Re: [dpdk-dev] [PATCH v3] ethdev: add sanity checks in control APIs

[Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru>]

On 4/14/21 2:11 PM, Min Hu (Connor) wrote:
> This patch adds more sanity checks in control path APIs.
> 
> Fixes: 214ed1acd125 ("ethdev: add iterator to match devargs input")
> Fixes: 3d98f921fbe9 ("ethdev: unify prefix for static functions and variables")
> Fixes: 0366137722a0 ("ethdev: check for invalid device name")
> Fixes: d948f596fee2 ("ethdev: fix port data mismatched in multiple process model")
> Fixes: 5b7ba31148a8 ("ethdev: add port ownership")
> Fixes: f8244c6399d9 ("ethdev: increase port id range")
> Cc: stable@dpdk.org

Please, see below. Error logging is missing in few cases and
I'd like to understand why.

> Signed-off-by: Min Hu (Connor) <humin29@huawei.com>
> ---
> v3:
> * set port_id checked first.
> * add error logging.
> 
> v2:
> * Removed unnecessary checks.
> * Deleted checks in internal API.
> * Added documentation in the header file.
> ---
>  lib/librte_ethdev/rte_ethdev.c | 274 ++++++++++++++++++++++++++++++++++++++---
>  lib/librte_ethdev/rte_ethdev.h |  20 ++-
>  2 files changed, 271 insertions(+), 23 deletions(-)
> 
> diff --git a/lib/librte_ethdev/rte_ethdev.c b/lib/librte_ethdev/rte_ethdev.c
> index 6b5cfd6..dfebcc9 100644
> --- a/lib/librte_ethdev/rte_ethdev.c
> +++ b/lib/librte_ethdev/rte_ethdev.c
> @@ -199,6 +199,9 @@ rte_eth_iterator_init(struct rte_dev_iterator *iter, const char *devargs_str)
>  	char *cls_str = NULL;
>  	int str_size;
>  
> +	if (iter == NULL || devargs_str == NULL)
> +		return -EINVAL;
> +

Is error logging skipped here intentially? Why?

>  	memset(iter, 0, sizeof(*iter));
>  
>  	/*
> @@ -293,7 +296,7 @@ rte_eth_iterator_init(struct rte_dev_iterator *iter, const char *devargs_str)
>  uint16_t
>  rte_eth_iterator_next(struct rte_dev_iterator *iter)
>  {
> -	if (iter->cls == NULL) /* invalid ethdev iterator */
> +	if (iter == NULL || iter->cls == NULL) /* invalid ethdev iterator */
>  		return RTE_MAX_ETHPORTS;

Is error logging skipped here intentially? Why?

>  
>  	do { /* loop to try all matching rte_device */
> @@ -322,7 +325,7 @@ rte_eth_iterator_next(struct rte_dev_iterator *iter)
>  void
>  rte_eth_iterator_cleanup(struct rte_dev_iterator *iter)
>  {
> -	if (iter->bus_str == NULL)
> +	if (iter == NULL || iter->bus_str == NULL)
>  		return; /* nothing to free in pure class filter */

Is error logging skipped here intentially? Why?

>  	free(RTE_CAST_FIELD(iter, bus_str, char *)); /* workaround const */
>  	free(RTE_CAST_FIELD(iter, cls_str, char *)); /* workaround const */

[snip]

> @@ -2491,6 +2536,12 @@ rte_eth_tx_done_cleanup(uint16_t port_id, uint16_t queue_id, uint32_t free_cnt)
>  	RTE_ETH_VALID_PORTID_OR_ERR_RET(port_id, -ENODEV);
>  	RTE_FUNC_PTR_OR_ERR_RET(*dev->dev_ops->tx_done_cleanup, -ENOTSUP);
>  
> +	if (queue_id >= dev->data->nb_tx_queues) {
> +		RTE_ETHDEV_LOG(ERR, "Queue id should be < %u.",
> +			       dev->data->nb_tx_queues);
> +		return -EINVAL;
> +	}
> +

Again, it is not always a control path. So, I'm not sure that we should
add the check there.

>  	/* Call driver to free pending mbufs. */
>  	ret = (*dev->dev_ops->tx_done_cleanup)(dev->data->tx_queues[queue_id],
>  					       free_cnt);

[snip]

> @@ -2667,6 +2732,9 @@ rte_eth_link_speed_to_str(uint32_t link_speed)
>  int
>  rte_eth_link_to_str(char *str, size_t len, const struct rte_eth_link *eth_link)
>  {
> +	if (str == NULL || eth_link == NULL)
> +		return -EINVAL;
> +

Is error logging skipped here intentionally? Why?

>  	if (eth_link->link_status == ETH_LINK_DOWN)
>  		return snprintf(str, len, "Link down");
>  	else

[snip]

> @@ -4602,6 +4784,9 @@ rte_eth_dma_zone_free(const struct rte_eth_dev *dev, const char *ring_name,
>  	const struct rte_memzone *mz;
>  	int rc = 0;
>  
> +	if (dev == NULL || ring_name == NULL)
> +		return -EINVAL;
> +

Same question about logging here.

>  	rc = eth_dev_dma_mzone_name(z_name, sizeof(z_name), dev->data->port_id,
>  			queue_id, ring_name);
>  	if (rc >= RTE_MEMZONE_NAMESIZE) {

[snip]

> @@ -5629,6 +5861,8 @@ rte_eth_representor_id_get(const struct rte_eth_dev *ethdev,
>  	struct rte_eth_representor_info *info = NULL;
>  	size_t size;
>  
> +	if (ethdev == NULL)
> +		return -EINVAL;

Question about logging here as well.

>  	if (type == RTE_ETH_REPRESENTOR_NONE)
>  		return 0;
>  	if (repr_id == NULL)

[snip]