[PATCH net 1/2] net: Use device model to get driver name in skb_gso_segment()

[PATCH net 1/2] net: Use device model to get driver name in skb_gso_segment()

[Ben Hutchings]

ethtool operations generally require the caller to hold RTNL and are
not safe to call in atomic context.  The device model provides this
information for most devices; we'll only lose it for some old ISA
drivers.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
---
 net/core/dev.c |    8 ++++----
 1 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/net/core/dev.c b/net/core/dev.c
index f494675..7e6b7dc 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -1962,13 +1962,13 @@ struct sk_buff *skb_gso_segment(struct sk_buff *skb,
 
 	if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
 		struct net_device *dev = skb->dev;
-		struct ethtool_drvinfo info = {};
+		const char *driver = "";
 
-		if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
-			dev->ethtool_ops->get_drvinfo(dev, &info);
+		if (dev && dev->dev.parent)
+			driver = dev_driver_string(dev->dev.parent);
 
 		WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d ip_summed=%d\n",
-		     info.driver, dev ? &dev->features : NULL,
+		     driver, dev ? &dev->features : NULL,
 		     skb->sk ? &skb->sk->sk_route_caps : NULL,
 		     skb->len, skb->data_len, skb->ip_summed);
 
-- 
1.7.7.5

Re: [PATCH net 1/2] net: Use device model to get driver name in skb_gso_segment()

[Herbert Xu]

On Mon, Jan 16, 2012 at 10:38:59PM +0000, Ben Hutchings wrote:
> ethtool operations generally require the caller to hold RTNL and are
> not safe to call in atomic context.  The device model provides this
> information for most devices; we'll only lose it for some old ISA
> drivers.
> 
> Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>

Acked-by: Herbert Xu <herbert@gondor.apana.org.au>

Thanks,
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

[PATCH net 2/2] net: WARN if skb_checksum_help() is called on skb requiring segmentation

[Ben Hutchings]

skb_checksum_help() has never done anything useful with skbs that
require segmentation.  Setting skb->ip_summed = CHECKSUM_NONE makes
them invalid and provokes a later WARNing in skb_gso_segment().

Passing such an skb to skb_checksum_help() indicates a bug, so we
should warn about it immediately.  Move the warning from
skb_gso_segment() into a shared function, and add the calling function
name, gso_type and gso_size to it.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
---
The price for writing the warning format only once is having to pass in
the calling function name.  Not sure whether it's a good trade-off.

Ben.

 net/core/dev.c |   31 +++++++++++++++++++------------
 1 files changed, 19 insertions(+), 12 deletions(-)

diff --git a/net/core/dev.c b/net/core/dev.c
index 7e6b7dc..6824393 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -1887,6 +1887,22 @@ void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
 EXPORT_SYMBOL(skb_set_dev);
 #endif /* CONFIG_NET_NS */
 
+static void skb_warn_bad_offload(const char *func, const struct sk_buff *skb)
+{
+	struct net_device *dev = skb->dev;
+	const char *driver = "";
+
+	if (dev && dev->dev.parent)
+		driver = dev_driver_string(dev->dev.parent);
+
+	WARN(1, "%s: in %s caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
+	     "gso_type=%d ip_summed=%d\n",
+	     driver, func, dev ? &dev->features : NULL,
+	     skb->sk ? &skb->sk->sk_route_caps : NULL,
+	     skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
+	     skb_shinfo(skb)->gso_type, skb->ip_summed);
+}
+
 /*
  * Invalidate hardware checksum when packet is to be mangled, and
  * complete checksum manually on outgoing path.
@@ -1900,8 +1916,8 @@ int skb_checksum_help(struct sk_buff *skb)
 		goto out_set_summed;
 
 	if (unlikely(skb_shinfo(skb)->gso_size)) {
-		/* Let GSO fix up the checksum. */
-		goto out_set_summed;
+		skb_warn_bad_offload(__func__, skb);
+		return -EINVAL;
 	}
 
 	offset = skb_checksum_start_offset(skb);
@@ -1961,16 +1977,7 @@ struct sk_buff *skb_gso_segment(struct sk_buff *skb,
 	__skb_pull(skb, skb->mac_len);
 
 	if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
-		struct net_device *dev = skb->dev;
-		const char *driver = "";
-
-		if (dev && dev->dev.parent)
-			driver = dev_driver_string(dev->dev.parent);
-
-		WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d ip_summed=%d\n",
-		     driver, dev ? &dev->features : NULL,
-		     skb->sk ? &skb->sk->sk_route_caps : NULL,
-		     skb->len, skb->data_len, skb->ip_summed);
+		skb_warn_bad_offload(__func__, skb);
 
 		if (skb_header_cloned(skb) &&
 		    (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
-- 
1.7.7.5


-- 
Ben Hutchings, Staff Engineer, Solarflare
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.

