Re: PATCH] firewire: add padding to some struct

From: Stefan Richter
Date: Sat Jul 19 2008 - 06:38:26 EST


JiSheng Zhang wrote:
On Fri, 18 Jul 2008 17:27:44 +0200
Mikael Pettersson <mikpe@xxxxxxxx> wrote:
JiSheng Zhang writes:
> --- old/drivers/firewire/fw-cdev.c 2008-07-14 05:51:29.000000000 +0800
> +++ new/drivers/firewire/fw-cdev.c 2008-07-18 20:20:45.841328585 +0800
> @@ -382,9 +382,9 @@
> > response->response.type = FW_CDEV_EVENT_RESPONSE;
> response->response.rcode = rcode;
> - queue_event(client, &response->event,
> - &response->response, sizeof(response->response),
> - response->response.data, response->response.length);
> + queue_event(client, &response->event, &response->response, > + sizeof(response->response) + response->response.length,
> + NULL, 0);
> }

Neither of these look correct.
If sizeof(struct ...) != offsetof(struct ..., data) as you claim is possible,
then the old code will copy too much to/from ->response but the correct amount
to/from ->response.data, and the new code will copy too much to/from ->response.data.

The old code will copy 4 extra bytes totally on some platforms, the new code
is correct. The old one queue like this:
struct ...(excluding the padding bytes)|4 padding bytes|4 padding bytes|data

That's why C has offsetof():

queue_event(client, &response->event,
&response->response,
offsetof(typeof(*response->responce), data), // I don't know the struct name
response->response.data, response->response.length);

sizeof(struct ...) != offsetof(struct ..., data) happens for example on x86-64.

This is what I get with the current firewire drivers in a block read response from firecontrol on i686:

Command: r . 0 0xfffff0000400 20
reading from node 0, bus 1023, offset 0XFFFFF0000400 20 bytes
read succeeded. Data follows (hex):
04 04 04 8F
31 33 39 34
F0 00 A2 22
00 10 DC 56
00 FE D2 D4
Ack code: complete

And the same on x86-64:

Command: r . 0 0xfffff0000400 20
reading from node 0, bus 1023, offset 0XFFFFF0000400 20 bytes
read succeeded. Data follows (hex):
04 04 04 8F
04 04 04 8F
31 33 39 34
F0 00 A2 22
00 10 DC 56
Ack code: complete

Command: r . 0 0xfffff0000400 24
reading from node 0, bus 1023, offset 0XFFFFF0000400 20 bytes
read succeeded. Data follows (hex):
04 04 04 8F
04 04 04 8F
31 33 39 34
F0 00 A2 22
00 10 DC 56
00 FE D2 D4
Ack code: complete

I used libraw1394 from Dan's git repo. Gscanbus shows exactly the same results. So, x86-64 and all other architectures where struct fw_cdev_event* are aligned on u64 boundaries are currently seriously broken... but nobody noticed it before. The only breakage which I saw (and is obviously the result of this bug) is that gscanbus shows a wrong bus topology on x86-64 but the correct one on i686. The damage from this bug is limited though because
- most applications send requests which get responses with 0 or 4
bytes payload,
- no application (which can run on both old and new stack without
change) uses address range mappings, i.e. get incoming requests.

I'll look further into your proposed fix.
--
Stefan Richter
-=====-==--- -=== =--==
http://arcgraph.de/sr/
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/