Skip to contentSkip to conversation

devlink: move DEVLINK_ATTR_MAX-sized array off stack

PATCH Superseded Arnd Bergmann 4 participants 4 messages Jul 8, 2025 – Jul 9, 2025

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 devlink: move DEVLINK_ATTR_MAX-sized array off stack

From: Arnd Bergmann <[email protected]>

There are many possible devlink attributes, so having an array of them on the stack can cause a warning for excessive stack usage:

net/devlink/rate.c: In function 'devlink_nl_rate_tc_bw_parse':
net/devlink/rate.c:382:1: error: the frame size of 1648 bytes is larger than 1536 bytes [-Werror=frame-larger-than=]

Change this to dynamic allocation instead.

Fixes 566e8f108fc7 ("devlink: Extend devlink rate API with traffic classes bandwidth management")

---
I see that only two of the many array entries are actually used in this
function: DEVLINK_ATTR_RATE_TC_INDEX and DEVLINK_ATTR_RATE_TC_BW. If there is an interface to extract just a single entry, using that would be a little easier than the kcalloc().

---
 net/devlink/rate.c | 18 ++++++++++++------
 1 file changed, 12 insertions(+), 6 deletions(-)
net/devlink/rate.c +12 -6
index d39300a9b3d4..e4083649748f 100644
@@ -346,19 +346,23 @@ static int devlink_nl_rate_tc_bw_parse(struct nlattr *parent_nest, u32 *tc_bw,
 				       unsigned long *bitmap,
 				       struct netlink_ext_ack *extack)
 {
-	struct nlattr *tb[DEVLINK_ATTR_MAX + 1];
+	struct nlattr **tb;
 	u8 tc_index;
 	int err;
 
+	tb = kcalloc(DEVLINK_ATTR_MAX + 1, sizeof(struct nlattr *), GFP_KERNEL);
+	if (!tb)
+		return -ENOMEM;
 	err = nla_parse_nested(tb, DEVLINK_ATTR_MAX, parent_nest,
 			       devlink_dl_rate_tc_bws_nl_policy, extack);
 	if (err)
-		return err;
+		goto out;
 
+	err = -EINVAL;
 	if (!tb[DEVLINK_ATTR_RATE_TC_INDEX]) {
 		NL_SET_ERR_ATTR_MISS(extack, parent_nest,
 				     DEVLINK_ATTR_RATE_TC_INDEX);
-		return -EINVAL;
+		goto out;
 	}
 
 	tc_index = nla_get_u8(tb[DEVLINK_ATTR_RATE_TC_INDEX]);
@@ -366,19 +370,21 @@ static int devlink_nl_rate_tc_bw_parse(struct nlattr *parent_nest, u32 *tc_bw,
 	if (!tb[DEVLINK_ATTR_RATE_TC_BW]) {
 		NL_SET_ERR_ATTR_MISS(extack, parent_nest,
 				     DEVLINK_ATTR_RATE_TC_BW);
-		return -EINVAL;
+		goto out;
 	}
 
 	if (test_and_set_bit(tc_index, bitmap)) {
 		NL_SET_ERR_MSG_FMT(extack,
 				   "Duplicate traffic class index specified (%u)",
 				   tc_index);
-		return -EINVAL;
+		goto out;
 	}
 
 	tc_bw[tc_index] = nla_get_u32(tb[DEVLINK_ATTR_RATE_TC_BW]);
 
-	return 0;
+out:
+	kfree(tb);
+	return err;
 }
 
 static int devlink_nl_rate_tc_bw_set(struct devlink_rate *devlink_rate,
Signature · 1 line
2.39.5
reviewer
#1 lore raw

On Tue, Jul 08, 2025 at 06:06:43PM +0200, Arnd Bergmann wrote:

...

Quoted 1 line
net/devlink/rate.c +0 -0

...

Quoted 1 line
net/devlink/rate.c +1 -0
+	err = -EINVAL;

...

net/devlink/rate.c static int devlink_nl_rate_tc_bw_parse(struct nlattr *parent_nest, u32 *tc_bw, · line 381

Quoted 3 lines
net/devlink/rate.c +0 -1
 	tc_bw[tc_index] = nla_get_u32(tb[DEVLINK_ATTR_RATE_TC_BW]);
 
-	return 0;

Hi Arnd,

It looks like err will be -EINVAL here, which doesn't seem desirable. I'd just return 0 here.

And, FWIIW, I'd only set err when it is needed, even if that means some duplication. I think that is an idiomatic approach, at least within Networking.