Re: [PATCH net 2/2] net: WARN if skb_checksum_help() is called on skb requiring segmentation

[Herbert Xu]

On Mon, Jan 16, 2012 at 10:43:38PM +0000, Ben Hutchings wrote:
> skb_checksum_help() has never done anything useful with skbs that
> require segmentation.  Setting skb->ip_summed = CHECKSUM_NONE makes
> them invalid and provokes a later WARNing in skb_gso_segment().
> 
> Passing such an skb to skb_checksum_help() indicates a bug, so we
> should warn about it immediately.  Move the warning from
> skb_gso_segment() into a shared function, and add the calling function
> name, gso_type and gso_size to it.
> 
> Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
> ---
> The price for writing the warning format only once is having to pass in
> the calling function name.  Not sure whether it's a good trade-off.

Do we really need the name since we should get a back trace?

Cheers,
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

Re: [PATCH net 1/2] net: Use device model to get driver name in skb_gso_segment()

[David Miller]

From: Herbert Xu <herbert@gondor.hengli.com.au>
Date: Tue, 17 Jan 2012 09:41:38 +1100

> On Mon, Jan 16, 2012 at 10:38:59PM +0000, Ben Hutchings wrote:
>> ethtool operations generally require the caller to hold RTNL and are
>> not safe to call in atomic context.  The device model provides this
>> information for most devices; we'll only lose it for some old ISA
>> drivers.
>> 
>> Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
> 
> Acked-by: Herbert Xu <herbert@gondor.apana.org.au>

Applied.

Re: [PATCH net 2/2] net: WARN if skb_checksum_help() is called on skb requiring segmentation

[David Miller]

From: Herbert Xu <herbert@gondor.hengli.com.au>
Date: Tue, 17 Jan 2012 17:11:15 +1100

> On Mon, Jan 16, 2012 at 10:43:38PM +0000, Ben Hutchings wrote:
>> skb_checksum_help() has never done anything useful with skbs that
>> require segmentation.  Setting skb->ip_summed = CHECKSUM_NONE makes
>> them invalid and provokes a later WARNing in skb_gso_segment().
>> 
>> Passing such an skb to skb_checksum_help() indicates a bug, so we
>> should warn about it immediately.  Move the warning from
>> skb_gso_segment() into a shared function, and add the calling function
>> name, gso_type and gso_size to it.
>> 
>> Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
>> ---
>> The price for writing the warning format only once is having to pass in
>> the calling function name.  Not sure whether it's a good trade-off.
> 
> Do we really need the name since we should get a back trace?

Agreed.

Re: [PATCH net 2/2] net: WARN if skb_checksum_help() is called on skb requiring segmentation

[Ben Hutchings]

On Tue, 2012-01-17 at 17:11 +1100, Herbert Xu wrote:
> On Mon, Jan 16, 2012 at 10:43:38PM +0000, Ben Hutchings wrote:
> > skb_checksum_help() has never done anything useful with skbs that
> > require segmentation.  Setting skb->ip_summed = CHECKSUM_NONE makes
> > them invalid and provokes a later WARNing in skb_gso_segment().
> > 
> > Passing such an skb to skb_checksum_help() indicates a bug, so we
> > should warn about it immediately.  Move the warning from
> > skb_gso_segment() into a shared function, and add the calling function
> > name, gso_type and gso_size to it.
> > 
> > Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
> > ---
> > The price for writing the warning format only once is having to pass in
> > the calling function name.  Not sure whether it's a good trade-off.
> 
> Do we really need the name since we should get a back trace?

If the compiler were to aggressively inline some of these functions then
it might no longer be clear where this was called from.  But I agree
that the name is normally redundant, so I'll take it out.

Ben.

-- 
Ben Hutchings, Staff Engineer, Solarflare
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.