Minki/linux
1
0
Fork 1
mirror of https://github.com/torvalds/linux.git synced 2024-11-25 13:41:51 +00:00
Code Issues Packages Projects Releases Wiki Activity

netfilter: xt_IDLETIMER: target v1 - match Android layout

Browse Source 
Android has long had an extension to IDLETIMER to send netlink
messages to userspace, see:
  https://android.googlesource.com/kernel/common/+/refs/heads/android-mainline/include/uapi/linux/netfilter/xt_IDLETIMER.h#42
Note: this is idletimer target rev 1, there is no rev 0 in
the Android common kernel sources, see registration at:
  https://android.googlesource.com/kernel/common/+/refs/heads/android-mainline/net/netfilter/xt_IDLETIMER.c#483

When we compare that to upstream's new idletimer target rev 1:
  https://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf-next.git/tree/include/uapi/linux/netfilter/xt_IDLETIMER.h#n46

We immediately notice that these two rev 1 structs are the
same size and layout, and that while timer_type and send_nl_msg
are differently named and serve a different purpose, they're
at the same offset.

This makes them impossible to tell apart - and thus one cannot
know in a mixed Android/vanilla environment whether one means
timer_type or send_nl_msg.

Since this is iptables/netfilter uapi it introduces a problem
between iptables (vanilla vs Android) userspace and kernel
(vanilla vs Android) if the two don't match each other.

Additionally when at some point in the future Android picks up
5.7+ it's not at all clear how to resolve the resulting merge
conflict.

Furthermore, since upgrading the kernel on old Android phones
is pretty much impossible there does not seem to be an easy way
out of this predicament.

The only thing I've been able to come up with is some super
disgusting kernel version >= 5.7 check in the iptables binary
to flip between different struct layouts.

By adding a dummy field to the vanilla Linux kernel header file
we can force the two structs to be compatible with each other.

Long term I think I would like to deprecate send_nl_msg out of
Android entirely, but I haven't quite been able to figure out
exactly how we depend on it.  It seems to be very similar to
sysfs notifications but with some extra info.

Currently it's actually always enabled whenever Android uses
the IDLETIMER target, so we could also probably entirely
remove it from the uapi in favour of just always enabling it,
but again we can't upgrade old kernels already in the field.

(Also note that this doesn't change the structure's size,
as it is simply fitting into the pre-existing padding, and
that since 5.7 hasn't been released yet, there's still time
to make this uapi visible change)

Cc: Manoj Basapathi <manojbm@codeaurora.org>
Cc: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This commit is contained in:
Maciej Żenczykowski 2020-03-31 09:35:59 -07:00 committed by Pablo Neira Ayuso
parent a26c1e49c8
commit bc9fe6143d
2 changed files with 4 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
include/uapi/linux/netfilter/xt_IDLETIMER.h
View File

@ -48,6 +48,7 @@ struct idletimer_tg_info_v1 {
char label[MAX_IDLETIMER_LABEL_SIZE];
__u8 send_nl_msg; /* unused: for compatibility with Android */
__u8 timer_type;
/* for kernel module internal use only */

3
net/netfilter/xt_IDLETIMER.c
View File

@ -346,6 +346,9 @@ static int idletimer_tg_checkentry_v1(const struct xt_tgchk_param *par)
pr_debug("checkentry targinfo%s\n", info->label);
if (info->send_nl_msg)
return -EOPNOTSUPP;
ret = idletimer_tg_helper((struct idletimer_tg_info *)info);
if(ret < 0)
{