From: Paul Guo <ggang@tilera.com>
To: <netfilter-devel@vger.kernel.org>
Subject: unaligned ip header due to ip_route_me_harder()->pskb_expand_head()
Date: Tue, 8 Nov 2011 23:02:53 +0800 [thread overview]
Message-ID: <4EB9449D.8020701@tilera.com> (raw)
Hi,
Recently on our chip (arch/tile), a panic caused by unaligned data access was observed. The stack looks like this:
frame 0: 0xfd247ef0 ipt_do_table+0x240/0x850 (sp 0xf6a6fb40)
frame 1: 0xfd4a44a8 nf_iterate+0xe8/0x188 (sp 0xf6a6fbd0)
frame 2: 0xfd4ab850 nf_hook_slow+0xa0/0x180 (sp 0xf6a6fbf8)
frame 3: 0xfd0c02a8 ip_local_out+0x28/0x88 (sp 0xf6a6fc30)
frame 4: 0xfd25b778 ip_push_pending_frames+0x4f8/0x798 (sp 0xf6a6fc40)
frame 5: 0xfd272298 icmp_send+0x670/0x700 (sp 0xf6a6fc60)
frame 6: 0xfd593d28 ipv4_link_failure+0x28/0xa8 (sp 0xf6a6fd50)
frame 7: 0xfd5aae90 arp_error_report+0x70/0x90 (sp 0xf6a6fd60)
frame 8: 0xfd483f28 neigh_invalidate+0x128/0x1b0 (sp 0xf6a6fd70)
frame 9: 0xfd3b5410 neigh_timer_handler.cold+0x2a0/0x2e0 (sp 0xf6a6fd98)
frame 10: 0xfd028cd8 run_timer_softirq+0x2d8/0x4a8 (sp 0xf6a6fdb0)
frame 11: 0xfd02bfe0 __do_softirq+0x1e0/0x330 (sp 0xf6a6fe08)
frame 12: 0xfd034808 do_softirq+0xc8/0x160 (sp 0xf6a6fe50)
frame 13: 0xfd034658 irq_exit+0x98/0x130 (sp 0xf6a6fe60)
frame 14: 0xfd031cc0 do_timer_interrupt+0xc0/0xf8 (sp 0xf6a6fe68)
frame 15: 0xfd3b1da8 handle_interrupt+0x2d8/0x2e0 (sp 0xf6a6fe80)
<interrupt 25 while in kernel mode>
frame 16: 0xfd5cb350 _cpu_idle_nap+0x0/0x10 (sp 0xf6a6ffd0)
frame 17: 0xfd06b020 cpu_idle+0x1a0/0x3b8 (sp 0xf6a6ffd0)
int ip_route_me_harder(struct sk_buff *skb, unsigned addr_type)
{
......
/* Change in oif may mean change in hh_len. */
hh_len = skb_dst(skb)->dev->hard_header_len;
if (skb_headroom(skb) < hh_len &&
pskb_expand_head(skb, hh_len - skb_headroom(skb), 0, GFP_ATOMIC))
return -1;
}
During testing, our chip acts as NAT. the interface for internal hosts are configured with vlan. When we try to ping a unknown external host from internal hosts, it panic immediately due to unaligned data access. In this case, hh_len is equal to 18 (vlan mac header length), headroom (skb_headroom(skb)) for this packet is 16 so after calling pskb_expand_head() the ip header becomes to be unaligned. This seems to be a bug in netfilter. Modifying pskb_expand_head() like this can resolve this issue.
+ pskb_expand_head(skb, LL_RESERVED_SPACE(skb_dst(skb)->dev) - skb_headroom(skb), 0, GFP_ATOMIC))
- pskb_expand_head(skb, hh_len - skb_headroom(skb), 0, GFP_ATOMIC))
Thanks,
Paul
next reply other threads:[~2011-11-08 15:08 UTC|newest]
Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top
2011-11-08 15:02 Paul Guo [this message]
2011-11-08 16:58 ` unaligned ip header due to ip_route_me_harder()->pskb_expand_head() Eric Dumazet
2011-11-09 9:33 ` Paul Guo
2011-11-09 9:48 ` Eric Dumazet
2011-11-10 7:33 ` Paul Guo
2011-11-13 17:02 ` Pablo Neira Ayuso
2011-11-14 8:33 ` [PATCH] possible unaligned packet header caused by ip_route_me_harder()->pskb_expand_head() Paul Guo
2011-11-14 9:46 ` Eric Dumazet
2011-11-14 11:00 ` Paul Guo
2011-11-21 17:47 ` Pablo Neira Ayuso
2011-11-16 16:54 ` Pablo Neira Ayuso
2011-11-21 10:30 ` Eric Dumazet
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=4EB9449D.8020701@tilera.com \
--to=ggang@tilera.com \
--cc=netfilter-devel@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).