Skip to contentSkip to conversation

net/mlx5: fix the modify field check of tag

PATCH In review Bing Zhao 4 participants 6 messages Jun 30, 2023 – Jul 10, 2023 Network drivers

lore
Share & tools
lore

Bing Zhao

author
#0 lore raw

PATCH 1/7 net/mlx5: fix the modify field check of tag

The new member "tag_index" was added into the structure "rte_flow_action_modify_data". It is the index of the header inside encapsulation level. Both the "RTE_FLOW_FIELD_TAG" and "MLX5_RTE_FLOW_FIELD_META_REG" types will use this member instead of the "level", but the "level" will still be supported for the compatibility.

In the validation stage, not only the "RTE_FLOW_FIELD_TAG" type, but also the "MLX5_RTE_FLOW_FIELD_META_REG" needs to be checked.

In the meanwhile, lowering down the log priority will help to prevent the flooding.

Fixes 68c513e861ca ("net/mlx5: align implementation with modify API")
---
 drivers/net/mlx5/mlx5_flow.c | 5 +++--
 drivers/net/mlx5/mlx5_flow.h | 7 ++++---
 2 files changed, 7 insertions(+), 5 deletions(-)
drivers/net/mlx5/mlx5_flow.c +3 -2
index cf83db7b60..97211fc209 100644
@@ -2411,7 +2411,8 @@ flow_validate_modify_field_level(const struct rte_flow_action_modify_data *data,
 {
 	if (data->level == 0)
 		return 0;
-	if (data->field != RTE_FLOW_FIELD_TAG)
+	if (data->field != RTE_FLOW_FIELD_TAG &&
+	    data->field != (enum rte_flow_field_id)MLX5_RTE_FLOW_FIELD_META_REG)
 		return rte_flow_error_set(error, ENOTSUP,
 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 					  "inner header fields modification is not supported");
@@ -2424,7 +2425,7 @@ flow_validate_modify_field_level(const struct rte_flow_action_modify_data *data,
 	 * 'tag_index' field. In old API, it was provided using 'level' field
 	 * and it is still supported for backwards compatibility.
 	 */
-	DRV_LOG(WARNING, "tag array provided in 'level' field instead of 'tag_index' field.");
+	DRV_LOG(DEBUG, "tag array provided in 'level' field instead of 'tag_index' field.");
 	return 0;
 }
 
drivers/net/mlx5/mlx5_flow.h +4 -3
index 003e7da3a6..d6e88425a3 100644
@@ -52,8 +52,8 @@ enum mlx5_rte_flow_action_type {
 
 /* Private (internal) Field IDs for MODIFY_FIELD action. */
 enum mlx5_rte_flow_field_id {
-		MLX5_RTE_FLOW_FIELD_END = INT_MIN,
-			MLX5_RTE_FLOW_FIELD_META_REG,
+	MLX5_RTE_FLOW_FIELD_END = INT_MIN,
+	MLX5_RTE_FLOW_FIELD_META_REG,
 };
 
 #define MLX5_INDIRECT_ACTION_TYPE_OFFSET 29
@@ -1117,9 +1117,10 @@ flow_dv_fetch_field(const uint8_t *data, uint32_t size)
 static inline bool
 flow_modify_field_support_tag_array(enum rte_flow_field_id field)
 {
-	switch (field) {
+	switch ((int)field) {
 	case RTE_FLOW_FIELD_TAG:
 	case RTE_FLOW_FIELD_MPLS:
+	case MLX5_RTE_FLOW_FIELD_META_REG:
 		return true;
 	default:
 		break;
Signature · 1 line
2.34.1
#1 lore raw

drivers/net/mlx5/mlx5_flow.h flow_dv_fetch_field(const uint8_t *data, uint32_t size) · line 1123

Quoted 9 lines · On Fri, 30 Jun 2023 08:43:03 +0300 Bing Zhao <[email protected]> wrote:
drivers/net/mlx5/mlx5_flow.h +2 -1
@@ -1117,9 +1117,10 @@ flow_dv_fetch_field(const uint8_t *data, uint32_t size)
 static inline bool
 flow_modify_field_support_tag_array(enum rte_flow_field_id field)
 {
-	switch (field) {
+	switch ((int)field) {
 	case RTE_FLOW_FIELD_TAG:
 	case RTE_FLOW_FIELD_MPLS:
+	case MLX5_RTE_FLOW_FIELD_META_REG:

Mixing internal and API fields seems like something that could get easily broken by changes to rte_flow.

author
#2 lore raw

Hi Stephen,
If I understand correctly, do you mean that the internal value and rte_flow API value may have some conflict?
All the MLX5 internal enum values start from INT_MIN. When treating it as a int value, it would not have the same value with rte_flow enums, unless all the 2^^32 are defined. But yes, this has some risk since there is no limitation of the values in the rte_flow API.

BR. Bing

Quoted 27 lines

-----Original Message-----
From: Stephen Hemminger <[email protected]>
Sent: Friday, June 30, 2023 2:09 PM
To: Bing Zhao <[email protected]>
Cc: Matan Azrad <[email protected]>; Slava Ovsiienko <[email protected]>; Ori Kam <[email protected]>; Suanming Mou <[email protected]>; Raslan Darawsheh <[email protected]>; [email protected]; Michael Baum <[email protected]>
Subject: Re: [PATCH 1/7] net/mlx5: fix the modify field check of tag

External email: Use caution opening links or attachments

drivers/net/mlx5/mlx5_flow.h flow_dv_fetch_field(const uint8_t *data, uint32_t size) · line 1123

Quoted 8 lines · On Fri, 30 Jun 2023 08:43:03 +0300 Bing Zhao <[email protected]> wrote:
(quoted diff) +0 -0
@@ -1117,9 +1117,10 @@ flow_dv_fetch_field(const uint8_t *data,

uint32_t size) static inline bool flow_modify_field_support_tag_array(enum rte_flow_field_id field) {

drivers/net/mlx5/mlx5_flow.h +2 -1
-     switch (field) {
+     switch ((int)field) {
      case RTE_FLOW_FIELD_TAG:
      case RTE_FLOW_FIELD_MPLS:
+     case MLX5_RTE_FLOW_FIELD_META_REG:

Mixing internal and API fields seems like something that could get easily broken by changes to rte_flow.

#3 lore raw

03/07/2023 15:31, Bing Zhao:

Quoted 4 lines

Hi Stephen,
If I understand correctly, do you mean that the internal value and rte_flow API value may have some conflict?
All the MLX5 internal enum values start from INT_MIN. When treating it as a int value, it would not have the same value with rte_flow enums, unless all the 2^^32 are defined. But yes, this has some risk since there is no limitation of the values in the rte_flow API.

We can assume it will never happen.
This is good to go.

Quoted 28 lines
Quoted 27 lines

-----Original Message-----
From: Stephen Hemminger <[email protected]>
Sent: Friday, June 30, 2023 2:09 PM
To: Bing Zhao <[email protected]>
Cc: Matan Azrad <[email protected]>; Slava Ovsiienko <[email protected]>; Ori Kam <[email protected]>; Suanming Mou <[email protected]>; Raslan Darawsheh <[email protected]>; [email protected]; Michael Baum <[email protected]>
Subject: Re: [PATCH 1/7] net/mlx5: fix the modify field check of tag

External email: Use caution opening links or attachments

drivers/net/mlx5/mlx5_flow.h flow_dv_fetch_field(const uint8_t *data, uint32_t size) · line 1123

Quoted 8 lines · On Fri, 30 Jun 2023 08:43:03 +0300 Bing Zhao <[email protected]> wrote:
(quoted diff) +0 -0
@@ -1117,9 +1117,10 @@ flow_dv_fetch_field(const uint8_t *data,

uint32_t size) static inline bool flow_modify_field_support_tag_array(enum rte_flow_field_id field) {

drivers/net/mlx5/mlx5_flow.h +2 -1
-     switch (field) {
+     switch ((int)field) {
      case RTE_FLOW_FIELD_TAG:
      case RTE_FLOW_FIELD_MPLS:
+     case MLX5_RTE_FLOW_FIELD_META_REG:

Mixing internal and API fields seems like something that could get easily broken by changes to rte_flow.

author
#4 lore raw

Thank you.

