netfilter: cttimeout: remove VLA in ctnl_timeout_parse_policy

RFC In review Gustavo A. R. Silva 2 participants 5 messages Mar 6, 2018 – Mar 11, 2018

Subsystems (3)

Networking Network drivers netfilter

lore

Share & tools

lore

Gustavo A. R. Silva

author

2018-03-06 18:47 UTC #0 lore raw

RFC 1/1 netfilter: cttimeout: remove VLA in ctnl_timeout_parse_policy

In preparation to enabling -Wvla, remove VLA and replace it with dynamic memory allocation.

Signed-off-by Gustavo A. R. Silva <[email protected]>

---
 net/netfilter/nfnetlink_cttimeout.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

net/netfilter/nfnetlink_cttimeout.c +10 -2

index 95b0470..a2f7d92 100644
@@ -52,18 +52,26 @@ ctnl_timeout_parse_policy(void *timeouts,
 			  struct net *net, const struct nlattr *attr)
 {
 	int ret = 0;
+	struct nlattr **tb = NULL;
 
 	if (likely(l4proto->ctnl_timeout.nlattr_to_obj)) {
-		struct nlattr *tb[l4proto->ctnl_timeout.nlattr_max+1];
+		tb = kcalloc(l4proto->ctnl_timeout.nlattr_max + 1, sizeof(*tb),
+			     GFP_KERNEL);
+
+		if (!tb)
+			return -ENOMEM;
 
 		ret = nla_parse_nested(tb, l4proto->ctnl_timeout.nlattr_max,
 				       attr, l4proto->ctnl_timeout.nla_policy,
 				       NULL);
 		if (ret < 0)
-			return ret;
+			goto err;
 
 		ret = l4proto->ctnl_timeout.nlattr_to_obj(tb, net, timeouts);
 	}
+
+err:
+	kfree(tb);
 	return ret;
 }

Signature · 1 line

2.7.4

Pablo Neira Ayuso

maintainer

2018-03-11 22:04 UTC #1 lore raw

Quoted 2 lines · Gustavo A. R. Silva

In preparation to enabling -Wvla, remove VLA and replace it with dynamic memory allocation.

Looks good but...

Quoted 10 lines

Signed-off-by Gustavo A. R. Silva <[email protected]>
---
 net/netfilter/nfnetlink_cttimeout.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)
net/netfilter/nfnetlink_cttimeout.c +0 -0
index 95b0470..a2f7d92 100644

net/netfilter/nfnetlink_cttimeout.c ctnl_timeout_parse_policy(void *timeouts, · line 55

Quoted 4 lines

net/netfilter/nfnetlink_cttimeout.c +1 -0

@@ -52,18 +52,26 @@ ctnl_timeout_parse_policy(void *timeouts,
 			  struct net *net, const struct nlattr *attr)
 {
 	int ret = 0;
+	struct nlattr **tb = NULL;

I think we don't need to initialize this, right?

Quoted 27 lines

net/netfilter/nfnetlink_cttimeout.c +9 -2

 	if (likely(l4proto->ctnl_timeout.nlattr_to_obj)) {
-		struct nlattr *tb[l4proto->ctnl_timeout.nlattr_max+1];
+		tb = kcalloc(l4proto->ctnl_timeout.nlattr_max + 1, sizeof(*tb),
+			     GFP_KERNEL);
+
+		if (!tb)
+			return -ENOMEM;
 
 		ret = nla_parse_nested(tb, l4proto->ctnl_timeout.nlattr_max,
 				       attr, l4proto->ctnl_timeout.nla_policy,
 				       NULL);
 		if (ret < 0)
-			return ret;
+			goto err;
 
 		ret = l4proto->ctnl_timeout.nlattr_to_obj(tb, net, timeouts);
 	}
+
+err:
+	kfree(tb);
 	return ret;
 }

Signature · 1 line

2.7.4

Gustavo A. R. Silva

author

2018-03-11 22:12 UTC #2 lore raw

Hi Pablo,

Quoted 11 lines · Pablo Neira Ayuso

Quoted 2 lines · Gustavo A. R. Silva
In preparation to enabling -Wvla, remove VLA and replace it with dynamic memory allocation.
Looks good but...
Quoted 5 lines
Signed-off-by Gustavo A. R. Silva <[email protected]>
---
  net/netfilter/nfnetlink_cttimeout.c | 12 ++++++++++--
  1 file changed, 10 insertions(+), 2 deletions(-)

Quoted 12 lines

net/netfilter/nfnetlink_cttimeout.c ctnl_timeout_parse_policy(void *timeouts, · line 55
Quoted 9 lines
net/netfilter/nfnetlink_cttimeout.c +1 -0
index 95b0470..a2f7d92 100644
@@ -52,18 +52,26 @@ ctnl_timeout_parse_policy(void *timeouts,
  			  struct net *net, const struct nlattr *attr)
  {
  	int ret = 0;
+	struct nlattr **tb = NULL;
I think we don't need to initialize this, right?

