Skip to contentSkip to conversation

net/kni: initialize rte_kni_conf to 0 before using it

PATCH In review Harold Huang 2 participants 6 messages Mar 2, 2022 – Mar 4, 2022 Network drivers

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 net/kni: initialize rte_kni_conf to 0 before using it

When kni driver calls eth_kni_start to start device, some fields such as min_mtu and max_mtu of rte_kni_conf are not initialized. It will cause kni_ioctl_create create a kni netdevice with a random min_mtu and max_mtu value. This is unexpected and in some time we could not change the kni device mtu with ip link command.

Fixes ff1e35fb5f8 ("kni: calculate MTU from mbuf size")
---
 drivers/net/kni/rte_eth_kni.c | 1 +
 1 file changed, 1 insertion(+)
drivers/net/kni/rte_eth_kni.c +1 -0
index c428caf441..23b15edfac 100644
@@ -128,6 +128,7 @@ eth_kni_start(struct rte_eth_dev *dev)
 	const char *name = dev->device->name + 4; /* remove net_ */
 
 	mb_pool = internals->rx_queues[0].mb_pool;
+	memset(&conf, 0, sizeof(conf));
 	strlcpy(conf.name, name, RTE_KNI_NAMESIZE);
 	conf.force_bind = 0;
 	conf.group_id = port_id;
Signature · 1 line
2.27.0
#1 lore raw
Quoted 6 lines · Harold Huang

When kni driver calls eth_kni_start to start device, some fields such as min_mtu and max_mtu of rte_kni_conf are not initialized. It will cause kni_ioctl_create create a kni netdevice with a random min_mtu and max_mtu value. This is unexpected and in some time we could not change the kni device mtu with ip link command.

Agree on the problem and the solution, thanks for the fix.

Quoted 11 lines
Fixes ff1e35fb5f8 ("kni: calculate MTU from mbuf size")
---
  drivers/net/kni/rte_eth_kni.c | 1 +
  1 file changed, 1 insertion(+)
drivers/net/kni/rte_eth_kni.c +0 -0
index c428caf441..23b15edfac 100644

drivers/net/kni/rte_eth_kni.c eth_kni_start(struct rte_eth_dev *dev) · line 131

Quoted 4 lines
drivers/net/kni/rte_eth_kni.c +1 -0
@@ -128,6 +128,7 @@ eth_kni_start(struct rte_eth_dev *dev)
  	const char *name = dev->device->name + 4; /* remove net_ */
 
  	mb_pool = internals->rx_queues[0].mb_pool;
+	memset(&conf, 0, sizeof(conf));

Can you prefer initialize to zero, instead of 'memset', I think it is more clear that way:

- struct rte_kni_conf conf;
+ struct rte_kni_conf conf = { 0 };

Quoted 3 lines
drivers/net/kni/rte_eth_kni.c +0 -0
  	strlcpy(conf.name, name, RTE_KNI_NAMESIZE);
  	conf.force_bind = 0;
  	conf.group_id = port_id;
#2 lore raw
Quoted 24 lines · Ferruh Yigit
Quoted 6 lines · Harold Huang

When kni driver calls eth_kni_start to start device, some fields such as min_mtu and max_mtu of rte_kni_conf are not initialized. It will cause kni_ioctl_create create a kni netdevice with a random min_mtu and max_mtu value. This is unexpected and in some time we could not change the kni device mtu with ip link command.

Agree on the problem and the solution, thanks for the fix.

Quoted 13 lines
Fixes ff1e35fb5f8 ("kni: calculate MTU from mbuf size")
---
  drivers/net/kni/rte_eth_kni.c | 1 +
  1 file changed, 1 insertion(+)
drivers/net/kni/rte_eth_kni.c +0 -0
index c428caf441..23b15edfac 100644
@@ -128,6 +128,7 @@ eth_kni_start(struct rte_eth_dev *dev)
      const char *name = dev->device->name + 4; /* remove net_ */
 
Quoted 5 lines

drivers/net/kni/rte_eth_kni.c eth_kni_start(struct rte_eth_dev *dev) · line 131

Quoted 2 lines
drivers/net/kni/rte_eth_kni.c +1 -0
      mb_pool = internals->rx_queues[0].mb_pool;
+     memset(&conf, 0, sizeof(conf));

Can you prefer initialize to zero, instead of 'memset', I think it is more clear that way:

Thanks. Sounds good, fix it.