Quoted 60 lines

-----Original Message-----
From: Thomas Monjalon <[email protected]>
Sent: Thursday, July 6, 2023 5:37 PM
To: Stephen Hemminger <[email protected]>; Bing Zhao <[email protected]>
Cc: [email protected]; Matan Azrad <[email protected]>; Slava Ovsiienko <[email protected]>; Ori Kam <[email protected]>; Suanming Mou <[email protected]>; Raslan Darawsheh <[email protected]>; [email protected]; Michael Baum <[email protected]>
Subject: Re: [PATCH 1/7] net/mlx5: fix the modify field check of tag

External email: Use caution opening links or attachments

03/07/2023 15:31, Bing Zhao:

Quoted 2 lines

Hi Stephen,
If I understand correctly, do you mean that the internal value and rte_flow

API value may have some conflict?

Quoted 1 line

All the MLX5 internal enum values start from INT_MIN. When treating it as a

int value, it would not have the same value with rte_flow enums, unless all the 2^^32 are defined.

Quoted 1 line

But yes, this has some risk since there is no limitation of the values in the

rte_flow API.

We can assume it will never happen.
This is good to go.

Quoted 28 lines
Quoted 27 lines

-----Original Message-----
From: Stephen Hemminger <[email protected]>
Sent: Friday, June 30, 2023 2:09 PM
To: Bing Zhao <[email protected]>
Cc: Matan Azrad <[email protected]>; Slava Ovsiienko <[email protected]>; Ori Kam <[email protected]>; Suanming Mou <[email protected]>; Raslan Darawsheh <[email protected]>; [email protected]; Michael Baum <[email protected]>
Subject: Re: [PATCH 1/7] net/mlx5: fix the modify field check of tag

External email: Use caution opening links or attachments

drivers/net/mlx5/mlx5_flow.h flow_dv_fetch_field(const uint8_t *data, uint32_t size) · line 1123

Quoted 8 lines · On Fri, 30 Jun 2023 08:43:03 +0300 Bing Zhao <[email protected]> wrote:
(quoted diff) +0 -0
@@ -1117,9 +1117,10 @@ flow_dv_fetch_field(const uint8_t *data,

uint32_t size) static inline bool flow_modify_field_support_tag_array(enum rte_flow_field_id field) {

drivers/net/mlx5/mlx5_flow.h +2 -1
-     switch (field) {
+     switch ((int)field) {
      case RTE_FLOW_FIELD_TAG:
      case RTE_FLOW_FIELD_MPLS:
+     case MLX5_RTE_FLOW_FIELD_META_REG:

Mixing internal and API fields seems like something that could get easily broken by changes to rte_flow.

#5 lore raw
In reply to Bing Zhao

Hi,

Quoted 18 lines

-----Original Message-----
From: Bing Zhao <[email protected]>
Sent: Friday, June 30, 2023 8:43 AM
To: Matan Azrad <[email protected]>; Slava Ovsiienko <[email protected]>; Ori Kam <[email protected]>; Suanming Mou <[email protected]>; Raslan Darawsheh <[email protected]>
Cc: [email protected]; Michael Baum <[email protected]>
Subject: [PATCH 1/7] net/mlx5: fix the modify field check of tag

The new member "tag_index" was added into the structure "rte_flow_action_modify_data". It is the index of the header inside encapsulation level. Both the "RTE_FLOW_FIELD_TAG" and "MLX5_RTE_FLOW_FIELD_META_REG" types will use this member instead of the "level", but the "level" will still be supported for the compatibility.

In the validation stage, not only the "RTE_FLOW_FIELD_TAG" type, but also the "MLX5_RTE_FLOW_FIELD_META_REG" needs to be checked.

Quoted 8 lines

In the meanwhile, lowering down the log priority will help to prevent the flooding.

Fixes 68c513e861ca ("net/mlx5: align implementation with modify API")

Series applied to next-net-mlx,

Kindest regards,
Raslan Darawsheh