Skip to contentSkip to conversation

net: control the length of the altname list

PATCH Applied Jakub Kicinski 3 participants 7 messages Mar 9, 2022 – Mar 11, 2022

Subsystems (2)
lore
Share & tools
lore

Jakub Kicinski

author maintainer
#0 lore raw

cover net: control the length of the altname list

Count the memory used for altnames and don't let user overflow the property nlattr. This was reported by George:
https://lore.kernel.org/all/[email protected]/

Targeting net-next because I think we generally don't consider tightening "root controlled" memory accounting to be a fix. There's also some risk of breakage.

Jakub Kicinski (2):
net: account alternate interface name memory
net: limit altnames to 64k total

 net/core/rtnetlink.c | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)
Signature · 1 line
2.34.1
author maintainer
#2 lore raw

PATCH 2/2 net: limit altnames to 64k total

Property list (altname is a link "property") is wrapped in a nlattr. nlattrs length is 16bit so practically speaking the list of properties can't be longer than that, otherwise user space would have to interpret broken netlink messages.

Prevent the problem from occurring by checking the length of the property list before adding new entries.

---
 net/core/rtnetlink.c | 11 +++++++++++
 1 file changed, 11 insertions(+)
net/core/rtnetlink.c +11 -0
index aa05e89cc47c..159c9c61e6af 100644
@@ -3652,12 +3652,23 @@ static int rtnl_alt_ifname(int cmd, struct net_device *dev, struct nlattr *attr,
 			   bool *changed, struct netlink_ext_ack *extack)
 {
 	char *alt_ifname;
+	size_t size;
 	int err;
 
 	err = nla_validate(attr, attr->nla_len, IFLA_MAX, ifla_policy, extack);
 	if (err)
 		return err;
 
+	if (cmd == RTM_NEWLINKPROP) {
+		size = rtnl_prop_list_size(dev);
+		size += nla_total_size(ALTIFNAMSIZ);
+		if (size >= U16_MAX) {
+			NL_SET_ERR_MSG(extack,
+				       "effective property list too long");
+			return -EINVAL;
+		}
+	}
+
 	alt_ifname = nla_strdup(attr, GFP_KERNEL_ACCOUNT);
 	if (!alt_ifname)
 		return -ENOMEM;
Signature · 1 line
2.34.1
author maintainer
#1 lore raw

PATCH 1/2 net: account alternate interface name memory

In reply to Jakub Kicinski

George reports that altnames can eat up kernel memory. We should charge that memory appropriately.

---
 net/core/rtnetlink.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
net/core/rtnetlink.c +1 -1
index a759f9e0a847..aa05e89cc47c 100644
@@ -3658,7 +3658,7 @@ static int rtnl_alt_ifname(int cmd, struct net_device *dev, struct nlattr *attr,
 	if (err)
 		return err;
 
-	alt_ifname = nla_strdup(attr, GFP_KERNEL);
+	alt_ifname = nla_strdup(attr, GFP_KERNEL_ACCOUNT);
 	if (!alt_ifname)
 		return -ENOMEM;
 
Signature · 1 line
2.34.1
#3 lore raw

Re: [PATCH net-next 2/2] net: limit altnames to 64k total

In reply to Jakub Kicinski
Quoted 18 lines · Jakub Kicinski
net/core/rtnetlink.c +4 -0
index aa05e89cc47c..159c9c61e6af 100644
@@ -3652,12 +3652,23 @@ static int rtnl_alt_ifname(int cmd, struct net_device *dev, struct nlattr *attr,
 			   bool *changed, struct netlink_ext_ack *extack)
 {
 	char *alt_ifname;
+	size_t size;
 	int err;
 
 	err = nla_validate(attr, attr->nla_len, IFLA_MAX, ifla_policy, extack);
 	if (err)
 		return err;
 
+	if (cmd == RTM_NEWLINKPROP) {
+		size = rtnl_prop_list_size(dev);
+		size += nla_total_size(ALTIFNAMSIZ);

net/core/rtnetlink.c static int rtnl_alt_ifname(int cmd, struct net_device *dev, struct nlattr *attr, · line 3674

Quoted 10 lines
net/core/rtnetlink.c +7 -0
+		if (size >= U16_MAX) {
+			NL_SET_ERR_MSG(extack,
+				       "effective property list too long");
+			return -EINVAL;
+		}
+	}
+
 	alt_ifname = nla_strdup(attr, GFP_KERNEL_ACCOUNT);
 	if (!alt_ifname)
 		return -ENOMEM;

this tests the existing property size. Don't you want to test the size with the alt_ifname - does it make the list go over 64kB?