We actually do have to initialized it because in the unlikely case that the code block inside the 'if' below is not executed, then we will end up freeing an uninitialized pointer.

Thanks
--
Gustavo

Quoted 27 lines

net/netfilter/nfnetlink_cttimeout.c +9 -2

  	if (likely(l4proto->ctnl_timeout.nlattr_to_obj)) {
-		struct nlattr *tb[l4proto->ctnl_timeout.nlattr_max+1];
+		tb = kcalloc(l4proto->ctnl_timeout.nlattr_max + 1, sizeof(*tb),
+			     GFP_KERNEL);
+
+		if (!tb)
+			return -ENOMEM;
 
  		ret = nla_parse_nested(tb, l4proto->ctnl_timeout.nlattr_max,
  				       attr, l4proto->ctnl_timeout.nla_policy,
  				       NULL);
  		if (ret < 0)
-			return ret;
+			goto err;
 
  		ret = l4proto->ctnl_timeout.nlattr_to_obj(tb, net, timeouts);
  	}
+
+err:
+	kfree(tb);
  	return ret;
  }

Signature · 1 line

2.7.4

Pablo Neira Ayuso

maintainer

2018-03-11 22:21 UTC #3 lore raw

Quoted 27 lines · Gustavo A. R. Silva

Hi Pablo,
Quoted 11 lines · Pablo Neira Ayuso
Quoted 2 lines · Gustavo A. R. Silva
In preparation to enabling -Wvla, remove VLA and replace it with dynamic memory allocation.
Looks good but...
Quoted 5 lines
Signed-off-by Gustavo A. R. Silva <[email protected]>
---
  net/netfilter/nfnetlink_cttimeout.c | 12 ++++++++++--
  1 file changed, 10 insertions(+), 2 deletions(-)
Quoted 12 lines
net/netfilter/nfnetlink_cttimeout.c ctnl_timeout_parse_policy(void *timeouts, · line 55
Quoted 9 lines
net/netfilter/nfnetlink_cttimeout.c +1 -0
index 95b0470..a2f7d92 100644
@@ -52,18 +52,26 @@ ctnl_timeout_parse_policy(void *timeouts,
  			  struct net *net, const struct nlattr *attr)
  {
  	int ret = 0;
+	struct nlattr **tb = NULL;
I think we don't need to initialize this, right?

Quoted 3 lines

We actually do have to initialized it because in the unlikely case that the code block inside the 'if' below is not executed, then we will end up freeing an uninitialized pointer.

I see, you're right indeed.

We can probably simplify this code, but just doing:

        if (!l4proto->ctnl_timeout.nlattr_to_obj))
                return 0;

        netlink attribute parsing here.

You could even remove the likely() thing, which doesn't make much sense for control plane code.

I understand this is a larger change, but I think this function will look better while we're removing VLA.

Would you mind having a look? I'd appreciate if so.

Thanks.

Gustavo A. R. Silva

author

2018-03-11 22:45 UTC #4 lore raw

Quoted 32 lines · Pablo Neira Ayuso

Quoted 27 lines · Gustavo A. R. Silva
Hi Pablo,
Quoted 11 lines · Pablo Neira Ayuso
Quoted 2 lines · Gustavo A. R. Silva
In preparation to enabling -Wvla, remove VLA and replace it with dynamic memory allocation.
Looks good but...
Quoted 5 lines
Signed-off-by Gustavo A. R. Silva <[email protected]>
---
   net/netfilter/nfnetlink_cttimeout.c | 12 ++++++++++--
   1 file changed, 10 insertions(+), 2 deletions(-)
Quoted 12 lines
net/netfilter/nfnetlink_cttimeout.c ctnl_timeout_parse_policy(void *timeouts, · line 55
Quoted 9 lines
net/netfilter/nfnetlink_cttimeout.c +1 -0
index 95b0470..a2f7d92 100644
@@ -52,18 +52,26 @@ ctnl_timeout_parse_policy(void *timeouts,
   			  struct net *net, const struct nlattr *attr)
   {
   	int ret = 0;
+	struct nlattr **tb = NULL;
I think we don't need to initialize this, right?
Quoted 3 lines
We actually do have to initialized it because in the unlikely case that the code block inside the 'if' below is not executed, then we will end up freeing an uninitialized pointer.

Quoted 7 lines

I see, you're right indeed.
We can probably simplify this code, but just doing:
         if (!l4proto->ctnl_timeout.nlattr_to_obj))
                 return 0;

I wonder if it is better to code this instead:

if (unlikely(!l4proto->ctnl_timeout.nlattr_to_obj)))

	return 0;

Quoted 5 lines

netlink attribute parsing here.
You could even remove the likely() thing, which doesn't make much sense for control plane code.

Why is that?

Quoted 5 lines

I understand this is a larger change, but I think this function will look better while we're removing VLA.
Would you mind having a look? I'd appreciate if so.

I can do that. No problem.

Thanks
--
Gustavo