net: llc: explicitly set skb->transport_header
Subsystems (3)
PATCH 1/1 net: llc: explicitly set skb->transport_header
802.2+LLC+SNAP frames received by napi_complete_done with GRO and DSA have skb->transport_header set two bytes short, or pointing 2 bytes before network_header & skb->data. As snap_rcv expects transport_header to point to SNAP header (OID:PID) after LLC processing advances offset over LLC header (llc_rcv & llc_fixup_skb), code doesn't find a match and packet is dropped.
Between napi_complete_done and snap_rcv, transport_header is not used until __netif_receive_skb_core, where originally it was being reset. Commit fda55eca5a33 ("net: introduce skb_transport_header_was_set()") only does so if not set, on the assumption the value was set correctly by GRO (and also on assumption that "network stacks usually reset the transport header anyway"). Afterwards it is moved forward by llc_fixup_skb.
Locally generated traffic shows up at __netif_receive_skb_core with no transport_header set and is processed without issue. On a setup with GRO but no DSA, transport_header and network_header are both set to point to skb->data which is also correct.
As issue is LLC specific, to avoid impacting non-LLC traffic, and to follow up on original assumption made on previous code change, llc_fixup_skb to reset and advance the offset. llc_fixup_skb already assumes the LLC header is at skb->data, and by definition SNAP header immediately follows.
--- net/llc/llc_input.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
net/llc/llc_input.c +1 -1
@@ -124,7 +124,7 @@ static inline int llc_fixup_skb(struct sk_buff *skb)if (unlikely (!pskb_may_pull (skb, llc_len)))return 0 ; - skb->transport_header += llc_len; +skb_set_transport_header (skb, llc_len);skb_pull (skb, llc_len);if (skb->protocol ==htons (ETH_P_802_2 )) {__be16 pdulen;
Signature · 1 line
2.43.0
PATCH v2 1/1 net: llc: explicitly set skb->transport_header
802.2+LLC+SNAP frames received by napi_complete_done with GRO and DSA have skb->transport_header set two bytes short, or pointing 2 bytes before network_header & skb->data. As snap_rcv expects transport_header to point to SNAP header (OID:PID) after LLC processing advances offset over LLC header (llc_rcv & llc_fixup_skb), code doesn't find a match and packet is dropped.
Between napi_complete_done and snap_rcv, transport_header is not used until __netif_receive_skb_core, where originally it was being reset. Commit fda55eca5a33 ("net: introduce skb_transport_header_was_set()") only does so if not set, on the assumption the value was set correctly by GRO (and also on assumption that "network stacks usually reset the transport header anyway"). Afterwards it is moved forward by llc_fixup_skb.
Locally generated traffic shows up at __netif_receive_skb_core with no transport_header set and is processed without issue. On a setup with GRO but no DSA, transport_header and network_header are both set to point to skb->data which is also correct.
As issue is LLC specific, to avoid impacting non-LLC traffic, and to follow up on original assumption made on previous code change, llc_fixup_skb to reset and advance the offset. llc_fixup_skb already assumes the LLC header is at skb->data, and by definition SNAP header immediately follows.
--- net/llc/llc_input.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
net/llc/llc_input.c +1 -1
@@ -124,7 +124,7 @@ static inline int llc_fixup_skb(struct sk_buff *skb)if (unlikely (!pskb_may_pull (skb, llc_len)))return 0 ; - skb->transport_header += llc_len; +skb_set_transport_header (skb, llc_len);skb_pull (skb, llc_len);if (skb->protocol ==htons (ETH_P_802_2 )) {__be16 pdulen;
Signature · 1 line
2.43.0
Quoted 21 lines · Antonio Pastor
802.2+LLC+SNAP frames received by
napi_complete_donewith GRO and DSA haveskb->transport_headerset two bytes short, or pointing 2 bytes beforenetwork_header&skb->data. Assnap_rcvexpectstransport_headerto point to SNAP header (OID:PID) after LLC processing advances offset over LLC header (llc_rcv&llc_fixup_skb), code doesn't find a match and packet is dropped.Between
napi_complete_doneandsnap_rcv,transport_headeris not used until__netif_receive_skb_core, where originally it was being reset. Commitfda55eca5a33("net: introduceskb_transport_header_was_set()") only does so if not set, on the assumption the value was set correctly by GRO (and also on assumption that "network stacks usually reset the transport header anyway"). Afterwards it is moved forward byllc_fixup_skb.Locally generated traffic shows up at
__netif_receive_skb_corewith notransport_headerset and is processed without issue. On a setup with GRO but no DSA,transport_headerandnetwork_headerare both set to point toskb->datawhich is also correct.