author maintainer
#4 lore raw

Re: [PATCH net-next 2/2] net: limit altnames to 64k total

Quoted 16 lines · On Wed, 9 Mar 2022 19:51:07 -0700 David Ahern wrote:
Quoted 15 lines · Jakub Kicinski
net/core/rtnetlink.c +1 -0
index aa05e89cc47c..159c9c61e6af 100644
@@ -3652,12 +3652,23 @@ static int rtnl_alt_ifname(int cmd, struct net_device *dev, struct nlattr *attr,
 			   bool *changed, struct netlink_ext_ack *extack)
 {
 	char *alt_ifname;
+	size_t size;
 	int err;
 
 	err = nla_validate(attr, attr->nla_len, IFLA_MAX, ifla_policy, extack);
 	if (err)
 		return err;
 
Quoted 16 lines

net/core/rtnetlink.c static int rtnl_alt_ifname(int cmd, struct net_device *dev, struct nlattr *attr, · line 3674

Quoted 13 lines
net/core/rtnetlink.c +10 -0
+	if (cmd == RTM_NEWLINKPROP) {
+		size = rtnl_prop_list_size(dev);
+		size += nla_total_size(ALTIFNAMSIZ);
+		if (size >= U16_MAX) {
+			NL_SET_ERR_MSG(extack,
+				       "effective property list too long");
+			return -EINVAL;
+		}
+	}
+
 	alt_ifname = nla_strdup(attr, GFP_KERNEL_ACCOUNT);
 	if (!alt_ifname)
 		return -ENOMEM;  

this tests the existing property size. Don't you want to test the size with the alt_ifname - does it make the list go over 64kB?

Do you mean counting the exact length of the string?

Or that I'm counting pre-add? That's why I added:

	size += nla_total_size(ALTIFNAMSIZ);

I like coding things up as prepare (validate) + commit, granted it doesn't exactly look pretty here so I can recode if you prefer. But there's no bug, right? (other than maybe

Quoted 1 line

= could have been > but whatever).

Reviewed-by
#5 lore raw

Re: [PATCH net-next 2/2] net: limit altnames to 64k total

Quoted 36 lines · Jakub Kicinski
Quoted 16 lines · On Wed, 9 Mar 2022 19:51:07 -0700 David Ahern wrote:
Quoted 15 lines · Jakub Kicinski
net/core/rtnetlink.c +1 -0
index aa05e89cc47c..159c9c61e6af 100644
@@ -3652,12 +3652,23 @@ static int rtnl_alt_ifname(int cmd, struct net_device *dev, struct nlattr *attr,
 			   bool *changed, struct netlink_ext_ack *extack)
 {
 	char *alt_ifname;
+	size_t size;
 	int err;
 
 	err = nla_validate(attr, attr->nla_len, IFLA_MAX, ifla_policy, extack);
 	if (err)
 		return err;
 
Quoted 16 lines

net/core/rtnetlink.c static int rtnl_alt_ifname(int cmd, struct net_device *dev, struct nlattr *attr, · line 3674

Quoted 13 lines
net/core/rtnetlink.c +10 -0
+	if (cmd == RTM_NEWLINKPROP) {
+		size = rtnl_prop_list_size(dev);
+		size += nla_total_size(ALTIFNAMSIZ);
+		if (size >= U16_MAX) {
+			NL_SET_ERR_MSG(extack,
+				       "effective property list too long");
+			return -EINVAL;
+		}
+	}
+
 	alt_ifname = nla_strdup(attr, GFP_KERNEL_ACCOUNT);
 	if (!alt_ifname)
 		return -ENOMEM;  

this tests the existing property size. Don't you want to test the size with the alt_ifname - does it make the list go over 64kB?

Do you mean counting the exact length of the string?

Quoted 8 lines

Or that I'm counting pre-add? That's why I added:

	size += nla_total_size(ALTIFNAMSIZ);

I like coding things up as prepare (validate) + commit, granted it doesn't exactly look pretty here so I can recode if you prefer. But there's no bug, right? (other than maybe

Quoted 1 line

= could have been > but whatever).

right. It's a worst case size estimation versus taking into account the actual space used for the name (rtnl_prop_list_size does that for each name so this is really conservative in space use).

#6 lore raw
In reply to Jakub Kicinski
This series was applied to netdev/net-next.git (master) by Jakub Kicinski <[email protected]>:

Hello:

This series was applied to netdev/net-next.git (master) by Jakub Kicinski <[email protected]>:

Quoted 9 lines · On Wed, 9 Mar 2022 10:29:12 -0800 you wrote:

Count the memory used for altnames and don't let user overflow the property nlattr. This was reported by George:
https://lore.kernel.org/all/[email protected]/

Targeting net-next because I think we generally don't consider tightening "root controlled" memory accounting to be a fix. There's also some risk of breakage.

[...]

Here is the summary with links:
- [net-next,1/2] net: account alternate interface name memory

    https://git.kernel.org/netdev/net-next/c/5d26cff5bdbe
  - [net-next,2/2] net: limit altnames to 64k total
    https://git.kernel.org/netdev/net-next/c/155fb43b70b5

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html