Quoted 4 lines
net/devlink/rate.c +3 -0
+out:
+	kfree(tb);
+	return err;
 }

...

Signature · 1 line
pw-bot: changes-requested
maintainer
#2 lore raw
In reply to Arnd Bergmann
Quoted 64 lines · Tue, Jul 08, 2025 at 06:06:43PM +0200, [email protected] wrote:

From: Arnd Bergmann <[email protected]>

There are many possible devlink attributes, so having an array of them on the stack can cause a warning for excessive stack usage:

net/devlink/rate.c: In function 'devlink_nl_rate_tc_bw_parse':
net/devlink/rate.c:382:1: error: the frame size of 1648 bytes is larger than 1536 bytes [-Werror=frame-larger-than=]

Change this to dynamic allocation instead.

Fixes 566e8f108fc7 ("devlink: Extend devlink rate API with traffic classes bandwidth management")

---
I see that only two of the many array entries are actually used in this
function: DEVLINK_ATTR_RATE_TC_INDEX and DEVLINK_ATTR_RATE_TC_BW. If there is an interface to extract just a single entry, using that would be a little easier than the kcalloc().
---
net/devlink/rate.c | 18 ++++++++++++------

1 file changed, 12 insertions(+), 6 deletions(-)
net/devlink/rate.c +8 -4
index d39300a9b3d4..e4083649748f 100644
@@ -346,19 +346,23 @@ static int devlink_nl_rate_tc_bw_parse(struct nlattr *parent_nest, u32 *tc_bw,
 				       unsigned long *bitmap,
 				       struct netlink_ext_ack *extack)
 {
-	struct nlattr *tb[DEVLINK_ATTR_MAX + 1];
+	struct nlattr **tb;
 	u8 tc_index;
 	int err;
 
+	tb = kcalloc(DEVLINK_ATTR_MAX + 1, sizeof(struct nlattr *), GFP_KERNEL);
+	if (!tb)
+		return -ENOMEM;
 	err = nla_parse_nested(tb, DEVLINK_ATTR_MAX, parent_nest,
 			       devlink_dl_rate_tc_bws_nl_policy, extack);
 	if (err)
-		return err;
+		goto out;
 
+	err = -EINVAL;
 	if (!tb[DEVLINK_ATTR_RATE_TC_INDEX]) {
 		NL_SET_ERR_ATTR_MISS(extack, parent_nest,
 				     DEVLINK_ATTR_RATE_TC_INDEX);
-		return -EINVAL;
+		goto out;
 	}
 
 	tc_index = nla_get_u8(tb[DEVLINK_ATTR_RATE_TC_INDEX]);
@@ -366,19 +370,21 @@ static int devlink_nl_rate_tc_bw_parse(struct nlattr *parent_nest, u32 *tc_bw,
 	if (!tb[DEVLINK_ATTR_RATE_TC_BW]) {
 		NL_SET_ERR_ATTR_MISS(extack, parent_nest,
 				     DEVLINK_ATTR_RATE_TC_BW);
-		return -EINVAL;
+		goto out;
 	}
 
 	if (test_and_set_bit(tc_index, bitmap)) {
 		NL_SET_ERR_MSG_FMT(extack,
 				   "Duplicate traffic class index specified (%u)",
 				   tc_index);

net/devlink/rate.c static int devlink_nl_rate_tc_bw_parse(struct nlattr *parent_nest, u32 *tc_bw, · line 387

Quoted 10 lines
net/devlink/rate.c +4 -2
-		return -EINVAL;
+		goto out;
 	}
 
 	tc_bw[tc_index] = nla_get_u32(tb[DEVLINK_ATTR_RATE_TC_BW]);
 
-	return 0;
+out:
+	kfree(tb);
+	return err;

Isn't it about the time to start to leverage or cleanup infrastructure for things like this? /me covers against all the eggs flying in I mean, there are subsystems where it's perfectly fine to use it. Can we start with frees like this one and avoid the silly gotos?

#3 lore raw
Quoted 6 lines · Jiri Pirko

Tue, Jul 08, 2025 at 06:06:43PM +0200, [email protected] wrote:

Isn't it about the time to start to leverage or cleanup infrastructure for things like this? /me covers against all the eggs flying in I mean, there are subsystems where it's perfectly fine to use it. Can we start with frees like this one and avoid the silly gotos?

I had so far resisted learning about how those actually work, thanks for pushing me to finally trying it out ;-)

I've sent a v2 now.

Arnd