 	return ret;
···
 		return PTR_ERR(ctx);
 
 	ucma_copy_conn_param(ctx->cm_id, &conn_param, &cmd.conn_param);
+	mutex_lock(&ctx->mutex);
 	ret = rdma_connect(ctx->cm_id, &conn_param);
+	mutex_unlock(&ctx->mutex);
 	ucma_put_ctx(ctx);
 	return ret;
 }
···
 
 	ctx->backlog = cmd.backlog > 0 && cmd.backlog < max_backlog ?
 		       cmd.backlog : max_backlog;
+	mutex_lock(&ctx->mutex);
 	ret = rdma_listen(ctx->cm_id, ctx->backlog);
+	mutex_unlock(&ctx->mutex);
 	ucma_put_ctx(ctx);
 	return ret;
 }
···
 	if (cmd.conn_param.valid) {
 		ucma_copy_conn_param(ctx->cm_id, &conn_param, &cmd.conn_param);
 		mutex_lock(&file->mut);
+		mutex_lock(&ctx->mutex);
 		ret = __rdma_accept(ctx->cm_id, &conn_param, NULL);
+		mutex_unlock(&ctx->mutex);
 		if (!ret)
 			ctx->uid = cmd.uid;
 		mutex_unlock(&file->mut);
-	} else
+	} else {
+		mutex_lock(&ctx->mutex);
 		ret = __rdma_accept(ctx->cm_id, NULL, NULL);
-
+		mutex_unlock(&ctx->mutex);
+	}
 	ucma_put_ctx(ctx);
 	return ret;
 }
···
 	if (IS_ERR(ctx))
 		return PTR_ERR(ctx);
 
+	mutex_lock(&ctx->mutex);
 	ret = rdma_reject(ctx->cm_id, cmd.private_data, cmd.private_data_len);
+	mutex_unlock(&ctx->mutex);
 	ucma_put_ctx(ctx);
 	return ret;
 }
···
 	if (IS_ERR(ctx))
 		return PTR_ERR(ctx);
 
+	mutex_lock(&ctx->mutex);
 	ret = rdma_disconnect(ctx->cm_id);
+	mutex_unlock(&ctx->mutex);
 	ucma_put_ctx(ctx);
 	return ret;
 }
···
 	resp.qp_attr_mask = 0;
 	memset(&qp_attr, 0, sizeof qp_attr);
 	qp_attr.qp_state = cmd.qp_state;
+	mutex_lock(&ctx->mutex);
 	ret = rdma_init_qp_attr(ctx->cm_id, &qp_attr, &resp.qp_attr_mask);
+	mutex_unlock(&ctx->mutex);
 	if (ret)
 		goto out;
 
···
 		struct sa_path_rec opa;
 
 		sa_convert_path_ib_to_opa(&opa, &sa_path);
+		mutex_lock(&ctx->mutex);
 		ret = rdma_set_ib_path(ctx->cm_id, &opa);
+		mutex_unlock(&ctx->mutex);
 	} else {
+		mutex_lock(&ctx->mutex);
 		ret = rdma_set_ib_path(ctx->cm_id, &sa_path);
+		mutex_unlock(&ctx->mutex);
 	}
 	if (ret)
 		return ret;
···
 
 	switch (level) {
 	case RDMA_OPTION_ID:
+		mutex_lock(&ctx->mutex);
 		ret = ucma_set_option_id(ctx, optname, optval, optlen);
+		mutex_unlock(&ctx->mutex);
 		break;
 	case RDMA_OPTION_IB:
 		ret = ucma_set_option_ib(ctx, optname, optval, optlen);
···
 	if (IS_ERR(ctx))
 		return PTR_ERR(ctx);
 
+	mutex_lock(&ctx->mutex);
 	if (ctx->cm_id->device)
 		ret = rdma_notify(ctx->cm_id, (enum ib_event_type)cmd.event);
+	mutex_unlock(&ctx->mutex);
 
 	ucma_put_ctx(ctx);
 	return ret;
···
 	mc->join_state = join_state;
 	mc->uid = cmd->uid;
 	memcpy(&mc->addr, addr, cmd->addr_size);
+	mutex_lock(&ctx->mutex);
 	ret = rdma_join_multicast(ctx->cm_id, (struct sockaddr *)&mc->addr,
 				  join_state, mc);
+	mutex_unlock(&ctx->mutex);
 	if (ret)
 		goto err2;
 
···
 		goto out;
 	}
 
+	mutex_lock(&mc->ctx->mutex);
 	rdma_leave_multicast(mc->ctx->cm_id, (struct sockaddr *) &mc->addr);
+	mutex_unlock(&mc->ctx->mutex);
+
 	mutex_lock(&mc->ctx->file->mut);
 	ucma_cleanup_mc_events(mc);
 	list_del(&mc->list);