Quoted 8 lines
As issue is LLC specific, to avoid impacting non-LLC traffic, and to follow up on original assumption made on previous code change,
llc_fixup_skbto reset and advance the offset.llc_fixup_skbalready assumes the LLC header is atskb->data, and by definition SNAP header immediately follows.Fixes fda55eca5a33 ("net: introduce skb_transport_header_was_set()")Signed-off-by Antonio Pastor <[email protected]>
11 years with this stuff being broken...
I wonder if we could remove it from the kernel, given nobody cared.
Can you at the same time fix net/802/psnap.c,snap_rcv() is probably having the same issue ?
I would also use skb_reset_transport_header() as in
net/llc/llc_input.c +0 -0
100644
net/llc/llc_input.c +1 -1
@@ -124,8 +124,8 @@ static inline int llc_fixup_skb(struct sk_buff *skb)if (unlikely (!pskb_may_pull (skb, llc_len)))return 0 ; - skb->transport_header += llc_len;skb_pull (skb, llc_len); +skb_reset_transport_header (skb);if (skb->protocol ==htons (ETH_P_802_2 )) {__be16 pdulen;s32 data_size;
A message that is not in the archive (sent off-list or not yet received): [email protected]
Quoted 53 lines · Antonio Pastor
On 2024-12-23 08:54, Eric Dumazet wrote:
On Mon, Dec 23, 2024 at 2:40 PM Antonio Pastor <[email protected]> wrote:
802.2+LLC+SNAP frames received by
napi_complete_donewith GRO and DSA haveskb->transport_headerset two bytes short, or pointing 2 bytes beforenetwork_header&skb->data. Assnap_rcvexpectstransport_headerto point to SNAP header (OID:PID) after LLC processing advances offset over LLC header (llc_rcv&llc_fixup_skb), code doesn't find a match and packet is dropped.Between
napi_complete_doneandsnap_rcv,transport_headeris not used until__netif_receive_skb_core, where originally it was being reset. Commitfda55eca5a33("net: introduceskb_transport_header_was_set()") only does so if not set, on the assumption the value was set correctly by GRO (and also on assumption that "network stacks usually reset the transport header anyway"). Afterwards it is moved forward byllc_fixup_skb.Locally generated traffic shows up at
__netif_receive_skb_corewith notransport_headerset and is processed without issue. On a setup with GRO but no DSA,transport_headerandnetwork_headerare both set to point toskb->datawhich is also correct.As issue is LLC specific, to avoid impacting non-LLC traffic, and to follow up on original assumption made on previous code change,
llc_fixup_skbto reset and advance the offset.llc_fixup_skbalready assumes the LLC header is atskb->data, and by definition SNAP header immediately follows.Fixes fda55eca5a33 ("net: introduce skb_transport_header_was_set()")Signed-off-by Antonio Pastor <[email protected]>Thanks for your reply.
11 years with this stuff being broken...
It works great without DSA thrown in the mix, for some reason I have not been able to determine, not for the lack of trying. Wonder how many configurations use DSA and care about SNAP frames.
DSA was added recently to the implementation I'm working on, so for practical purposes this is just broken since 1-2 years ago. Tripped on it about 6 months ago. It's fresh. Or well aged. :)I wonder if we could remove it from the kernel, given nobody cared.
