Re: [PATCH v6 04/25] rtrs: core: lib functions shared between client and server modules

[Bart Van Assche <bvanassche@acm.org>]

On 2019-12-30 02:29, Jack Wang wrote:
> + * InfiniBand Transport Layer

Is RTRS an InfiniBand or an RDMA transport layer?

> +MODULE_DESCRIPTION("RTRS Core");

Please write out RTRS in full and consider changing the word "Core" into
"client and server".

> +	WARN_ON(!queue_size);
> +	ius = kcalloc(queue_size, sizeof(*ius), gfp_mask);
> +
> +	if (unlikely(!ius))
> +		return NULL;

No blank line between the 'ius' assignment and the 'ius' check please.

> +int rtrs_iu_post_recv(struct rtrs_con *con, struct rtrs_iu *iu)
> +{
> +	struct rtrs_sess *sess = con->sess;
> +	struct ib_recv_wr wr;
> +	const struct ib_recv_wr *bad_wr;
> +	struct ib_sge list;
> +
> +	list.addr   = iu->dma_addr;
> +	list.length = iu->size;
> +	list.lkey   = sess->dev->ib_pd->local_dma_lkey;
> +
> +	if (WARN_ON(list.length == 0)) {
> +		rtrs_wrn(con->sess,
> +			  "Posting receive work request failed, sg list is empty\n");
> +		return -EINVAL;
> +	}
> +
> +	wr.next    = NULL;
> +	wr.wr_cqe  = &iu->cqe;
> +	wr.sg_list = &list;
> +	wr.num_sge = 1;
> +
> +	return ib_post_recv(con->qp, &wr, &bad_wr);
> +}
> +EXPORT_SYMBOL_GPL(rtrs_iu_post_recv);

The above code is fragile: although this is unlikely, if a member would
be added in struct ib_sge or in struct ib_recv_wr then the above code
will leave some member variables uninitialized. Has it been considered
to initialize these structures using a single assignment statement, e.g.
as follows:

	wr = (struct ib_recv_wr) {
		.wr_cqe = ...,
		.sg_list = ...,
		.num_sge = 1,
	};

> +int rtrs_post_recv_empty(struct rtrs_con *con, struct ib_cqe *cqe)
> +{
> +	struct ib_recv_wr wr;
> +	const struct ib_recv_wr *bad_wr;
> +
> +	wr.next    = NULL;
> +	wr.wr_cqe  = cqe;
> +	wr.sg_list = NULL;
> +	wr.num_sge = 0;
> +
> +	return ib_post_recv(con->qp, &wr, &bad_wr);
> +}
> +EXPORT_SYMBOL_GPL(rtrs_post_recv_empty);

Same comment for this function.

> +int rtrs_post_recv_empty_x2(struct rtrs_con *con, struct ib_cqe *cqe)
> +{
> +	struct ib_recv_wr wr_arr[2], *wr;
> +	const struct ib_recv_wr *bad_wr;
> +	int i;
> +
> +	memset(wr_arr, 0, sizeof(wr_arr));
> +	for (i = 0; i < ARRAY_SIZE(wr_arr); i++) {
> +		wr = &wr_arr[i];
> +		wr->wr_cqe  = cqe;
> +		if (i)
> +			/* Chain backwards */
> +			wr->next = &wr_arr[i - 1];
> +	}
> +
> +	return ib_post_recv(con->qp, wr, &bad_wr);
> +}
> +EXPORT_SYMBOL_GPL(rtrs_post_recv_empty_x2);

I have not yet seen any other RDMA code that is similar to the above
function. A comment above this function that explains its purpose would
be more than welcome.

> +int rtrs_iu_post_send(struct rtrs_con *con, struct rtrs_iu *iu, size_t size,
> +		       struct ib_send_wr *head)
> +{
> +	struct rtrs_sess *sess = con->sess;
> +	struct ib_send_wr wr;
> +	const struct ib_send_wr *bad_wr;
> +	struct ib_sge list;
> +
> +	if ((WARN_ON(size == 0)))
> +		return -EINVAL;

No superfluous parentheses please.

