On Sun, 2008-05-04 at 20:12 +0200, Johannes Berg wrote:
> This patch makes the truesize warning be printed when the truesize
> actually changed, not just when the header was increased and the
> additional size actually used.
> 
> Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
> Cc: Herbert Xu <herbert@gondor.apana.org.au>
> ---
> It'll trigger with mac80211, should hold it until I fixed that.
> 
>  include/linux/skbuff.h |    8 ++++++--
>  net/core/skbuff.c      |   10 ++++++++--
>  2 files changed, 14 insertions(+), 4 deletions(-)
> 
> --- everything.orig/include/linux/skbuff.h	2008-05-03 15:47:00.000000000 +0200
> +++ everything/include/linux/skbuff.h	2008-05-04 00:30:34.000000000 +0200
> @@ -387,9 +387,13 @@ extern void	      skb_truesize_bug(struc
>  
>  static inline void skb_truesize_check(struct sk_buff *skb)
>  {
> -	int len = sizeof(struct sk_buff) + skb->len;
> +#ifdef NET_SKBUFF_DATA_USES_OFFSET
> +	int len = sizeof(struct sk_buff) + skb->end;
> +#else
> +	int len = sizeof(struct sk_buff) + (skb->end - skb->head);
> +#endif

Umm, is this even correct? Should it check data_len? I seem to get the
truesize warning a bit now:

[11381.081709] SKB BUG: Invalid truesize (16864) size=4112, sizeof(sk_buff)=272
[11381.081725] last reallocate at:
[11381.081729]  [<c0274194>] __alloc_skb+0xdc/0x140
[11381.081745]  [<c02aeca8>] sk_stream_alloc_skb+0x38/0x134
[11381.081758]  [<c02af16c>] tcp_sendmsg+0x3c8/0xcbc
[11381.081767]  [<c026bcc4>] sock_sendmsg+0xac/0xe4
[11381.081780]  [<c026c058>] sys_sendto+0xbc/0xec
[11381.081790]  [<c026cbc8>] sys_socketcall+0x14c/0x1dc
[11381.081800]  [<c00124cc>] ret_from_syscall+0x0/0x38


johannes