Skip to contentSkip to conversation

IPv6:remove duplicate check of optlen when setsockopt with IPV6_PKTINFO option

PATCH In review Yang Hongyang 5 participants 18 messages Jan 13, 2009 – Jan 15, 2009

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 IPv6:remove duplicate check of optlen when setsockopt with IPV6_PKTINFO option

Actually the condition (optlen == 0) is included in (optlen < sizeof(struct in6_pktinfo)), so we do not need to check it separately.

---
 net/ipv6/ipv6_sockglue.c |    4 +---
 1 files changed, 1 insertions(+), 3 deletions(-)
net/ipv6/ipv6_sockglue.c +1 -3
index 40f3246..e6affa7 100644
@@ -399,9 +399,7 @@ sticky_done:
 	{
 		struct in6_pktinfo pkt;
 
-		if (optlen == 0)
-			goto e_inval;
-		else if (optlen < sizeof(struct in6_pktinfo) || optval == NULL)
+		if (optlen < sizeof(struct in6_pktinfo) || optval == NULL)
 			goto e_inval;
 
 		if (copy_from_user(&pkt, optval, sizeof(struct in6_pktinfo))) {
Signature · 4 lines
1.6.0.3
-- 
Regards
Yang Hongyang
#1 lore raw
Quoted 2 lines · Yang Hongyang

Actually the condition (optlen == 0) is included in (optlen < sizeof(struct in6_pktinfo)), so we do not need to check it separately.

You don't need to check optval == NULL either since that's the job of copy_from_user.

Cheers,

Signature · 4 lines
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
#2 lore raw
Quoted 6 lines · Herbert Xu
Quoted 2 lines · Yang Hongyang

Actually the condition (optlen == 0) is included in (optlen < sizeof(struct in6_pktinfo)), so we do not need to check it separately.

You don't need to check optval == NULL either since that's the job of copy_from_user.

If optval==NULL, what we should return?EINVAL or EFAULT? If it is EINVAL,then we should check it .otherwise it's the job of copy_from_user

Quoted 2 lines

Cheers,

Signature · 2 lines
Regards
Yang Hongyang
#3 lore raw
Quoted 7 lines · Yang Hongyang
Quoted 5 lines · Herbert Xu
Quoted 3 lines · Yang Hongyang

Actually the condition (optlen == 0) is included in (optlen < sizeof(struct in6_pktinfo)), so we do not need to check it separately.

Quoted 7 lines
Quoted 3 lines

You don't need to check optval == NULL either since that's the job of copy_from_user.

If optval==NULL, what we should return?EINVAL or EFAULT?
If it is EINVAL,then we should check it .otherwise it's the job of copy_from_user

I think if optval==NULL, the in6_pktinfo which is set should be remove. So, you should handle optval==NULL. Not just return error.

Quoted 6 lines
Quoted 3 lines

Cheers,

#7 lore raw
In reply to Yang Hongyang
Quoted 4 lines · Yang Hongyang

If optval==NULL, what we should return?EINVAL or EFAULT? If it is EINVAL,then we should check it .otherwise it's the job of copy_from_user

I think EFAULT is fine. We return that elsewhere as well.

Cheers,

Signature · 4 lines
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
#4 lore raw
In reply to Wei Yongjun
Quoted 13 lines · Wei Yongjun
Quoted 12 lines · Yang Hongyang
Quoted 9 lines · Herbert Xu
Quoted 4 lines · Yang Hongyang

Actually the condition (optlen == 0) is included in (optlen < sizeof(struct in6_pktinfo)),
so we do not need to check it separately.

You don't need to check optval == NULL either since that's the job of copy_from_user.

Quoted 7 lines
Quoted 4 lines

If optval==NULL, what we should return?EINVAL or EFAULT? If it is EINVAL,then we should check it .otherwise it's the job of copy_from_user

I think if optval==NULL, the in6_pktinfo which is set should be remove. So, you should handle optval==NULL. Not just return error.

There's no RFC defines the behavior above,but:
RFC3542 said The application can remove any sticky Routing header or sticky
Destination options header or sticky Hop-by-Hop options header by calling setsockopt() with a zero option length.

So,do we need to allow remove any sticky pktinfo option by calling setsockopt() with a zero option length?

Quoted 10 lines
Quoted 7 lines
Quoted 4 lines

Cheers,

Signature · 2 lines
Regards
Yang Hongyang
#5 lore raw
Quoted 20 lines · Yang Hongyang
Quoted 12 lines · Wei Yongjun
Quoted 11 lines · Yang Hongyang
Quoted 9 lines · Herbert Xu
Quoted 4 lines · Yang Hongyang

Actually the condition (optlen == 0) is included in (optlen < sizeof(struct in6_pktinfo)),
so we do not need to check it separately.

