[dpdk-dev] [PATCH v3] net/tap: fix promiscuous rules double insertions

[Ophir Munk <ophirmu@mellanox.com>]

Running testpmd command "port stop all" followed by command "port start
all" may result in a TAP error:
PMD: Kernel refused TC filter rule creation (17): File exists

Root cause analysis: during the execution of "port start all" command
testpmd calls rte_eth_promiscuous_enable() while during the execution
of "port stop all" command testpmd does not call
rte_eth_promiscuous_disable().
As a result the TAP PMD is trying to add tc (traffic control command)
promiscuous rules to the remote netvsc device consecutively. From the
kernel point of view it is seen as an attempt to add the same rule more
than once. In recent kernels (e.g. version 4.13) this attempt is rejected
with a "File exists" error. In less recent kernels (e.g. version 4.4) the
same rule may have been successfully accepted twice, which is undesirable.

In the corrupted code every tc promiscuous rule included a different
handle number parameter. If instead an identical handle number is
used for all tc promiscuous rules - all kernels will reject the second
identical rule with a "File exists" error, which is easy to identify and
to silently ignore.

Fixes: 2bc06869cd94 ("net/tap: add remote netdevice traffic capture")
Cc: stable@dpdk.org

Signed-off-by: Ophir Munk <ophirmu@mellanox.com>
---
v1: initial version
v2: add detailed commit message
v3: textual fixes to commit message and code comments

 drivers/net/tap/tap_flow.c | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/drivers/net/tap/tap_flow.c b/drivers/net/tap/tap_flow.c
index 65657f0..551b2d8 100644
--- a/drivers/net/tap/tap_flow.c
+++ b/drivers/net/tap/tap_flow.c
@@ -123,6 +123,7 @@ enum key_status_e {
 };
 
 #define ISOLATE_HANDLE 1
+#define REMOTE_PROMISCUOUS_HANDLE 2
 
 struct rte_flow {
 	LIST_ENTRY(rte_flow) next; /* Pointer to the next rte_flow structure */
@@ -1692,9 +1693,15 @@ int tap_flow_implicit_create(struct pmd_internals *pmd,
 	 * The ISOLATE rule is always present and must have a static handle, as
 	 * the action is changed whether the feature is enabled (DROP) or
 	 * disabled (PASSTHRU).
+	 * There is just one REMOTE_PROMISCUOUS rule in all cases. It should
+	 * have a static handle such that adding it twice will fail with EEXIST
+	 * with any kernel version. Remark: old kernels may falsely accept the
+	 * same REMOTE_PROMISCUOUS rules if they had different handles.
 	 */
 	if (idx == TAP_ISOLATE)
 		remote_flow->msg.t.tcm_handle = ISOLATE_HANDLE;
+	else if (idx == TAP_REMOTE_PROMISC)
+		remote_flow->msg.t.tcm_handle = REMOTE_PROMISCUOUS_HANDLE;
 	else
 		tap_flow_set_handle(remote_flow);
 	if (priv_flow_process(pmd, attr, items, actions, NULL,
@@ -1709,12 +1716,16 @@ 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)
+			goto success;
 		RTE_LOG(ERR, PMD,
 			"Kernel refused TC filter rule creation (%d): %s\n",
 			errno, strerror(errno));
 		goto fail;
 	}
 	LIST_INSERT_HEAD(&pmd->implicit_flows, remote_flow, next);
+success:
 	return 0;
 fail:
 	if (remote_flow)
-- 
2.7.4