[dpdk-dev] [PATCH v1] net/tap: fix isolation mode toggling

[Wiles, Keith]

> On May 7, 2018, at 3:36 AM, Ophir Munk <ophirmu at mellanox.com> wrote:
> 
> Running testpmd command "flow isolae <port> 0" (i.e. disabling flow
> isolation) followed by command "flow isolate <port> 1" (i.e. enabling
> flow isolation) may result in a TAP error:
> PMD: Kernel refused TC filter rule creation (17): File exists
> 
> Root cause analysis: when disabling flow isolation we keep the local
> rule to redirect packets on TX (TAP_REMOTE_TX index) while we add it
> again when enabling flow isolation. As a result this rule is added
> two times in a raw which results in "File exists" error.

/raw/row
> The fix is to identify the "File exists" error and silently ignore it.
> 
> Another issue occurs when enabling isolation mode several times in a
> raw in which case the same tc rules are added consecutively and

/raw/row
> rte_flow structs are added to a linked list before removing the
> previous rte_flow structs.
> The fix is to act upon isolation mode command only when there is a
> change from "0" to "1" (or vice versa).
> 
> Fixes: f503d2694825 ("net/tap: support flow API isolated mode")
> Cc: stable at dpdk.org
> 
> Signed-off-by: Ophir Munk <ophirmu at mellanox.com>
> ---
> drivers/net/tap/tap_flow.c | 37 +++++++++++++++++++++++--------------
> 1 file changed, 23 insertions(+), 14 deletions(-)
> 
> diff --git a/drivers/net/tap/tap_flow.c b/drivers/net/tap/tap_flow.c
> index aab9eef..91f15f6 100644
> --- a/drivers/net/tap/tap_flow.c
> +++ b/drivers/net/tap/tap_flow.c
> @@ -1568,10 +1568,10 @@ tap_flow_isolate(struct rte_eth_dev *dev,
> {
> 	struct pmd_internals *pmd = dev->data->dev_private;
> 
> -	if (set)
> -		pmd->flow_isolate = 1;
> -	else
> -		pmd->flow_isolate = 0;
> +	/* if already in the right isolation mode - nothing to do */
> +	if ((!!set ^ pmd->flow_isolate) == 0)
> +		return 0;
> +	pmd->flow_isolate = !!set;

Using double negation is not very readable IMO, I would prefer this converted to a true boolean type if required.

variable ‘set' here should a 0 or 1 already, please expand this code to not use !!, using this in modern compilers should not be required. I understand this maybe shorted to write, but not very readable IMO and we need to make DPDK readable.

> 	/*
> 	 * If netdevice is there, setup appropriate flow rules immediately.
> 	 * Otherwise it will be set when bringing up the netdevice (tun_alloc).
> @@ -1579,21 +1579,30 @@ tap_flow_isolate(struct rte_eth_dev *dev,
> 	if (!pmd->rxq[0].fd)
> 		return 0;
> 	if (set) {
> -		struct rte_flow *flow;
> +		struct rte_flow *remote_flow;
> 
> -		while (1) {
> -			flow = LIST_FIRST(&pmd->implicit_flows);
> -			if (!flow)
> +		while (!LIST_EMPTY(&pmd->implicit_flows)) {
> +			remote_flow = LIST_FIRST(&pmd->implicit_flows);
> +			if (!remote_flow)
> 				break;
> 			/*
> 			 * Remove all implicit rules on the remote.
> 			 * Keep the local rule to redirect packets on TX.
> 			 * Keep also the last implicit local rule: ISOLATE.
> 			 */
> -			if (flow->msg.t.tcm_ifindex == pmd->if_index)
> -				break;
> -			if (tap_flow_destroy_pmd(pmd, flow, NULL) < 0)
> -				goto error;
> +			if (remote_flow->msg.t.tcm_ifindex != pmd->if_index) {
> +				/*
> +				 * remove TC from kernel and
> +				 * remote_flow from list
> +				 */
> +				if (tap_flow_destroy_pmd(pmd, remote_flow,
> +						NULL) < 0)
> +					goto error;
> +			} else {
> +				/* remove remote_flow from list */
> +				LIST_REMOVE(remote_flow, next);
> +				rte_free(remote_flow);
> +			}
> 		}
> 		/* Switch the TC rule according to pmd->flow_isolate */
> 		if (tap_flow_implicit_create(pmd, TAP_ISOLATE) == -1)
> @@ -1739,8 +1748,8 @@ int tap_flow_implicit_create(struct pmd_internals *pmd,
> 	}
> 	err = tap_nl_recv_ack(pmd->nlsk_fd);
> 	if (err < 0) {
> -		/* Silently ignore re-entering remote promiscuous rule */
> -		if (errno == EEXIST && idx == TAP_REMOTE_PROMISC)
> +		/* Silently ignore re-entering existing rule */
> +		if (errno == EEXIST)
> 			goto success;
> 		TAP_LOG(ERR,
> 			"Kernel refused TC filter rule creation (%d): %s",
> -- 
> 2.7.4
> 

Regards,
Keith