tcp: add missing tcp_skb_can_collapse() test in tcp_shift_skb_data()

PATCH Applied Eric Dumazet 5 participants 5 messages Feb 1, 2022 – Feb 3, 2022

Subsystems (3)

Networking Network drivers TCP

lore

Share & tools

lore

Eric Dumazet

author

2022-02-01 18:46 UTC #0 lore raw

PATCH 1/1 tcp: add missing tcp_skb_can_collapse() test in tcp_shift_skb_data()

From: Eric Dumazet <[email protected]>

tcp_shift_skb_data() might collapse three packets into a larger one.

P_A, P_B, P_C -> P_ABC

Historically, it used a single tcp_skb_can_collapse_to(P_A) call, because it was enough.

In commit 85712484110d ("tcp: coalesce/collapse must respect MPTCP extensions"), this call was replaced by a call to tcp_skb_can_collapse(P_A, P_B)

But the now needed test over P_C has been missed.

This probably broke MPTCP.

Then later, commit 9b65b17db723 ("net: avoid double accounting for pure zerocopy skbs") added an extra condition to tcp_skb_can_collapse(), but the missing call from tcp_shift_skb_data() is also breaking TCP zerocopy, because P_A and P_C might have different skb_zcopy_pure() status.

Fixes 85712484110d ("tcp: coalesce/collapse must respect MPTCP extensions")

Fixes 9b65b17db723 ("net: avoid double accounting for pure zerocopy skbs")

Signed-off-by Eric Dumazet <[email protected]>

Cc Paolo Abeni <[email protected]>

Cc Mat Martineau <[email protected]>

Cc Talal Ahmad <[email protected]>

Cc Arjun Roy <[email protected]>

Cc Soheil Hassas Yeganeh <[email protected]>

Cc Willem de Bruijn <[email protected]>

---
 net/ipv4/tcp_input.c | 2 ++
 1 file changed, 2 insertions(+)

net/ipv4/tcp_input.c +2 -0

index dc49a3d551eb919baf5ad812ef21698c5c7b9679..bfe4112e000c09ba9d7d8b64392f52337b9053e9 100644
@@ -1660,6 +1660,8 @@ static struct sk_buff *tcp_shift_skb_data(struct sock *sk, struct sk_buff *skb,
 	    (mss != tcp_skb_seglen(skb)))
 		goto out;
 
+	if (!tcp_skb_can_collapse(prev, skb))
+		goto out;
 	len = skb->len;
 	pcount = tcp_skb_pcount(skb);
 	if (tcp_skb_shift(prev, skb, pcount, len))

Signature · 1 line

2.35.0.rc2.247.g8bbb082509-goog

Soheil Hassas Yeganeh

Acked-by on v1 1/1

2022-02-01 18:48 UTC #1 lore raw

Quoted 22 lines · Eric Dumazet

From: Eric Dumazet <[email protected]>
tcp_shift_skb_data() might collapse three packets into a larger one.
P_A, P_B, P_C -> P_ABC
Historically, it used a single tcp_skb_can_collapse_to(P_A) call, because it was enough.
In commit 85712484110d ("tcp: coalesce/collapse must respect MPTCP extensions"), this call was replaced by a call to tcp_skb_can_collapse(P_A, P_B)
But the now needed test over P_C has been missed.
This probably broke MPTCP.
Then later, commit 9b65b17db723 ("net: avoid double accounting for pure zerocopy skbs") added an extra condition to tcp_skb_can_collapse(), but the missing call from tcp_shift_skb_data() is also breaking TCP zerocopy, because P_A and P_C might have different skb_zcopy_pure() status.

Quoted 9 lines

Fixes 85712484110d ("tcp: coalesce/collapse must respect MPTCP extensions")
Fixes 9b65b17db723 ("net: avoid double accounting for pure zerocopy skbs")
Signed-off-by Eric Dumazet <[email protected]>
Cc Paolo Abeni <[email protected]>
Cc Mat Martineau <[email protected]>
Cc Talal Ahmad <[email protected]>
Cc Arjun Roy <[email protected]>
Cc Soheil Hassas Yeganeh <[email protected]>
Cc Willem de Bruijn <[email protected]>

Acked-by Soheil Hassas Yeganeh <[email protected]>

I wish there were some packetdrill tests for MPTCP. Thank you for the fix!

Quoted 20 lines

---
 net/ipv4/tcp_input.c | 2 ++
 1 file changed, 2 insertions(+)

net/ipv4/tcp_input.c +2 -1

index dc49a3d551eb919baf5ad812ef21698c5c7b9679..bfe4112e000c09ba9d7d8b64392f52337b9053e9 100644
@@ -1660,6 +1660,8 @@ static struct sk_buff *tcp_shift_skb_data(struct sock *sk, struct sk_buff *skb,
            (mss != tcp_skb_seglen(skb)))
                goto out;
 
+       if (!tcp_skb_can_collapse(prev, skb))
+               goto out;
        len = skb->len;
        pcount = tcp_skb_pcount(skb);
        if (tcp_skb_shift(prev, skb, pcount, len))
--

