[dpdk-dev] [PATCH v4 06/12] net/failsafe: add flexible device definition

[Stephen Hemminger]

On Mon, 29 May 2017 15:42:18 +0200
Gaetan Rivet <gaetan.rivet at 6wind.com> wrote:

>  
> +- **exec(<shell command>)** parameter
> +
> +  This parameter allows the user to provide a command to the fail-safe PMD to
> +  execute and define a sub-device.
> +  It is done within a regular shell context.
> +  The first line of its output is read by the fail-safe PMD and otherwise
> +  interpreted as if passed by the regular **dev** parameter.
> +  Any other line is discarded.
> +  If the command fail or output an incorrect string, the sub-device is not
> +  initialized.
> +  All commas within the ``shell command`` are replaced by spaces before
> +  executing the command. This helps using scripts to specify devices.
> +

Exec from a DPDK application seems like possible security hole since most DPDK applications
have to run as root.


> static int
> +fs_execute_cmd(struct sub_device *sdev, char *cmdline)
> +{
> +	FILE *fp;
> +	/* store possible newline as well */
> +	char output[DEVARGS_MAXLEN + 1];
> +	size_t len;
> +	int old_err;
> +	int ret;
> +
> +	assert(cmdline != NULL || sdev->cmdline != NULL);
> +	if (sdev->cmdline == NULL) {
> +		char *new_str;
> +		size_t i;
> +
> +		len = strlen(cmdline) + 1;
> +		new_str = rte_realloc(sdev->cmdline, len,
> +				RTE_CACHE_LINE_SIZE);
> +		if (new_str == NULL) {
> +			ERROR("Command line allocation failed");
> +			return -ENOMEM;
> +		}

Using rte_malloc for cmdline is way over optimizing. rte_malloc comes from huge page area
which is limited. The only reason to use it is if the memory needs to be shared by primary/slave.
Also rte_malloc has much less protection (memleak checkers, guards etc) compared to regular malloc.