Quoted 8 lines

- struct rte_kni_conf conf;
+ struct rte_kni_conf conf = { 0 };

Quoted 3 lines
drivers/net/kni/rte_eth_kni.c +0 -0
      strlcpy(conf.name, name, RTE_KNI_NAMESIZE);
      conf.force_bind = 0;
      conf.group_id = port_id;
#3 lore raw

PATCH 1/1 net/kni: initialize rte_kni_conf to 0 before using it

In reply to Harold Huang

When kni driver calls eth_kni_start to start device, some fields such as min_mtu and max_mtu of rte_kni_conf are not initialized. It will cause kni_ioctl_create create a kni netdevice with a random min_mtu and max_mtu value. This is unexpected and sometimes we could not change the kni device mtu with ip link command.

Fixes ff1e35fb5f83 ("kni: calculate MTU from mbuf size")
---
 drivers/net/kni/rte_eth_kni.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/net/kni/rte_eth_kni.c +1 -1
index c428caf441..ba58d9dbae 100644
@@ -124,7 +124,7 @@ eth_kni_start(struct rte_eth_dev *dev)
 	struct pmd_internals *internals = dev->data->dev_private;
 	uint16_t port_id = dev->data->port_id;
 	struct rte_mempool *mb_pool;
-	struct rte_kni_conf conf;
+	struct rte_kni_conf conf = { 0 };
 	const char *name = dev->device->name + 4; /* remove net_ */
 
 	mb_pool = internals->rx_queues[0].mb_pool;
Signature · 1 line
2.27.0
Reviewed-by on v1 1/1 Applied
#4 lore raw
Quoted 9 lines · Harold Huang

When kni driver calls eth_kni_start to start device, some fields such as min_mtu and max_mtu of rte_kni_conf are not initialized. It will cause kni_ioctl_create create a kni netdevice with a random min_mtu and max_mtu value. This is unexpected and sometimes we could not change the kni device mtu with ip link command.

Fixes ff1e35fb5f83 ("kni: calculate MTU from mbuf size")

Applied to dpdk-next-net/main, thanks.

#5 lore raw
In reply to Harold Huang
Quoted 18 lines · Harold Huang

When kni driver calls eth_kni_start to start device, some fields such as min_mtu and max_mtu of rte_kni_conf are not initialized. It will cause kni_ioctl_create create a kni netdevice with a random min_mtu and max_mtu value. This is unexpected and sometimes we could not change the kni device mtu with ip link command.

Fixes ff1e35fb5f83 ("kni: calculate MTU from mbuf size")
---
  drivers/net/kni/rte_eth_kni.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
drivers/net/kni/rte_eth_kni.c +0 -0
index c428caf441..ba58d9dbae 100644

drivers/net/kni/rte_eth_kni.c eth_kni_start(struct rte_eth_dev *dev) · line 127

Quoted 5 lines
drivers/net/kni/rte_eth_kni.c +1 -1
@@ -124,7 +124,7 @@ eth_kni_start(struct rte_eth_dev *dev)
  	struct pmd_internals *internals = dev->data->dev_private;
  	uint16_t port_id = dev->data->port_id;
  	struct rte_mempool *mb_pool;
-	struct rte_kni_conf conf;
+	struct rte_kni_conf conf = { 0 };

This is causing build error [1], updated in next-net [2] to fix.

Normally '{ 0 }' should be universal zero-initializer, but I guess it fails because of old compiler version:
gcc 4.8.5 "cc (GCC) 4.8.5 20150623 (Red Hat 4.8.5-44)"

[1]
FAILED: drivers/a715181@@tmp_rte_net_kni@sta/net_kni_rte_eth_kni.c.o ../drivers/net/kni/rte_eth_kni.c: In function 'eth_kni_start':
../drivers/net/kni/rte_eth_kni.c:127:9:
error: missing braces around initializer [-Werror=missing-braces]

  struct rte_kni_conf conf = { 0 };
         ^

[2]
struct rte_kni_conf conf = {{0}};

Quoted 3 lines
drivers/net/kni/rte_eth_kni.c +0 -0
  	const char *name = dev->device->name + 4; /* remove net_ */
 
  	mb_pool = internals->rx_queues[0].mb_pool;