2024-03-05 22:43:29

by Juntong Deng

[permalink] [raw]
Subject: [PATCH net-next] net: ipv4, ipv6: Fix incorrect skb->data_len caused by __ip_append_data

When __ip_append_data allocate the first skb in the queue, or when the
size of the data in the skb exceed the MTU and require a new fragment
and allocate a new skb, both cause the size of the data increased by
this __ip_append_data to not be added to skb->data_len.

This is because in the current __ip_append_data, skb_put is used when
putting in the data for the new skb, but skb_put only increase skb->len,
but not skb->data_len, resulting in skb->data_len missing this part of
the data size.

All skb processed by __ip_append_data are unable to obtain the accurate
data size based on skb->data_len for the above reason.

Also __ip6_append_data has the same problem.

This patch fixes the bug.

Signed-off-by: Juntong Deng <[email protected]>
---
net/ipv4/ip_output.c | 3 +++
net/ipv6/ip6_output.c | 3 +++
2 files changed, 6 insertions(+)

diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c
index 1fe794967211..42686be0a843 100644
--- a/net/ipv4/ip_output.c
+++ b/net/ipv4/ip_output.c
@@ -1171,6 +1171,9 @@ static int __ip_append_data(struct sock *sk,
copy = 0;
}

+ if (copy >= 0)
+ skb->data_len += copy;
+
offset += copy;
length -= copy + transhdrlen;
transhdrlen = 0;
diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c
index 31b86fe661aa..2091b91513f0 100644
--- a/net/ipv6/ip6_output.c
+++ b/net/ipv6/ip6_output.c
@@ -1689,6 +1689,9 @@ static int __ip6_append_data(struct sock *sk,
copy = 0;
}

+ if (copy >= 0)
+ skb->data_len += copy;
+
offset += copy;
length -= copy + transhdrlen;
transhdrlen = 0;
--
2.39.2



2024-03-06 03:18:41

by Jakub Kicinski

[permalink] [raw]
Subject: Re: [PATCH net-next] net: ipv4, ipv6: Fix incorrect skb->data_len caused by __ip_append_data

On Tue, 5 Mar 2024 22:42:05 +0000 Juntong Deng wrote:
> When __ip_append_data allocate the first skb in the queue, or when the
> size of the data in the skb exceed the MTU and require a new fragment
> and allocate a new skb, both cause the size of the data increased by
> this __ip_append_data to not be added to skb->data_len.
>
> This is because in the current __ip_append_data, skb_put is used when
> putting in the data for the new skb, but skb_put only increase skb->len,
> but not skb->data_len, resulting in skb->data_len missing this part of
> the data size.
>
> All skb processed by __ip_append_data are unable to obtain the accurate
> data size based on skb->data_len for the above reason.
>
> Also __ip6_append_data has the same problem.
>
> This patch fixes the bug.

data_len is the amount of data in the non-linear parts of the skb.
Please run some tests before submitting patches.
--
pw-bot: cr
pv-bot: s