2.35.0.rc2.247.g8bbb082509-goog

Mat Martineau

2022-02-01 20:01 UTC #2 lore raw

Quoted 33 lines · Soheil Hassas Yeganeh

Quoted 22 lines · Eric Dumazet
From: Eric Dumazet <[email protected]>
tcp_shift_skb_data() might collapse three packets into a larger one.
P_A, P_B, P_C -> P_ABC
Historically, it used a single tcp_skb_can_collapse_to(P_A) call, because it was enough.
In commit 85712484110d ("tcp: coalesce/collapse must respect MPTCP extensions"), this call was replaced by a call to tcp_skb_can_collapse(P_A, P_B)
But the now needed test over P_C has been missed.
This probably broke MPTCP.
Then later, commit 9b65b17db723 ("net: avoid double accounting for pure zerocopy skbs") added an extra condition to tcp_skb_can_collapse(), but the missing call from tcp_shift_skb_data() is also breaking TCP zerocopy, because P_A and P_C might have different skb_zcopy_pure() status.
Quoted 9 lines
Fixes 85712484110d ("tcp: coalesce/collapse must respect MPTCP extensions")
Fixes 9b65b17db723 ("net: avoid double accounting for pure zerocopy skbs")
Signed-off-by Eric Dumazet <[email protected]>
Cc Paolo Abeni <[email protected]>
Cc Mat Martineau <[email protected]>
Cc Talal Ahmad <[email protected]>
Cc Arjun Roy <[email protected]>
Cc Soheil Hassas Yeganeh <[email protected]>
Cc Willem de Bruijn <[email protected]>

Quoted 4 lines

Acked-by Soheil Hassas Yeganeh <[email protected]>
I wish there were some packetdrill tests for MPTCP. Thank you for the fix!

Soheil -

I have good news, there are packetdrill tests for MPTCP:

https://github.com/multipath-tcp/packetdrill

This is still in a fork. I think Davide has talked to Neal about upstreaming the MPTCP changes before but there may be some code that needs refactoring before that could happen.

Quoted 12 lines

---
 net/ipv4/tcp_input.c | 2 ++
 1 file changed, 2 insertions(+)

net/ipv4/tcp_input.c +0 -0

index dc49a3d551eb919baf5ad812ef21698c5c7b9679..bfe4112e000c09ba9d7d8b64392f52337b9053e9 100644
@@ -1660,6 +1660,8 @@ static struct sk_buff *tcp_shift_skb_data(struct sock *sk, struct sk_buff *skb,
            (mss != tcp_skb_seglen(skb)))
                goto out;

Quoted 9 lines

Quoted 8 lines

net/ipv4/tcp_input.c +2 -1

+       if (!tcp_skb_can_collapse(prev, skb))
+               goto out;
        len = skb->len;
        pcount = tcp_skb_pcount(skb);
        if (tcp_skb_shift(prev, skb, pcount, len))
--

2.35.0.rc2.247.g8bbb082509-goog

--
Mat Martineau
Intel

Paolo Abeni

maintainer Acked-by

2022-02-02 09:38 UTC #3 lore raw

Quoted 13 lines · Mat Martineau

Quoted 11 lines · Soheil Hassas Yeganeh
Quoted 10 lines · Eric Dumazet
From: Eric Dumazet <[email protected]>
tcp_shift_skb_data() might collapse three packets into a larger one.
P_A, P_B, P_C -> P_ABC
Historically, it used a single tcp_skb_can_collapse_to(P_A) call, because it was enough.

Quoted 6 lines

Quoted 6 lines
Quoted 6 lines
In commit 85712484110d ("tcp: coalesce/collapse must respect MPTCP extensions"), this call was replaced by a call to tcp_skb_can_collapse(P_A, P_B)
But the now needed test over P_C has been missed.
This probably broke MPTCP.

Indeed it looks like it could cause MPTCP data stream corruption, in case of multiple substreams, if we hit this code-path. Thanks for catching and fixing it!

Quoted 15 lines

Quoted 15 lines
Quoted 14 lines
Then later, commit 9b65b17db723 ("net: avoid double accounting for pure zerocopy skbs") added an extra condition to tcp_skb_can_collapse(), but the missing call from tcp_shift_skb_data() is also breaking TCP zerocopy, because P_A and P_C might have different skb_zcopy_pure() status.
Fixes 85712484110d ("tcp: coalesce/collapse must respect MPTCP extensions")
Fixes 9b65b17db723 ("net: avoid double accounting for pure zerocopy skbs")
Signed-off-by Eric Dumazet <[email protected]>
Cc Paolo Abeni <[email protected]>
Cc Mat Martineau <[email protected]>
Cc Talal Ahmad <[email protected]>
Cc Arjun Roy <[email protected]>
Cc Soheil Hassas Yeganeh <[email protected]>
Cc Willem de Bruijn <[email protected]>

Quoted 1 line

Quoted 1 line
Acked-by Soheil Hassas Yeganeh <[email protected]>

Acked-by Paolo Abeni <[email protected]>

Quoted 2 lines

Quoted 2 lines
I wish there were some packetdrill tests for MPTCP. Thank you for the fix!

Do you have by chance a drill for the zero-copy case? it may help creating the MPTCP one, too.

Thanks!

Paolo

[email protected]

CI Applied

2022-02-03 00:40 UTC #4 lore raw

In reply to Eric Dumazet

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

Hello:

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

Quoted 10 lines · On Tue, 1 Feb 2022 10:46:40 -0800 you wrote:

From: Eric Dumazet <[email protected]>
tcp_shift_skb_data() might collapse three packets into a larger one.
P_A, P_B, P_C -> P_ABC
Historically, it used a single tcp_skb_can_collapse_to(P_A) call, because it was enough.
[...]

Here is the summary with links:
- [net] tcp: add missing tcp_skb_can_collapse() test in tcp_shift_skb_data()

    https://git.kernel.org/netdev/net/c/b67985be4009

You are awesome, thank you!

Signature · 2 lines

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