> +	list.addr   = iu->dma_addr;
> +	list.length = size;
> +	list.lkey   = sess->dev->ib_pd->local_dma_lkey;
> +
> +	memset(&wr, 0, sizeof(wr));
> +	wr.next       = NULL;
> +	wr.wr_cqe     = &iu->cqe;
> +	wr.sg_list    = &list;
> +	wr.num_sge    = 1;
> +	wr.opcode     = IB_WR_SEND;
> +	wr.send_flags = IB_SEND_SIGNALED;

Has it been considered to use designated initializers instead of a
memset() followed by multiple assignments? Same question for
rtrs_iu_post_rdma_write_imm() and rtrs_post_rdma_write_imm_empty().

> +static int create_qp(struct rtrs_con *con, struct ib_pd *pd,
> +		     u16 wr_queue_size, u32 max_sge)
> +{
> +	struct ib_qp_init_attr init_attr = {NULL};
> +	struct rdma_cm_id *cm_id = con->cm_id;
> +	int ret;
> +
> +	init_attr.cap.max_send_wr = wr_queue_size;
> +	init_attr.cap.max_recv_wr = wr_queue_size;

What code is responsible for ensuring that neither max_send_wr nor
max_recv_wr exceeds the device limits? Please document this in a comment
above this function.

> +	init_attr.cap.max_recv_sge = 1;
> +	init_attr.event_handler = qp_event_handler;
> +	init_attr.qp_context = con;
> +#undef max_send_sge
> +	init_attr.cap.max_send_sge = max_sge;

Is the "undef max_send_sge" really necessary? If so, please add a
comment that explains why it is necessary.

> +static int rtrs_str_gid_to_sockaddr(const char *addr, size_t len,
> +				     short port, struct sockaddr_storage *dst)
> +{
> +	struct sockaddr_ib *dst_ib = (struct sockaddr_ib *)dst;
> +	int ret;
> +
> +	/*
> +	 * We can use some of the I6 functions since GID is a valid
> +	 * IPv6 address format
> +	 */
> +	ret = in6_pton(addr, len, dst_ib->sib_addr.sib_raw, '\0', NULL);
> +	if (ret == 0)
> +		return -EINVAL;

What is "I6"?

Is the fourth argument to this function correct? From the comment above
in6_pton(): "@delim: the delimiter of the IPv6 address in @src, -1 means
no delimiter".

> +int sockaddr_to_str(const struct sockaddr *addr, char *buf, size_t len)
> +{
> +	int cnt;
> +
> +	switch (addr->sa_family) {
> +	case AF_IB:
> +		cnt = scnprintf(buf, len, "gid:%pI6",
> +			&((struct sockaddr_ib *)addr)->sib_addr.sib_raw);
> +		return cnt;
> +	case AF_INET:
> +		cnt = scnprintf(buf, len, "ip:%pI4",
> +			&((struct sockaddr_in *)addr)->sin_addr);
> +		return cnt;
> +	case AF_INET6:
> +		cnt = scnprintf(buf, len, "ip:%pI6c",
> +			  &((struct sockaddr_in6 *)addr)->sin6_addr);
> +		return cnt;
> +	}
> +	cnt = scnprintf(buf, len, "<invalid address family>");
> +	pr_err("Invalid address family\n");
> +	return cnt;
> +}
> +EXPORT_SYMBOL(sockaddr_to_str);

Is the pr_err() statement in the above function useful? Will anyone be
able to figure out what is going on if the "Invalid address family"
string appears in the system log? Please consider changing that pr_err()
statement into a WARN_ON_ONCE() statement.

> +	ret = rtrs_str_to_sockaddr(str, len, port, addr->dst);
> +
> +	return ret;

Please change this into a single return statement.

> +EXPORT_SYMBOL(rtrs_addr_to_sockaddr);
> +
> +void rtrs_ib_dev_pool_init(enum ib_pd_flags pd_flags,
> +			    struct rtrs_ib_dev_pool *pool)
> +{
> +	WARN_ON(pool->ops && (!pool->ops->alloc ^ !pool->ops->free));
> +	INIT_LIST_HEAD(&pool->list);
> +	mutex_init(&pool->mutex);
> +	pool->pd_flags = pd_flags;
> +}
> +EXPORT_SYMBOL(rtrs_ib_dev_pool_init);
> +
> +void rtrs_ib_dev_pool_deinit(struct rtrs_ib_dev_pool *pool)
> +{
> +	WARN_ON(!list_empty(&pool->list));
> +}
> +EXPORT_SYMBOL(rtrs_ib_dev_pool_deinit);

Since rtrs_ib_dev_pool_init() calls mutex_init(), should
rtrs_ib_dev_pool_deinit() call mutex_destroy()?

Thanks,

Bart.