Well, at least I do, and ballpark 40 other people for what I can tell. Some others might be lurking in the sidelines.
Can you at the same time fix
net/802/psnap.c,snap_rcv()is probably having the same issue ?
Quoted 8 lines
That's how I got here, because of
snap_rcv(). But no fix is required atsnap_rcv(). This patch is to fix problem there.snap_rcv()doesn't advance thetransport_headeroffset or pull the skb as far as I saw so adding a reset there, that would be redundant. Once this piece of code determines LLC header size SNAP header always follows for SNAP frames.
I'll double-check and submit a v3 w/that included if I think it makes sense but honestly I don't think it does.
I see skb->transport_header being advanced at line 61 :
/* Pass the frame on. */
skb->transport_header += 5;Note that setting the transport header (conditionally or not) in __netif_receive_skb()
is probably a mistake. Only network (ipv4, ipv6) handlers can possibly know the concept
of transport header.
Hopefully at some point we can remove this defensive code.
net/core/dev.c +0 -0
100644
net/core/dev.c +0 -0
@@ -5474,8 +5474,6 @@ static int __netif_receive_skb_core(struct
sk_buff **pskb, bool pfmemalloc,
orig_dev = skb->dev;
skb_reset_network_header(skb);
- if (!skb_transport_header_was_set(skb))
- skb_reset_transport_header(skb);
skb_reset_mac_len(skb);
pt_prev = NULL;Quoted 4 lines · Eric Dumazet
I see
skb->transport_headerbeing advanced at line 61 :/* Pass the frame on. */ skb->transport_header += 5;
Same treatment as before? Reset after pull?
@@ -58,8 +58,8 @@ static int snap_rcv(struct sk_buff *skb, struct net_device *dev, proto = find_snap_client(skb_transport_header(skb));
if (proto) {
/* Pass the frame on. */
- skb->transport_header += 5;
skb_pull_rcsum(skb, 5);
+ skb_reset_transport_header(skb);
rc = proto->rcvfunc(skb, dev, &snap_packet_type, orig_dev); }
rcu_read_unlock();
I'll submit that as a separate patch.
Quoted 5 lines
Note that setting the transport header (conditionally or not) in
__netif_receive_skb()
is probably a mistake. Only network (ipv4, ipv6) handlers can possibly know the concept
of transport header.
Not too sure about that, but I don't have specifics. Non-IP stacks are probably all ancient, but some might care.
Quoted 16 lines
Hopefully at some point we can remove this defensive code.
net/core/dev.c +0
-0100644
net/core/dev.c +0
-2@@ -5474,8 +5474,6 @@ static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, orig_dev = skb->dev;skb_reset_network_header (skb); -if (!skb_transport_header_was_set (skb)) -skb_reset_transport_header (skb);skb_reset_mac_len (skb);
Quoted 1 line
pt_prev = NULL;
I don't know the code well enough to identify all possible paths after this point to ensure all of them set transport header. I'd expect IPv4/IPv6 are OK and we are making LLC whole, but don't know what else to test beyond that. My testing is limited to SNAP... taking that out requires regression testing at a level I'm not comfortable running.
PATCH 1/1 net: 802: reset skb->transport_header
802.2+LLC+SNAP frames received by napi_complete_done() with GRO and DSA have skb->transport_header set two bytes short, or pointing 2 bytes before network_header & skb->data. This was an issue as snap_rcv() expected offset to point to SNAP header (OID:PID), causing packet to be dropped.
A fix at llc_fixup_skb() (a024e377efed) resets transport_header, addressing the issue. This patch is additional clean up to snap_rcv() so that it resets the offset after pulling the skb instead of incrementing it to match the pull.