You don't need to check optval == NULL either since that's the job of copy_from_user.

Quoted 6 lines
Quoted 4 lines

If optval==NULL, what we should return?EINVAL or EFAULT? If it is EINVAL,then we should check it .otherwise it's the job of copy_from_user

I think if optval==NULL, the in6_pktinfo which is set should be remove. So, you should handle optval==NULL. Not just return error.

Quoted 8 lines

There's no RFC defines the behavior above,but:
RFC3542 said The application can remove any sticky Routing header or sticky
Destination options header or sticky Hop-by-Hop options header by calling setsockopt() with a zero option length.

So,do we need to allow remove any sticky pktinfo option by calling setsockopt() with a zero option length?

Can remove the option using seting in6_pktinfo struct wiht ipi6_ifindex=0,ipi6_ifindex=IN6ADDR_ANY_INIT.

If no RFC definition, not to reset with optlen=0,
for example IPV6_TCLASS, PV6_2292DSTOPTS.

Signature · 2 lines
Best Regards
Shan Wei
#6 lore raw

Re: [PATCHv2]IPv6:remove duplicate check when setsockopt with IPV6_PKTINFO option

Actually the condition (optlen == 0) is included in (optlen < sizeof(struct in6_pktinfo)), so we do not need to check it separately.
We don't need to check (optval==NULL) either,because copy_from_user() will do the job.

---
 net/ipv6/ipv6_sockglue.c |    4 +---
 1 files changed, 1 insertions(+), 3 deletions(-)
net/ipv6/ipv6_sockglue.c +1 -3
index 40f3246..1c203b6 100644
@@ -399,9 +399,7 @@ sticky_done:
 	{
 		struct in6_pktinfo pkt;
 
-		if (optlen == 0)
-			goto e_inval;
-		else if (optlen < sizeof(struct in6_pktinfo) || optval == NULL)
+		if (optlen < sizeof(struct in6_pktinfo))
 			goto e_inval;
 
 		if (copy_from_user(&pkt, optval, sizeof(struct in6_pktinfo))) {
Signature · 1 line
1.6.0.3
Quoted 35 lines · Shan Wei
Quoted 19 lines · Yang Hongyang
Quoted 12 lines · Wei Yongjun
Quoted 11 lines · Yang Hongyang
Quoted 9 lines · Herbert Xu
Quoted 4 lines · Yang Hongyang

Actually the condition (optlen == 0) is included in (optlen < sizeof(struct in6_pktinfo)),
so we do not need to check it separately.

You don't need to check optval == NULL either since that's the job of copy_from_user.

Quoted 6 lines
Quoted 4 lines

If optval==NULL, what we should return?EINVAL or EFAULT? If it is EINVAL,then we should check it .otherwise it's the job of copy_from_user

I think if optval==NULL, the in6_pktinfo which is set should be remove. So, you should handle optval==NULL. Not just return error.

Quoted 8 lines

There's no RFC defines the behavior above,but:
RFC3542 said The application can remove any sticky Routing header or sticky
Destination options header or sticky Hop-by-Hop options header by calling setsockopt() with a zero option length.

So,do we need to allow remove any sticky pktinfo option by calling setsockopt() with a zero option length?

Can remove the option using seting in6_pktinfo struct wiht ipi6_ifindex=0,ipi6_ifindex=IN6ADDR_ANY_INIT.

If no RFC definition, not to reset with optlen=0,
for example IPV6_TCLASS, PV6_2292DSTOPTS.

Signature · 2 lines
Regards
Yang Hongyang
maintainer
#8 lore raw
In reply to Herbert Xu

From: Herbert Xu <[email protected]>
Date: Wed, 14 Jan 2009 15:48:56 +1100

Quoted 7 lines
Quoted 4 lines · Yang Hongyang

If optval==NULL, what we should return?EINVAL or EFAULT? If it is EINVAL,then we should check it .otherwise it's the job of copy_from_user

I think EFAULT is fine. We return that elsewhere as well.

Actually, we return EINVAL just a few lines above this code block for some other socket option cases when optval==NULL.

So for consistency I'm applying Yang's original patch to net-next-2.6

Thanks.

#9 lore raw
Quoted 3 lines · David Miller

Actually, we return EINVAL just a few lines above this code block for some other socket option cases when optval==NULL.

Well the very next option IPV6_2292PKTOPTIONS returns EFAULT by virtue of not explicitly checking optval == NULL :)

Cheers,

Signature · 4 lines
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
#10 lore raw
Quoted 7 lines · Herbert Xu
Quoted 3 lines · David Miller

Actually, we return EINVAL just a few lines above this code block for some other socket option cases when optval==NULL.