--- net/802/psnap.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
net/802/psnap.c +1 -1
@@ -58,8 +58,8 @@ static int snap_rcv(struct sk_buff *skb, struct net_device *dev, proto =find_snap_client (skb_transport_header (skb));if (proto) {/* Pass the frame on. */ - skb->transport_header +=5 ;skb_pull_rcsum (skb,5 ); +skb_reset_transport_header (skb); rc = proto->rcvfunc (skb, dev, &snap_packet_type, orig_dev); }rcu_read_unlock ();
Signature · 1 line
2.43.0
Re: [PATCH net] net: 802: reset skb->transport_header
Quoted 33 lines · Antonio Pastor
802.2+LLC+SNAP frames received by
napi_complete_done()with GRO and DSA haveskb->transport_headerset two bytes short, or pointing 2 bytes beforenetwork_header&skb->data. This was an issue assnap_rcv()expected offset to point to SNAP header (OID:PID), causing packet to be dropped.A fix at
llc_fixup_skb()(a024e377efed) resetstransport_header, addressing the issue. This patch is additional clean up tosnap_rcv()so that it resets the offset after pulling the skb instead of incrementing it to match the pull.Fixes fda55eca5a33 ("net: introduce skb_transport_header_was_set()")Signed-off-by Antonio Pastor <[email protected]>--- net/802/psnap.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)net/802/psnap.c +1
-2@@ -58,8 +58,8 @@ static int snap_rcv(struct sk_buff *skb, struct net_device *dev, proto =find_snap_client (skb_transport_header (skb));if (proto) {/* Pass the frame on. */ - skb->transport_header +=5 ;skb_pull_rcsum (skb,5 ); +skb_reset_transport_header (skb); rc = proto->rcvfunc (skb, dev, &snap_packet_type, orig_dev); }rcu_read_unlock (); --
Quoted 2 lines
2.43.0
Sorry, this patch is wrong, it does not fix the potential issue yet.
Note how skb_transport_header(skb) is used in
find_snap_client(skb_transport_header(skb));
The proper way to fix the issue is to not rely on the transport header at all, only reset it after pulling the network header.
net/802/psnap.c +0 -0
100644
net/802/psnap.c +0 -0
@@ -55,11 +55,11 @@ static int snap_rcv(struct sk_buff *skb, struct
net_device *dev,
goto drop;
rcu_read_lock();
- proto = find_snap_client(skb_transport_header(skb));
+ proto = find_snap_client(skb->data);
if (proto) {
/* Pass the frame on. */
- skb->transport_header += 5;
skb_pull_rcsum(skb, 5);
+ skb_reset_transport_header(skb);
rc = proto->rcvfunc(skb, dev, &snap_packet_type, orig_dev);
}
rcu_read_unlock();Re: [PATCH net] net: 802: reset skb->transport_header
Quoted 1 line
Sorry, this patch is wrong, it does not fix the potential issue yet.
No worries! Thanks for your patience with this. Much appreciated.
Quoted 2 lines
Note how
skb_transport_header(skb)is used in
find_snap_client(skb_transport_header(skb));
I've spent so much time trying to figure out why the offset is wrong I lost sight that the core issue is that it is being used to begin with. Paolo Abeni hinted at that too.
Quoted 23 lines
The proper way to fix the issue is to not rely on the transport header at all, only reset it after pulling the network header.
net/802/psnap.c +0
-0100644
net/802/psnap.c +2
-2@@ -55,11 +55,11 @@ static int snap_rcv(struct sk_buff *skb, structnet_device *dev, goto drop;rcu_read_lock (); - proto =find_snap_client (skb_transport_header (skb)); + proto =find_snap_client (skb->data);if (proto) {/* Pass the frame on. */ - skb->transport_header +=5 ;skb_pull_rcsum (skb,5 ); +skb_reset_transport_header (skb); rc = proto->rcvfunc (skb, dev, &snap_packet_type, orig_dev); }
Quoted 1 line
rcu_read_unlock();
Will send V2.
PATCH v2 1/1 net: 802: LLC+SNAP OID:PID lookup on start of skb data
In reply to Eric Dumazet802.2+LLC+SNAP frames received by napi_complete_done() with GRO and DSA have skb->transport_header set two bytes short, or pointing 2 bytes before network_header & skb->data. This was an issue as snap_rcv() expected offset to point to SNAP header (OID:PID), causing packet to be dropped.
A fix at llc_fixup_skb() (a024e377efed) resets transport_header for any LLC consumers that may care about it, and stops SNAP packets from being dropped, but doesn't fix the problem which is that LLC and SNAP should not use transport_header offset.
Ths patch eliminates the use of transport_header offset for SNAP lookup of OID:PID so that SNAP does not rely on the offset at all. The offset is reset after pull for any SNAP packet consumers that may (but shouldn't) use it.
--- net/802/psnap.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
net/802/psnap.c +2 -2
@@ -55,11 +55,11 @@ static int snap_rcv(struct sk_buff *skb, struct net_device *dev, goto drop;rcu_read_lock (); - proto =find_snap_client (skb_transport_header (skb)); + proto =find_snap_client (skb->data);if (proto) {/* Pass the frame on. */ - skb->transport_header +=5 ;skb_pull_rcsum (skb,5 ); +skb_reset_transport_header (skb); rc = proto->rcvfunc (skb, dev, &snap_packet_type, orig_dev); }rcu_read_unlock ();
Signature · 1 line
2.43.0
Re: [PATCH net v2] net: 802: LLC+SNAP OID:PID lookup on start of skb data
Quoted 12 lines · Antonio Pastor
802.2+LLC+SNAP frames received by
napi_complete_done()with GRO and DSA haveskb->transport_headerset two bytes short, or pointing 2 bytes beforenetwork_header&skb->data. This was an issue assnap_rcv()expected offset to point to SNAP header (OID:PID), causing packet to be dropped.A fix at
llc_fixup_skb()(a024e377efed) resetstransport_headerfor any LLC consumers that may care about it, and stops SNAP packets from being dropped, but doesn't fix the problem which is that LLC and SNAP should not usetransport_headeroffset.
Quoted 7 lines
Ths patch eliminates the use of
transport_headeroffset for SNAP lookup of OID:PID so that SNAP does not rely on the offset at all. The offset is reset after pull for any SNAP packet consumers that may (but shouldn't) use it.Fixes fda55eca5a33 ("net: introduce skb_transport_header_was_set()")Signed-off-by Antonio Pastor <[email protected]>
Re: [PATCH net v2] net: 802: LLC+SNAP OID:PID lookup on start of skb data
In reply to Antonio PastorThis patch was applied to netdev/net.git (main) by Jakub Kicinski <[email protected]>:
Hello:
This patch was applied to netdev/net.git (main)
by Jakub Kicinski <[email protected]>:
Quoted 6 lines · On Thu, 2 Jan 2025 20:23:00 -0500 you wrote:
802.2+LLC+SNAP frames received by
napi_complete_done()with GRO and DSA haveskb->transport_headerset two bytes short, or pointing 2 bytes beforenetwork_header&skb->data. This was an issue assnap_rcv()expected offset to point to SNAP header (OID:PID), causing packet to be dropped.
Quoted 6 lines
A fix at
llc_fixup_skb()(a024e377efed) resetstransport_headerfor any LLC consumers that may care about it, and stops SNAP packets from being dropped, but doesn't fix the problem which is that LLC and SNAP should not usetransport_headeroffset.[...]
Here is the summary with links:
- [net,v2] net: 802: LLC+SNAP OID:PID lookup on start of skb data
https://git.kernel.org/netdev/net/c/1e9b0e1c550c
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html
PATCH v3 1/1 net: llc: reset skb->transport_header
In reply to Eric Dumazet802.2+LLC+SNAP frames received by napi_complete_done with GRO and DSA have skb->transport_header set two bytes short, or pointing 2 bytes before network_header & skb->data. As snap_rcv expects transport_header to point to SNAP header (OID:PID) after LLC processing advances offset over LLC header (llc_rcv & llc_fixup_skb), code doesn't find a match and packet is dropped.