Well the very next option IPV6_2292PKTOPTIONS returns EFAULT by virtue of not explicitly checking optval == NULL :)

In fact checking for a NULL pointer is strictly speaking wrong since NULL may actually have been mapped in user-space :)

Cheers,

Signature · 4 lines
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
#11 lore raw
Quoted 9 lines · Herbert Xu
Quoted 5 lines · Herbert Xu
Quoted 2 lines · David Miller

Actually, we return EINVAL just a few lines above this code block for some other socket option cases when optval==NULL.

Well the very next option IPV6_2292PKTOPTIONS returns EFAULT by virtue of not explicitly checking optval == NULL :)

In fact checking for a NULL pointer is strictly speaking wrong since NULL may actually have been mapped in user-space :)

But there are some cases that user-space passes a NULL pointer to the kernel,Otherwise,copy_from_user needn't to cheak the NULL pointer either.:)

By the way,I think most part of the ipv6 socket option implementation are kind of ugly:)

Quoted 2 lines

Cheers,

Signature · 2 lines
Regards
Yang Hongyang
#13 lore raw
In reply to David Miller
Quoted 7 lines · David Miller

From: Herbert Xu <[email protected]>
Date: Wed, 14 Jan 2009 15:48:56 +1100

Quoted 4 lines
Quoted 3 lines · Yang Hongyang

If optval==NULL, what we should return?EINVAL or EFAULT? If it is EINVAL,then we should check it .otherwise it's the job of copy_from_user

Quoted 7 lines
Quoted 1 line

I think EFAULT is fine. We return that elsewhere as well.

Actually, we return EINVAL just a few lines above this code block for some other socket option cases when optval==NULL.

So for consistency I'm applying Yang's original patch to net-next-2.6

Dave,there are some of the code elsewhere return EFAULT when optval=NULL,I will post a patch to fix them:)

Quoted 4 lines

Thanks.

Signature · 2 lines
Regards
Yang Hongyang
#14 lore raw
Quoted 3 lines · Yang Hongyang

Dave,there are some of the code elsewhere return EFAULT when optval=NULL,I will post a patch to fix them:)

I disagree. The concept of a user-space NULL pointer is different to that of a kernel-space NULL pointer, so the kernel should not be treating a user-space NULL pointer differently than any other pointer.

Cheers,

Signature · 4 lines
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
maintainer
#12 lore raw
In reply to Herbert Xu

From: Herbert Xu <[email protected]>
Date: Thu, 15 Jan 2009 16:04:49 +1100

Quoted 7 lines · David Miller
Quoted 3 lines

Actually, we return EINVAL just a few lines above this code block for some other socket option cases when optval==NULL.

Well the very next option IPV6_2292PKTOPTIONS returns EFAULT by virtue of not explicitly checking optval == NULL :)

Touche' :-)

maintainer
#15 lore raw
In reply to Herbert Xu

From: Herbert Xu <[email protected]>
Date: Thu, 15 Jan 2009 16:37:02 +1100

Quoted 9 lines · Yang Hongyang
Quoted 3 lines

Dave,there are some of the code elsewhere return EFAULT when optval=NULL,I will post a patch to fix them:)

I disagree. The concept of a user-space NULL pointer is different to that of a kernel-space NULL pointer, so the kernel should not be treating a user-space NULL pointer differently than any other pointer.

I agree with Herbert, but of course we have to adhere to cases where APIs actually define a user NULL pointer to have a specific meaning.

#16 lore raw
Quoted 4 lines · David Miller

I agree with Herbert, but of course we have to adhere to cases where APIs actually define a user NULL pointer to have a specific meaning.

Absolutely. Although AFAIK the relevant RFCs here only use the notion of a zero option length for such purposes rather than NULL pointers (presumably because the meaning of NULL pointers are context dependent).

Cheers,

Signature · 4 lines
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
#17 lore raw
In reply to David Miller
Quoted 6 lines · David Miller

From: Herbert Xu <[email protected]>
Date: Thu, 15 Jan 2009 16:37:02 +1100

Quoted 3 lines · Yang Hongyang
Quoted 2 lines

Dave,there are some of the code elsewhere return EFAULT when optval=NULL,I will post a patch to fix them:)

Quoted 10 lines
Quoted 4 lines

I disagree. The concept of a user-space NULL pointer is different to that of a kernel-space NULL pointer, so the kernel should not be treating a user-space NULL pointer differently than any other pointer.

I agree with Herbert, but of course we have to adhere to cases where APIs actually define a user NULL pointer to have a specific meaning.

Agreed.Seems that Patch v2 is correct according to the discussion.^_^

Signature · 2 lines
Regards
Yang Hongyang