Between napi_complete_done and snap_rcv, transport_header is not used until __netif_receive_skb_core, where originally it was being reset. Commit fda55eca5a33 ("net: introduce skb_transport_header_was_set()") only does so if not set, on the assumption the value was set correctly by GRO (and also on assumption that "network stacks usually reset the transport header anyway"). Afterwards it is moved forward by llc_fixup_skb.
Locally generated traffic shows up at __netif_receive_skb_core with no transport_header set and is processed without issue. On a setup with GRO but no DSA, transport_header and network_header are both set to point to skb->data which is also correct.
As issue is LLC specific, to avoid impacting non-LLC traffic, and to follow up on original assumption made on previous code change, llc_fixup_skb to reset the offset after skb pull. llc_fixup_skb assumes the LLC header is at skb->data, and by definition SNAP header immediately follows.
--- net/llc/llc_input.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
net/llc/llc_input.c +1 -1
@@ -124,8 +124,8 @@ static inline int llc_fixup_skb(struct sk_buff *skb)if (unlikely (!pskb_may_pull (skb, llc_len)))return 0 ; - skb->transport_header += llc_len;skb_pull (skb, llc_len); +skb_reset_transport_header (skb);if (skb->protocol ==htons (ETH_P_802_2 )) {__be16 pdulen;s32 data_size;
Signature · 1 line
2.43.0
Re: [PATCH net v3] net: llc: reset skb->transport_header
Quoted 21 lines · Antonio Pastor
802.2+LLC+SNAP frames received by
napi_complete_donewith GRO and DSA haveskb->transport_headerset two bytes short, or pointing 2 bytes beforenetwork_header&skb->data. Assnap_rcvexpectstransport_headerto point to SNAP header (OID:PID) after LLC processing advances offset over LLC header (llc_rcv&llc_fixup_skb), code doesn't find a match and packet is dropped.Between
napi_complete_doneandsnap_rcv,transport_headeris not used until__netif_receive_skb_core, where originally it was being reset. Commitfda55eca5a33("net: introduceskb_transport_header_was_set()") only does so if not set, on the assumption the value was set correctly by GRO (and also on assumption that "network stacks usually reset the transport header anyway"). Afterwards it is moved forward byllc_fixup_skb.Locally generated traffic shows up at
__netif_receive_skb_corewith notransport_headerset and is processed without issue. On a setup with GRO but no DSA,transport_headerandnetwork_headerare both set to point toskb->datawhich is also correct.
Quoted 8 lines
As issue is LLC specific, to avoid impacting non-LLC traffic, and to follow up on original assumption made on previous code change,
llc_fixup_skbto reset the offset after skb pull.llc_fixup_skbassumes the LLC header is atskb->data, and by definition SNAP header immediately follows.Fixes fda55eca5a33 ("net: introduce skb_transport_header_was_set()")Signed-off-by Antonio Pastor <[email protected]>
Re: [PATCH net v3] net: llc: reset skb->transport_header
In reply to Antonio PastorThis patch was applied to netdev/net.git (main) by Jakub Kicinski <[email protected]>:
Hello:
This patch was applied to netdev/net.git (main)
by Jakub Kicinski <[email protected]>:
Quoted 8 lines · On Tue, 24 Dec 2024 20:07:20 -0500 you wrote:
802.2+LLC+SNAP frames received by
napi_complete_donewith GRO and DSA haveskb->transport_headerset two bytes short, or pointing 2 bytes beforenetwork_header&skb->data. Assnap_rcvexpectstransport_headerto point to SNAP header (OID:PID) after LLC processing advances offset over LLC header (llc_rcv&llc_fixup_skb), code doesn't find a match and packet is dropped.[...]
Here is the summary with links:
- [net,v3] net: llc: reset skb->transport_header
https://git.kernel.org/netdev/net/c/a024e377efed
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html